JP2010056682A - E-mail receiver and method of receiving e-mail, e-mail transmitter and e-mail transmission method, mail transmission server - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a technique for preventing inappropriate exchange of e-mails and preventing occurrence of misunderstanding or trouble due to erroneous transmission.
An e-mail receiving apparatus that receives an e-mail from an external server via a communication network, which is connected to the communication network and receives an e-mail, and from the received e-mail Along with the e-mail, the greeting text extraction processing means 202 for extracting the greeting text, the greeting text feeling determination processing means 203 for determining the type of word feeling included in the greeting text using at least the phrase information included in the greeting text, And a received mail output processing unit 204 that outputs the sensation type information of the determined greeting.
[Selection] Figure 2

Description

  The present invention relates to a technique related to transmission / reception of an electronic mail by a computer, and more particularly to a technique for automatically determining a word feeling included in an electronic mail.

The use of e-mail has become widespread as a daily communication means, and business communication in companies and private mail exchange between individuals are performed.
Since e-mail can be sent and received with very simple operations and delivered immediately to the other party, the use of e-mail for even a small amount of communication and the frequent exchange of e-mails just like conversations are increasing.

On the other hand, there are not a few problems of contacting using e-mail. For example, there is a problem that confidential information is transmitted to an incorrect destination by mistake in an attached file, or that information is transmitted from a company illegally because information can be easily transmitted.
In addition, although it can be transmitted easily, misrepresentation occurs because the text is not examined carefully, and human relations may be impaired. In other words, since it is time-consuming to send a letter, it is common to carefully review the text, and to send a call while checking the other party's reaction. There are few.
However, in the case of e-mail, in extreme cases, a sentence that only answers “yes” or “no” may be sent, and the nuance and true intention will not be accurately conveyed to the other party, leading to an emotional misunderstanding. Sometimes.

  Furthermore, when sending an email to a partner with the same surname, the email may be inadvertently sent to another partner. In particular, when the other party is an important person, you may feel uncomfortable by sending an e-mail to a friend.

Even when receiving mail from the other party, you can know what kind of mail you normally write, so you can know if the other person's feelings have changed, I had to remember how to write and make judgments based on the memory.
When the number of opponents became large, it was impossible to grasp all of them, and it was difficult to detect subtle changes in the emotions of the opponents, which hindered smooth communication.

Conventionally, several techniques for preventing erroneous transmission have been proposed.
For example, according to Patent Document 1, a technique for outputting a warning when a combination of an e-mail destination and a subject or a combination of an address and a keyword matches is proposed.
Since this method only compares the address and keyword match on a rule basis, it only prevents sending another email to the wrong party, and the above-mentioned email writing method is inappropriate. It is not something that can be detected.

As techniques related to the present invention, the present inventors have disclosed techniques of Non-Patent Documents 1 to 4.
Non-Patent Document 1 describes the creation of an adverb dictionary for a Japanese adverb, in which three polarities of the word, affirmation, denial, and neutrality, are assigned as evaluation polarities.
Non-Patent Document 2 describes a technique for providing a classification tag for classifying Web text for each opinion.
Non-Patent Document 3 describes a technique for classifying a speaker's expression intention by type for a Japanese adverb, constructing an adverb dictionary, and automatically classifying adverbs using machine learning.
Non-Patent Document 4 describes a method of classifying opinions based on affirmation, denial, and neutrality using machine learning for newspaper articles.

JP 2008-083917

Toshiyuki Kanamaru, Masaki Murata, Hitoshi Isahara “Creation and Evaluation of Adverb Dictionary with Evaluation Polarity”, 13th Annual Conference of the Language Processing Society of Japan, Ryukoku University. March 20, 2007. Toshiyuki Kanamaru, Masaki Murata, Hitoshi Isahara “Development of Web Text Analysis System Using Language Resources and Classification Tagging Technology” The 14th Annual Conference of the Association for Natural Language Processing p. 213-216, The University of Tokyo Komaba Campus, March 2008. Toshiyuki Kanamaru, Masaki Murata, Hitoshi Isahara “Development of Adverb Dictionaries Concerning Speaker Intention”, 12th Annual Workshop of Language Processing Society of Japan (W1) “Emotion, Evaluation, Attitude and Language”, Keio University. 2006. Toshiyuki Kanamaru, Masaki Murata, and Hitoshi Isahara, "Japanese Opinion Extraction System for Japanese Newspapers Using Machine-Learning Method", The 6th NTCIR Workshop NII, Tokyo, 5 May 2007.

  The present invention was created in view of the above-described problems of the prior art, and is a technique for preventing inappropriate exchange of e-mails and preventing occurrence of misunderstandings or troubles due to erroneous transmission. The purpose is to provide.

The present invention uses the following means for solving the above problems.
The invention described in claim 1 provides an e-mail receiving apparatus for receiving e-mail from an external server via a communication network. The e-mail receiving apparatus includes a mail reception processing unit that receives an e-mail by connecting to a communication network, a greeting text extraction processing unit that extracts a greeting from the received e-mail, and at least phrase information included in the greeting Is used to determine the type of sensation included in the greeting text, and the received mail output processing means is used to output the sensation type information of the determined greeting text together with the e-mail.

  According to the invention of claim 2, in the greeting text extraction processing means, referring to the greeting text rule database in which the text information or the appearance location information of the greeting text in the e-mail is stored in advance, and the greeting text database, A greeting text rule extracting unit is provided that extracts a related greeting text by comparing text information and text included in an e-mail, or extracts a greeting text from text included in an e-mail according to appearance location information. And

  According to the invention of claim 3, in the greeting text extraction processing means, using teacher data that predefines which part in the e-mail is a greeting text, the appearance location information of the greeting text in the teacher data, Machine learning based on at least one of the following features: text information, part of speech, whether or not a word is included in a greeting expression dictionary with words used for greetings in advance, e-mail sender or recipient information, and e-mail subject words And a greeting text learning database storing the results. An electronic mail receiving apparatus including a greeting text machine learning extraction unit that inputs the received electronic mail and extracts a greeting text by machine learning processing with reference to a greeting text learning database.

  According to the invention of claim 4, in the greeting text word feeling determination processing means, greeting text word feeling rule database in which word feeling type information corresponding to each greeting text is stored in advance for text information of the greeting text in the e-mail; An e-mail receiving device comprising: a greeting text extracted by the greeting text extraction processing means and a greeting text word feeling rule determination unit for determining a greeting text word feeling type by a comparison process with a greeting text word feeling rule database or a predetermined calculation process Can be provided.

According to the fifth aspect of the present invention, the greeting text sensation determination processing means uses teacher data that predefines the type of linguistic feeling to be determined from the greeting text in the e-mail. Appearance location information, text information, part of speech, whether or not it is a word included in the greeting expression dictionary with words used for greetings in advance, information on the sender or recipient of the email, at least one of the words in the subject of the email Machine learning is used as a feature, and a greeting text learning database storing the results is used.
An electronic device comprising a greeting text / feeling machine learning determination unit that inputs at least the greeting text extracted by the greeting text extraction processing means and determines the speech type information by machine learning processing with reference to the greeting text / feeling learning database A mail receiving device is provided.

  According to the sixth aspect of the present invention, in the electronic mail receiving device, in the electronic mail received by the mail reception processing means, at least information on the sender of the electronic mail and word sense type information related to the electronic mail are included. History or change of sensation type information with reference to past history storage processing means to be stored in storage means, and past history in which the received e-mail and sender's information match at the time of e-mail reception or at a predetermined opportunity It is possible to provide an e-mail receiving device that includes a past history detecting means for detecting a history and a past history output means for outputting a history or a change together with at least the received e-mail.

According to a seventh aspect of the present invention, the electronic mail receiving apparatus includes an electronic mail utterance determination processing unit that determines the type of sensation included in the text using at least phrase information included in the text of the electronic mail. In this case, the e-mail word sense determination processing means performs a comparison process or a predetermined calculation process between the polarity dictionary database in which word type information corresponding to a plurality of words is stored in advance, the words included in the text, and the polarity dictionary database. An e-mail word sense rule judging unit for judging the word sense type.
In this configuration, the received mail output processing means outputs the text or the text sensation type.

According to an eighth aspect of the present invention, the electronic mail receiving device includes an electronic mail feeling determination processing means for determining a kind of word feeling included in the text using at least word / phrase information included in the text of the electronic mail.
In this configuration, the e-mail feeling determination processing means uses a polarity dictionary database that pre-stores feeling type information corresponding to a plurality of phrases, and teacher data that predefines the type information of the feeling to be determined from the e-mail, Machine learning using text information in teacher data, whether or not it contains words or phrases in the polarity dictionary, or its kind of feeling information, information on the sender or recipient of an email, at least one of words in the subject of an email as a feature, An e-mail word sense learning database storing the result, and an e-mail word sense machine learning determination unit that inputs at least the received e-mail and determines word type information by machine learning processing with reference to the e-mail word sense learning database; Is provided. The received mail output processing means outputs the text or the text sensation type.

According to the ninth aspect of the present invention, it is also possible to provide an e-mail transmission device that transmits an e-mail to an external server via a communication network.
The apparatus includes a transmission mail input processing means for inputting an electronic mail to be transmitted, a greeting text extraction processing means for extracting a greeting text from the electronic mail to be transmitted, and at least a phrase text included in the greeting text. A greeting sentence word feeling determination processing means for determining the kind of word feeling included in the voice message, and a word feeling output processing means for outputting the word feeling kind information of the determined greeting sentence.
In addition to the above, this apparatus includes known mail transmission processing means for transmitting an e-mail to a specified destination connected to a communication network.

  According to the invention described in claim 10, in the greeting text extraction processing means, referring to the greeting text rule database in which the text information or the appearance location information of the greeting text in the e-mail is stored in advance, and the greeting text database, E-mail provided with a greeting text rule extraction unit that extracts a related greeting text by comparing text information and text contained in the e-mail, or extracts a greeting text from the text included in the e-mail according to appearance information A transmission device can be provided.

According to the eleventh aspect of the present invention, the greeting text extraction processing means uses teacher data that predefines which part in the e-mail is a greeting text, Machine learning based on at least one of the following features: text information, part of speech, whether or not a word is included in a greeting expression dictionary with words used for greetings in advance, e-mail sender or recipient information, and e-mail subject words Then, a greeting text learning database storing the results is used.
A greeting text machine learning extraction unit is provided that inputs the e-mail to be transmitted and extracts a greeting text by machine learning processing with reference to the greeting text learning database.

  According to the invention described in claim 12, in the greeting text word feeling determination processing means, a greeting text word feeling rule database in which word feeling type information corresponding to each greeting text is stored in advance for text information of the greeting text in the e-mail; An e-mail transmission device comprising a greeting text word sense rule determination unit that determines a greeting text word feeling type by a comparison process of a greeting text extracted by a greeting text extraction processing means and a greeting text word feeling rule database or a predetermined calculation process Can be provided.

According to the invention of claim 13, the greeting text word feeling determination processing means uses teacher data in which the kind information of the word feeling to be determined from the greeting text in the email is defined in advance, and the greeting text in the teacher data Appearance location information, text information, part of speech, whether or not it is a word included in the greeting expression dictionary with words used for greetings in advance, information on the sender or recipient of the email, at least one of the words in the subject of the email Machine learning is used as a feature, and a greeting text learning database storing the results is used.
A greeting text / feeling machine learning determination unit that inputs at least the greeting text extracted by the greeting text extraction processing means and determines the word feeling type information by machine learning processing with reference to the greeting text word feeling learning database.

  According to the invention described in claim 14, in the electronic mail transmitting apparatus, in the electronic mail input by the transmitted mail input processing means, at least information on the sender of the electronic mail and linguistic type information related to the electronic mail are included. History or change of word sense type information with reference to past history storage processing means to be stored in storage means, and past history in which e-mail to be sent and recipient information match at the time of e-mail transmission or at a predetermined opportunity It is possible to provide an e-mail transmission apparatus including a past history detection unit that detects a history and a past history output unit that outputs a history or a change together with at least an e-mail to be transmitted.

  According to the fifteenth aspect of the present invention, the electronic mail transmitting device includes an electronic mail feeling determination processing means for determining a kind of word feeling included in the text using at least word / phrase information included in the text of the electronic mail. A configuration in which the e-mail word sense determination processing means stores a word dictionary type information corresponding to a plurality of words in advance, a comparison process between words included in the text and the polarity dictionary database, or a predetermined calculation process. And an e-mail sensation rule determination unit for determining the sensation type of the text, and the sensation output processing means outputs the sensation type of the text or text.

According to the sixteenth aspect of the present invention, the electronic mail transmitting device includes an electronic mail feeling determination processing means for determining a kind of word feeling included in the text by using at least phrase information included in the text of the electronic mail.
In this configuration, the e-mail feeling determination processing means uses a polarity dictionary database that pre-stores feeling type information corresponding to a plurality of phrases, and teacher data that predefines the type information of the feeling to be determined from the e-mail, Machine learning using text information in teacher data, whether or not it contains words or phrases in the polarity dictionary, or its kind of feeling information, information on the sender or recipient of an email, at least one of words in the subject of an email as a feature, An e-mail linguistic learning database storing the results is used.
Then, at least the received e-mail is input, and an e-mail word sense machine learning determination unit that determines word sense type information by machine learning processing with reference to an e-mail word sense learning database, It is characterized by outputting the text or the feeling type of the text.

  According to the seventeenth aspect of the present invention, the electronic mail transmitting device includes a greeting text presentation processing means for automatically presenting a greeting text to be added to the electronic mail to be transmitted. The means refers to the past history of the past history storage processing means, and extracts a past history greeting text extraction processing section for extracting a greeting text in the past history that matches at least the recipient information of the e-mail to be transmitted, and the extracted greeting text A greeting sentence presentation processing unit that presents a part or all of the message.

According to the eighteenth aspect of the present invention, it is possible to provide a mail transmission server that receives mail data from a client terminal and transmits mail toward a destination address included in the mail data.
The server includes the electronic mail transmitting device according to any one of claims 9 to 17, and all the output results from at least one of the above-mentioned speech output processing means, past history output means, and greeting sentence presentation processing means. Alternatively, a terminal inquiry processing means for notifying a client terminal or a predetermined management terminal of a part corresponding to a predetermined condition and inquiring whether or not to send at least an e-mail to a destination is provided.

According to the nineteenth aspect of the present invention, it is possible to provide an electronic mail receiving method for a computer that receives an electronic mail from an external server via a communication network.
In this method, a mail reception processing means connects to a communication network to receive an e-mail, a mail reception processing step, a greeting text extraction processing means extracts a greeting text from the received e-mail, a greeting text extraction processing step, a greeting The sentence feeling determination processing means uses at least the phrase information included in the greeting sentence, the greeting sentence word feeling determination processing step for determining the kind of word feeling included in the greeting sentence, and the received mail output processing means are determined together with the e-mail. It has the received mail output process step which outputs the sensation kind information of a greeting.

According to the twentieth aspect of the present invention, it is possible to provide an e-mail transmission method for a computer that transmits an e-mail to an external server via a communication network.
In this method, a transmission mail input processing means inputs a transmission mail input processing step for inputting an electronic mail to be transmitted; a greeting text extraction processing means extracts a greeting text from the electronic mail to be transmitted; The determination processing means uses at least the phrase information included in the greeting text to determine the type of sensation included in the greeting text, and the utterance output processing means determines the sensation type information of the determined greeting text. It has a word output process step for outputting.

The present invention has the following effects by providing the above configuration.
According to the invention described in claim 1 or 19, a greeting can be extracted from the received e-mail, and the type of sensation can be determined and shown to the user. This makes it possible to objectively grasp the e-mail word sense, such as whether it is an e-mail or a friendly e-mail, and contributes to knowing the reader's intention accurately.
In particular, the present invention is characterized in that the sensation is efficiently determined by focusing on greetings that are relatively easy to classify into patterns and often represent the relationship between the sender and the receiver.

  According to the second aspect of the present invention, in the greeting sentence extraction process, extraction based on a rule base defined in advance by the user can be performed, so that a simple extraction method can be realized. Further, by using the rule base, it is possible to accurately determine an e-mail including a phrase that expresses a sense of word.

According to the third aspect of the present invention, whether or not the word is included in the greeting expression dictionary having the appearance part information of the greeting sentence, the text information, the part of speech, and the phrase used for the greeting in advance as a feature using the machine learning module. By using the information of the sender or receiver of the e-mail and the word / phrase of the subject of the e-mail, it contributes to the extraction of a highly accurate greeting.
In particular, it is possible to make a good determination by using a grammatical information such as a phrase, syntax, part of speech, etc. used in a greeting sentence as a feature, because a part where a greeting sentence appears generally has many sentence heads and sentence endings. Furthermore, it may affect whether it is easy to include a greeting from the domain name of the e-mail, and this can be used as a feature.

According to the fourth aspect of the present invention, it is possible to determine the word feeling based on the rule base using the greeting sentence word feeling rule database in the greeting sentence word feeling determination processing means.
Many greetings are typical sentences and phrases, and a relatively good determination can be made by the rule base, and a simple and high-speed determination method can be realized.
In particular, in a configuration in which e-mails of a plurality of sending users are transmitted via this apparatus, it is possible to detect a mail with inappropriate content on a rule basis.

According to the fifth aspect of the present invention, whether or not the word is included in the greeting expression dictionary having the appearance part information of the greeting sentence, the text information, the part of speech, and the phrase used for greeting in advance as the feature using the machine learning module. By using the information of the sender or receiver of the e-mail and the phrase of the subject of the e-mail, it contributes to the determination of the word feeling with high accuracy.
In particular, it is possible to make a good determination by using as features the location where the greeting appears, the number of times, grammatical information such as phrases, syntax, parts of speech used in the greeting. Furthermore, since the relationship between the sender and the receiver may appear from the domain name of the e-mail, this can be used as a feature.

According to the invention described in claim 6, since the past history of the received mail can be stored, when the e-mail is received or specified by the user, the same sender's history is stored. You can refer to past feelings. For example, when an e-mail from the same sender arrives, the attitude of the sender (became intimate, no longer ridiculous, angry, etc.) compared to the e-mail received in the past is shown to the user be able to.
Thereby, even a user who receives a large amount of mail can easily grasp the change in the word feeling of the mail and contribute to smooth communication.

According to the seventh aspect of the present invention, it is possible to determine the sensation of texts other than the above greetings, for example, the body of a mail or the title.
At that time, by using the polarity dictionary etc. described in the prior art and making a determination by comparing the words and phrases registered in the dictionary with the text or by a predetermined calculation process, the text including a plurality of sentences is targeted. In addition, it is possible to easily determine the word feeling of the e-mail using a computer.

According to the eighth aspect of the present invention, the machine learning module is used, and as its feature, text information in the learning data, whether or not it contains a word or phrase in the polar dictionary, or its sensation type information, sender or reception of e-mail It contributes to the determination of word sense with high accuracy by using the words of the subject and the subject of the e-mail.
In particular, it is possible to make a good determination by using text information such as sentences and phrases included in the e-mail, whether or not to include words or phrases in the polar dictionary, or using the word sense type information. Furthermore, since the relationship between the sender and the receiver may appear from the domain name of the e-mail, this can be used as a feature.

According to the ninth or twentieth aspect of the present invention, since the greeting can be extracted from the e-mail to be transmitted, and the kind of feeling can be determined and shown to the user. Can be objectively grasped whether or not was appropriate. Thereby, an unnecessary trouble with a third party can be prevented. Furthermore, it contributes to preventing an email with an inappropriate language feeling from being misunderstood by the recipient of the email.
In particular, the present invention is characterized in that the sensation is efficiently determined by focusing on greetings that are relatively easy to classify into patterns and often represent the relationship between the sender and the receiver.

According to the tenth aspect of the present invention, in the greeting sentence extraction process, extraction based on a rule base defined in advance by the user can be performed, so that a simple extraction method can be realized. Further, by using the rule base, it is possible to accurately determine an e-mail including a phrase that expresses a sense of word.
In particular, in a configuration in which e-mails of a plurality of sending users are transmitted via this apparatus, it is possible to detect a mail with inappropriate content on a rule basis.

According to the invention of claim 11, whether or not the word is included in the greeting expression dictionary having the appearance part information of the greeting sentence, the text information, the part of speech, and the word / phrase used in advance for greetings using the machine learning module. By using the information of the sender or receiver of the e-mail and the word / phrase of the subject of the e-mail, it contributes to the extraction of a highly accurate greeting.
In particular, it is possible to make a good determination by using a grammatical information such as a phrase, syntax, part of speech, etc. used in a greeting sentence as a feature, because a part where a greeting sentence appears generally has many sentence heads and sentence endings. Furthermore, it may affect whether it is easy to include a greeting from the domain name of the e-mail, and this can be used as a feature.

According to the twelfth aspect of the present invention, it is possible to determine the sensation based on the rule base by using the greeting sentence sensation rule database in the greeting sentence sensation determination processing means.
Many greetings are typical sentences and phrases, and a relatively good determination can be made by the rule base, and a simple and high-speed determination method can be realized.

According to the invention of claim 13, whether or not the word is included in the greeting expression dictionary having the appearance part information of the greeting sentence, the text information, the part of speech, and the word / phrase used for the greeting in advance, using the machine learning module. By using the information of the sender or receiver of the e-mail and the phrase of the subject of the e-mail, it contributes to the determination of the word feeling with high accuracy.
In particular, it is possible to make a good determination by using as features the location where the greeting appears, the number of times, grammatical information such as phrases, syntax, parts of speech used in the greeting. Furthermore, since the relationship between the sender and the receiver may appear from the domain name of the e-mail, this can be used as a feature.

According to the invention described in claim 14, since the past history of the mail input to the mail transmitting device can be stored, when sending an e-mail or when a user designates, for example, It is possible to refer to feelings of past e-mails sent to the same recipient. For example, when e-mailing the same sender, it is useful to create and send a greeting similar to that sent in the past.
If the greetings are different, a message that prompts attention depending on the situation can be issued.

According to the fifteenth aspect of the present invention, it is possible to determine the sensation of texts other than the greetings described above, for example, the body and title of an email.
At that time, by using the polarity dictionary etc. described in the prior art and making a determination by comparing the words and phrases registered in the dictionary with the text or by a predetermined calculation process, the text including a plurality of sentences is targeted. In addition, it is possible to easily determine the word feeling of the e-mail using a computer.

According to the invention described in claim 16, using a machine learning module, the feature includes text information in the learning data, whether or not a word or phrase in the polarity dictionary is included, or its sensation type information, sender or receiver of an e-mail It contributes to the determination of word sense with high accuracy by using the words of the subject and the subject of the e-mail.
In particular, it is possible to make a good determination by using text information such as sentences and phrases included in the e-mail, whether or not to include words or phrases in the polar dictionary, or using the word sense type information. Furthermore, since the relationship between the sender and the receiver may appear from the domain name of the e-mail, this can be used as a feature.

According to the seventeenth aspect of the present invention, since a greeting sentence to be added to an email to be transmitted can be automatically presented by providing a greeting sentence presentation processing means, a necessary and appropriate greeting can be easily and electronically transmitted. Can be added to mail.
In addition, it contributes to smooth communication by adding sentences with the same level of feeling as the previous greeting, or by adding greetings that differ only in wording.

  According to the invention described in claim 18, the electronic mail to be transmitted at the client terminal is received by the electronic mail transmitting apparatus of the present invention, and it is determined whether or not the greeting included in the mail is appropriate. it can. When the electronic mail transmitting device according to any one of claims 9 to 13, 15, and 16 is incorporated, the sensation type information is notified to a client terminal or a management terminal for managing an employee's transmitted mail in a company or the like. Can do. As a result, it is possible for the person who is trying to send the e-mail, the administrator, etc. to know the inappropriate e-mail in advance and take action such as canceling the e-mail transmission.

In addition, when the electronic mail transmitting apparatus according to claim 14 is incorporated, the past history can be transmitted to the client terminal or the management terminal.
According to the seventeenth aspect of the present invention, it is possible to transmit the greeting sent from the previous transmission to the client terminal or the management terminal.
As a result, it is possible to compare with the previous greeting text or add a similar greeting text using the previous greeting text, contributing to smooth communication.

  Embodiments of the present invention will be described based on examples shown in the drawings. The embodiment is not limited to the following. The present invention provides an e-mail receiving device and transmitting device, which may be configured separately or by a single computer.

The configuration according to the first embodiment forms a receiving device (11) and a mail transmitting device (12) in one terminal (10) as shown in FIG. Such a configuration is the same as that of an electronic mail application software (hereinafter referred to as mail software) of a general personal computer, and the configuration of this embodiment is implemented as a function of the mail software.
As is well known, mail software has a reception function and a transmission function, and communicates with an external transmission server and reception server through the Internet.

  Hereinafter, the configuration of the receiving device (11) with respect to the mail receiving function and the configuration of the mail transmitting device (12) with respect to the mail transmitting function will be described. In order to clarify each structure, it shows in another figure, but the process part which performs the same process can be shared suitably.

The configuration and processing method of the receiving apparatus (11) according to claims 1, 2, 4 and 19 of the present invention will be described.
FIG. 2 is a block diagram of an electronic mail receiver (hereinafter referred to as receiver) (11) according to the present invention. A flowchart of processing in this apparatus is shown in FIG. The present invention can be easily realized by a known personal computer, and each step of the present invention is executed and processed by a CPU (20) that controls arithmetic processing, machine learning, text processing, and the like. As is well known, the CPU (20) operates in cooperation with a memory (not shown). In addition to input means such as a keyboard and a mouse (21), a monitor (22) for displaying output results and an external storage such as a hard disk. A device (23) is provided.
In addition, a network adapter (24) connected to a network such as the Internet can be provided to input / output data.

The CPU (20) is provided with a mail reception processing unit (201), a greeting text extraction processing unit (202), a greeting text word feeling determination processing unit (203), and a received mail output unit (204).
And the program described by the well-known programming language operates CPU (20) and the hardware linked with it, The function of each part (201)-(204) demonstrated below is implement | achieved.

  In the external storage device (23), the greeting text rule database (231) storing the text information or appearance location information of the greeting text in the e-mail, and the greeting text text information in the e-mail correspond to each greeting text. A greeting sentence word sense database (232) in which word kind information is stored in advance is recorded.

In the present embodiment, a configuration will be described in which a greeting sentence extraction process and a process for determining the sensation of a greeting sentence are performed based on a rule (rule base).
Hereinafter, each processing of the present invention will be described in detail using the processing flowchart shown in FIG.
First, the mail reception processing unit (201) receives mail data sent through the network adapter (24). (Mail reception processing step: S10) When the CPU (20) is executing mail client software, the user is logged in to the mail server by authenticating the user name and password, and data is sent from the mail server. come. If the CPU (20) is executing the mail receiving server software, the mail data sent from the sender's mail transmitting server is stored in the user storage area by this processing (S10). Is done.

The greeting text extraction processing unit (202) executes the greeting text extraction processing (S11) from the received text data of the mail.
In the present embodiment, a greeting text extraction process based on the rule base is performed here. That is, a data table such as Table 1 or Table 2 is stored in the greeting text rule database.

In Table 1, greetings that are commonly seen are registered, and the greeting text extraction processing unit (202) compares the text strings of these greeting texts with the text data text of the received mail and matches them. Extract what to do. For example, referring to the number 2 “I am always indebted”, the same or similar sentences (referred to as related sentences) are extracted from the text data.
The case of similarity includes, for example, blurring of whether or not to use kanji, differences in the position of punctuation marks, addition and removal of some words, and the like.

  If you ’re always very helpful, everything like “I ’m always indebted”, “I ’m always indebted”, “I ’m very indebted”, etc. are all related Applicable to sentences. For these, kanji and reading may be defined in advance, or words that may be added or removed may be defined so as to be able to deal with blurring of notation.

Whether or not it is a related sentence can also be determined by how many characters are matched in one sentence including sentences registered in the greeting sentence rule database. For example, when a 16-character greeting text “I am always indebted to you” is registered in the greeting text rule database, if you check the text data of the received e-mail in order, Suppose you have a sentence (19 characters).
Since the overlapping character strings are 13 characters “Thank you very much”, the similarity between them is the sum of the fractions with the total number of characters as the denominator and the common number of characters as the numerator (13/16) × (13/19) = 0.468
Can also be calculated.
In addition, the sum of the total number of characters is the denominator, the sum of the number of common characters is the numerator,
(13 + 13) / (16 + 19) = 0.743
Can also be calculated.

In the above description, the similarity is calculated based on the number of characters. However, morpheme analysis may be performed using a known technique, the similarity may be obtained from the number and ratio of matching morphemes, and may be determined to be similar when the value is equal to or greater than a threshold value.
Various methods for calculating the degree of similarity are known, and any of them may be used. And when such a similarity calculation is performed and a predetermined threshold value is exceeded, the one sentence can be extracted as a greeting sentence.

Table 2 shows a configuration in which a greeting text is extracted based on the appearance position of the greeting text. That is, the top of the e-mail, specifically, the first sentence of the body of the e-mail as number 7 is defined as a greeting, and any sentence can be defined as a greeting.
In this case, since the first sentence is extracted as a greeting sentence anyway, if it is a greeting sentence, it is possible to determine the word feeling without any trouble. Even if the sentence does not greet at all, it is inappropriate in the greeting sentence word feeling determination process at the later stage. It works well enough because it is judged to be a mail that contains a sense of word.

  In addition, in the case of words such as “Thank you” with the number 5, it is highly likely that it is a greeting when it appears near the beginning or end of the text, and otherwise it may be an acknowledgment for the specific content Therefore, greeting texts can be extracted effectively by limiting the appearance positions.

  The appearance position can be defined by the number of characters as shown in Table 2, a method of defining by the number of lines, a ratio in the whole text, for example, within 5% from the beginning of the sentence, and the like. Of course, an intermediate range such as 20% to 50% from the beginning of the sentence can be defined.

Next, the greeting sentence feeling determination process (S12) will be described.
In the present embodiment, a method of determining a feeling based on a rule base using a greeting sentence feeling rule database (232) is also shown. Table 3 shows an example of the greeting sentence feeling rule database (232).

In the greeting sentence word sense rule database (232), a greeting sentence and a corresponding word feeling are defined. As the word feeling, the kind of word feeling may be defined as “affirmation”, “neutral”, or “denial” simply by the overall impression.
In this embodiment, an example is shown in which word feeling is classified and defined on several scales. That is, the greeting with the number A has a positive impression, expresses joy, and indicates that it is a greeting to the outsider. Similarly, the B greeting expresses a neutral impression and excuse. The greeting of C shows empathy to the people concerned. D's greeting is neutral and shows excuses to the people involved.

As described in Non-Patent Document 2, the classification method as shown in Table 4 can also be used for such sensation types.
As the sensation type information of the present invention, these sensation types may be used in appropriate combination.

  The greeting text word feeling determination processing unit (203) determines the corresponding word feeling type by comparing the greeting text extracted by the greeting text extraction processing unit (202) and the contents of the greeting text word feeling rule database (232). For example, when “Thank you for your work” is extracted as a greeting, the greeting sentence feeling determination processing unit (203) determines “affirmation / sympathy / related party” as the kind of feeling.

  In the greeting sentence word feeling determination process (S12), the word feeling determination can be extended to a similar sentence as in the greeting sentence extraction process (S11). In this case as well, the degree of similarity may be calculated so that the feeling of similar greetings can be determined.

Finally, the type of sensation determined by the process of the received mail output unit (204) is displayed to the user from the monitor (22). (Received mail output processing: S13)
The user can grasp what kind of nuance and meaning the received mail has objectively by looking at the displayed sensation type.

  As an output method, the data may be displayed on the monitor (22) or may be recorded as a log in the external storage device (23). The log may be periodically transmitted to a predetermined administrator.

  As a second embodiment, a configuration and processing method of a receiving apparatus (11) according to claim 6 of the present invention will be described. That is, as shown in FIG. 4, a past history storage processing unit (205) that stores an e-mail sender and word sense type information in the past history table (233) of the external storage device (23) in the receiving device (11). The past history (233) for detecting the history or change of the sensation type information with reference to the past history (233) in which the sent email and the sender information match, and the past history (233) together with the received email And a past history output unit (207) for outputting changes in sensation type information.

FIG. 5 is a flowchart of the process of the present embodiment. After the greeting sentence and sensation determination process (S12) of the first embodiment, the sensation type information determined by the past history storage processing unit (205) is the information of the sender of the mail. For example, the past history table (233) is stored together with the e-mail address and name. (Past history storage step: S14)
An example of the past history table (233) is shown in Table 5.

Here, the reception history and the greeting itself are also stored in the past history table (233). When the next e-mail reception process (S10) is performed, the past history detection unit (206) detects the past history of the same sender from the past history table (233).
Specifically, the mail address in the sender column is extracted from the text data received by the mail reception processing unit (201), and the matching data is extracted in the sender column of the past history table.
The detected data is a greeting text in the past mail and its word feeling.

Then, simultaneously with the received mail output process (S13), the past history output process (S16) is also performed. The user can display and compare greetings and speech from the same sender while viewing the received mail and the display of the determined speech.
It is difficult for users who receive a lot of emails to remember the nuances of past emails of all senders, but according to the present invention they are displayed in a comparable form. Thus, it is possible to clearly grasp the change in the sentiment of the sender.

In addition to such a configuration, a change in word feeling may be detected by the past history detection unit (206). In this case, the sensation determined in the greeting sentence sensation determination process (S12) is input to the past history detection unit (206) (dashed line in FIG. 5) and compared with the contents of the past history table (233).
For example, if any one of the words “affirmation / joy / external” has changed, the process proceeds to the past history output step (S 16) assuming that there has been a change.

  Alternatively, it may be output only when there is a change with respect to a specific scale. That is, since it is unnatural that the scale of the viewpoint changes for the same sender, the past history may be output when the viewpoint changes.

  In the attitude scale, a distance may be set between each type of sensation, and a past history may be output when the distance has changed beyond a threshold value. For example, if the distance threshold is 5 and the distance between “joy” and “anger” is 10, and the sentiment of the sender who is “joy” changes to “anger”, the change is The user can be warned.

It is also possible to compare greetings.
When comparing greetings, there are a method of outputting if there is a slight change, and a method of outputting when greetings differ to some extent based on the degree of matching.

Whether or not the character strings are completely matched can be determined by a simple character string comparison process without calculating the degree of coincidence. In the case of greetings, the nuance is often changed by a slight change of words even though the meaning is the same.
For example, in the example in the upper part of Table 5, if there is a past history of “I am always indebted to you” and the received greeting is “I am always indebted”, the degree of agreement is quite high. However, there is a slight change in nuances.

  Using the same method as described above, the degree of coincidence between the greeting text extracted from the received mail and the greeting text extracted from the past history table (233) is compared. A history output process (S16) may be performed.

  At this time, a synonym expression dictionary is prepared in advance, and the degree of coincidence is calculated assuming that synonymous expressions coincide. Also, a special expression dictionary is prepared in advance, and when the expressions in the dictionary do not match, the calculation is performed so that the degree of matching decreases particularly. It is also possible to calculate the degree of coincidence with respect to.

  The warning mode may be changed according to the degree of coincidence. For example, when the degree of matching is relatively high, “Caution: The wording of the greeting is changing” is displayed. When the degree of matching is low, “Warning: The wording is changing greatly.” Needs attention. ”, The user can be alerted.

It is not limited to the method of detecting whether or not the sensation changes based on the above-described predetermined calculation or rule, but can also be detected by machine learning described later. In this case, a combination of a greeting text or sensation type and a result of giving whether or not to detect in the case of the combination is prepared as teacher data. For the feature, a character string of a greeting text or a feeling type is used.
Machine learning is first performed by the machine learning module provided in the past history detection unit (206), and the learning result is stored in the external storage device (23).
Next, when the past history detection process (S15) is executed, the learning result is referred to, and the machine learning module performs a predetermined calculation to perform the detection process. Use machine learning like this

  The sender's information may be used for the feature. In this case, it is possible to make a difference depending on whether or not the change is easily detected depending on the sender, and it is possible to distinguish and detect a sender that requires special attention and a sender that does not require much attention even if it changes. .

As a third embodiment, the configuration and processing method of the receiving apparatus (11) according to claims 1, 3, and 5 of the present invention will be described. In this embodiment, greeting text extraction processing (S11) and greeting text feeling determination processing (S12) are performed using machine learning.
FIG. 6 is a detailed block diagram of the greeting text extraction processing unit (202) in this embodiment, FIG. 7 is a processing flowchart when machine learning is used in the greeting text extraction processing (S11), and FIG. 203) is a detailed configuration diagram, and FIG. 9 is a processing flowchart in the case of using machine learning in the greeting text word feeling determination processing (S12).

First, greeting text extraction processing will be described with reference to FIGS. After the mail reception process (S10), features used for machine learning calculation are extracted from the text data. (Feature extraction step: S11a)
The feature will be described in detail in the machine learning process described later. For example, the text of the data content is divided into morphemes by morphological analysis, and a predetermined feature is extracted. This feature extraction is performed in the same way during machine learning.

  The machine learning processing unit (2023) uses the greeting text learning result (2025) of the external storage device (23) prepared in advance, and uses the feature extracted by the feature extraction unit as an input for greeting text machine learning extraction processing (S11b) Execute. The greeting text learning result (2025) is, for example, a set of parameters that dynamically change during the learning process, although the data format differs depending on the machine learning model used.

The greeting text extracted as a result is used in the next greeting text feeling determination process (S12).
Next, greeting sentence feeling determination processing will be described with reference to FIGS. Features used for machine learning operations are extracted from the text data of the received mail. (Feature Extraction Step: S12a) Regarding this feature extraction, if machine learning is used in the greeting text extraction processing (S11), the features may be extracted simultaneously in the feature extraction processing (S11a).

  The machine learning processing unit (2033) uses the word learning result (2035) of the external storage device (23) prepared in advance, and uses the feature extracted by the feature extraction unit as an input for greeting text word feeling machine learning determination processing (S12b). Execute. The linguistic learning result (2035) is a set of parameters that dynamically change during the learning process, for example, although the data format differs depending on the machine learning model used.

The process proceeds to the received mail output process (S13) by the above two processes.
In the present embodiment, a configuration in which machine learning processing is used in both the greeting text extraction processing (S11) and the greeting text word feeling determination processing (S12) is shown, but any one of them may be based on a rule base. Further, for example, the execution processing may be performed together with the rule base method, and after the processing is first performed based on the rule base, the remaining text data may be processed again by machine learning.

Next, a machine learning method that can be used in the implementation of the present invention will be described mainly by taking an example of greeting text extraction processing.
The greeting text extraction processing unit in the present embodiment includes a teacher data input unit (2021), an answer-feature pair extraction unit (2022), a machine learning processing unit (2023), and an extraction unit (2024).

  In the machine learning method, a learning process in a known machine learning module and a solution estimation process using the learning process are integrated. In the implementation of the present invention, the learning process is not necessarily required, and the external storage device (23) may be provided only with the greeting text learning result (2025) formed as a result of machine learning. In that case, the greeting sentence extraction processing unit (202) only needs an extraction unit (2024) that determines the order according to each machine learning method.

  Various methods of machine learning are known, but each method will be briefly described here. Prepare many sets of problem-solution pairs, learn with them, learn what kind of solution is the problem, and use the learning results to guess the solution even for new problems (For example, see Non-Patent Document 5 to Non-Patent Document 7 below).

Murata, Language processing based on machine learning, Faculty of Science and Engineering, Ryukoku University. Invited lecture. 2004. http: // www2. nict. go. jp / x / x161 / member / murata / ps / kougi-ml-siryou-new2. pdf Japanese-English translation of tense aspect modality using support vector machine, Masaki Murata, Mao, Kiyochi Uchimoto, Hitoshi Isahara, IEICE Language Understanding and Communication Study NLC2000-78, 2001. CRL in the SENSEVAL2J dictionary task, Maki Murata, Masao Uchiyama, Kiyotaka Uchimoto, Mao Uchimoto, Hitoshi Isahara, IEICE Language Understanding and Communication Study NLC2001-40, 2001.

In order to convey to the machine the situation of the problem such as what kind of problem, features (each element constituting the problem with information used for analysis) are required. The problem is expressed by the feature. For example, in the problem of estimating the tense of Japanese sentence ending expressions,
Problem: "He speaks." ---- "Current"
Is given as an example, “He speaks.” “Speaks.” “Speaks.” “Su.” “.”.

  In other words, the machine learning method prepares many sets of feature set-solution pairs, performs learning, learns what kind of solution the feature set becomes, and uses the learning result. This is a method of extracting a set of features from a new problem and inferring a solution in the case of the feature.

The machine learning process obtains the classification result (solution) as to how the distributed text data is classified as shown in FIG. In the present invention, if morphemes in text data distributed in input / output data are used as features, the data types are classified from them. That is, the data type determined by the classification destination.
As a machine learning method in the machine learning processing unit (2023), for example, a k neighborhood method, a simple Bayes method, a decision list method, a maximum entropy method, a support vector machine method, or the like is used.

  The k-nearest neighbor method is a method for obtaining a classification destination (solution) by using the k most similar cases instead of the most similar case, and by majority decision of the k cases. k is a predetermined integer number, and generally an odd number between 1 and 9 is used.

  The Simple Bayes method is a method of estimating the probability of each classification based on Bayes' theorem and determining the classification having the highest probability value as a classification destination.

  In the Simple Bayes method, the probability of outputting the classification a in the context b is given by the following formula 1.

Here, the context b is a set of features f _j (εF, 1 ≦ j ≦ k) set in advance. p (b) is the appearance probability of the context b. Here, since it is independent of the classification a and is a constant, it is not calculated. P (a) (where P is a tilde at the top of p) and P (f _i | a) are probabilities estimated from the teacher data, respectively, and the appearance probability of class a, and the feature f for class a means the probability of having _i . When the value obtained by performing maximum likelihood estimation as P (f _i | a) is used, the value often becomes zero, and it may be difficult to determine the classification destination because the value of Equation 2 is zero. Therefore, smoothing is performed. Here, the smoothing using the following equation 3 is used.

However, freq _(f i, a), the number of cases has a feature _{f i} and classification is a, freq (a), the classification means the number of cases is a.

  The decision list method uses features and combinations of classification destinations as rules, stores them in the list in a predetermined priority order, and when input to be detected is given, from the highest priority in the list This is a method in which input data is compared with the feature of the rule, and the classification destination of the rule having the same feature is set as the classification destination of the input.

In the decision list method, the probability value of each classification is obtained using only one of the features f _j (εF, 1 ≦ j ≦ k) set in advance as the context. The probability of outputting classification a in a context b is given by

(Equation 4)
p (a | b) = p (a | fmax)
However, fmax is given by the following equation 5.

Further, P (a _i | f _j ) (where P is a tilde at the top of p) is a rate of appearance of the classification a _i when the feature f _j is included in the context.

In the maximum entropy method, assuming that a set of preset features f _j (1 ≦ j ≦ k) is F, the number 7 representing entropy is maximized while satisfying a predetermined conditional expression (Equation 6) below. This is a method of obtaining a probability distribution p (a, b) for the classification and obtaining a classification having the largest probability value among the respective classification probabilities obtained according to the probability distribution.

However, A and B mean a set of classifications and contexts, and g _j (a, b) is a function that is 1 if the context b has a feature f _j and the classification is a, and is 0 otherwise. means. Further, P (a _i | f _j ) (where P is a tilde at the top of p) means the rate of occurrence of (a, b) in the known data.

  Equation 6 is to obtain the expected value of the frequency of the output and feature pair by multiplying the probability p and the function g meaning the appearance of the pair of output and feature, The entropy maximization (smoothing of the probability distribution) is performed on the assumption that the expected values calculated based on the probability distribution obtained on the left side are equal, and the probability distribution of the output and the context is obtained. Details of the maximum entropy method are described in Non-Patent Document 8 below.

Eric Sven Ristad, Maximum Entropy Modeling for Natural Language, (ACL / EACL Tutoral Program, Madrid, 1997)

  The support vector machine method is a method of classifying data composed of two classifications by dividing a space by a hyperplane. FIG. 11 shows the concept of margin maximization in the support vector machine method. In FIG. 11, a white circle means a positive example, a black circle means a negative example, a solid line means a hyperplane that divides the space, and a broken line means a surface that represents the boundary of the margin area. FIG. 11A is a conceptual diagram when the interval between the positive example and the negative example is narrow (small margin), and FIG. 11B is a conceptual diagram when the interval between the positive example and the negative example is wide (large margin). is there.

  At this time, assuming that the two classifications consist of positive examples and negative examples, the larger the interval (margin) between the positive examples and the negative examples in the teacher data, the lower the possibility of incorrect classification with open data. As shown in FIG. 11B, a hyperplane that maximizes this margin is obtained, and classification is performed using the hyperplane.

  Basically, it is as described above. Usually, an extension of the technique that a small number of cases may be included in the inner area of the margin in the teacher data, or an extension that makes the linear part of the hyperplane nonlinear ( Kernel function introduced) is used.

  This extended method is equivalent to classification using the following discriminant function (Equation 8), and the two classes can be discriminated depending on whether the output value of the discriminant function is positive or negative.

Where x is the context (set of features) to be identified, and x _i and y _j (i = 1,..., L, y _j ε {1, -1}) are the context of the teacher data and the classification destination And the function sgn is
sgn (x) = 1 (x ≧ 0)
-1 (otherwise)
Further, each α _i is for the case where the number 9 is maximized under the constraints of the numbers 10 and 11.

  The function K is called a kernel function, and various functions are used. In this embodiment, the following polynomial is used.

(Equation 12)
K (x, y) = (x · y + 1) d

C and d are constants set experimentally. For example, C was fixed at 1 throughout all treatments. Moreover, two types of 1 and 2 are tried for d. Here, x _i satisfying α _i > 0 is called a support vector, and the portion taking the sum of Expression 8 is usually calculated using only this case. That is, only actual cases called support vectors are used for actual analysis.

  Details of the extended support vector machine method are described in Non-Patent Document 9 and Non-Patent Document 10 below.

Nello Cristianini and John Shawe-Taylor, An Introduction to Support Vector Machines and other kernel-based learning methods, (CambridgeUs, United States) Taka Kudoh, Tinysvm: Support Vector machines, (http://chasen.org/~taku/software/TinySVM/, 2002)

  The support vector machine method handles data with two classifications. Therefore, when handling cases with three or more classifications, a pair-wise method or a one-VS rest method is usually used in combination with this.

  In the pairwise method, in the case of data having n classifications, every pair (n (n-1) / 2) of two different classification destinations is generated, and a binary classifier indicates which is better for each pair. That is, it is obtained by the support vector machine method processing module and finally obtains the classification destination by majority decision of the classification destination by n (n−1) / 2 binary classification.

  For example, when there are three classification destinations a, b, and c, the one VS rest method generates three sets of classification destination a and other, classification destination b and other, classification destination c and other, The learning process is performed on the set of the support vector machine method. Then, in the estimation process based on the learning result, the learning results of the three sets of support vector machines are used. See how the three support vector machines are estimated as candidates to be estimated. Of the three support vector machines, it is the non-other classification target and the separation plane of the support vector machine. This is a method for obtaining a classification destination of a thing farthest from the object. For example, when a candidate is farthest from the separation plane in the support vector machine created by the learning process of “classification destination a and others”, the candidate classification destination is estimated as a.

With the machine learning model of the machine learning processing method as described above, the present invention extracts greetings. That is, the received e-mail text is determined in turn as to whether or not it is a greeting text, and what is determined as a greeting text is extracted.
As a result of the machine learning model, it can be classified into two types of greetings or whether it is a greeting.
Moreover, when classifying into the sensation types of greetings, it becomes a problem to classify into many sensation types as shown in Table 4.

  As a machine learning model in such language processing, it is known that the classification result by the support vector machine or the maximum entropy method is relatively good. In the present invention, the support vector machine or the maximum entropy method is used for the machine learning model. May be.

For example, the accuracy of the classification destination is calculated by the calculation process of the extraction unit (2024) using the greeting text learning result (2025) created as shown in FIG.
In that case, the extraction unit (2024) performs a threshold comparison process with reference to a predetermined threshold (2027) defined in the external storage device (23) in order to output any one determination result, and the threshold is exceeded. The largest classification destination is output as the judgment result.

  Although not shown, the same process as described above is performed for the greeting text language sense machine learning determination process (12b).

Next, the process of learning a greeting will be described in detail with reference to FIG.
At the time of learning, teacher data input unit (2021) inputs teacher data prepared in advance. The teacher data may be stored in an external storage device or the like. It is assumed that the teacher data includes a text whose data type is known in advance and is not a greeting, which is a greeting as a solution. The method of determining the solution can be designed as appropriate according to the machine learning method and the required result.
In addition, the method of obtaining a solution differs as follows depending on any of the machine learning methods described above.

  For example, in the embodiment of the present invention, when the k-nearest neighbor method is used as the machine learning method, the feature-feature pair extraction unit (2022) extracts features from the teacher data input by the teacher data input unit (2021) (S20). ) The machine learning processing unit (2023) defines the similarity between cases based on the ratio of overlapping features (the ratio of how many of the same features) in the extracted feature set (S21). The similarity and the case are stored as learning result information in the greeting text learning result (2025) (S22).

  The extraction unit (2024) refers to the data type probability defined in the greeting text learning result (2025) and the feature of the data extracted by the feature extraction unit from the e-mail text data. K features are selected from the example of the greeting text learning result (2025) in descending order of the possibility that the answer is correct, and the most probable classification destination (not greeting text or greeting text) is determined by a majority decision based on the selected k features. Is estimated as the solution.

  That is, in the extraction unit (2024), the degree of what kind of solution (classification destination) is likely to be obtained for each piece of extracted data is determined based on the number of votes of majority vote with the selected k features, here a greeting It is the number of votes obtained by the classification of whether it is a sentence or a greeting.

  Further, when the simple Bayes method is used as the machine learning method, the machine learning processing unit (2023) performs a machine learning process (S21) to set a pair of a solution of the case and a set of features as an example of the teacher data. It memorize | stores in a greeting sentence learning result (2025) as learning result information.

  Then, when the mail reception processing unit (201) receives the data content (S10), the extraction unit (2024) has a set of a solution of the learning result information of the greeting text learning result (2025) and a set of features. For the feature set acquired by the feature extraction unit based on Bayes' theorem, calculate the probability of classifying the greeting as a greeting or not as a greeting and classify the classification with the highest probability as the data The feature classification (solution) for is estimated.

  That is, in the extraction unit (2024), the probability of becoming a certain solution in the case of a set of features about the extracted data is a probability of being classified into each classification, here, a classification of greeting text or not greeting text. Probability.

  When the decision list method is used as the machine learning method, the machine learning processing unit (2023) arranges the rules of the features and the classification destinations in a predetermined priority order as examples of the teacher data as the machine learning process (S21). The list is stored in the greeting text learning result (2025). Then, when the mail reception processing unit (201) receives the data contents (S10), the extraction unit (2024) extracts the expression pairs extracted in descending order of priority from the list of greeting text learning results (2025). The feature of the candidate is compared with the feature of the rule, and the classification destination of the rule having the same feature is estimated as the classification destination (solution) of the candidate.

  That is, in the extraction unit (2024), the degree of the likelihood of being a solution in the case of a set of features of the extracted data is determined according to a predetermined priority or a numerical value or scale corresponding thereto, a greeting sentence or a greeting sentence here. The priority in the list of probabilities that the classification is not.

  When the maximum entropy method is used as the machine learning method, the machine learning processing unit (2023) specifies a class that can be a solution from the example of the teacher data, satisfies a predetermined conditional expression, and indicates an entropy. A probability distribution composed of a set of features when maximizing and a binomial classification that can be a solution is obtained and stored in the greeting text learning result (2025). Then, when the mail reception processing unit (201) receives the data contents (S10), the extraction unit (2024) uses the probability distribution of the greeting text learning result (2025) to identify the feature of the extracted file. The probability of the classification that can be the solution for the set of is determined, the classification that can be the solution having the largest probability value is identified, and the identified classification is estimated as the candidate solution. That is, in the extraction unit (2024), the degree of the likelihood of becoming a certain solution in the case of the set of features of the extracted data is the probability of being classified into each classification, here, the classification of whether it is a greeting sentence or not a greeting sentence. Probability.

  When the support vector machine method is used as the machine learning method, the machine learning processing unit (2023) specifies a class that can be a solution from the example of the teacher data, divides the class into a positive example and a negative example, In a space whose dimension is a set of case features according to a predetermined execution function using a function, the interval between the positive example and the negative example of the case is maximized, and the positive example and the negative example are divided by a hyperplane. The plane is obtained and stored in the greeting text learning result (2025).

  When the mail reception processing unit (201) receives the data content (S10), the extraction unit (2024) uses the hyperplane of the greeting text learning result (2025) to identify the feature of the extracted data. Whether the set is on the positive example side or the negative example side in the space divided by the hyperplane is specified, and the classification determined based on the specified result is estimated as the candidate solution.

  That is, in the extraction unit (2024), the degree of the likelihood of becoming a solution in the case of the set of features of the extracted data is determined based on the distance from the separation plane to the space of the positive example (data in which the appearance order is correct). Say it. More specifically, when a greeting is a positive example and a non-greeting is a negative example, the data located in the space on the positive example side with respect to the separation plane is determined as a greeting and the case is separated. Let the distance from the plane be the degree that the data is a greeting.

Furthermore, in the present invention, a method using a known neural network or a method using multiple regression analysis can be used as a machine learning method.
For example, if there are two types of classification to be obtained, multiple regression analysis can be used. The method of executing the multiple regression analysis on a computer is detailed in Non-Patent Document 11.

“Time Series Analysis and Forecasting with Excel”, Chapter 3, Ohm

  In the case of multiple regression analysis, as many explanatory variables x as the number of features are prepared, and the presence / absence of the features is expressed by 1, 0. The objective variable (explained variable) may be obtained with a value of 1 for a certain classification and a value of 0 for another classification.

  As described above, the present invention generates a greeting text learning result (2025) by using a machine learning module equipped with any known machine learning technique, and the extraction unit (2024) Determine the type accurately.

8 and 9 is the same as the above, and the teacher data input unit (2021) of FIG. 6 is replaced with the teacher data input unit (2031), and the feature-feature pair extraction unit (2022) is replaced. The above description is valid by replacing the feature-feature pair extraction unit (2032), the machine learning processing unit (2023) with the machine learning processing unit (2033), and the extraction unit (2024) with the determination unit (2034). The feature extraction process (S20) to learning result output (S22), teacher data (2026), and greeting text learning result (2025) in FIG. 7 are respectively converted into (S30) to learning result output (S32) and teacher data ( 2036), corresponding to the greeting text learning result (2035).
The predetermined threshold (2027) shown in FIG. 12 may be provided in the external storage device (23) also in the greeting sentence feeling determination process (S12b).

  In the present invention, the appearance location information of the greeting text in the teacher data, the text information, the part of speech, whether the word is included in the greeting expression dictionary provided with the words used in advance for greetings, information on the sender or receiver of the e-mail Some or all of the words in the subject line of the e-mail can be used.

  The case where the appearance information of the greeting text is used as the feature is based on the fact that in the greeting text extraction process, the greeting text tends to appear biased to a specific place as described above. In the greeting sentence feeling determination process, attention is paid to the fact that the nuances of the same words may differ depending on the appearance location.

The text information and the part of speech of the greeting are used for the feature. In the greeting extraction process, paying attention to the fact that the greeting is composed of a certain type of phrase, Contributes to extracting greetings. In the greeting sentence feeling determination process, attention is paid to the fact that the word feeling is determined by the phrase, and that the meaning and feeling differ depending on the part of speech even for the same character string.
The same is true for the email subject.

The case of using whether or not a word is included in the greeting expression dictionary means that in the greeting sentence extraction process, pay attention to the fact that the greeting sentence is composed of a certain standard phrase, especially used for other than the greeting expression By registering words that cannot be used as a dictionary, it is possible to easily extract words that can be said to be surely greetings if the keywords are included.
In the greeting sentence feeling determination process, the word feeling is determined by the phrase.

When email sender or recipient information is used as a feature, in the greeting text extraction process, whether the sender is a person or a greeting text depending on the relationship between the sender and the receiver We focus on how things change. For example, a partner who frequently discusses topics related to care should be determined not to be a greeting even if a word such as “care” is included. Such a determination can be made by using the information of the sender and the receiver.
In the greeting sentence feeling determination process, the relationship between the sender and the receiver is particularly important. The same “Thank you”, but when the sender is a boss and the receiver is in a subordinate relationship, there is a strong sense of gratitude, but when the sender is a store and the receiver is a customer, Become.

The present invention aims at high-precision processing by performing machine learning processing by paying attention to the above features.
Any one of the features may be used, but the text information and the part of speech of the greeting are particularly important, and the appearance location information may be combined in addition thereto. The present invention is preferably carried out with these features as the center.

(1) A method that uses machine learning to extract proper expressions
When extracting the features as described above, the extraction process can be automatically performed by using the following specific expression extraction technique without being manually defined in advance.
There is a method of extracting a specific expression using machine learning (see, for example, Non-Patent Document 12 below).

Masayuki Asahara, Yuji Matsumoto, Use of Redundant Morphological Analysis in Japanese Named Expression Extraction Information Processing Society of Japan NL153-7 2002

First, for example, the sentence “The prime minister of Japan is Mr. Koizumi” is divided into each character, and the correct characters such as B-LOCATION and I-LOCATION are assigned to the divided characters as follows. To set the correct answer. The first column below is each divided character, and the correct tag of each character is the second column.

B-LOCATION
I-LOCATION
O
Neck O
Phase O
Is O
Fortune B-PERSON
I-PERSON
O
N
At O
O
. O
In the above, B-? ? ? Is a tag that signifies the start of the type of proper expression below the hyphen. For example, B-LOCATION means the beginning of a unique expression called a place name, and B-PERSON means the beginning of a unique expression called a person name. I-? ? ? Is a tag that means other than the beginning of the type of proper expression below the hyphen, and O is other than these. Therefore, for example, the character “day” is a character that corresponds to the beginning of the unique name “place name”, and the character “book” is the unique name “place name”.

  In this way, the correct answer of each character is set, learned from such data, this correct answer is estimated with new data, and from this correct answer tag, the beginning of each proper expression and how far it is. Is recognized and the proper expression is estimated.

When learning from the correct data set for each character, the system uses various information in the form of features. For example,
B-LOCATION
Part of
Japan-B Noun-B
Such information is used. Japan-B means the beginning of the word Japan, and noun-B means the beginning of the noun. For recognition of words and parts of speech, for example, morphological analysis by Chasen described above is used. Since Chasen described above can also estimate the part of speech of each word, the following result is obtained when "going to school" is input.

School Gakkou School Noun-General
To He To Particles-Case particles-General
Go Iku Go Verb-Independence
EOS
In this way, each line is divided so that one word is included, and reading and part-of-speech information are given to each word.

  For example, in the non-patent document 12 described above, as features, characters themselves (for example, “small”), character types (for example, hiragana and katakana), part-of-speech information, tags, Information (for example, “B-PERSON” or the like) is used.

  Learning using these features. Investigate what features appear in the characters that estimate the tag and the surrounding characters, learn what features are likely to appear when the features appear, and use the learning results to create new data Perform tag estimation. For machine learning, for example, a support vector machine is used.

  In addition to the above-described method, there are various methods for extracting the proper expression. For example, there is a technique for extracting a specific expression using a maximum entropy model and a rewrite rule (see Non-Patent Document 13).

Uchimoto Kiyotaka, Ma Aoi, Murata Maki, Osaku Hiromi, Uchiyama Masao, Isahara Hitoshi, Named Expression Extraction Based on Maximum Entropy Model and Rewrite Rules, Journal of the Language Processing Society, Vol. 7, no. 2, 2000

Also, for example, in the following Non-Patent Document 14, using a support vector machine,
It describes a technique for extracting expressions.

Hiroyasu Yamada, Taku Kudo, Yuji Matsumoto, Japanese Named Expression Extraction using Support Vector Machine, Journal of Information Processing Society of Japan, Vol. 43, no. 1 ”, 2002

(2) Method using the created rule
There is also a method of extracting a specific expression by manually creating a rule.
For example,
If it is noun + "san", it will be named
Noun + “Prime Minister” is the name
Noun + “corporation” is the name of the company + “town” is the place name
For example, a noun + “city” is a place name.

  The unique expression is extracted by the above method, and among the extracted expressions, for example, a person name or a company name can be extracted by the solution feature pair extraction unit (2022).

In the first embodiment, a method using a rule base is described, in the second embodiment, a method of presenting to a user with reference to a past history, and in a third embodiment, a method using machine learning is described. In the present invention, it is of course possible to use Example 2 in combination with Example 3.
In addition, the rule-based method and the machine learning method can be implemented at the same time, and the user can select and use them together by specifying options.
Greeting texts extracted on a rule basis can be added as machine learning teacher data to perform machine learning processing, thereby enhancing learning results.

  The receiving device (11) of the present invention is to determine the feeling of the greeting as described above to understand the sentiment of the sender and to facilitate the communication. As described, it is possible to determine the utterance with higher accuracy by considering the vocabulary of the text other than the greeting. Examples are shown below.

  FIG. 13 is a configuration diagram of the receiving device (11) in the present embodiment. The configuration is the same as the above except that the CPU (20) includes an e-mail word sense determination processing unit (208) and the external storage device (23) includes a polarity dictionary database (234).

Here, for the polarity dictionary database, an adverb dictionary with evaluation polarity as proposed by the present inventors in Non-Patent Document 1 can also be used.
For example, as shown in Table 6, adverbs are evaluated by affirmation, denial, and neutrality.

An example will be described in which a sentence can be evaluated by focusing on adverbs. In the case of “clean” in Table 6, in the case of “cleanly stored the light source in a panel frame with a thickness of 20 mm”, if there is no adverb “clean”, simply state the fact and evaluate “neutral” It can be received.
However, by using the adverb “clean” in this sentence, it can be determined that the speaker of this sentence is positively evaluating the situation represented by the sentence.

  Next, an example will be described in which evaluations by sentence elements other than adverbs and adverb evaluations are correlated. This is the case for a sentence whose evaluation is determined only by sentence elements other than adverbs. In the case of the example of “Is n’t the Carpel coach's trap between sets effective?” In the case of this sentence, even if there is no adverb “Kitchi” in this sentence, It can be judged as a positive evaluation.

  On the other hand, in the case of “This time, there is a degree of difference inside the hair and it will be damaged by any means.” In this sentence, even if there is no adverb “Absolutely” in this sentence, It can be judged as negative evaluation. At this time, if the evaluation value of the sentence and the evaluation value of the adverb in the sentence are the same, it is possible to classify the evaluation of the sentence only by paying attention to the adverb.

As described above, by using the adverb dictionary in the polar dictionary database, the rule determination unit (not shown) of the email word sense determination processing unit (208) extracts words that match the adverb dictionary from the entire email, and includes adverb E-mail wording can be determined from the evaluation polarity.
For example, in the case where a plurality of adverbs are included, the wording of the e-mail may be output as “affirmed” if there are many positive evaluations by majority vote.
Alternatively, a value indicating the degree of affirmation is defined for each adverb, and an average value is obtained by calculation. If it is affirmative using an appropriate threshold, it is “affirmative”, if it is close to the intermediate value, “neutral”, negative If it is close, it may be configured to output “No”.

  Furthermore, in addition to the determination by the greeting text word feeling determination processing unit (203), the determination by the electronic mail word feeling determination processing unit (208) may be taken into consideration to make a comprehensive determination. In this case, when the type of sensation is represented by a numerical value, it can be calculated by a total value or the like. Moreover, a weight may be defined for each, and comprehensive determination according to the weight may be performed.

Machine learning can also be used for the determination by the e-mail word sense determination processing unit (208). FIG. 14 is a processing flowchart relating to this processing.
Since the machine learning method is the same as in the above embodiment, the feature extraction process (S14a) in the e-mail word sense determination process (S14) corresponds to S11a in FIG. The mail word sensation machine learning determination process (S14b) determined by the word sensation learning database corresponds to S11b in FIG.
In addition, the greeting text learning result (2025) and the teacher data (2026) are sent to the mail speech feeling learning result (2045), and the teacher data (2046) are sent to the feature extraction process (S20) to the learning result output process (S22). This corresponds to the processing (S40) to learning result output processing (S42).
In the teacher data, type information of the sensation to be determined from the electronic mail is defined in advance.

  In the present embodiment, text information in teacher data, whether or not to include words / phrases in the polarity dictionary, or feeling type information thereof, information on the sender / receiver of email, and words / phrases in the subject of email are used as features.

  Using text information as a feature is a method that uses any word or phrase in the body of an e-mail as a feature and judges the word feeling based on information such as how and how many such words appear. The feeling of the e-mail is determined.

When using whether or not to include a word or phrase in the polar dictionary as a feature, it is close to the rule-based determination method, but it is not a majority vote or average value as described above, but a method that is determined by machine learning in combination with other features It is. The polarity dictionary does not define the polarity for each word, but, for example, in the case of a dictionary indicating “affirmation”, only “whether or not a word in the polarity dictionary is included” can be used as a feature.
In the case of a dictionary including sensation type information as shown in Table 6, it is necessary to use the sensation type information for the feature.

  When the information of the sender or receiver of the e-mail and the word / phrase of the subject of the e-mail are used for the feature, the meaning is the same as that in the greeting sentence feeling determination process.

Depending on the sensation type determination of the e-mail text, the mail text may be classified and presented to the user in the received mail output process.
That is, from the content of the text, whether it is an email that requires an urgent response, whether it is an email that contains notes for the email recipient, whether it is a thank-you email or a complaint email, etc. Machine learning can be performed by the method, and the learning result can be presented to the user.

  When machine learning is performed to determine whether an urgent response is required, the learning database uses emails that require urgent responses and emails that do not need them. There are two types: “No urgent answer is required”. A plurality of classification destinations may be used, and a value indicating the degree to which an urgent answer is required, for example, a configuration in which classification is made into five levels may be used.

  Similarly, in the case of whether or not an e-mail contains precautions, the learning database uses e-mails that contain precautions and e-mails that are not described. Yes "and" Notices are not described ". A plurality of classification destinations may be used to classify the precautions in five levels.

  The same applies to whether it is a thank-you mail or a complaint mail, and machine learning may be performed using a thank-you mail and other mails, or a complaint mail and other mails in the learning database. In addition, since the thank-you mail and the claim mail are opposite to each other, the learning may be performed by “thank-you mail”, “claim mail”, and “other mail” and classified into three classification destinations.

For the features in these cases, it is preferable to use a word string, a character string, information on a mail sender / receiver, a word in a mail subject, and a character string.
In addition, important expressions corresponding to each problem can be extracted by feature analysis such as chi-square test, and the expressions can also be used for features.

In the first embodiment, it has been described that the present invention can integrate the receiving device (11) and the mail transmitting device (12) into one personal computer.
As a fifth embodiment, the configuration and processing method of the mail transmission device (12) according to claims 9, 10, 12 and 20 of the present invention will be described.
FIG. 15 is a block diagram of an electronic mail transmission device (hereinafter referred to as a transmission device) (12) according to the present invention. A flowchart of processing in this apparatus is shown in FIG. The present invention can be easily realized by a known personal computer. Like the receiving device (11), the CPU (30), input means such as a keyboard and a mouse (31), and a monitor (32) for displaying output results. An external storage device (33) such as a hard disk, and a network adapter (34).

The CPU (30) includes a transmission mail input processing unit (301), a greeting sentence extraction processing unit (302), a greeting sentence word feeling determination processing part (303), a word feeling output processing part (304), and a mail transmission processing part (305). ) Is provided.
And the program described by the well-known programming language operates CPU (30) and the hardware linked with it, The function of each part (301)-(305) demonstrated below is implement | achieved.

  In the external storage device (33), the greeting text rule database (331) storing the text information of the greeting text in the e-mail or the appearance location information, and the text information of the greeting text in the e-mail correspond to each greeting text. A greeting text word sense rule database (332) in which word sense type information is stored in advance is recorded.

In the present embodiment, a configuration will be described in which a greeting sentence extraction process and a process for determining the sensation of a greeting sentence are performed based on a rule (rule base).
Hereinafter, each processing of the present invention will be described in detail with reference to a processing flowchart shown in FIG.
First, the outgoing mail input processing unit (301) acquires the electronic mail input by the user using the keyboard (31) and the mail data sent from the user terminal through the network adapter (24), and the processing in the CPU (30) Used for. (Sent mail input processing step: S50)

A greeting text extraction processing unit (302) executes a greeting text extraction process (S51) from the text data of the received mail.
In this embodiment, as in the first embodiment, the greeting text is extracted based on the rule base. The greeting sentence rule database (331) used here has the same configuration as the database (231) of the first embodiment, and may be shared in the case of an integrated configuration with the receiving device (11).

  The next greeting text / feeling determination process (S52) is the same as the processing (S12) of the first embodiment, and the greeting text / feeling rule database (332) is shared with the database (232) of the receiving device (11). Can do.

Then, in the transmission device, the word feeling determined by the word feeling output processing unit (304) is output to the user (S53). Although the output form is arbitrary, for example, the linguistic information can be displayed together on the screen of an e-mail input by the user and used for transmission.
Further, the portion of the greeting may be clearly indicated by coloring or highlighting the portion of the greeting, and the linguistic information may be presented.

As the transmission device (12), the mail transmission processing unit (305) performs transmission processing (S54) of the electronic mail input by a known method to a designated destination. Although FIG. 16 shows a configuration that is performed simultaneously with the speech output process (S53), it may be performed after the speech output process (S53) as will be described later, or transmitted when a predetermined condition is satisfied. Also good.

  As a sixth embodiment, a configuration and processing method of a transmission device (12) (shown in FIG. 17) according to claim 14 of the present invention will be described. That is, the past history storage processing unit (305) for storing the e-mail recipient and speech sense type information in the past history table (333) of the external storage device (33), the e-mail to be sent and the sender The past history detection unit (307) that detects the history or change of the sensation type information with reference to the past history (333) with the same information, and the past history that outputs the change of the past history (333) and sensation type information An output unit (308) is provided.

FIG. 18 is a flowchart of the process of the present embodiment. After the greeting sentence feeling determination process (S52), the feeling type information determined by the past history storage processing unit (305) is changed to the information of the mail recipient (destination), For example, the past history table (333) is stored together with the e-mail address and name. (Past history storage step: S55)
An example of the past history table (333) is shown in Table 7.

Here, the transmission history and the greeting itself are also stored in the past history table (333). When the outgoing mail input process (S50) is performed next, the past history detection unit (307) detects the past history of the same sender from the past history table (333).
Specifically, the mail address in the destination (To, Cc, Bcc) column is extracted from the text data acquired by the transmission mail input processing unit (301), and the matching data is extracted in the recipient column of the past history table. .
The detected data is a greeting text in the past mail and its word feeling.

Then, the past history output process (S57) is also performed simultaneously with the speech output process (S53). The user can grasp whether the greeting attached to the outgoing mail that he / she wants to send is comparable to the past greeting or whether it has changed.
Also, even if an attempt is made to send the wrong address, the greetings given to the opponents in different positions are often different, so that the greetings are very different, so it is easy to notice. This also contributes to prevention of misdelivery.

In addition to such a configuration, a change in word feeling may be detected by the past history detection unit (307). In this case, the sensation determined in the greeting sentence sensation determination process (S52) is input to the past history detection unit (307) (the chain line in FIG. 18) and compared with the contents of the past history table (333). (Past history detection step: S56)
For example, if any one of the word feelings “affirmation / joy / external” has changed, the process proceeds to the past history output step (S57) assuming that there has been a change.

  Alternatively, it may be output only when there is a change with respect to a specific scale. That is, since it is unnatural that the viewpoint scale changes for the same destination, the past history may be output when the viewpoint changes.

  In the attitude scale, a distance may be set between each type of sensation, and a past history may be output when the distance has changed beyond a threshold value. For example, when the distance threshold is 5 and the distance between “joy” and “anger” is 10, the user is warned that the emotion of the user who has been “joy” has changed to “anger”. , Which also contributes to encouraging the creation of cool emails.

A configuration for comparing greetings, a configuration using a synonym expression dictionary, a configuration for changing a warning mode according to the degree of coincidence, a configuration for detecting a change in linguistic sense by machine learning, and the like, are all the same as the receiving device (11) Configuration can be taken.
In the transmission device (12), a configuration that outputs a past history of greetings is particularly useful. As mentioned above, even a slight change in the greeting may affect the impression of the other party. However, if you send it to many other parties, you may forget what kind of greeting you wrote before and change it to a less polite word than the previous one. Therefore, it is preferable that the transmission device (12) extracts the past history and warns when it does not completely match. This configuration is also suitable for finding typographical errors.

As described in claim 17 of the present invention, in the transmission device (12), the past history output unit (308) acting as a greeting text presentation processing means presents an appropriate greeting text to be added to the user. It is good also as composition to do.
That is, when the past history detection unit (307) detects that a greeting is not included, or when it is determined that there is a large change in speech feeling and is inappropriate, a past history greeting text extraction processing unit (not shown) A greeting text of the same destination is extracted, and a past history is presented to the user from a greeting text presentation processing unit (not shown). This presentation method may be configured to display directly on the monitor (32) or to be displayed on the user terminal through the network.
When the user selects from the input means or the like, a greeting is inserted into the text.

  Furthermore, a configuration may be adopted in which greetings are automatically added to all mails without prompting the user to select. In that case, the past history output process (S57) is a process of adding a greeting to the mail.

  As shown in FIG. 18, in the present embodiment, the mail transmission process (S54) is performed after the speech output process (S53) and the past history output process (S57). In this method, it is also possible to send an e-mail after selecting an appropriate greeting by the processing of the greeting text presentation processing section.

As a seventh embodiment, a configuration and a processing method of the transmission device (12) according to claims 9, 11, and 13 of the present invention will be described. In this embodiment, greeting text extraction processing (S11) and greeting text feeling determination processing (S12) are performed using machine learning.
6 is the same as FIG. 6 in the detailed configuration diagram of the greeting sentence extraction processing unit (302) in this embodiment, and the same as that of FIG. 8 in the detailed configuration diagram of the greeting sentence word feeling determination processing unit (303). The greeting sentence extraction process (S51) is the same as the greeting sentence extraction process (S11) and the feature extraction process (S20) to the learning result output process (S22) in FIG. The greeting text word feeling determination process (S52) is also the same as the greeting text extraction process (S12) and the feature extraction process (S30) to the learning result output process (S32) in FIG.

First, greeting text extraction processing will be described with reference to FIGS. After the outgoing mail input process (S50), the features used for the machine learning calculation are extracted from the text data. (Feature extraction step: S11a)
The machine learning processing unit (2023) uses the greeting text learning result (2025) of the external storage device (23) prepared in advance, and uses the feature extracted by the feature extraction unit as an input for greeting text machine learning extraction processing (S11b) Execute. The greeting text learning result (2025) is, for example, a set of parameters that dynamically change during the learning process, although the data format differs depending on the machine learning model used.

The greeting text extracted as a result is used in the next greeting text word feeling determination process (S52).
Next, greeting sentence feeling determination processing will be described with reference to FIGS. Features used for machine learning operations are extracted from text data of emails to be sent. (Feature Extraction Step: S12a) Regarding this feature extraction, if machine learning is used in the greeting text extraction processing (S11), the features may be extracted simultaneously in the feature extraction processing (S11a).

  The machine learning processing unit (2033) uses the word learning result (2035) of the external storage device (23) prepared in advance, and uses the feature extracted by the feature extraction unit as an input for greeting text word feeling machine learning determination processing (S12b). Execute. The linguistic learning result (2035) is a set of parameters that dynamically change during the learning process, for example, although the data format differs depending on the machine learning model used.

The sensation output process (S53) is performed by the above two processes.
In the present embodiment, a configuration in which machine learning processing is used in both the greeting text extraction processing (S51) and the greeting text word feeling determination processing (S52) is shown, but any one of them may be based on a rule base. Further, for example, the execution processing may be performed together with the rule base method, and after the processing is first performed based on the rule base, the remaining text data may be processed again by machine learning.

As described above, the transmission device (12) of the present invention determines the sensation of the greeting, thereby suppressing the transmission of inappropriate mail to the user who transmits the mail, thereby facilitating communication. However, as described in the fifteenth or sixteenth aspect, it is possible to determine the utterance with higher accuracy by considering the utterance of the text other than the greeting.
Since this technique is the same as that of the fourth embodiment of the receiving apparatus (11), description thereof is omitted.

As described above, the present invention is characterized in that the sensation of a sentence is detected from an e-mail greeting or text and presented to the user. The “sense of sensation” as used in the present invention is a subtle impression or subjective impression of a word, and the nuance of a sentence, the type of sentence, the meaning of a sentence, etc. can be classified by the sense of word.
At the time of reception, the sender's intention can be known, and the type of electronic mail (whether an urgent reply is necessary, thank you mail, etc.) can be clearly indicated to the user who is the receiver. Even at the time of transmission, an objective evaluation can be transmitted to the user who is the sender, and mail transmission without trouble can be promoted.

The present invention can also be provided as a mail transmission server (claim 18) provided with the transmission device (12) as shown in the fifth to eighth embodiments.
That is, as shown in FIG. 19, a mail transmission server (40) provided with a transmission device (12) is arranged, and a mail transmitted from the client terminal (41) is sent to the transmission device (12) by a transmission mail input process. The unit (301) (see FIG. 15) inputs and executes each process such as a greeting sentence extraction process (S51), a greeting sentence word feeling determination process (S52), and the like (see FIG. 16). Then, the sensation type information output from the sensation output processing unit (304) is transmitted from the terminal inquiry processing unit (401) to the client terminal (41) and the management terminal (42).

  Each of these terminals (41) and (42) can be constituted by a known personal computer, and known mail software is executed at the client terminal (41). Since the operations of the client terminal (41) and the mail transmission server (40) are well known as general Internet mail transmission processing, they will not be described in detail.

As an operation of the terminal inquiry processing unit (401), all of the determined sensation type information may be notified, or only in the case of the sensation type information corresponding to a predetermined condition. As an example of the predetermined condition, a case where the sensation type is “No” or a case where the sensation type changes can be used.
A configuration may be employed in which transmission is always made to the client terminal (41), while transmission is made only to the management terminal (42) when the word is “No”.
The client terminal (41) or the management terminal (42) is inquired whether or not the mail can be transmitted as it is, and the result is received. If transmission is possible, the mail transmission processing unit (305) transmits the mail to the destination.

  As described above, the mail transmission server (40) of the present invention interrupts the transmission of the electronic mail to be transmitted from the client terminal according to the predetermined condition, and the determination result is sent from the speech output processing unit (304) to the user terminal. You can also send and ask for user confirmation. This configuration is useful for, for example, mounting this apparatus on a corporate mail transmission server and preventing employees from sending inappropriate mail.

In the client terminal (41), for example, “send as it is” or “modify” is selected, and in the case of the former, an instruction signal to that effect is sent to the terminal inquiry processing unit (401) and sent from the mail sending device (12). A mail is transmitted to the external PC (43). In the latter case, the sensation of the corrected outgoing mail is determined again.
The terminal inquiry processing unit (401) may transmit a past history output from the past history output unit (308), a past greeting, or the like without being limited to the sensation type information.
When a past greeting is transmitted, the client terminal (41) may be able to select a greeting according to the presentation.

  Next, technologies that can be used in the processing of each of the above embodiments will be listed. In addition to exemplifying application examples, each technique can be used as appropriate within the scope of design matters.

(Search technology for similar documents)
First, similar mail data is detected from the past history table when determining the relevance between the greeting text stored in the greeting text rule in the greeting text extraction process and the greeting text to be extracted. In this case, a similar document search technique can be used.
After data collection, the collected data morphological analysis is performed, and then the word group A (in the present invention, the word group refers to one or a plurality of words) constituting the data obtained by the analysis is similar. A method for extracting information data will be described.

(1) Explanation of basic method (TF / IDF method)
(Equation 13)
score (D) = Σ (tf (w, D) * log (N / df (w)))
Add by w ∈ W
W is a set of keywords, tf (w, D) is the number of occurrences of w in the collected data, df (w) is the number of documents in which W appears in all documents, and N is the total number of documents. The document data having a high score (D) is output as similar similar data.

(2) Explanation of Oka wetting by Robertson et al.
This method is described in Non-Patent Document 15.

Murata Masaki, Ma Aoi, Uchimoto Kiyotaka, Osaku Hiromi, Uchiyama Masao, Isahara Hitoshi "Information Retrieval Using Location Information and Field Information" Natural Language Processing (Journal of the Language Processing Society) April 2000, Volume 7, Issue 2 , P. 141-p. 160 It is known that the number 14 in Non-Patent Document 15 has good performance. The product of the tf term and the idf term before taking the product by Σ is the Okapi weighting method, and this value is used as the word weight.

If it is an expression of Okapi (Equation 14)
score (D) = Σ (tf (w, D) / (tf (w, D) + length / delta) * log (N / df (w)))
Add by w ∈ W

The length is the length of the data D, the delta is the average length of the data, and the length of the data is the number of bytes of the data, the number of words included in the data, or the like.

Further, the following information search can be performed.
(Okapi reference)
The Okapi formula and SMART formula as disclosed in Non-Patent Documents 16 and 17 can also be used. As a more advanced information retrieval method, these Okapi SMART formulas may be used instead of formulas using only tf · idf.

S. E. Robertson, S.M. Walker, S.W. Jones, M.M. M.M. Hancock-Beaulieu, and M.M. Gatford Okapi at TREC-3, TREC-3, 1994 Amit Singhal AT & T at TREC-6, TREC-6, 1997

  In these methods, more accurate information retrieval can be performed using not only tf · idf but also the length of an article.

  In this method of extracting articles containing more word groups A, Rocchio's formula (Non-patent Document 18) can be used.

J. et al. J. et al. Rocchio, Relevance feedback back in information retrieval, The SMART retry system, Edited by G. Salton, Prentice Hall, Inc. , Pages 313-323, 1971

In this method, instead of log (N / df (w))
(Equation 15)
{E (t) + k_af * (RatioC (t) −RatioD (t))} * log (N / df (w))
use.

E (t) = 1 (keyword in the original search)
= 0 (other than that)
RatioC (t) is the occurrence rate of t in data group B
RatioD (t) is the occurrence rate of t in the article group C
The score (D) is obtained by replacing the log (N / df (w)) with the above equation, and the larger the value, the more the word group A is extracted.

  A set W of words w to be added when Σ of score (D) is added is assumed to be both the original keyword and the word group A. However, the original keyword and the word group A should not overlap.

  As another method, it is added when Σ of score (D) is added. The set W of words w is only the word group A. However, the original keyword and the word group A should not overlap.

  Here, a complicated method is used with the formula of roccio. However, as the sum of the number of appearances of the words in the word group A is larger, it may be taken out as an article containing more word groups A. Alternatively, the larger the difference in the appearance of the word group A, the more the word group A may be taken out as an article.

Next, a method for extracting similar sentences in the collected data group B will be described.
Define the similarity between articles. The similarity may be tf · idf, okapi, or smart. Two articles x and y that compare an article D and a query in tf.idf, okapi, smart, or the like may be used. A word included in both x and y is preferably w.

  A vector having each word as a dimension and the score of each word as an element is created, the vector of article x is changed to a vector (vector_x) using words contained in article x, and the vector of article y is changed to article y. It is good also considering the value of the cosine (cos (vector_x, vector_y)) of those vectors as the similarity of articles | goods using the word contained, and making it a vector (vector_y). For the calculation of the score of each word, tf · idf, okapi, and smart may be used.

  The expression after the Σ of those expressions is the expression for calculating the score. The value of the expression is the score for each word.

If tf · idf, then tf (w, D) * log (N / df (w))
If it is okapi, tf (w, D) / (tf (w, D) + length / delta) * log (N / df (w))
Is the formula.

  In addition, even in the extraction of articles containing more word groups A, the value of the cosine (cos (vector_x, vector_y)) of this vector is obtained, and an article with a larger word value is judged to be an article containing more word groups A. Good. In this case, the word included in the word group A is used as a vector (vector_x), and the word included in the article is used as a vector (vector_y).

The similarity between the data group B and the article information data x includes the following methods.
(1) Data group B that is most similar to article information data x and a method that uses the similarity of article information data x as the similarity (2) Article that is most similar to article information data x in data group B And (3) a method in which the average of the similarities of all articles in the data group B and the article information data x is used as the similarity.
Although other methods may be used, the similarity between the data group B and the article information data x can be obtained in this way, and the article having a high similarity can be extracted as a similar article.

  As another method, a word that appears biased in the data group B is extracted by the previous method, and score (D) based on Rocchio's formula is calculated using the word, and score (D ) May be extracted as similar data.

By this method, a similar sentence can be taken out, and in the present invention, this may be extracted as a similar mail group.
Note that the article may be extracted by executing both the above-described method of extracting articles including the word group A and the method of extracting similar articles described here.

  Here, in the case of the method based on extraction of articles including the word group A, normalization is required. As normalization, it is assumed that there is a document having the same keyword group as the input keyword group, and the score of the document is calculated. Then, the normalized similarity can be calculated by dividing the similarity by the score. Note that the normalization method is arbitrary.

Similarity determination can also be performed using a paraphrase technique disclosed in Non-Patent Document 19. In this method, first, a large number of synonym paraphrasing transformation rules are prepared, and the paraphrasing is performed using these rules. Paraphrasing with increasing similarity is performed, and the similarity is obtained between sentences having the highest similarity. The similarity can be calculated as a score, for example, how many keywords included in one sentence and the pattern appearing in the vicinity are extracted in the other sentence, but not limited to this. The similarity can be calculated by this method.
This paraphrase may be paraphrased for both sentences for which the degree of similarity is obtained.

  In this way, when the degree of similarity is calculated after paraphrasing, the sentences are in a similar state, and thus the degree of similarity between sentences can be calculated more accurately.

Mata Murata and Hitoshi Isahara, “Unified Paraphrasing Model-Utilization of Scale-Based Transformation”, Natural Language Processing, Vol. 11, No. 5, p. 113-133, The Language Processing Society of Japan, October 2004

(Method of using machine learning to extract related greetings)
Next, there is a method of using machine learning to extract a related greeting text in the greeting text rule extraction unit.
In the above, whether or not it is related is defined by reading kanji in advance and extracting related greetings on a rule basis. The machine learning model is used for this extraction itself to determine the validity of the extraction. You can also

The CPU (20) includes a similarity calculation unit (not shown) having a method for calculating the similarity of the above greetings, and a similar greeting evaluation unit (evaluation of the validity as a similar greeting using a machine learning model) (Not shown).
The machine learning module included in the similar greeting evaluation unit performs machine learning in the following manner and stores the result on the hard disk as machine learning result data.

For machine learning, a large amount of data extracted from the greeting text extraction processing unit (202) and a data group of the greeting text rule database (231) are prepared, and the input data for learning is input to the machine learning module. And
It is manually determined whether or not the data extracted by the greeting text extraction processing unit (202) is really valid as a greeting text, and the result is used as output data for learning.

At the same time, the similarity calculator calculates the similarity between the learning greeting text and the text in the learning greeting text rule database. The method for calculating the similarity is as described above. The similarity may be a binary result that is similar or not, or may be a numerical value indicating the degree of similarity.
And as a feature in machine learning, a word string constituting input data acquisition data and a word string constituting an article data group are used together with similarity.

Provided with such a similar greeting text evaluation unit, it is appropriate as an object to be extracted by the greeting text group by inputting the greeting text group extracted from the greeting text rule database (231) and the data determined to be relevant. It is determined whether or not. The greeting sentence extraction unit (202) outputs the result according to the determination result.
Any of the various methods described above can be used as the machine learning method in the machine learning module, and the judgment result at this time is output as “valid” or “invalid”, and the probability of being valid. May be output. In the former case, an “appropriate” article data group may be output, and in the latter case, it may be output when the probability exceeds a predetermined threshold.

(Method of extracting by similarity after paraphrasing)
In the present invention, a word string replacement unit (not shown) provided with the above paraphrasing technique and a similarity calculation unit are provided in the CPU, and the word included in the greeting sentence extracted from the e-mail in the word string replacement unit The similarity calculation unit calculates the similarity with the text in the greeting text rule database (331) while sequentially replacing the columns. Then, the word string is replaced so that the degree of similarity always increases, and it can be extracted as a greeting when the degree of similarity exceeds the predetermined threshold when the degree of similarity is highest.

(Output method)
In the present invention, the following technique can be provided when a speech is output or when a past greeting is output.
One of them is to collectively output speech feelings and greetings by clustering processing.

(A) Description of clustering There are various methods for clustering. The general ones are described below.

(Explanation of hierarchical clustering (bottom-up clustering))
Connect the closest members together to create a cluster. Clusters and clusters
No (clusters and members) also connect the nearest clusters.
Since there are various definitions of the distance between clusters, it will be described below.

・ The distance between cluster A and cluster B is the distance between members of cluster A and cluster B.
The method of setting the smallest distance among the distances
・ The distance between cluster A and cluster B is the distance between members of cluster A and cluster B.
The method of taking the largest distance among the distances
・ Distance between cluster A and cluster B, all members of cluster A and cluster B
The average of the members' distance is the distance
・ The distance between cluster A and cluster B, and the average position of all members of cluster A
The cluster position, and the average of all cluster B member positions
And the average of the distances between the positions is the distance.

There is also a method called the Ward method. Hereinafter, the Ward method will be described.
(Equation 16)
W = ΣΣ (x (i, j) −ave_x (i)) ^ 2
^ Means exponent.

First
Σ is an addition from i = 1 to i = g
Second
Σ is an addition from j = 1 to j = ni
x (i, j) is the position of j-th member of i-th cluster
ave_x (i) is the average of the positions of all members of the i-th cluster. When the clusters are joined together, the value of W increases, but in the Ward method, the clusters are separated so that the value of W is not as large as possible. I will stick together.

  The position of the member is a word extracted from the article, the type of the word is a vector dimension, the value of the vector element of each word is the frequency of the word, tf · idf of the word (ie, tf (w, D)) * Log (N / df (w))), Okapi's formula for the word (ie, tf (w, D) / (tf (w, D) + length / delta) * log (N / df (w))) And make it a member's position.

(Description of top-down clustering (non-hierarchical clustering))
Hereinafter, a method of top-down clustering (non-hierarchical clustering) will be described.

(Explanation of maximum distance algorithm)
Take a member. Next, take the member farthest from that member. These members will be the center of each cluster. The minimum distance between each cluster center and the member is taken as the distance of each member, and the member with the largest distance is the center of the new cluster. Repeat this. When the number of clusters reaches a predetermined number, stop repeating. Moreover, the repetition is stopped when the distance between the clusters is equal to or less than a predetermined number. In addition, there is a method in which the goodness of the cluster is evaluated based on an AIC information amount standard and the repetition is stopped using the value. Each member becomes the closest cluster-centered member.

(Explanation of K-mean method)
Consider clustering into a predetermined number k. Choose k members at random, and use it as the center of the cluster. Each member becomes the closest cluster-centered member. The average of each member in the cluster is the center of each cluster. Each member becomes the closest cluster-centered member. Moreover, the average of each member in a cluster is made into the center of each cluster. Repeat these. When the center of the cluster stops moving, it stops repeating. Or, repeat it a predetermined number of times. The cluster is obtained using the cluster center at the time of the final cluster center. Each member becomes the closest cluster-centered member.

(Explanation of clustering by word group)
As a document classification method similar to clustering, there is a method in which one or a plurality of word groups are defined in advance for each classification destination and classified into classification destinations based on whether or not the input information includes the word group. . The document classification method is also included in the clustering in the present invention.
When the input information includes a plurality of classification target word groups, the input information may be classified into the word group classification destinations with a large number of classifications, and a weight value is assigned to each word group. The words may be classified into the word group having a large weight.

  In this way, clustering is performed. Since various methods other than these are known as clustering methods, they may be used.

  In the present invention, the display mode when outputting from the monitor (22) in the past history output unit (207) can be changed as follows. That is, for some or all of the emails to be transmitted / received, the distribution date can be displayed in a graph with the date of transmission / reception and the type of sensation as the coordinate axes.

  For example, if the time series of received date is plotted on the horizontal axis and the higher the degree of affirmation on the vertical axis, it can be visualized how the positive evaluation changes according to the time series. In companies that receive a large amount of mail, it is easy to grasp the consumer's reaction as an overview.

  As a graph display method, in addition to displaying a set of points by plotting, a line graph showing the transition of the number of positive cases, a contour line graph with a high degree of positiveness, a display method using a bubble chart, and the like can be used.

(Explanation using significant difference test)
In each of the above embodiments, a threshold value setting method when using a predetermined threshold value, and when determining whether or not it is a characteristic word when extracting a greeting sentence / greeting expression, etc., a significant difference test Can also be used. This will be described below.
・ Explanation for binomial test
Let N be the number of occurrences of C in A. Let N1 be the number of occurrences of A at B.
Let N2 = N−N1.

  Assuming that the probability that A will appear in B when it appears in C is 0.5, the probability that A will appear in C and not appear in B out of N total occurrences, N2 times or less. Ask.

This probability is
P1 = ΣC (N1 + N2, x) * 0.5 ^ (x) * 0.5 ^ (N1 + N2-x)
(Where Σ is the sum of x = 0 to x = N2)
(However, C (A, B) is the number for extracting B objects from A different objects)
(However, ^ means exponent.)
If the value of this probability is sufficiently small, it can be determined that N1 and N2 are not equivalent probabilities, that is, N1 is significantly larger than N2.

If 5% test
P1 is less than 5%, and if 10% test, P1 is less than 10%.
It will be a criterion for judging whether or not you want.

  Words in which N1 is determined to be significantly larger than N2 are words that appear biased in the article group B. Also, the smaller P1 is, the more frequently the word appears in the article group B.

・ Explanation for chi-square test
The number of occurrences of A in B is N1, the total number of occurrences of words in B is F1,
The number of occurrences of A that is in C and not in B is N2,
Let F2 be the total number of words that are in C but not in B.

As N = N1 + N2,
Chi-square value = (N * (F1 * (N2−F2) − (N1−F1) * F2) ^ 2) / ((F1 + F2) * (N− (F1 + F2)) * N1 * N2)
Ask for.

And as this chi-square value is larger, it can be said that there is a significant difference between R1 and R2, and the chi-square value is
If it is greater than 3.84, it can be said that there is a significant difference of 5% in risk rate, and the chi-square value is greater than 6.63.
It can be said that there is a significant difference of 1% in the risk rate.

  As N1> N2 and the chi-square value is larger, the word appears more biased in the article group B.

・ Ratio test, to be exact, explanation of ratio difference test
p = (F1 + F2) / (N1 + N2)
p1 = R1
p2 = R2
As
Z = | p1-p2 | / sqrt (p * (1-p) * (1 / N1 + 1 / N2))
(Where sqrt means the root) And, as Z is larger, R1 and R2 are more significant, and when Z is greater than 1.96, there is a significant difference of 5% risk. When Z is larger than 2.58, it can be said that there is a significant difference of 1% of the risk rate.
A word that appears more biased in the article group B as N1> N2 and Z is larger.

  These three test methods may be combined with the method of simply determining the appearance rate of A in B / the appearance rate of A in C.

  For example, among those having a significant difference of 5% or more in the risk rate, the higher the appearance rate of A in B / the appearance rate of A in C, the more likely to appear in the article group B.

In the present embodiment, in the determination using machine learning, a predetermined number can be extracted in order from the one with the highest probability or the highest one. It is also possible to set a certain threshold value and extract a value exceeding the threshold value. Only a predetermined number may be extracted from those having a predetermined probability or higher and a higher probability. In this way, the criteria for extracting data in this way are arbitrary.
The threshold value and the predetermined number may be fixed in advance in the apparatus, or may be changed by the user.

(Support for multiple languages)
The language of the e-mail of the present invention may be any language. Known translation software may be provided in the mail reception processing unit (201) or the like and machine-translated to obtain data contents in Japanese. For example, the present invention processing can be performed after an English e-mail is translated into Japanese by translation software. In addition, since the processing of the present invention is not limited to Japanese, processing may be performed in English.
A configuration may also be adopted in which word processing is determined by translation processing at the stage of extracting a greeting. In this way, even when mail is sent and received in both Japanese and English languages, the previous Japanese mail feeling may be used for comparison with the later English mail feeling.

When outputting in a plurality of languages, the translation result may be displayed, or only the determination result of the feeling may be displayed.
Instead of using advanced translation software, you can simply match a dictionary and replace a word.
In particular, greetings are often typical even in foreign language mails, and even if they are replaced by words or sentences, the effects of the present invention are sufficiently achieved.

  In the present invention, when detecting the sensation of the body text of the electronic mail, the detection process can be performed after the summary process. That is, it is difficult to accurately detect the word feeling as a whole when a long mail is included and various topics are included. Therefore, by detecting the word feeling after performing the summarization process, it becomes possible to detect the word feeling of an important part.

First, various known methods are known for the summarization process. For example, the summarization methods disclosed in the methods of Patent Document 2 and Patent Document 3 by the present inventors can be used.
That is, according to the method of Patent Document 2, as summarization apparatus, solution data that stores solution data composed of a combination of a sentence and a problem that is a summary result thereof and a plurality of classification destination solutions that indicate evaluation on the summary result Based on the storage means and the sentences and summary results that are the problem of the solution data, for example, the information indicating the smoothness of the sentences of the summary results and the predetermined information including the information indicating whether or not the summary results display the contents of the sentences Extract as

  A solution-feature pair extraction unit that generates a set of the solution and a set of features; a machine learning unit that stores a set of the solution and the set of features as a learning result in a learning result storage unit; and a solution-feature Based on a set of feature extraction means for extracting the set of features from the input text and information extracted by the pair extraction means, and the solution and the set of features as a learning result, And a probability estimation unit that obtains the probability of each classification in the case of the set of text features obtained from the feature extraction unit based on the theorem, and uses the category having the largest probability value as an estimated solution to be obtained.

  The method described in Patent Document 3 is a solution data editing processing device that edits solution data used in processing for automatically summarizing text by a machine learning method, and displays summary results of text on a display device. Means for accepting an evaluation input for the summary result and evaluating the summary result; and a solution data output process for outputting solution data with the text and the summary result as problems and using the evaluation as a solution Means.

Japanese Patent No. 3682529 JP 2003-248676 A

All of the above-mentioned methods are intended to improve the accuracy of machine learning by evaluating or feeding back the results of the methods to known summarization methods. Contribute.
Of course, the machine learning method used here is not limited to the simple Bayes method, and any model such as a k-nearest neighbor method, a decision list method, a maximum entropy method, a support vector machine method, or a neural network method may be used.

Note that the summarization processing in the present invention is not limited to the method based on machine learning, and any known summarization method can be used.
For example, the score of each sentence is obtained using the information on the position of the document, the number of words including the title word, and the value of tfidf of the word that appears in the sentence, and the summary result is the one with the highest score. There is a way.
In general, since the title word is important, a sentence including many title words can be extracted and used as a summary result.
More simply, the first sentence such as the first sentence of the document may be a summary sentence.

It is explanatory drawing explaining the structure of Example 1 of this invention. 1 is a configuration diagram of an electronic mail receiving device according to Embodiment 1. FIG. 3 is a processing flowchart of the processing method. 6 is a configuration diagram of an electronic mail receiving apparatus according to Embodiment 2. FIG. 3 is a processing flowchart of the processing method. It is a detailed block diagram of the greeting sentence extraction process part which concerns on Example 3. FIG. It is a process flowchart at the time of using machine learning in a greeting sentence extraction process. It is a detailed block diagram of the greeting text word feeling determination processing part which concerns on Example 3. FIG. It is a process flowchart at the time of using machine learning in the greeting sentence feeling determination process. It is explanatory drawing explaining how the distributed text data is classified. It is a conceptual diagram of margin maximization of the support vector machine method. It is a process flowchart of a greeting sentence machine learning extraction process. FIG. 10 is a configuration diagram of an electronic mail receiving apparatus according to a fourth embodiment. It is a process flowchart in the case of using machine learning for the determination of an e-mail word sense determination processing part. FIG. 10 is a configuration diagram of an electronic mail transmission device according to a fifth embodiment. 3 is a processing flowchart of the processing method. FIG. 10 is a configuration diagram of an electronic mail transmitting apparatus according to a sixth embodiment. 3 is a processing flowchart of the processing method. It is a block diagram of the mail transmission server which concerns on Example 9. FIG.

Explanation of symbols

11 Mail receiver 20 CPU
DESCRIPTION OF SYMBOLS 201 Mail reception process part 202 Greeting sentence extraction process part 203 Greeting sentence word feeling judgment processing part 204 Reception mail output part 21 Keyboard / mouse 22 Monitor 23 External storage device 231 Greeting sentence rule database 232 Greeting sentence word feeling rule database 24 Network adapter

Claims (20)

An e-mail receiving device that receives e-mail from an external server via a communication network,
Mail reception processing means for connecting to the communication network and receiving e-mail;
Greeting text extraction processing means for extracting a greeting text from the received e-mail;
A greeting sentence sensation determination processing means for determining a type of sensation included in the greeting sentence using at least phrase information included in the greeting sentence;
An e-mail receiving apparatus comprising: a received e-mail output processing unit that outputs the determined sensation type information of the greeting text together with the e-mail. In the greeting sentence extraction processing means,
Greeting text rule database pre-stored text information or appearance location information of the greeting text in the email,
By referring to the greeting text database, a related greeting text is extracted by comparing the text information with the text included in the e-mail, or a greeting text is extracted from the text included in the e-mail according to the appearance information. The e-mail receiving device according to claim 1, further comprising: a greeting sentence rule extracting unit for extracting. In the greeting sentence extraction processing means,
Included in the greeting expression dictionary that uses pre-defined teacher data to determine which part of the e-mail is a greeting text, and includes information on the appearance of the greeting text in the teacher data, text information, parts of speech, and phrases used for greetings in advance Whether or not it is a word, machine information of the sender or recipient of the email, at least one of the words of the subject of the email as a feature, and a greeting text learning database storing the results,
3. A greeting text machine learning extraction unit that inputs the received e-mail and extracts a greeting text by machine learning processing with reference to the greeting text learning database. E-mail receiver. In the greeting sentence feeling determination processing means,
About the text information of the greeting text in the e-mail, the greeting text word sense rule database in which the word feeling type information corresponding to each greeting text is stored in advance,
A greeting text word sense rule determination unit that determines a greeting text word feeling type by comparing the greeting text extracted by the greeting text extraction processing means with the greeting text word feeling rule database or by a predetermined calculation process. The electronic mail receiving device according to any one of claims 1 to 3. In the greeting sentence feeling determination processing means,
Greeting data with greetings appearance information, text information, parts of speech, and phrases used for greetings in advance, using teacher data that pre-defines the type of sensation to be determined from greetings in email Whether it is a word included in the dictionary, information of the sender or recipient of the email, machine learning as a feature of at least one of the words of the subject of the email, and a greeting sentence feeling learning database storing the result,
A greeting sentence feeling machine learning determination unit that inputs at least the greeting sentence extracted by the greeting sentence extraction processing means and determines the word type information by machine learning processing with reference to the greeting sentence word feeling learning database; The electronic mail receiving device according to claim 1, wherein In the electronic mail receiving device,
In the electronic mail received by the mail reception processing means, at least information of the sender of the electronic mail, and past history storage processing means for storing in the storage means the sensation type information related to the electronic mail,
A past history detecting means for detecting a history or change of the sensation type information with reference to a past history in which the received email and the sender's information match at the time of receiving the email or at a predetermined opportunity;
6. The electronic mail receiving apparatus according to claim 1, further comprising: past history output means for outputting the history or change together with at least the received electronic mail. In the electronic mail receiving device,
Using at least word / phrase information included in the text of the e-mail, comprising e-mail word feeling determination processing means for determining the kind of word feeling included in the text,
The e-mail word sense determination processing means
A polarity dictionary database pre-stored word sense type information corresponding to a plurality of phrases;
An e-mail word sense rule determining unit that determines the word sense type of the text by comparison processing with the polar dictionary database or predetermined arithmetic processing with the words included in the text,
The electronic mail receiving apparatus according to any one of claims 1 to 6, wherein the received mail output processing means outputs the text or a feeling type of the text. In the electronic mail receiving device,
Using at least word / phrase information included in the text of the e-mail, comprising e-mail word feeling determination processing means for determining the kind of word feeling included in the text,
The e-mail word sense determination processing means
A polarity dictionary database pre-stored word sense type information corresponding to a plurality of phrases;
Using teacher data in which the type information of the sensation to be determined from the e-mail is defined in advance, the text information in the teacher data, whether or not it contains a phrase in the polarity dictionary, or the sensation type information, the sender or reception of the e-mail Machine learning with at least one of the user ’s information and / or the subject of the email as a feature,
An e-mail word sense machine learning determination unit that inputs at least the received e-mail and determines the word type information by machine learning processing with reference to the e-mail word sense learning database;
The electronic mail receiving apparatus according to any one of claims 1 to 7, wherein the received mail output processing means outputs the text or a feeling type of the text. An e-mail transmission device that transmits an e-mail to an external server via a communication network,
An outgoing mail input processing means for inputting an email to be sent;
Greeting text extraction processing means for extracting a greeting text from an email to be transmitted;
A greeting sentence sensation determination processing means for determining a type of sensation included in the greeting sentence using at least phrase information included in the greeting sentence;
A sensation output processing means for outputting the sensation type information of the determined greeting;
An e-mail transmission apparatus comprising: mail transmission processing means for transmitting an e-mail to a designated destination connected to the communication network. In the greeting sentence extraction processing means,
Greeting text rule database pre-stored text information or appearance location information of the greeting text in the email,
By referring to the greeting text database, a related greeting text is extracted by comparing the text information with the text included in the e-mail, or a greeting text is extracted from the text included in the e-mail according to the appearance information. The e-mail transmission device according to claim 9, further comprising: a greeting sentence rule extracting unit for extracting. In the greeting sentence extraction processing means,
Included in the greeting expression dictionary that uses pre-defined teacher data to determine which part of the e-mail is a greeting text, and includes information on the appearance of the greeting text in the teacher data, text information, parts of speech, and phrases used for greetings in advance Whether or not it is a word, machine information of the sender or recipient of the email, at least one of the words of the subject of the email as a feature, and a greeting text learning database storing the results,
11. A greeting text machine learning extraction unit that inputs the e-mail to be transmitted and extracts a greeting text by machine learning processing with reference to the greeting text learning database. E-mail sending device. In the greeting sentence feeling determination processing means,
About the text information of the greeting text in the e-mail, the greeting text word sense rule database in which the word feeling type information corresponding to each greeting text is stored in advance,
A greeting text word sense rule determination unit that determines a greeting text word feeling type by comparing the greeting text extracted by the greeting text extraction processing means with the greeting text word feeling rule database or by a predetermined calculation process. The electronic mail transmitting device according to any one of claims 9 to 11. In the greeting sentence feeling determination processing means,
Greeting data with greetings appearance information, text information, parts of speech, and phrases used for greetings in advance, using teacher data that pre-defines the type of sensation to be determined from greetings in email Whether it is a word included in the dictionary, information of the sender or recipient of the email, machine learning as a feature of at least one of the words of the subject of the email, and a greeting sentence feeling learning database storing the result,
A greeting sentence feeling machine learning determination unit that inputs at least the greeting sentence extracted by the greeting sentence extraction processing means and determines the word type information by machine learning processing with reference to the greeting sentence word feeling learning database; The electronic mail transmitting apparatus according to claim 9, wherein the electronic mail transmitting apparatus is an electronic mail transmitting apparatus. In the electronic mail transmitting device,
In the electronic mail input by the outgoing mail input processing means, past history storage processing means for storing in the storage means at least information of the sender of the electronic mail and linguistic type information related to the electronic mail,
A past history detecting means for detecting a history or change of the sensation type information with reference to a past history in which the information of the recipient and the email to be transmitted coincide with when the email is transmitted or at a predetermined opportunity;
14. The electronic mail transmitting device according to claim 9, further comprising: past history output means for outputting the history or change together with at least the electronic mail to be transmitted. In the electronic mail transmitting device,
Using at least word / phrase information included in the text of the e-mail, comprising e-mail word feeling determination processing means for determining the kind of word feeling included in the text,
The e-mail word sense determination processing means
A polarity dictionary database pre-stored word sense type information corresponding to a plurality of phrases;
An e-mail word sense rule determining unit that determines the word sense type of the text by comparison processing with the polar dictionary database or predetermined arithmetic processing with the words included in the text,
The electronic mail transmitting device according to any one of claims 9 to 14, wherein the word output processing means outputs the text or a word type of the text. In the electronic mail transmitting device,
Using at least word / phrase information included in the text of the e-mail, comprising e-mail word feeling determination processing means for determining the kind of word feeling included in the text,
The e-mail word sense determination processing means
A polarity dictionary database pre-stored word sense type information corresponding to a plurality of phrases;
Using teacher data that predefines the type of sensation to be determined from the e-mail, text information in the teacher data, whether or not it contains words in the polarity dictionary, or the sensation type information, sender or reception of the e-mail Machine learning with at least one of the user ’s information and / or the subject of the email as a feature,
An e-mail word sense machine learning determination unit that inputs at least the received e-mail and determines the word type information by machine learning processing with reference to the e-mail word sense learning database;
The electronic mail transmitting device according to any one of claims 9 to 15, wherein the word output processing means outputs the text or a word type of the text. In the electronic mail transmitting device,
A configuration comprising greeting text presentation processing means for automatically presenting a greeting text to be added to the e-mail to be transmitted,
The greeting sentence presentation processing means
A past history greeting text extraction processing unit that refers to the past history of the past history storage processing means and extracts a greeting text in the past history that matches at least the recipient information of the e-mail to be transmitted, and a part of the extracted greeting text An e-mail transmission device according to any one of claims 14 to 16, further comprising: a greeting text presentation processing unit for presenting all of them. A mail transmission server that receives mail data from a client terminal and transmits the mail toward a destination address included in the mail data,
The electronic mail transmitting device according to claim 9 to 17 is provided.
Notifying the client terminal or a predetermined management terminal of all the output results from at least one of the speech output processing means, the past history output means, and the greeting sentence presentation processing means, or a part corresponding to a predetermined condition. , Comprising at least terminal inquiry processing means for inquiring whether or not to send an e-mail to a destination,
A mail transmission server, wherein the mail transmission processing means of the electronic mail transmission device performs an electronic mail transmission process according to the inquiry result. An electronic mail receiving method for a computer that receives an electronic mail from an external server via a communication network,
A mail reception processing step, wherein the mail reception processing means connects to the communication network and receives an email;
A greeting text extraction processing means for extracting a greeting text from the received e-mail;
A greeting sentence feeling determination processing means, using at least phrase information included in the greeting sentence, a greeting sentence word feeling determination processing step of determining a kind of feeling included in the greeting sentence;
A received mail output processing means, comprising: a received mail output processing step of outputting, together with the electronic mail, the determined sensation type information of the greeting. A computer e-mail transmission method for sending an e-mail to an external server via a communication network,
An outgoing mail input processing means for an outgoing email input processing step for inputting an email to be sent;
Greeting text extraction processing means for extracting a greeting text from the email to be transmitted,
A greeting sentence feeling determination processing means, using at least phrase information included in the greeting sentence, a greeting sentence word feeling determination processing step of determining a kind of feeling included in the greeting sentence;
An electronic mail transmission method, wherein the speech output processing means includes a speech output processing step of outputting the determined speech type information of the greeting.

Images