JP2009042950A - Information retrieval system - Google Patents

Abstract

【Task】
In the prior art, it is necessary to intentionally prepare specific information in order to specify data to be acquired. In addition, a dedicated procedure for this is required.
[Solution]
When searching for information, master data, pattern data, and tabulation tables are used to reduce the load on both the inquirer and the respondent when extracting answers, to provide advice information to the inquirer, and to find just-fitted answers As provided.
In addition, even if it is impossible to answer, it also has a knowledge DB with a temporary area for storing useful questions.
[Selection] Figure 1

Description

  The present invention relates to the field related to information retrieval using so-called natural language. The present invention also relates to a technique for making search results more suitable. This includes a technique for optimizing search results and stocking useful information overflowed without hitting for the purpose of further aging and upgrading the system. In particular, the present invention relates to a maintenance support system for performing maintenance of equipment such as a factory using the retrieved information.

  The actual situation of on-site maintenance in electric power companies varies depending on self-knowledge, following manuals, and relying on experts. In either case, depending on the knowledge level of the personnel involved in maintenance, there is a large difference in the quality and quantity of information to be searched and the time required to find the information. Due to such factors, there are many cases in which maintenance work is not completed within the time limit. That is, there is a high need for information retrieval.

  Therefore, there is a need for advanced information retrieval techniques that are more convenient and easy to use. As a technique for this purpose, Patent Document 1 exists.

JP 2004-139507 A

  In Patent Document 1, it is necessary to “prepare specific information intentionally in order to specify data to be acquired”. In addition, a dedicated procedure for this is required.

  Often, complicated and complicated inquiries require manual complementation processing, generating extra load on both the inquiring side and the answering side.

  Furthermore, depending on the precision of the inquiry content, there may be a case where an enormous number of answer results including unnecessary information are obtained, and there is a case where a truly satisfying answer cannot be obtained for the inquirer.

  When searching for information, no special dedicated procedure is required, and a natural language input device is used.

  For this reason, in the present invention, it is determined whether or not the inquiry content, that is, the input information in the information search satisfies a predetermined condition (for example, presence or absence of stored value), and if it is determined, the information is stored. To do. Here, in the present invention, the processing as described above is included when it is determined that there is no answer corresponding to the inquiry content.

  Therefore, in the present invention, master data, pattern data, and tabulation are used to reduce the load on both the inquirer and the respondent when extracting answers, to provide advice information to the inquirer, and to search for just-fitted answers. A table may be provided as a means.

  Furthermore, even if it is impossible to answer, it also has a knowledge DB with a temporary area to accumulate useful questions.

  By incorporating the human interaction style using natural language of the present invention into the system, it is possible to promptly give an accurate answer to the question of the maintenance staff at the site. In addition to the personnel responsible for on-site maintenance, the knowledge of human resources who are indirectly involved in the field can be leveled. Furthermore, the newly accumulated information and knowledge will lead to an improvement in the technical capabilities of the entire company.

  The best mode of the present invention is shown in FIG. 1 configuration diagram of maintenance support system using knowledge DB.

  This system is connected to the analysis / answer server 200 via an inquiry terminal 600 and a network 100. The analysis / answer server 200, which is the core of this system, includes an input processing unit 201 that processes external inquiry data, an output processing unit 202 that transmits processing results to the outside, and a normalization that analyzes the content of the query Processing unit 203, whether or not there is a response to the query content, and a sorting processing unit 204 for judging the value of the query content, answer extraction, query / storage processing unit 205 for stocking the query content, answering the inquirer, Or, it comprises a text editing processing unit 206 for providing advice information for narrowing down.

  Then, the analysis / answer server 200 is queried with the corpus DB 300 that holds the master data 301 as information for analyzing the query contents and extracting the answer, and the pattern data 302 that is the configuration management of the equipment. In order to store the analysis result of the contents, the totaling table 500 is connected.

  Further, an attribute area 401 that holds an answer linked to the corpus DB 300 and a knowledge DB 400 that has a temporary area 402 for storing useful questions that cannot be answered are also connected.

  In order for the inquirer to maintain the knowledge DB 400 through a general terminal operation with the natural language input device 601, the text editing mechanism 602, and the display device 603, for dialogue with human language using natural language The operation management terminal 700 is provided.

  An example of maintenance work of an electric power company will be described. By using the system of the present invention, on-site maintenance personnel can inquire questions related to on-site maintenance in natural language, and can derive optimum answers from the knowledge DB at the same time as advice.

  If there is no answer, it is judged as a raw question of the maintenance staff at the site, and the question is stored and accumulated for the purpose of growing the question into a more mature knowledge DB.

Examples of the present invention are shown below.
FIG. 1 shows a configuration diagram of a maintenance support system using a knowledge DB. In FIG. 1, an analysis / answer server 200 forms the core of this system, and performs transaction processing from the inquiry terminal 600 via the network 100.

  First, the flow of transaction processing will be described with reference to FIG. 1 maintenance support system configuration diagram 2 using the knowledge DB.

  A transaction is normally generated as an inquiry content from the inquiry terminal 600, enters the internal processing unit 201 through the input processing unit 201 of the analysis / answer server 200 via the network 100, and in the normalization processing unit 203 Compare the content with the device configuration information of the corpus DB300 and the pattern of event synonyms, determine whether they match, and prepare to extract answers that correspond to the query content.

  From the result, the distribution processing unit 204 determines processing distribution for extracting an answer or stocking the contents of the inquiry.

  If there is an answer, the inquiry / accumulation processing unit 205 extracts the corresponding answer from the attribute area 401 of the knowledge DB 400.

  If there are multiple answers due to ambiguous inquiry content, the text editing processing unit 206 provides the inquiry information (maintenance staff) with advice information for narrowing down answers in order to extract accurate answers. Narrow down the answers by making the content of the inquiry concrete.

  The inquirer (maintenance staff) provides multiple answers to the inquirer (maintenance staff) in the form of a list in response to requests to refer to and confirm multiple answers without narrowing down the answers. A method for finding the optimum answer from the answers on the list.

  When there is no answer (when it is determined that there is no answer), the distribution processing unit 204 determines the value of this inquiry based on whether there is a contradiction in the configuration of the equipment, and there is no contradiction. In this case, it is determined that the query has value, and the query content is stocked in the temporary area 402 of the knowledge DB 400 in the query / accumulation processing unit 205. Then, inquiries that have no stocked answers are scrutinized by the operation management terminal 700, and a formal answer text is added. The above is the flow of a series of transactions.

  Next, the processing contents of the analysis / answer server 200 will be described using the general flow of the internal processing unit in FIG.

<S010>
After the inquirer (maintenance staff) sends the query content created in natural language to the analysis / answer server 200, the corpus DB300 is used to confirm that the query content is consistent with the equipment configuration. And extracting the answer extraction conditions and advice information for narrowing down the answers.

Next, details of this process <S010> will be described using the normalization process flow of FIG.
<< S01001 >>
In this step, the inquiry content is once read into the work area.
The inquiry should be “Tell me about the turbine frequency range” as shown in the screen transition (advice for narrowing down) in FIG.

<< S01002 >>
The process of this step is a preparatory process for speeding up the comparison between the device configuration information and the event synonym included in the work area and the pattern data in the corpus DB 300.

Details of specific processing are shown below.
The data in the work area, that is, “Tell me the range of the turbine frequency shown in << S01001 >>” and “Turbine” in the large classification column of the master data 301 in the corpus DB300, “ For “frequency” and “range” in the event synonym column, for example, SQL is used to extract codes corresponding to the respective ranges. Then, the code is stored in the aggregation table 500.

Here, means for speeding up the comparison with the pattern data in the corpus DB 300 will be described below.
(1) As master data 301, large / medium / small classification and event synonyms are coded and held.
(2) As the pattern data 302, all combination patterns generated from the master data are systematically managed and registered.
In addition, there should be an answer key for linking to the answer text in the knowledge DB400.
(3) The above-mentioned aggregation table 500 must be held.
By performing the above pre-preparation process, the subsequent search process can be speeded up, and a proper advice can be returned to the inquirer.

<< S01003 >>
In this step, information related to the answer to the inquiry from the inquirer (maintenance staff) is extracted.

The specific processing contents are as follows.
Of the major category, middle category, minor category, and event synonym column edited in << S01002 >>, the category for which the code is set, that is, “turbine” in the major category column, minor category Using the “frequency” in the field and the “range” in the event synonym field as keys, extract the corresponding record from the pattern data 302 using, for example, SQL, and the number of extracted records and the answer key in the record Column and answer flag column data are extracted. (In the case of the corpus DB record format in FIG. 4, “the number of records is 2, the answer keys are 1 and 2, and the answer presence flag is 1 and 1”.)
<< S01004 >>
In this step, in the contents of the inquiry from the inquirer (maintenance staff), it is determined whether or not the inquiry contents are inconsistent in the configuration of the equipment.

The specific processing contents are as follows.
If the number of records extracted in << S01003 >> is 0 (zero), it is determined that “the contents of this inquiry are inconsistent”. (If the record format of the corpus DB in Fig. 4 is taken as an example, "the number of records is 2 and there is no contradiction")
<< S01005 >>
In this step, the number of responses to the inquiry from the inquirer (maintenance staff) is confirmed.
The specific processing contents are as follows.
The answer presence / absence flag corresponding to the answer key extracted in << S01003 >> is checked. When the answer presence / absence flag is 1, it is determined that there is an answer, and the cumulative value of the numerical value in the answer presence / absence flag column is calculated as the number of answers. (In the case of the record format of the corpus DB in Fig. 4, "the number of responses is 2")
<< S01006 >>
The purpose of this step is to prompt the inquirer (maintenance staff) to ask a question in more detail! For this reason, the vocabulary as a key, that is, “high pressure” and “low pressure” shown in the screen transition (narrowing advice) in FIG. 7 is specified so that the inquirer can narrow down easily.

Details of specific processing are shown below.
The classification and event synonym whose code is not set (blank) among the major classification, middle classification, minor classification, and event synonym fields in the aggregation table 500 are extracted from the pattern data 302.
(In the case of the record format of the corpus DB in Fig. 4, it becomes “high pressure (1) in the middle category, low pressure (2) in the middle category”)
By executing this processing, appropriate advice to the inquirer can be automatically given.

<< S01007 >>
In this step, information is returned to the host process when there is a response to the inquiry from the inquirer (maintenance staff). The specific processing contents are as follows.
The classification for advice extracted in << S01006 >> and the number of answers and the answer key determined in << S01005 >> that the inquiry content determined in << S01004 >> is consistent. Return the code to the upper process <S020>. (For example, in the case of the record format of the corpus DB in FIG. 4, there is no inconsistency in the inquiry contents, the number of answers is 2, the answer keys are 1 and 2, and the advice classification code is 1 and 2 in the middle classification). )
<< S01008 >>
In this step, in the inquiry from the inquirer (maintenance staff), there is no contradiction in the inquiry contents, but when there is no answer, it is determined whether or not the inquiry contents are stored in the knowledge DB 400.

The specific processing contents are as follows.
If all the codes are set, for example, in the major classification, middle classification, minor classification, and event synonym fields in the tabulation table 500 (even if all codes are not set, the predetermined condition is satisfied) However, it is determined that the content of the query is not ambiguous, and it is determined that the inquiry can be stored in the temporary area 402 of the knowledge DB 400.

<< S01009 >>
In this step, when the inquiry from the inquirer (maintenance staff) is received, the information is returned to the upper process. That is, the contents of the work area read in << S01001 >>, the answer key extracted in << S01003 >>, and the number of answers 0 (zero) determined in << S01005 >> Return the inquiry contents determined in << S01008 >> to <S020> to the upper process.

<< S01010 >>
In this step, information when there is a contradiction in the inquiry from the inquirer (maintenance staff) is returned to the upper process. In other words, the number of records (number of responses 0 (zero)) extracted in << S01003 >> and the determination in << S01004 >> are inconsistent or the determination is based on << S01008 >> Returned to the upper process <S020>.

<S020>
In the presence / absence of an answer to the inquiry of the inquirer (maintenance staff), a branch is made to the subsequent processing corresponding to the presence / absence of the answer from the return result of the number of answers in <S010>.

<S030>
As for the number of answers to the inquiry of the inquirer (maintenance staff), a branch to the subsequent processing corresponding to the number of answers is performed from the return result of the number of answers of <S010>.

<S040>
In the answer to the inquiry of the inquirer (maintenance staff), the answer data corresponding to the inquiry is extracted from the return result of the answer key of <S010> from the attribute area 401 of the knowledge DB 400 using the answer key.

<S050>
The answer data corresponding to the inquiry extracted in <S040> is edited in a format that can be referred to by the inquirer (maintenance staff).

<S060>
The middle classification code extracted in <S010> is converted into a name using the middle classification of the master data 301.

Next, the name is edited in a format that can be referred to by the inquirer (maintenance staff), and a notification message is output. (In the case of Fig. 7 screen transition (advice for narrowing down), for example, “Middle category: [Please add one of the words [High pressure] [Low pressure].”)
<S070>
In the notification message of <S060>, the inquirer (maintenance staff) determines whether or not to refer to a plurality of answers, and branches to subsequent processing corresponding to the presence or absence of reference to the plurality of answers.

<S080>
From the return result of whether there is a contradiction to the query of <S010>, if there is a contradiction in the query content, this query has no value, and if there is no contradiction in the query content, this query has a value, And branch to the subsequent processing corresponding to the presence / absence of the value of the query.

<S090>
The processing of this step is a message to the inquirer (maintenance staff) that "There is no answer to the inquiry, but the inquiry content is valuable, so this inquiry content will be stored as knowledge data!" The purpose of this is to collect the questions on the site and the raw questions. To that end, the contents of the inquiry returned from <S010> are accumulated, that is, the notification of the message “I will keep this inquiry” shown in the screen transition (acquisition function) in FIG. .

Details of specific processing are shown below.
If there is no inconsistency in the query content determined in <S010>, if there is no inconsistency, it is determined that the query is a valuable query, and this query content is stored in the temporary area 402 of the knowledge DB 400. It is additionally registered in the inquiry column of the answer key extracted in S010>.

  By implementing this process, it is possible to accumulate live voices and questions from the field, and to develop the knowledge database into a more useful one.

<S100>
After the storage of the inquiry content by the storage processing of <S090> is completed, a completion message “Inquiry has been received” is notified to the inquirer (maintenance staff).

<S110>
If it is determined in the <S010> normalization process that there is a contradiction in the content of the inquiry, to the inquirer (maintenance staff), “There is a conflict in the content of the inquiry. Check the content of the inquiry. A message is sent to prompt the review of the inquiry content.

  As described above, the overview of the transaction processing by the processing of the present invention is the screen transition of FIG. 7 (advice for narrowing down), the screen transition of FIG. 8 (acquisition function), and the screen transition of FIG.

  In other words, by incorporating human language dialogues using natural language into the system, guidance is provided while providing quick and accurate answers to the questions of field maintenance personnel. In addition to the personnel responsible for on-site maintenance, we will try to standardize the knowledge of personnel who are indirectly involved in the field. Furthermore, by using the newly accumulated questions as knowledge, it is possible to improve the technical capabilities of the entire company.

  In the present embodiment, the input resource is a sentence, but there is no particular limitation, and voice input may be used.

  The present invention can naturally be applied not only to industries that manage and maintain equipment but also to other industries that operate under a similar management system.

An overall configuration diagram of the maintenance support system using the knowledge DB is shown. The general flow of the analysis / answer server is shown. The analysis and investigation of the contents of the query and the preparation process of the knowledge DB query are shown. Indicates the corpus DB record format. Indicates the record format of the knowledge DB. A summary table is shown. Shows the screen transition of the refinement advice. The screen transition of the custody function is shown. Shows the screen transition of inquiries with no value.

Explanation of symbols

100: Network
200: Analysis / answer server
201: Input processing section
202: Output processing section
203: Normalization processing part
204: Distribution processing section
205: Inquiry / accumulation processor
206: Text editing processor
300: Corpus DB
301: Master data
302: Pattern data
400: Knowledge DB
401: Attribute area
402: Temporary area
500: Total table
600: Inquiry terminal
601: Natural language input device
602: Text editing mechanism
603: Display device
700: Operation management terminal

Claims (4)

In an information retrieval system for retrieving desired information from a database based on inquiry information,
Means for receiving input of the inquiry information;
Means for retrieving response information corresponding to the inquiry information from the database;
Means for determining whether the content of the inquiry information satisfies a predetermined condition;
And a means for storing the inquiry information in a storage device when it is determined that the predetermined condition is satisfied. The information search system according to claim 1,
The information search system according to claim 1, wherein the determining means determines whether or not the predetermined condition is satisfied when the answer information is not searched by the searching means. The information search system according to claim 2,
The information retrieval system constitutes a maintenance support system for performing equipment maintenance,
The database has a corpus DB that performs hierarchical management of device configuration information indicating the configuration of the device, and a knowledge DB that holds replies,
The information search system, wherein the searching means executes a search using a normalization processing unit that analyzes the query information and the analyzed query information. In the information search system according to claim 3,
The means for determining uses the configuration information of the equipment and the known answer information to determine the value of the inquiry information,
The information search system according to claim 1, wherein the storage unit executes the storage when it is determined that the storage unit is valuable.

Images