JP2002278982A - Information extraction method and information retrieval method - Google Patents

Abstract

(57) [Summary] [Problem] To focus on evaluation expressions in a document and extract semantic classification or named expressions or evaluation information. Unknown words to the user,
Making unknown information or unknown evaluation information available to users for document search. SOLUTION: An evaluation expression dictionary 3 in which an evaluation expression "new" and a semantic classification "person" that can co-occur with the evaluation expression 3 are described in association with each other.
And a pattern dictionary 4 that describes the order of appearance 4a of the evaluation expression and the co-occurrence expression on the sentence.
And the dictionary 3 is used to identify a word that matches the pattern in the word string. The dictionary 4 searches for a pattern corresponding to the identified word string, and the evaluation target item “ The meaning classification “person” of “〇〇” is obtained from the dictionary 3. In addition to the semantic classification, an evaluation value, a date, a title, and the like are added. The document and the word “@” are registered in the database 10 in association with the extracted semantic classification and the like, and the semantic classification and the like are given to search for the document or the word.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

The present invention relates to a method for extracting information from an electronic document or an electronic dictionary and a method for retrieving information.

[0002]

2. Description of the Related Art In recent years, the amount of electronic documents distributed over the Internet and the like has increased dramatically, making it impossible to read all documents. Therefore, a search technology, an information analysis technology, and an information extraction technology for searching a user's desired document from a document database have been studied.

[0003] In a search, a method generally used is a full-text search in which a word input by a user is used as a keyword to collate the entire document, or a method in which a keyword is previously assigned to the document and collated with the keyword to retrieve the document. It is.

There is also a technique for extracting important expressions (proper nouns, dates, etc.) in a text from a document in advance. In particular, proper nouns can be used as keywords for extracting document contents and for searching with higher accuracy by estimating the meaning related to them.

[0005]

However, proper nouns are:
It is very difficult to register everything in a dictionary in advance because new organizations and products are created every day, and the people who are noticed change.

In Japanese Patent Application Laid-Open No. 7-85095, a document is subjected to syntax analysis and context processing to extract a specific expression and its relation, and to estimate the semantic classification of the specific expression from the relation. In this case, the meaning is presumed from a verb having a specific relational name of interest and a word having a modification relation with the verb. Therefore, when there is no modification relationship with the verb, the named entity cannot be extracted.

In the case of the above-described extraction in a format close to a slot in which a relation centered on a verb or a relation between words is described in advance, it is necessary to prepare a slot or a relation name in advance, and it is necessary to prepare a predetermined name. Only the information that has been extracted can be extracted.

Although information on facts in a document can be extracted based on slots and modification relations, information (e.g., good, bad, new, old, etc.) about the extracted named entity and evaluation of matters can be extracted from actual search and use of information. There is a problem that it is sometimes impossible for a user to extract data even though it is important.

[0009] A first object of the present invention is to extract information about a named entity and an evaluation by an expression about an evaluation. A second object is to make words unknown to a user, unknown information or evaluation information available for document search.

[0010]

Means for Solving the Problems (1) An evaluation expression dictionary in which an expression representing an evaluation and a semantic classification in which the expression representing the evaluation can co-occur is described, and an expression expression dictionary that co-occurs with the expression representing the evaluation Using a pattern dictionary that describes the order of appearance in a sentence, assigning semantic classification to specific expressions in text or words in a database or dictionary,
Semantic information extraction method.

According to this, "development" is performed as in the prior art.
Other than the predicates and their relations extracted in advance, etc., the nature of the word that modifies the word to be extracted even if it is unclear such as case relation or the co-occurrence restriction that the predicate can co-occur with various words is loose It is possible to extract by using.

In particular, words that modify / express the properties and states of adjectives, modifier nouns, adjective verbs, and the like are limited by the words that can be modified by the words, and semantic classification can be obtained along with extraction. In particular, it is effective for proper nouns such as names of persons and organizations that are not registered in the dictionary. Further, in addition to adjectives, modifier nouns, and adjective verbs, the words to be modified can be irrespective of part of speech such as adverbs, adverbs, and conjunctions.

[0013] Even when an unclear thing such as a case relationship or a predicate can co-occur with various words, and the co-occurrence restrictions are loose, it is necessary to extract words and obtain their semantic classification according to the properties of the words that modify the words to be extracted. Can be.

[0014]

DESCRIPTION OF THE PREFERRED EMBODIMENTS (1v) An evaluation expression dictionary in which expressions representing evaluations and meaning classifications in which expressions representing evaluations can co-occur is described, and texts of expressions co-occurring with expressions representing evaluations. Using a pattern dictionary that describes the order in which
Match the pattern of the pattern dictionary with the word string in the sentence, identify the word matched with the pattern of the word string using the evaluation expression dictionary, search the pattern dictionary, the pattern corresponding to the identified word string, An evaluation information extracting method for obtaining a semantic classification of an evaluation target item on a corresponding pattern from an evaluation expression dictionary.

(2) An evaluation expression dictionary in which expressions representing evaluations are associated with semantic classifications in which expressions representing evaluations can co-occur, and the order in which the expressions expressing evaluations and expressions co-occurring appear in the text An evaluation information extraction method for assigning an expression representing an evaluation to a specific expression in a text or a word in a database or a dictionary, using a pattern dictionary that describes the evaluation.

According to this, information on the evaluation can be automatically collected. In particular, information on evaluation can be obtained by using words such as adjectives and adjective verbs that have conventionally been involved in evaluation but have not been extracted much. In a specific example 2 described later, evaluation information on a certain person's name can be automatically collected. Further, evaluation information from a plurality of documents for the same target (the same person in Example 2) can be collected, and information for the same target can be efficiently collected.

That is, it is possible to automatically collect evaluation information on the extracted information. In particular, information related to evaluation cannot be obtained from static data, but must be created based on information of various documents, and can be created automatically.

(2v) Using an evaluation expression dictionary in which expressions representing evaluations and evaluation information are described in association with each other, and a pattern dictionary in which the order of appearances of expressions co-occurring with the expressions representing evaluations appear in sentences are described. Then, a pattern in a pattern dictionary is compared with a word string in a sentence, a word in the word string is compared with the pattern using the evaluation expression dictionary, and a pattern corresponding to the identified word string in the pattern dictionary is searched. A method for extracting evaluation information on a corresponding pattern from an evaluation expression dictionary.

(2v-1) The evaluation information extracting method according to (2v), wherein the evaluation information is an evaluation type.

(2v-2) The evaluation information extracting method according to the above (2v), wherein the evaluation information is an evaluation value.

(3) The evaluation expression dictionary is a description in which an expression representing an evaluation and an expression representing the evaluation are co-occurred with a semantic classification and evaluation information, and are described in a specific expression in a text or in a database. Alternatively, an expression representing an evaluation and an evaluation value are assigned to a word in the dictionary.
Method for extracting evaluation information.

According to this, it is possible to automatically collect evaluation information (advanced) about a certain object (in one example, a person's name), and evaluate values (+,-, +,-) of whether the evaluation information is good or bad.
0) can be collected. According to the information good (+) and the toe (-), for example, when a plurality of documents are found when a document is searched by the search system and the keyword specified by the search system user is also included, the evaluation information ( New, +
/ − / 0), in particular, by using the evaluation value (+ / − / 0), valuable information can be obtained.

It is possible to automatically determine whether the information of the evaluation regarding the extracted information is a good evaluation or a bad evaluation.
It can be used in the order of display when there are multiple evaluations of the same semantic classification or the same object, and efficient information presentation is possible.

(4) A specific expression in the text or
The evaluation information extraction method according to (2) or (3), wherein the date in which the document in which the evaluation expression appears is created and associated with a word in the database or the dictionary.

According to this, similarly to the above (2) or (3), evaluation information on a certain object (for example, a person's name) can be automatically collected, and the time element (for example, 1986.4) is also collected at the same time. Evaluation information is relative, and in particular, factors such as “new / old” or “good / bad” may change over time. By adding time information to the evaluation information, it can be used for reliability of the evaluation information itself and ranking of search documents at the time of search. Further, when the evaluation and the date information for the extracted word are added, the date information is compared with the stored date information, and if only the newer information is stored, only the latest evaluation information at the present time can be reflected.

By adding date information, reliability and novelty of collected evaluation information can be obtained. In particular, when words and evaluation information are extracted, only the latest evaluation information can be stored by checking the date information.

(5) A specific expression in the text or
(2), (3) or (4), wherein the source information or the title of the document to which the extracted word or expression belongs is added to the word in the database or dictionary in association with the extracted word. ) Described evaluation information extraction method.

According to this, similarly to the above (3), it is possible to automatically collect evaluation information on a certain object (in one example, a person's name), and obtain the source information of the document in which the evaluation appears,
Add a title. As a result, it is possible to extract the context, the scene, and the field in which the evaluation information for a certain object is made. By using this information, it is possible to determine in what context the evaluation of a certain term is an evaluation.

The contents of the evaluation can be described more accurately by adding the information on the word and the evaluation and the document title in which the information on the evaluation appears. In particular, in a search system or the like, when a user uses the system to know an unknown word or a specific word, the user can obtain a context for evaluation of a desired word and provide accurate information.

(6) A specific expression in the text or
The evaluation information extracting method according to the above (2), (3), (4) or (5), wherein information relating to a subject who has evaluated or said the extracted evaluation expression is further added to a word in the database or the dictionary. .

According to this, the contents of the evaluation can be provided more accurately by adding the information on the evaluation and the information on the subject (person name, organization) that performed the evaluation.

(7) The semantic information of the input word is extracted and registered in the word semantic information database in a form corresponding to the word by the semantic information extracting method of (1), and is associated with the input word. An information retrieval method for retrieving and reading the obtained semantic classification from the word semantic information database.

According to this, when the user wants to know an unknown named expression, it is possible to present the semantic information on the word from the automatically extracted word and its meaning information.

(8) By the evaluation information extraction method of (2), an expression representing the evaluation of the input word is extracted and registered in the word meaning information database in a form corresponding to the word, and An information search method for searching an expression representing an evaluation from the word meaning information database.

According to this, the user can provide automatically extracted words and information on the evaluation of unknown named entities and known named entities. In particular, information relating to evaluation cannot be obtained from static data and needs to be created based on information of various documents. This information can be created automatically and can be easily provided by search.

(9) The information search method according to the above (8), wherein an expression relating to evaluation can be input as a search key at the time of search.

According to this, conventionally, words related to evaluation have not been used as a search key of the search system, but are now available, so that the search efficiency of the user can be improved.

(10) The evaluation information of the input word is extracted by the evaluation information extraction method of (3) and registered in the word meaning information database in a form corresponding to the word, and the evaluation of the input word is performed. An information retrieval method for retrieving information from the word meaning information database.

According to this, the user can present automatically extracted words, evaluation information, and evaluation values for unknown and known named expressions. In particular, the order in which information is presented and the display information can be selected based on the evaluation value, and information can be provided efficiently.

(11) According to the evaluation information extracting method described in (4) above, the creation date of the document containing the input word is registered in the word meaning information database in a form corresponding to the word, and the input word exists. An information retrieval method for retrieving a document creation date and time from the word meaning information database.

According to this, by retaining the evaluation information and the date information, the display order of the search results can be selected so that the user can grasp the data more efficiently. The reliability of the information on the evaluation varies greatly depending on the date, and the user can efficiently judge the reliability of the information on the evaluation.

(12) According to the evaluation information extracting method of (5), the source information or the title of the document having the input word is registered in the word meaning information database in a form corresponding to the word, and the input word is registered. An information retrieval method for retrieving the source information or title of a document from the word semantic information database.

According to this, the user can efficiently use the evaluation information by displaying the source information and the title together with the evaluation information.

(13) Information about the subject who evaluated or said the input word in the above (6) is registered in the word meaning information database in a form associated with the word, and the subject who evaluated or said the input word. An information search method for searching for information on the word meaning information database.

According to this, the user can efficiently judge the reliability of the evaluation information by obtaining the information of the subject who performed the evaluation together with the evaluation information.

(14) For the words of the document collected through the network, the information described in any of the above (7) to (13) is extracted and registered in the word meaning information database, and is input by the user. A network information search method for searching registered information corresponding to a word from the word meaning information database.

According to this, a search dictionary is constructed based on a group of documents collected by a user through a network, and words or expressions that make it easy to search for documents on the network are entered.
It can be easily obtained from a search dictionary, and a document search on a network becomes easy. (15) A program for executing the method according to any one of (1) to (14) on a computer. (16) A medium on which the program of (15) is recorded.

Other objects and features of the present invention will become apparent from the following description of embodiments with reference to the drawings.

[0049]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS First Embodiment FIG. 1 shows an outline of a document processing system for implementing the present invention in one embodiment. This system includes a word dictionary 1 (database), a connection table 2 (database), an evaluation expression dictionary 3 (database), a pattern dictionary 4 (database), a morphological analysis unit 5 (a morphological analysis function using a computer and a program), and a semantic estimation unit. 6 (meaning estimation function by computer and program).

The morphological analysis unit 5 identifies a word unit and a part of speech of the text in the input document. For this morphological analysis, the word dictionary 1 and the connection table 2 are used. This morphological analysis is generally known.

The semantic estimation unit 6 converts the words and parts of speech divided by the morphological analysis into the evaluation expression dictionary 3 and the pattern dictionary 4
, It is determined whether the word in the pattern dictionary 4 is a word string that matches the part of speech, and if the word string matches, the meaning information described in the evaluation expression dictionary 3 is added to the word.

FIG. 2 shows a part 3a of the contents of the evaluation expression dictionary 3.
Is shown. This includes a word expressing evaluation (the column of “notation” on 3a), its part of speech (the column of “part of speech” on 3a), and the semantic classification of co-occurring words (the “semantic classification” on 3a). Column) are described in association with each other. Part 3a of the evaluation expression dictionary means, for example, a correspondence that the word “new sharp” has a part of speech “modification noun” which is a kind of noun, and the meaning of the co-occurring word is “person name”. ing.
The evaluation expression dictionary 3 associates a word (e.g., new) with an evaluation function that describes a property and an attribute of the co-occurred word, in particular, with a specific word (e.g., a person's name) and its part of speech and meaning classification. Registered.

FIG. 3 shows an example 4a of a pattern registered in the pattern dictionary 4. One pattern is described in one line, delimited for each word unit. Notation of morphologically analyzed words and part of speech on text in input document (Fig. 2)
If there is a description or part of speech described in the pattern dictionary 4, the part marked with "*" in the pattern dictionary 4 (4a) is extracted as a corresponding word. In the pattern dictionary 4, the arrangement of the parts of speech is specified. When a notation is described, "notation:" is described. In the part marked with “#”, the word of the corresponding part of speech in the evaluation expression dictionary 3 can be entered. Also, if multiple parts of speech or notation can be applied, "()"
Are delimited by “|”. If the same part of speech or notation can be repeated multiple times, ", +" is described after the part of speech or notation. "!" Indicates negation.

Example 4a shows that, in the evaluation expression dictionary 3, after the qualified noun, the notation “no” is followed by one or more “proper nouns” or “unregistered words” in the part of speech, followed by the noun And a part of speech other than unregistered words come.
Third, the noun continuation part (word in the block of!) Of the description is a word to be added / extracted with information at the time of matching. Example 4a is a pattern consisting of four items in total.

FIG. 4 shows an outline of the processing of the meaning estimation unit 6. The morphologically analyzed words of the text in the input document are collated with the pattern described in the pattern dictionary 4 while referring to the words in the evaluation expression dictionary 3, and the pattern described in one line of the pattern dictionary 4 is checked. If all of the items of the pattern apply (match), the part of “*” of the pattern is extracted.

A specific description will be given. First, 0 is set in a pointer LASP for storing a collated position (step 1). In the following, the word “step” is omitted in parentheses, and only the step number is described.

Next, a value obtained by adding 1 to the pointer LASTP is set to the pointer P to indicate the position of the sentence on the input document subjected to the morphological analysis (2). Next, as an end condition, P
Is larger than the number of input words, the process ends (3-end).

If P is not larger than the number of input words, the initial value 1 is set to the pointer I indicating the collation position of the pattern dictionary (4). As an end condition, when the pointer I becomes larger than the number of patterns, it is determined that the matching of all the patterns has been completed, and a position obtained by adding 1 to the last position of the last matching with the pattern LASTP is stored in P as a target position of the input word. (6).

If I is not greater than the number of patterns, each item of the I-th pattern is checked against the input word. The item pointer is set to J, and first, J = 1 is set (7). It is checked whether the J-th item of the I-th pattern matches the part-of-speech or notation of the P-th input word (8). At this time,
If the pattern has "#", the word of the part of speech of the description is obtained from the evaluation expression dictionary 3, and it is checked whether or not it matches the input word (Pth word). If they do not match, it is assumed that the pattern of interest does not match, and the pointer of the pattern is recommended next (I = I + 1), and the pointer P of the input target word is a value obtained by adding one to the last matching part lastp. Set (9,1
0). However, if the J-th item is repeatable and has already been collated once, it recommends one pattern item to be collated (J = J + 1) and continues collation (11).

When the pattern descriptions match, if the J-th item can be repeated, the input target pointer P
Is incremented by 1 (12-14), and if repetition is not possible, P and the pointer J of the item are incremented by 1 (P = P + 1, J = J + 1) (12-13).

The above is repeated until J becomes large in the number of pattern items (15-8). If it becomes larger, it is determined that all the items match, and a portion with "*" in the pattern is extracted and output (16). The last word of the collation word in the input word is stored in the pointer LASP (17). Where P
LASTP = LASTP-1 because is advanced by one before that.

As described above, patterns are sequentially collated for words in the target document (4-10), and processing is advanced to the next word at the stage of collation (7-2). After matching all patterns, if they do not match, recommend one word in the target document (6-2), and match the patterns in order (4-
10).

Output of word in case of match (16)
At this time, not only the corresponding word but also the information of the evaluation expression dictionary 3 is output together.

Example 1: The sentence "I'm holding up on the topic of Mr. New XXX" This sentence is analyzed by the morphological analyzer 5 as follows: (Notation) New | | | Ms | | Topic | In (Part of speech) Modified noun Case particle unregistered word Suffix Case particle noun Case particle (continued) | Excited | Ta (Part of speech) Verb ) Auxiliary verb.

The description of the first line of the pattern description is collated in order from the first word "new". The first part of the first line of the pattern is a part-of-speech “modifying noun” and “#”, which means that words in the evaluation expression dictionary 3 are to be collated. "Fastest" and "lightweight" are set as matching candidates.

Since "new" matches, the collation is continued.
The next item in the pattern is "notation: no", which also matches. The next item is “* (unregistered word | proper noun), +”, which means that there is at least one unregistered word or proper noun. Since the next target word is a part of speech “unregistered word”, it matches. Since it can be repeated, it matches with the next word "Mr.", but it does not match, but since it has already been matched once, the pointer of the item
(J) is recommended to be "! (Noun | unregistered word)", and there is no other word than noun or unregistered word. The next word,
Since the number of items exceeds (4), the current pattern is matched, and the item with "*" in the pattern is used as the target expression, and semantic classification is performed from the corresponding portion of the notation in the evaluation expression dictionary 3. And output. FIG. 5A shows an example 6a of the output information.

The output information is written in the word dictionary 1 used in the morphological analysis or in a larger knowledge dictionary.
By registering as a "person name", the vocabulary and its meaning can be automatically acquired. In addition, in the document in which the above sentence appears, the notation “XXXXX” is registered as a keyword in association with the document when the document including the sentence is registered in the search system. Keywords can be used.

In this embodiment, the "development"
Other than the predicates and their relations extracted in advance, such as those in the examples, even if the uncertainties such as case relations or predicates can co-occur with various words are loose, the words to be extracted Can be extracted using the property of the word that modifies. In particular, words that modify / express the properties and states of things such as adjectives, modified nouns, and adjective verbs are limited by the words that can be modified by the words, and can be extracted and semantically classified. In particular, it is effective for proper nouns such as names of persons and organizations that are not registered in the dictionary. Also,
In addition to adjectives, modifier nouns, and adjective verbs, the words to be modified can be irrespective of part of speech such as adverbs, adverbs, and conjunctions. Even in the case of unclear things such as case relationships or loose co-occurrence restrictions in which predicates can co-occur with various words, word extraction and semantic classification can be obtained depending on the nature of the word that modifies the word to be extracted.

Second Embodiment The system configuration of the second embodiment is the same as that of the first embodiment.
Word dictionary 1, connection table 2, morphological analysis unit 5, evaluation expression dictionary 3 and pattern dictionary 4 are the same, but semantic estimation unit 6
There is an addition to the function. That is, when the meaning estimation unit 6 of the second embodiment determines that the pattern matches in the target word string, the extracted word, the contents of the meaning classification and the evaluation of the meaning classification associated with the word in the evaluation expression dictionary 3 are determined. Is output as additional information of the extracted word and added.

Example 2: The sentence "I'm holding up on the topic of the newest Mr. XXX" This sentence is the same as Example 1, and is analyzed by the morphological analyzer 5 as in Example 1 as follows: ( Notation) New | | | ○○○○ | Mr. | | | Topic | de (part of speech) Modified noun Case particle unregistered word Suffix case particle noun Case particle-continued-(notation) | ) Verb (Ra row 5 steps sound form) Auxiliary verb.

The progress of the comparison with the pattern in the pattern dictionary 4 is the same as that in the first example, and matches the pattern in the first row of the pattern dictionary 4. The item with “*” in the pattern is used as the target expression, and the semantic classification and the evaluation expression itself are output from the corresponding portion of the notation in the evaluation expression dictionary 3: output “OOOO, semantic classification: person name, evaluation : New cutting edge.

FIG. 5B shows an example 6b of the output information. The output of the semantic classification is not essential. The output information is stored in the word dictionary 1 or a larger knowledge dictionary used by the morphological analysis unit 5, where the notation “OOOO” means “personal name”,
By registering as a rating "new", vocabulary, its meaning, and rating can be automatically obtained.

In the document in which this sentence appears,
By registering the output information in association with the document as a keyword when registering the document in the search system,
This keyword can be used when searching the search system.

According to the second embodiment, information on evaluation can be automatically collected. In particular, information on evaluation can be obtained by using words such as adjectives and adjective verbs that have conventionally been involved in evaluation but have not been extracted much.
In Example 2 above, evaluation information on a certain person's name can be automatically collected. Further, evaluation information from a plurality of documents for the same target (the same person in Example 2) can be collected, and information for the same target can be efficiently collected.

That is, it is possible to automatically collect evaluation information on the extracted information. In particular, information related to evaluation cannot be obtained from static data, but must be created based on information of various documents, and can be created automatically.

Third Embodiment The system configuration of the third embodiment is the same as that of the first embodiment.
The word dictionary 1, the connection table 2, the morphological analysis unit 5 and the pattern dictionary 4 are the same, but the contents of the evaluation expression dictionary 3 are added, and correspondingly, the function of the meaning estimation unit 6 is also added.

FIG. 5C shows a part 3 b of the contents of the evaluation expression dictionary 3. The evaluation expression dictionary 3 of the third embodiment is obtained by adding “+, −, 0” as an evaluation value (evaluation type). This is "+" for an evaluation expression having a generally good meaning as an evaluation of words in the evaluation expression dictionary 3, and "-" for an evaluation expression having an unfavorable meaning. "0" if neither
Is described below. Here, three types of evaluation types are set as evaluation values, but numerical values or rank information may be used as evaluation values.

The semantic estimating unit 6 of the third embodiment, when determining the match with the pattern in the target word string, as in the first embodiment, finds the extracted word and the corresponding word in the evaluation expression dictionary 3. The attached semantic classification contents, evaluation contents, and evaluation type contents are output as additional information of the extracted words and added.

Example 3: The sentence "I'm holding up on the topic of the newest Mr. XXX" This sentence is the same as in Example 1, and is analyzed by the morphological analyzer 5 as in Example 1 as follows: ( Notation) New | | | ○○○○ | Mr. | | | ) Verb (Ra row 5 steps sound form) Auxiliary verb.

The progress of the comparison with the pattern in the pattern dictionary 4 is the same as that in the first embodiment, and matches the pattern in the first row of the pattern dictionary 4. The item with “*” in the pattern is used as the target expression, and the semantic classification, the evaluation expression and the evaluation value are output from the corresponding part of the notation in the evaluation expression dictionary 3: Output “○○○○” Semantic classification: person name Evaluation: Newest Evaluation value: +.

FIG. 5D shows an example 6c of the output information. The semantic classification and the evaluation expression itself are not essential, and the following output is also possible: Output “OOOO” Evaluation value: +.

In the third embodiment, similarly to the second embodiment, it is possible to automatically collect evaluation information (advanced) for a certain object (in one example, a person's name) and determine whether the evaluation information is good or bad. Information (+,-, 0) can be collected. According to the information good (+) and the toe (-), for example, when a plurality of documents are found when a document is searched by the search system and the keyword specified by the search system user is also included, the evaluation information ( More valuable information can be obtained by using the newest (+ /-/ 0), especially the evaluation value (+ /-/ 0).

In the case where a plurality of documents have been evaluated with respect to the object "XXX", the following examples are extracted, and the meaning is classified as: ○○○ Meaning: Machine / System Evaluation: Fastest Evaluation value: + Evaluation : Light weight Evaluation value: + Evaluation: High value Evaluation value:-By calculating each evaluation value of the same object as one point, ○○○○ Meaning classification: Machine / System Evaluation: Fastest, light weight, high value Evaluation value: 2 Is obtained. This can be displayed in ascending order and descending order of the evaluation value in the same classification.

In the third embodiment, as described above, it is possible to automatically determine whether the information on the extracted information is good or bad. Of the same semantic classification,
Alternatively, the information can be used in the order of display when there are a plurality of evaluations for the same object, and efficient information presentation is possible.

-Fourth Embodiment- The system configuration of the fourth embodiment is the same as that of the third embodiment.
The word dictionary 1, the connection table 2, the morphological analysis unit 5 and the pattern dictionary 4 are the same, but the contents of the evaluation expression dictionary 3 are added, and correspondingly, the function of the meaning estimation unit 6 is also added. That is, in the fourth embodiment, the creation date (creation year, creation date) of the document to be analyzed is added together with the evaluation information. This describes when the evaluation is performed for an object having evaluation information from the evaluation expression dictionary 3.

The extracted word when the pattern is matched in the target word string, the semantic classification obtained by searching the evaluation expression dictionary 3 with the word, the evaluation expression itself and the evaluation type of the evaluation expression dictionary 3, and the analysis target document Is output as additional information of the extracted word and added. Note that semantic classification is not essential.
Further, either the evaluation expression itself or the evaluation type may be given.

Example 4: The sentence "I'm holding up on the topic of the newest Mr. XXX" This sentence is the same as in Example 1, and is analyzed by the morphological analyzer 5 as in Example 1 as follows: ( Notation) New | | | ○○○○ | Mr. | | | Topic | de (part of speech) Modified noun Case particle unregistered word Suffix case particle noun Case particle-continued-(notation) | ) Verb (Ra row 5 steps sound form) Auxiliary verb.

The progress of the comparison with the pattern is the same as in the first embodiment. The pattern is checked against the pattern on the first line of the pattern dictionary. The item having "*" in the pattern is set as the target expression, and the semantic classification, the evaluation expression and the evaluation type are output from the corresponding portion of the notation of the evaluation expression dictionary. In addition, the creation date (date) of the document to which the sentence to be analyzed belongs is added to the data.
The semantic classification is not essential, and at least one piece of evaluation information is sufficient.

Output Example 1) "○○○○" Meaning Classification: Person Name Evaluation: New Evaluation Type: + Date: 1986.4. This is shown in FIG. 5 (e) as an example of output information 6d.

Output Example 2) "○○○○" Evaluation: New Evaluation Type: + Year: 19
86.4 Output example 3) "○○○○" Evaluation: New year and date: 1986.4 Output example 1) "○○○○" Evaluation type: + year and month: 1986.4 In the fourth embodiment, the second and third embodiments Similarly, reputation information on a subject (eg, a person's name) can be automatically collected, and the time component of such reputation (1986.4)
Collect at the same time.

The evaluation information is relative, and in particular, “new
Factors such as "/ old" or "good / bad" can change over time. By adding time information to the evaluation information, it can be used for reliability of the evaluation information itself and ranking of search documents at the time of search.

Further, when the evaluation and the date information for the extracted word are given, they are compared with the stored date information and are stored only when they are newer, so that only the latest evaluation information at the present time can be reflected. .

By adding date information, reliability and novelty of collected evaluation information can be obtained. In particular, when words and evaluation information are extracted, only the latest evaluation information can be stored by checking the date information.

Fifth Embodiment The system configuration of the fifth embodiment is the same as that of the fourth embodiment.
The word dictionary 1, the connection table 2, the morphological analysis unit 5 and the pattern dictionary 4 are the same, but the contents of the evaluation expression dictionary 3 are added, and correspondingly, the function of the meaning estimation unit 6 is also added. That is, in the fifth embodiment, when the pattern is matched in the target word string, the extracted word, information on the meaning classification and evaluation obtained by searching the evaluation expression dictionary with the word, and the source information of the document in which the target sentence appears Alternatively, a title is added. Next, an example in which the source information is a title will be described.

Example 5: Title "Food, Sales Growth" Sentence "The company's product" XXX "is performing well" Analyzed by morphological analysis as follows: (Notation) ~ company | | “○○○○” | (Part of speech) Unregistered word suffix Noun of case particle Open parenthesis Unregistered word Close parenthesis −Continue− (Notation) is |

The outline of the comparison with the pattern is described in the first embodiment.
Is the same as In this example, the pattern does not match the pattern in the first row of the pattern dictionary 4, but matches the second pattern. "Product" to "good" match.

In the present embodiment, an item having "*" in the pattern is used as a target expression, and a semantic classification, an evaluation expression, an evaluation type, a document creation date and the like are analyzed in the present embodiment. The entire title of the document in which the target document appears is extracted and added. The creation date and the evaluation type are not essential in this embodiment. Next, an example of information addition will be described.

Output "○○○○" Meaning classification: Product evaluation:
Good Evaluation type: + Year: 2000.4 Title: Food,
This is shown as an example of output information 6e in FIG. 5 (f).

In the fifth embodiment, as in the third embodiment, evaluation information on a certain object (in one example, a person's name) can be automatically collected, and the source information of a document in which the evaluation appears, Add a title. As a result, it is possible to extract the context, the scene, and the field in which the evaluation information for a certain object is made. By using this information, it is possible to determine in what context the evaluation of a certain term is evaluated.

By adding information about words and evaluations and a document title in which the evaluation information appears, the contents of the evaluation can be described more accurately. In particular, in a search system or the like, when a user uses the system to know an unknown word or a specific word, the user can obtain a context for evaluation of a desired word and provide accurate information.

Sixth Embodiment The system configuration of the sixth embodiment is the same as that of the first embodiment.
The word dictionary 1, the connection table 2, and the morphological analysis unit 5 are the same, but the content of the evaluation expression dictionary 3 is slightly different, and the content of the pattern dictionary 4 and the function of the meaning estimation unit 6 are slightly different correspondingly.

FIG. 6 shows a part 3c of the contents of the evaluation expression dictionary 3.
Is shown. The contents of the evaluation expression dictionary 3 are words related to evaluations and remarks and their notations.

FIGS. 7A and 7B show two examples 4b and 4c of the contents of the pattern dictionary 4, respectively. The outline of these is the same as that of the first embodiment, but in the sixth embodiment, when the target word string matches a certain pattern, the item name for storing information in the database is described. I can do it. In the examples shown in FIGS. 7A and 7B, <> immediately after * is an item name, and a portion enclosed by {} is a portion to be extracted. The part surrounded by () is a notation for repetition and selection of a word. This is one realization method.
~ 5.

[0104] Regarding evaluation information, not words,
Sentences and phrases at the level of continuous words can be stored as evaluation information. The pattern 4b on the first line is a suffix (post title) that is followed by a noun or proper noun, “case particle”, followed by the notation “about”, and a proper noun or unregistered word followed by a personal name. Etc.) are connected, and the notation is "ha""ga""mo". Further, after the opening parenthesis, a sequence of character strings other than the closing parenthesis is located, and after the closing parenthesis is located, the notation “to” and the verb or the intransitive noun in the evaluation expression dictionary are located.

Items enclosed in parentheses {} (<speaker> consists of two items) are to be extracted, and in this example, are extracted with meanings enclosed in <>. The meaning estimation unit 6 has the same processing function as that of the first embodiment.

In the sixth embodiment, it is possible to extract the evaluation information and the information on the entity that has performed the evaluation. The sixth embodiment can be implemented in combination with the second to fifth embodiments.

In Example 6 shown below, when the pattern dictionary matches the target word string, a plurality of targets can be extracted, and the extracted targets can be described in a database or the like in association with each other.

Example 6: The sentence "Regarding the 2000 Economy, Prime Minister XX stated," We can expect a slight recovery in the future. " This is analyzed by morphological analysis as follows: (Notation) 2000 | | | Business | About | ○ | Prime | (Part of speech) − (Notation) is | “| Future |, | Business | Recovery | || (Part of speech) Adjuncts are open parentheses Adverbial nouns Reading nouns Sa-variable nouns Case particles -Continued-(Notation) Slight | Expectation (part of speech) Adjective verb Secondary particle Auxiliary verb Secondary particle is a verb Connective particle is a sub-noun-continued-(notation) | can | (Part of speech) Auxiliary verb Close parenthesis Case particle and utterance verb Auxiliary verb Punctuation The meaning estimation unit 6 compares the target word example with a pattern in the pattern dictionary. In the pattern 4b to which the above sentence matches, after one or more nouns, proper nouns, unregistered words, case particles, and numbers, there is the notation “about”, and after one or more proper nouns or unregistered words, Followed by a personal name affix (the affix that follows the personal name), followed by the notation "ga", "ha", "mo"
Any of which are located. Next, there is an opening parenthesis, and after that, anything other than the closing parenthesis can be matched, followed by the closing parenthesis.

The example sentence matches this pattern 4b. In the pattern 4b, a portion enclosed by parentheses {} and marked with “*” is an extraction portion, and in the present embodiment, an example is also given to what kind of extracted information is given. The type of information taken out (the part enclosed by <>) is not essential. The output is shown below.

Target Remarks Evaluation Output It is possible to take out "Economics in 2000 XX Prime Minister can expect a slight recovery in the future." FIG. 7C shows an output example 6f.
Is shown.

As described above, it is possible to automatically obtain the information of the comment and the evaluation and the information of the speaker and the evaluator for a certain target.

The sixth embodiment can be combined with the second to fourth embodiments, and when the date information of the fourth embodiment is added, the following is obtained. In the future, economic recovery can be expected, albeit slightly, on 1999.12 ". This output information 6h is shown in FIG. 7 (e).

Another example in which the sixth embodiment is combined with the third embodiment is shown below. FIG. 8 shows a part 3d of the contents of the evaluation expression dictionary 3 used in this combination. The contents of the evaluation expression dictionary 3 include words related to evaluation and remarks and their notations,
It consists of evaluation values: Example sentence "The secretary XX criticized the Upper House election as saying," The issues of policy of each party are unclear. " ] Morphological analysis result (notation) ○○ | Secretary | About | The Upper House election | About | "| Each party | | Policy || Issue |||||||||||||||| (Part-of-speech) Proper noun Personal name affixing auxiliary particle is noun case particle open parenthesis noun case particle noun case particle is adverb verb auxiliary verb no closing parenthesis case particle and sa-variable noun auxiliary verb auxiliary verb phrase target output evaluation The evaluation value is not clear. "FIG. 7D shows the output information 6g.

According to the sixth embodiment, the contents of the evaluation can be provided more accurately by adding the information on the evaluation and the information on the subject (person name, organization) that performed the evaluation.

Seventh Embodiment FIG. 9 shows an information retrieval system according to a seventh embodiment. The document analysis unit 30 includes the document processing systems (1 to 6) shown in FIG. When a document is input, the document analysis unit 30
However, the semantic information extraction method of the first embodiment or the second to
According to the evaluation information extracting method of the sixth embodiment, the evaluation expression dictionary 3
And the pattern dictionary 4 to extract relevant words from the document and information on the meaning or evaluation of the words, associate the extracted words with the meaning and the evaluation information, and store them in the word meaning information database 10. Also, these can be stored in association with the extracted document.

FIG. 10 shows the document analysis 10 of the document analysis unit 30.
The outline of a is shown. The document analysis unit 30 first performs a morphological analysis of a sentence in the target document (21), and extracts a meaning or an evaluation expression for all words in the target document (22). Then, the extracted words are associated with the evaluation information and the meaning classification associated with the extracted words and stored in the word meaning information database 10 (23).

FIG. 11 shows an information retrieval section 40 of the information retrieval section 40.
The outline of a is shown. If there is a search input, the information search unit 40 performs a morphological analysis on the input expression (31, 32).
The word meaning information database 10 is searched (34) using each word subjected to morphological analysis as a search key, and the meaning and evaluation expression information associated with the word are read out and displayed (3).
5). The process ends when all words have been searched using the search key (33-processing end).

Example 7: An analysis example and a storage example in which the following sentence is present in the document: The sentence “I'm holding up on the topic of the newest Mr. XXX” The document analysis unit 30 performs the following by morphological analysis. Analyze to: (Notation) New | | | Excitement | Ta (part of speech) verb (ra line 5 steps sound form)

Example of storage in the semantic information database 10: "OOOO" semantic classification: person name The semantic classification "person name" for the word "OOOO" is stored in the semantic information database.

A search is performed when the user knows that there is a word "OOO" but does not know what the word is. First, the user is asked to
○ ”is input to the information search unit 40. The information search unit 40
The word / semantic information database 10 is searched using the input “OOOO” as a search key, and “person name” is displayed as the meaning classification. If necessary, the document from which the semantic information data is based is also displayed.

According to the seventh embodiment, when a user wants to know an unknown named expression, semantic information on a word can be presented from automatically extracted words and their meaning information.

-Eighth Embodiment- The system configuration is that shown in FIG. The document analysis unit 30 of this embodiment extracts the evaluation information of the second embodiment.

Example 8: An analysis example and a storage example in which the following sentence is present in the document: The sentence “I'm holding it up on the topic of Mr. OOOO” The document analysis unit 30 performs the following by morphological analysis. Analyze to: (Notation) New | || ○○○○ | Mr. | | | | (Part of speech) Modified noun Case particle unregistered word Suffix Case particle noun Case particle-Continuation-(Notation) | Excitement | Ta (part of speech) verb (ra line 5 steps sound form)

Example of storage in the semantic information database 10: "OOOO" semantic classification: person name Evaluation: new word The semantic classification "person name" and the evaluation information "new sharp" for the word "OOOO" The evaluation information is stored in the word meaning information database 10.

A search is performed when the user knows that there is a word "OOO", but does not know what the word is, or knows the meaning of the word itself. When the user inputs “OOOO” to the information search unit 40, the information search unit 40 outputs the input “OOOO”
Is used as a search key to search the word semantic information database 10, and the semantic classification "person name" and the evaluation "new" are displayed. If necessary, the document from which the semantic information data is based is also displayed.

According to the eighth embodiment, the user can provide automatically extracted words and information on the evaluation of unknown named entities and known named entities. In particular, information relating to evaluation cannot be obtained from static data and needs to be created based on information of various documents. This information can be created automatically and can be easily provided by search.

Ninth Embodiment A system configuration is as shown in FIG. The document analysis unit 30 of this embodiment extracts the evaluation information of the second embodiment, as in the eighth embodiment. However, slightly different from the eighth embodiment, an expression relating to evaluation can be input as a search key. That is, the information search unit 40 also searches for expressions related to evaluation.

Example 9: If the following sentence is present in the document, the sentence “The fastest XXXXX has been developed” sentence “The lightweight XXXXX has appeared” sentence “The fastest XXXXX has been developed The document analysis unit 30 performs a morphological analysis and stores the following information in the word semantic information database 10: ○○○○ semantic classification: machine / system Evaluation: fastest ○○○○ semantic classification: machine / system Evaluation: Lightweight ×××× Meaning classification: Machine / System Evaluation: Fastest.

When the user inputs, for example, the words “fastest” and “machine” into the information search unit 40, the information search unit 40 searches the word meaning information database 10 using the input words as search keys. Search and read ○○○○ Meaning Class: Machine / System Evaluation: Fastest ×××× Meaning Classification: Machine / System Evaluation: Fastest Read and display.

According to the ninth embodiment, words related to evaluation have not conventionally been used as a search key of a search system. However, since the words can be used, search efficiency of the user can be improved.

Tenth Embodiment The system configuration is as shown in FIG. The document analysis unit 30 of this embodiment extracts the evaluation information of the third embodiment.
That is, the feature is that the evaluation value is “+ −0” or a numerical value as the evaluation information.

The semantic classification “person name” evaluation “new” evaluation value “+” is displayed. If necessary, the document from which the semantic information data is based is also displayed. Also, when there is a plurality of pieces of information about the word “OOOO” to be searched, the information is arranged and displayed in order of the evaluation value, so that the word “OOOO” is displayed.
Is presented to the user in an easy-to-understand manner.

In the example, there are three types of evaluation values, but it is also possible to display information in the database as numerical values, for example, in descending order and ascending order of numerical values. If the number of search results is large, for example, Example 10 Display example ○○○○ Meaning classification: Machine / System Evaluation: Fastest Evaluation value: + Evaluation: Light weight Evaluation value: + Evaluation: High value Evaluation value:- indicate. In addition, an evaluation value for a certain target can be presented by setting the value obtained by adding only the evaluation value to the same target as the evaluation value. As a display example, ○○○○ Semantic classification: machine / system Evaluation: fastest, light weight, expensive Evaluation value: 2 For example, it is also possible to display data in the database in the order of evaluation values in the same semantic classification .

According to the tenth embodiment, the user can present automatically extracted words, evaluation information, and evaluation values for unknown or known named entities.
In particular, the order in which information is presented and the display information can be selected based on the evaluation value, and information can be provided efficiently.

-Eleventh Embodiment- The system configuration is as shown in FIG. The document analysis unit 30 of this embodiment extracts the evaluation information of the fourth embodiment.
That is, the feature is that date information is added together with the evaluation information.

Example 11: If the following sentence is present in the document, the sentence “The fastest XXXXX has been developed” sentence “The lightweight XXXXX has appeared” sentence “The fastest XXXXX has been developed The document analysis unit 30 performs a morphological analysis and stores the following information in the word semantic information database 10: ○○○○ semantic classification: machine / system Evaluation: fastest date: 2000.4 ○○○○ semantic classification: Machine / System Evaluation: Lightweight Date: 2000.4 ×××× Meaning Classification: Machine / System Evaluation: Fastest Date: 2000.4.

At the time of retrieval, if the user knows the name of the system "XXX", but does not know what it is, the user can search for "XXX" in the information. Input to the section 40. The information search unit 40 searches the word semantic information database 10 using the input “OOOO” as a search key, reads out the next information, and displays it. If necessary, the document from which this semantic information data is based is also displayed: ○○○○ Semantic classification: Machine / System Evaluation: Fastest Date: 2000.4 ○○○○ Semantic classification: Machine / System Evaluation: Lightweight Date: 2000.5 When there are multiple results, they are displayed side by side in the order of the date when the evaluation was expressed, or in the order of oldest.

When the search key is set to “system” or “fastest”
, The information retrieval unit 40 retrieves and reads the following information: ○○○○ Semantic classification: Machine / System Evaluation: Fastest Date: 2000.4 ×××× Semantic classification: Machine / System Evaluation: Fastest Date: 1999.12 In this case, a plurality of search results are output. In this case, the data having the desired "system" and "fastest" of the user can be presented in an easy-to-understand manner by chronological order or the like. Even when there are many search results, information can be presented in the order of newest date and oldest date. Further, when the evaluation information about the same object that has already been registered is extracted, the evaluation information can be changed to data with a new date, and only the latest evaluation data can be used.

According to the eleventh embodiment, by holding the evaluation information and the date information, the display order of the search results can be selected so that the user can grasp the data more efficiently. The reliability of rating information can vary significantly from date to date,
The user can efficiently judge the reliability of the information regarding the evaluation.

Twelfth Embodiment The system configuration is as shown in FIG. The document analysis unit 30 of this embodiment extracts the evaluation information of the fifth embodiment.
That is, the feature is that the extracted evaluation information and source information are held and used for search.

Example 12: If the following sentence is present in the document, the sentence "The fastest OOOO has been developed" sentence "The lightweight OOOO has appeared" sentence "The fastest XXXXXX has been developed The document analysis unit 30 performs morphological analysis and stores the following information in the word semantic information database 10: ○○○○ semantic classification: machine / system Evaluation: fastest Source: Δnewspaper ×××× semantic classification : Machine / System Evaluation: Lightweight Source: Δ Newspaper ○○○○ Meaning classification: Machine / System Evaluation: Average
Source: × Newspaper.

At the time of search, if “○○○○” is used as a search key, the following is obtained: ○○○○ Meaning classification: Machine / System Evaluation: Fastest Source: Δ newspaper ○○○○ Meaning classification: Machine System evaluation: average
Source: × Newspaper.

When a plurality of data are obtained as a search result for the same object, the user can judge the reliability of the data by looking at the evaluation and the source of the information. In the above example, it is the source information, but if it is a title, the evaluated context can be understood, and the judgment can be made more accurately.

According to the twelfth embodiment, by displaying the source information and the title together with the evaluation information, the user can efficiently use the evaluation information.

Thirteenth Embodiment A system configuration is shown in FIG. The document analyzer 30 of this embodiment extracts the evaluation information of the sixth embodiment.
That is, the feature is that the extracted evaluation information and information on the evaluated subject are held and used for the search.

Example 13: If the following sentence is present in the document, the sentence "
We can expect a modest economic recovery, "he said. Statement “Professor XX criticized the economy in 2000, saying,“ I must say that the confusion is deepening. ” Statement) "About the economy 10 years from now, the XX Agency has announced that it will have a good economic outlook." The document analysis unit 30 extracts the following evaluation information by the evaluation information extraction of the sixth embodiment and stores it in the word semantic information database 10: Target Remark Evaluation 2000 Business in XX Prime Minister XX Despite the expectations, we can expect the economy in 2000. ○ Professor We must say that the confusion is deepening. The economy in 2010 ○○ Agency The outlook is for a good economy.

At the time of search, the search key is “business” or “2”.
Assuming that "the economy of 000 years", the information retrieval unit 40 reads out the following information from the word meaning information database 10 and displays it.

Target Remarks Evaluation Economy in 2000 XX Prime Minister Economy in 2000 is expected to recover, albeit slightly, in the future. Professor XX Inevitably, confusion will deepen.

As described above, evaluation information for the same object can be obtained, and the user can efficiently judge the reliability of the information based on the information of the subject who has performed the evaluation.

According to the thirteenth embodiment, the user can efficiently judge the reliability of the evaluation information by obtaining the information of the subject who has performed the evaluation together with the evaluation information.

Fourteenth Embodiment FIG. 12 shows a search system according to a fourteenth embodiment. This has the same basic configuration as the information search system shown in FIG. 9, but the fourteenth embodiment is characterized by collecting documents from a network (WWW) and extracting evaluation information. For example, suppose the following expression is present in a document: Example 14 sentence "The fastest XXXXX has been developed" sentence "Lightweight XXXXX has appeared" sentence "The fastest XXXXXX has been developed The following information is obtained by extracting the evaluation information: ○○○○ Semantic classification: Machine / System Evaluation: Fastest Source: Δ Newspaper Date: 1999.4 ×× XXX Semantic classification: Machine / System Evaluation: Fastest Source: ○ Newspaper Date: 2000.4 ○○○○ Meaning classification: Machine / System Evaluation: Lightweight Source: × Newspaper Date: 1999.4.

For example, if the search system user is "XXXXX"
If you want to know about this, specify "○○○○" as a search key, and you will get: ○○○○ Meaning: Machine / System Evaluation: Fastest Source: ΔNewspaper Date: 1999.4 ○○○○ Semantic classification: Machine / system Evaluation: Light weight Source: × Newspaper Date: 1999.4.

[0153] The user can obtain evaluation information on the desired "XXX" without referring to documents on various networks.
It can be obtained by searching only the word meaning information database 10p.

If "system" and "fastest" are used as search keys, the following results can be obtained as search results: ○○○○ Meaning classification: machine / system Evaluation: fastest Source: Δ newspaper Date: 1999.4 ×××× Semantic classification: Machine / System Evaluation: Fastest Source: ○ Newspaper Date: 2000.4.

Although there are a plurality of results, it is possible to easily determine the reliability of the evaluation information and the new regulations from information such as the date, source, and speaker. In addition, even when the user does not know the word "OOO" and wants to check, the present example can check the semantic classification.

[0156]

According to the present invention, even in the case of an unclear thing such as a case relation or a loose co-occurrence restriction in which a predicate can co-occur with various words,
Word extraction and its semantic classification can be obtained according to the nature of the word that modifies the word to be extracted.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of a document processing system common to first to sixth embodiments of the present invention.

FIG. 2 shows an example 3a of the contents of the evaluation expression dictionary 3 shown in FIG.
FIG.

FIG. 3 is a plan view showing an example of a pattern that is the content of the pattern dictionary 4 shown in FIG.

FIG. 4 is a flowchart showing a meaning estimating process according to the first embodiment of the meaning estimating unit 6 shown in FIG. 1;

FIG. 5A is a plan view showing the contents of a semantic classification extracted in association with a certain word on an input sentence in the first embodiment, and FIG. 5B is a plan view showing the contents of the input sentence in the second embodiment. A plan view showing the contents of the evaluation information extracted in association with a certain word, (c) is a plan view showing another example 3b of the contents of the evaluation expression dictionary 3 shown in FIG. 1, and (d) is a third embodiment. In the example, a plan view showing the contents of the evaluation type extracted in association with a certain word on the input sentence, (e) shows a date assigned in association with a certain word on the input sentence in the fourth embodiment. FIG. 17F is a plan view showing a title assigned in association with a certain word on an input sentence in the fifth embodiment.

FIG. 6 is a plan view showing another example 3c of the contents of the evaluation expression dictionary 3 shown in FIG.

FIGS. 7A and 7B are plan views showing another example of a pattern that is the content of the pattern dictionary 4 shown in FIG. 1;
(C) is a plan view showing evaluation information provided in association with a certain word on an input sentence in the sixth embodiment, and (d) is a combination of the sixth embodiment and the third embodiment. FIG. 9E is a plan view showing evaluation information assigned to a certain word above, and FIG. 10E shows an evaluation assigned to a certain word on an input sentence in the combination of the sixth embodiment and the fourth embodiment. It is a top view showing information and a date.

8 is a plan view showing another example 3d of the contents of the evaluation expression dictionary 3 shown in FIG.

FIG. 9 is a block diagram showing a configuration of an information search system common to the seventh to thirteenth embodiments.

FIG. 10 is a flowchart showing a flow of document analysis in a seventh embodiment by the document analysis unit 30 shown in FIG. 9;

FIG. 11 is a flowchart showing a flow of information search in an information search unit 40 shown in FIG. 9 in a seventh embodiment.

FIG. 12 is a block diagram illustrating a configuration of a network information search system for performing information search according to a fourteenth embodiment.

[Explanation of symbols]

 PC: PC

Claims (14)

[Claims] 1. An evaluation expression dictionary in which an expression representing an evaluation is associated with a semantic classification in which the expression representing an evaluation can co-occur, and A semantic information extraction method that assigns a semantic classification to a specific expression in text or a word in a database or dictionary using a described pattern dictionary. 2. An evaluation expression dictionary in which an expression representing an evaluation and a semantic class that can co-occur with an expression representing an evaluation are described in association with each other. By using the described pattern dictionary, an expression representing evaluation is given to a specific expression in text or a word in a database or dictionary.
Evaluation information extraction method. 3. An evaluation expression dictionary is a description in which an expression expressing an evaluation and an expression expressing an evaluation co-occur with a semantic classification and evaluation information. 3. The evaluation information extraction method according to claim 2, wherein an expression representing an evaluation and an evaluation value are assigned to a word in the dictionary. 4. The method according to claim 2, wherein a specific expression in the text or a word in the database or the dictionary is further associated with the date of creation of the document in which the evaluation expression appears. The described evaluation information extraction method. 5. A specific expression in a text, or a word in a database or a dictionary, further provided with source information or a title of a document to which the extracted word or expression belongs in association with the extracted word. The evaluation information extracting method according to claim 2, wherein the evaluation information is extracted. 6. A method according to claim 2, further comprising: adding a specific expression in a text or a word in a database or a dictionary to a subject who has evaluated or said the extracted evaluation expression. The evaluation information extracting method according to claim 4 or 5. 7. The method according to claim 1, wherein the semantic information of the input word is extracted and registered in the word semantic information database in a form corresponding to the word, and the word is associated with the input word. An information retrieval method for retrieving and reading a meaning classification from the word meaning information database. 8. An evaluation information extracting method according to claim 2, wherein an expression representing the evaluation of the input word is extracted and registered in a word meaning information database in a form corresponding to the word, and the evaluation of the input word is evaluated. An information search method for searching the word meaning information database for an expression representing "." 9. The information search method according to claim 8, wherein an expression representing an evaluation is also used as a search key for a document containing the word. 10. The method for extracting evaluation information according to claim 3,
An information search method for extracting evaluation information of an input word, registering the evaluation information in a form associated with the word in a word meaning information database, and searching for evaluation information of the input word from the word meaning information database. 11. The evaluation information extracting method according to claim 4,
An information search, in which a creation date of a document having an input word is registered in a word meaning information database in a form corresponding to the word, and a creation date and time of a document having the input word is searched from the word meaning information database. Method. 12. The evaluation information extracting method according to claim 5,
Registers the source information or title of the document with the input word in the word meaning information database in a form corresponding to the word, and searches the word meaning information database for the source information or title of the document with the input word. The information retrieval method. 13. The information on the subject who evaluated or said the input word in claim 6 in a form associated with the word in a word meaning information database, and the information on the subject who evaluated or said the input word. From the word meaning information database. 14. A method for extracting information according to claim 7 and registering it in a word semantic information database for words of a document collected through a network. A network information search method for searching for corresponding registration information from the word meaning information database.