JP2000276487A - Method and device for instance storage and retrieval, computer readable recording medium for recording instance storage program, and computer readable recording medium for recording instance retrieval program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To obtain instance storage and retrieval techniques which facilitate the solution of a problem by similar instance retrieval even in the case of a complicated instance consisting of plural instance sentences where an instance storage and retrieval object is described in a natural language. SOLUTION: A processing object sentence extraction means 45, a similar sentence clustering means 49, and a similar instance retrieval means 52 are provided. The means 45 segments individual instance sentences in each classification from an electronic document, and the means 49 classifies similar instance sentences on the basis of syntax structural and semantic similarities between instance sentences, which a similar sentence collation means 48 uses an area ontology (instance storage and retrieval object area-dependent knowledge) 44 to obtain, and generates similar instance data to store it in an instance data base, and the means 52 retrieves similar instance sentences resembling the retrieval sentence from the instance data base on the basis of similarities, which are obtained by the similar sentence collation means 48, between the input retrieval sentence and individual instance sentences.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a document storing a pair of a description of a problem that occurred in the past and a solution to the problem, for example, a "case" created based on an inquiry record at a call center. The present invention relates to a case accumulation / search technique for accumulating cases and searching for similar cases for use in a help desk system used for solving a problem.

[0002]

2. Description of the Related Art There is a strong demand for reusing data stored by electronic means in exchanges for inquiries at a consultation desk. In response to these demands,
A technique of registering a case expressed in a natural language in a case database in advance and searching for a case similar to a newly input search case from the case database is disclosed in, for example, Japanese Patent Application Laid-Open No. 9-73364. Similar Case Search Device "
Is disclosed. Hereinafter, the present technology will be referred to as Conventional Technology 1.

FIG. 26 is a configuration diagram showing a configuration of a similar case search apparatus according to the prior art 1. As shown in FIG. In FIG. 26, 1 is a case described in a natural language and composed of a problem part and a solution method part, 2 is a case database storing case 1 and 3 is generated from case 1, and a case number and a category number are generated. Attribute information 4 including attribute information, keyword numbers, and weights is an attribute database in which attribute information 3 is stored.

A keyword number table 5 stores each keyword extracted from the case 1 with a unique number attached thereto, and a keyword 6 corresponding to the case number and a keyword extracted from the cases having the case number. A keyword table 7 stores the case numbers in association with each other. The case table 7 stores the case numbers and the category numbers corresponding to the categories representing the classification of the cases having the case numbers (category types are set in advance). It is a category table.

[0005] Reference numeral 8 denotes an example of a case in which each column of a table having a category number in the row direction and a keyword number in the column direction includes a keyword corresponding to the keyword number included in the category corresponding to the category number. Number of cases table storing numbers, 9 is keyword number table 5
, A keyword table 6, a category table 7,
A memory storing the case number table 8 and 10 is attribute information generating means for generating attribute information 3 of case 1, and 11 is a similarity generating means for generating similarity between cases stored in the case database 2.

Next, the operation of the similar case search apparatus according to the prior art 1 will be described with reference to FIG. First, the operation at the time of case registration will be described.

As shown in FIG. 26, the number-of-cases table 8 stores the number of cases including the category corresponding to the category number and the keyword corresponding to the keyword number in each column of the table. The sum of the number of cases for each keyword stored in each column in each row of the table 8 (total number of cases including the keyword corresponding to the keyword number), the number of cases for each category stored in each column in each column (The total number of cases included in the category corresponding to the category number) and the total number of cases stored in the case database 2.

In the case number table 8, in the initial state, only the category number corresponding to the category type set in advance exists, but each time the case 1 is accumulated (stored), it is extracted from the case 1. The keyword number corresponding to the keyword is added in the row direction, the number of cases in the column corresponding to the category number and the keyword number, the sum of the number of cases in each category corresponding to the category, The sum of the number of cases and the total number of cases of all the cases are updated.

At the time of case registration, the attribute information generating means 10 executes the following processing. That is, first, a keyword is extracted from the problem part of the given case, a keyword number is assigned, and the keyword is stored in the keyword number table 5. Next, the keyword numbers stored in the keyword number table 5 are extracted for the extracted keywords, and stored in the keyword table 6 in association with the case numbers assigned to the cases.

Next, according to a user's instruction, the category number of the category assigned to the case is stored in the category table 7 in association with the case number of the case. next,
A keyword number is added to the case number table 8 as needed, and the number of cases in the corresponding column in the case table 8
The data of the sum of the number of cases for each keyword, the sum of the number of cases of each category, and the total number of cases of all the cases are updated.

Next, based on the case number table 8, weights are
Generate. The category stored in the case number table 8
Number n, number m of keywords, total number s of all cases, category
Ri c _iSum s of cases belonging to (1 ≦ i ≦ n)_i,keyword
Sum of cases where the keyword of number j (1 ≦ j ≦ m) appears
t_j, Category c_iBelongs to and has the keyword number m.
-Number of cases where the word appears t_ijAnd generate t_ij≠ 0
Category c for the keyword with keyword number j
_iWeight for keyword with keyword number j in
ω_ijIs ω_ij= (T_ij/ T_j−si / s) + (t_ij/ S_i-T_j/
s) is generated. Next, a case number for the case,
Attribute consisting of category number and weight for each keyword
Information is created and stored in the attribute database 4.

Next, the operation at the time of case search will be described. At the time of case search, the similarity generation means 11 executes the following processing. That is, first, a keyword is extracted from the “problem” portion of the newly input search case. Next, the attribute database 4 is searched to extract attribute information including the category number of the case where the same keyword as the extracted keyword exists and the weight of each keyword.

Next, a similarity is generated based on the extracted attribute information. The similarity between the given case and the case corresponding to the extracted attribute information is obtained by calculating the similarity ω = (sum of the weights of the matched keywords) × ( The number of matching keywords / the length of the keyword list extracted from the attribute database 4) is calculated and generated. Next, sorting is performed in descending order of the generated similarity ω, and the case category, similarity, data content, and the like are output from the cases with the highest similarity ω and displayed or printed.

A technology for providing a case base for retaining past failure contents, failure response contents, failure attribute data, and the like in a work of claim processing of a device or the like is disclosed in, for example, Japanese Patent Application Laid-Open No. 6-309172. Base Apparatus and Knowledge Base Construction Method ". Hereinafter, the present technology will be referred to as Conventional Technology 2.

FIG. 27 is a configuration diagram showing a configuration of a knowledge base device according to the conventional technology 2. As shown in FIG.

In FIG. 27, reference numeral 12 denotes inference, similarity search,
Instruction and input means for the user to input data such as inference conditions and search conditions while the user instructs case editing and knowledge editing, and display means for presenting data such as search results and menus to the user. Provided user interface, 13 is a knowledge base storing knowledge composed of a condition part, a conclusion part, and a case number, 14 is a synonym dictionary storing a general thesaurus, and 15 is a management item (such as a customer name). , A failure content, a keyword group, a failure cause, and failure handling content, case data 16, a case base storing case data 15, 17 a keyword, and the number of occurrences of the keyword in the case data. The network structure is composed of the simultaneous appearance relationship between keywords in one case data and the number of simultaneous appearances of keywords in one case data. Is a keyword relationship database that Tsu.

Reference numeral 18 denotes a case database 1 which is exchanged with a user via the user interface 12.
A case editing means for editing the case data 15 stored in the user interface 12; an inference means 19 for receiving an inference condition from the user interface 12 to make an inference using the knowledge stored in the knowledge base 13; A similarity search means for receiving a search condition from a general thesaurus using a general thesaurus stored in a synonym dictionary 14 and performing a similarity search of the case data 15 stored in the case base 16; A keyword relation data creating unit that acquires keyword data from the case data 15 stored in the base 16 and creates relation data between keywords using the acquired keyword data.

Reference numeral 22 denotes menu creation means for creating a menu from the keyword-related data stored in the keyword-related database 17, and 23 denotes keyword-related data stored in the keyword-related database 17 and case data stored in the case database 16. A case / knowledge connect means for generating knowledge 24 from 15; 25 edits the knowledge 24 generated by the case / knowledge connect means 23 according to a knowledge editing instruction from the user interface 12;
This is a knowledge editing means stored in the knowledge base 13.

Next, the operation of the knowledge base device according to the prior art 2 will be described with reference to FIG. First, the operation at the time of case registration will be described.

At the time of case registration, first, the case editing means 18
Thus, the customer name (one of the management items), the failure content, and the keyword group are input, and the case data 15 is created and stored in the case base 16. The keyword group is input either directly or by specifying a menu created by the menu creating means 22. Also, the keyword, the number of appearances of the keyword, and the number of simultaneous appearances between keywords are stored in the keyword relation database 21 by the keyword relation data generation means 21.

Next, the case / knowledge connecting means 23 generates a combination of keywords having a high simultaneous appearance frequency stored in the keyword relation database 21 as a condition part of the knowledge 24 as a candidate. From the case base 16 with case data having
A knowledge 24 having a failure cause, a failure response content, and a case number of the retrieved case data as a conclusion part is generated, and the generated knowledge 24 is edited by a knowledge editing unit 25 by a user and stored in the knowledge base 13. Keep it.

Next, the operation at the time of case retrieval will be described. At the time of case search, first, the inference means 19 receives search condition data from the user via the user interface 12 and based on the received search condition data, the knowledge base 1.
3 is searched for knowledge. When the received search condition data is in the condition part of the knowledge 24, a conclusion part of the knowledge is presented to the user via the user interface 12. The received search condition data is "condition part" in knowledge 24
Is not found in the knowledge base 13, the conditions of the search condition data are expanded by the similarity search means 20 using the synonym dictionary 14 storing the general thesaurus, and the case stored in the case base 30 is obtained. A similarity search is performed on the data 15.

Further, a technique aimed at facilitating the determination of the validity of the search result in the document search is described in, for example,
It is disclosed in Japanese Patent Application Laid-Open No. 7-192020 "Document Information Searching Apparatus". Hereinafter, the present technology will be referred to as Conventional Technology 3.

In the document information search apparatus according to the prior art 3, the input sentence is morphologically analyzed and syntax-analyzed to generate a search command including a list of keywords and "viewpoints", and the document data is searched by the search command. By doing so, the degree of similarity between the search command and the document data is displayed with a plurality of viewpoints as axes, thereby making it easy to determine the validity of the search result. The prior art 3 performs syntax analysis of a natural language, but uses the analysis result only for viewpoint extraction, and uses only information composed of keywords and viewpoints during an actual search.

FIG. 28 is a configuration diagram showing a configuration of a document information retrieval apparatus according to the conventional technique 3. In FIG. 28, 26
Is an input unit for inputting a keyword or a sentence composed of a natural language instructing the user to search, and 28 is for inputting a search command 29 obtained by analyzing the input sentence 27 input by the user and converting it into a search command. This is an input analysis unit that outputs.

Reference numeral 30 denotes a sentence data recording unit for storing sentence data, and 31 denotes a search for a related document by inputting a search command 29 from the input unit 26 and referring to the sentence data stored in the document data recording unit 30. A search unit 33 that outputs the search result 32 obtained by the search; a search history storage unit 33 that receives the search command 29 from the input unit 26 and receives the search result 32 output by the search unit 31 and stores it as a search history; This is a search history display unit that inputs the search history stored in the search history storage unit 33 and displays it in a tree structure.

A search unit 36 inputs a search result 32 obtained by a search performed by the search unit 31 and displays a set of documents of the search result in a multidimensional manner (for example, a title or a bibliographic item of each document is displayed in a tree). A search result display unit for selecting a document from a set of documents displayed together with a structure, a table, a list, etc.), and a browse 37 displays the contents (full text) of the document selected on the search result display unit 36 Reference numeral 34 denotes a history management unit that controls the search history storage unit 33, the search history display unit 35, and the search result display unit 36 to control and manage the storage and display of the search history.

FIG. 29 shows the input analysis unit 28 in FIG.
FIG. 2 is a configuration diagram showing the configuration of FIG. In FIG. 29, reference numeral 38 denotes a morphological analysis unit which inputs individual sentences, performs morphological analysis, and outputs the analysis result, and 39 denotes a syntax analysis which inputs the result of the morphological analysis, performs syntax analysis, and outputs the result. Part 4,
Reference numeral 0 denotes an input analysis control unit that inputs the input sentence 27, activates the morphological analysis 38 and the syntax analysis 39, and obtains a syntax analysis result.

Reference numeral 41 denotes a viewpoint extraction rule describing a rule used for extracting a viewpoint. Reference numeral 42 denotes a viewpoint extraction unit for inputting the result of the syntax analysis and referring to the viewpoint extraction rule to create viewpoint information. Reference numeral 43 denotes a search command generation unit which inputs the viewpoint information extracted by the viewpoint extraction unit 42 and extracts a self-sustaining word from each sentence to form a search command. In addition,
29, those parts which are the same as or correspond to those in FIG. 28 are given the same reference numerals, and descriptions thereof will be omitted.

Next, the operation of the document information retrieval apparatus according to the prior art 3 will be described with reference to FIGS.
In the input analysis unit 28, the morphological analysis unit 38 and the syntax analysis unit 39 are controlled by the input analysis control unit 40 shown in FIG.
Is activated, and the result of parsing the input sentence 27 is obtained.
Next, the viewpoint extracting unit 42 (predicate, modality, case structure)
A viewpoint list is generated using a viewpoint extraction rule 41 that assigns a pattern to a viewpoint, and a search command generation unit 43 extracts an independent word from the result of the syntax analysis, thereby obtaining a list from the <keyword> and the <view list>. Is generated.

The retrieval unit 31 shown in FIG. 28 retrieves the document data stored in the document data storage unit 30 using the retrieval command 29, and obtains a similar document for each viewpoint as the retrieval result 32. The search result display section 36
, A search result (a set of documents) for each viewpoint can be displayed. The history management unit 34, the search history storage unit 33, and the search history display unit 35 manage and store the history of past search commands and search results, and store the history of the stored search commands and search results. Can be displayed. Further, by selecting a document from a set of documents that are search results displayed on the search result display unit 36, the browse unit 37 is selected.
The full text of the selected document can be displayed.

[0032]

In the prior arts 1 and 2, a simple similarity search in which a description in a natural language in a case is a simple sentence is assumed. In a case where a part is composed of a plurality of sentences, there is a problem that it is not possible to extract and classify sentences from a case in advance and to easily solve a problem by similarity search.

In the prior art 1, an index is generated only by keyword extraction, and similarity calculation is performed only by keyword category classification and frequency information. Therefore, modal expressions (negation, guesswork) of a sentence constituting a case are obtained. Etc.) cannot be determined in detail.

Further, in the prior art 2, a synonym dictionary storing a general thesaurus is used. However, an index is generated only by a combination of keywords having a high frequency of appearance, and an input for similarity search is also extracted from the input contents. Because only keywords that have been used are used, detailed processing related to the similarity of natural languages, including the modal expressions of sentences, is not considered, and cases that include sentences that are syntactically and semantically similar to the input sentence are searched. There was a problem that it was not possible.

Further, in the prior art 3, the syntax analysis of the search input sentence is performed, but the syntax analysis result information is used only for extracting the viewpoint, and only the information composed of the keyword and the viewpoint is used at the time of the actual search. ing. Therefore, similarly to the related art 1 and the related art 2, there is a problem that a document including a sentence that is syntactically and semantically similar to the input sentence cannot be searched.

Further, in the prior art 3, since there is no means for pre-accumulating the case including the relationship between the document data as in the present invention, it is not possible to perform a search conscious of the relationship between the data. was there.

The present invention has been made in order to solve the above-described problems, and it is easy to solve a problem by similarity search even for a complex case where a storage / search target is composed of a plurality of case sentences. Case storage / retrieval apparatus, case storage method, case search method, computer readable recording medium recording case storage program, and computer readable recording medium recording case search program The purpose is to:

Further, the present invention considers detailed similarity corresponding to various natural language expressions including modal expressions (negation, guesswork, etc.) of sentences constituting a case, and considers syntactic and semantic similarity. The purpose of the present invention is to make it possible to search for cases that include a sentence.

Another object of the present invention is to make it possible to perform a search in consideration of the relationship between document data such as similarity or contradiction.

[0040]

A first case accumulation / retrieval apparatus according to the present invention is an area accumulation / retrieval apparatus which prestores knowledge relating to terms dependent on an area to be accumulated / retrieved and relations between the terms in advance. An ontology, a processing target sentence extracting means for extracting each case sentence to be stored and searched for from the digitized document described in a natural language, and a search for searching for a case sentence desired by the user A morpheme by referring to knowledge stored in the area ontology with a search input unit for inputting a sentence, and each case sentence extracted by the processing target extraction unit or the search sentence input by the search input unit as an input. Sentence analysis means for performing analysis and syntax analysis to create the structure of each of the case sentences or the search sentence, and each of the case sentences or the search sentence created by the sentence analysis means Similar sentence matching means for obtaining a similarity between each of the case sentences or a similarity between each of the case sentences and the search sentence by referring to knowledge stored in the area ontology with a structure as an input; Based on the similarity between the case sentences obtained by the similar sentence matching means, the respective case sentences are classified to form a case cluster, and information on the case clusters and relation information between the case clusters are used. Similar sentence clustering means for creating constituted case data; a case database for storing case data created by the similar sentence clustering means; each of the case sentences and the search sentence obtained by the similar sentence matching means A similar case sentence similar to the search sentence input by the search sentence input means is stored in the case database based on the similarity between Wherein the similar case retrieving means for retrieving from the case data, that a search result display means for displaying the retrieved similar cases statements by the analogous case retrieval means.

A first case storing method according to the present invention includes a processing target sentence extracting step of cutting out each case sentence to be stored in a digitized document described in a natural language.
A morphological analysis is performed by referring to an area ontology in which each case sentence extracted by the processing target extraction step is input and knowledge about terms dependent on the area to be stored in the case and the relation between the terms is stored in advance. And a sentence analysis step of performing a syntax analysis to create the structure of each of the case sentences, and referring to the knowledge stored in the area ontology with the structure of each of the case sentences created by the sentence analysis step as an input. A similarity sentence matching step of calculating the similarity between the case sentences, and classifying each of the case sentences based on the similarity degree between the case sentence obtained by the similar sentence matching step. And creates case data composed of the case cluster information and the relationship information between the case clusters, and stores the case data in the case database. Characterized in that it consists of a similar sentence clustering step of.

A second case accumulation method according to the present invention is characterized in that, in the processing object sentence extracting step, a case sentence type is assigned to the extracted case sentence.

In the third case accumulating method according to the present invention, in the similar sentence clustering step, the number of hierarchical levels of clusters configured by classifying the respective case sentences and / or the case sentences are similar to each other. It is characterized in that a threshold value of the similarity between clusters to be used when it is determined that there is a presence is designated.

In the fourth case accumulation method according to the present invention, in the area ontology, IS-A relation knowledge describing a semantic upper-lower relation and HAS-A describing a semantic part-whole relation are described. At least one of relational knowledge, case relation knowledge that describes the relationship between concepts, paraphrase knowledge that describes one meaning, concept, and relation in multiple expressions, and knowledge that describes a conflict that cannot occur at the same time It is characterized by describing knowledge.

In a fifth case accumulating method according to the present invention, in the similar sentence matching step, each of the case sentences is connected to each other based on an attribute of a syntactic element in a sentence structure created in the sentence analyzing step. It is characterized in that a similarity is obtained by collating a semantic structure.

A sixth case accumulation method according to the present invention is characterized in that, in the similar sentence matching step, a degree of detail of matching used when obtaining the similarity between the case sentences is designated.

In a seventh case accumulating method according to the present invention, in the sentence analyzing step, the structure of the case sentence is created as a tree structure, and in the similar sentence matching step, the similarity between the case sentences is determined. It is characterized in that the degree of detail of the collation used for obtaining is specified by the depth of the tree structure of the case sentence.

An eighth case accumulation method according to the present invention is characterized in that the case database describes a similarity relationship or a contradiction relationship as a relationship between case clusters.

In the first case search method according to the present invention, a search input step of inputting a search sentence for searching a case sentence desired by a user from a case database storing case data to be searched. With reference to the knowledge stored in the area ontology in which knowledge relating to the term depending on the area to be searched for the case and the relationship between the terms is previously stored with the search sentence input in the search input step as an input. A sentence analysis step of performing the morphological analysis and syntax analysis to create the structure of the search sentence, and referring to the knowledge stored in the area ontology with the structure of the search sentence created by the sentence analysis step as input. A similar sentence matching step for obtaining a similarity between each of the case sentences and the search sentence; A similarity search for searching a similar case sentence similar to the search sentence input in the search sentence input step from the case data stored in the case database based on the similarity between each case sentence and the search sentence. The method is characterized by comprising a case search step and a search result display step of displaying a similar case sentence searched by the similar case search step.

The computer-readable storage medium storing the first case storage program according to the present invention cuts out each case sentence to be stored in a digitized document described in a natural language. An area in which knowledge relating to a processing target sentence extraction procedure, a term depending on an area to be subjected to the case accumulation as an input of each case sentence extracted by the processing target extraction procedure, and a relationship between the terms are stored in advance. A sentence analysis procedure for creating a structure of each of the case sentences by performing morphological analysis and syntax analysis with reference to an ontology; and inputting the structure of each of the case sentences created by the sentence analysis procedure to the area ontology as input. A similar sentence matching procedure for obtaining a similarity between the case sentences with reference to the stored knowledge; and a similar sentence matching procedure obtained by the similar sentence matching procedure. Classifying each of the case sentences based on the similarity of the case clusters to form a case cluster, creating case data including information on the case clusters and information on the relationship between the case clusters, and storing the created case data in a case database A case accumulation program comprising a similar sentence clustering procedure is recorded.

The computer-readable storage medium storing the first case search program according to the present invention searches for a case sentence desired by a user from a case database storing case data to be searched. A search input procedure for inputting a search sentence for inputting, and a search sentence input by the search input procedure as input, and knowledge about a term depending on an area to be searched for the case and a relationship between the terms are stored in advance. A sentence analysis procedure for creating a structure of the search sentence by performing morphological analysis and syntax analysis with reference to an area ontology, and the structure of the search sentence created by the sentence analysis procedure is stored in the area ontology as an input. A similar sentence matching procedure for obtaining a similarity between each of the case sentences and the search sentence with reference to knowledge; A similar case sentence similar to the search sentence input by the search sentence input procedure based on the similarity between each of the obtained case sentences and the search sentence is selected from among the case data stored in the case database. A case search program comprising a similar case search procedure for searching and a search result display procedure for displaying a similar case sentence searched by the similar case search procedure is recorded.

[0052]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Embodiment 1 Hereinafter, the case accumulation / retrieval technology according to the present invention will be described with reference to the drawings.

FIG. 1 is a configuration diagram showing the configuration of the case storage / retrieval apparatus according to Embodiment 1 of the present invention.

In FIG. 1, reference numeral 44 denotes an area ontology in which terms dependent on a target area of a case to be stored and searched by the case storage / retrieval apparatus and knowledge about a relationship between the terms are previously described and stored. , 45, each sentence to be subjected to the case accumulation processing for each field (hereinafter referred to as “case sentence”) from an electronic document file (not shown) in which a document in which a case is delimited for each field and described in a natural language is stored. ) Is extracted.

Reference numeral 46 denotes a search sentence input unit for inputting a search sentence (hereinafter referred to as “search sentence”) for case search processing, and reference numeral 47 denotes each case sentence extracted by the processing target sentence extraction unit 45. Or a search sentence input in the search sentence input unit 46, a word dictionary, grammar rules (not shown),
The sentence analyzer 48 performs morphological analysis and syntax analysis with reference to the area ontology 44 and the like. The sentence analyzer 48 receives two sentence structures as input, and refers to the area ontology 44, rules for calculating similarity (not shown), and the like. This is a similar sentence matching unit that obtains a similarity between the two.

Reference numeral 49 denotes a sentence analyzing unit 47 which receives a set of case sentence structures obtained by morphological analysis and syntax analysis of each case sentence and inputs any two of the input set of case sentence structures. Similar sentence matching unit 48 for the structure of two case sentences
Based on the degree of similarity between the case sentences obtained by the above, sentences similar to each other are classified (clustered) to form a case cluster for each field, and information on the formed case clusters and relation information between the case clusters are created. A similar sentence clustering unit 50 for storing case cluster information and related information between case clusters created for each field by the similar sentence clustering unit 49 with an index;
Reference numeral 51 denotes a case cluster editing unit that edits a hierarchical structure of clusters of cases as a problem solving tree based on information on clusters stored in the case database 50 and relation information between clusters.

Reference numeral 52 denotes a sentence analysis unit 47 for inputting the structure of a search sentence obtained by morphological analysis and syntax analysis of the search sentence, and the structure of each case sentence stored in the case database 50 and the input search sentence. A similar case search unit that searches for a case sentence similar to the search sentence based on the similarity between the case sentence and the search sentence obtained by the similar sentence matching unit 48 for the sentence structure;
Reference numeral 3 denotes a search result display unit that displays the search results in the similar case search unit 52 and the contents of the case cluster stored in the case database 50 for confirmation.

Next, the operation of the case accumulating and searching device according to the first embodiment of the present invention will be described.

The functions of the case storage / retrieval apparatus according to the first embodiment of the present invention are roughly classified into a case storage function of extracting a case from an electronic document and storing it in a case database and a case storage function of storing the case in a case database. There are two functions of a case search function for searching for similar cases from the performed cases.

First, the outline of the operation of the case accumulation function in the case accumulation / retrieval apparatus according to Embodiment 1 of the present invention will be described with reference to FIG.

FIG. 2 is a flowchart showing the flow of processing of the case accumulation function in the case accumulation / retrieval apparatus according to the first embodiment of the present invention. In FIG. 2, the same or corresponding parts as those in FIG.

In FIG. 2, the processing target sentence extraction process (step S1) is executed by the processing target sentence extraction unit 45 in the configuration of the case accumulation / retrieval apparatus shown in FIG. The sentence analysis process (step S2) is executed by the sentence analysis unit 47 in the configuration of the case accumulation / search device shown in FIG. Further, the similar sentence clustering process (step S3) is performed as shown in FIG.
Is executed in the similar sentence clustering unit 49 in the configuration of the case accumulation / search device shown in FIG. The case cluster editing process (step S4) is executed by the case cluster editing unit 51 in the configuration of the case accumulation / search device shown in FIG.

First, the target sentence extraction process (step S
In 1), an electronic document file is input, and each case sentence to be subjected to case accumulation / search processing is cut out. next,
In the sentence analysis process (step S2), each case sentence extracted by the process target sentence extraction process (step S1) is input, and morphological analysis and syntax analysis are performed, and the structure of each case sentence (specifically described later) (Dependency structure). Next, similar sentence clustering processing (step S3)
In step S2, the similar sentence matching unit 47 is called for the structure of each case sentence generated by the sentence analysis process (step S2) to obtain the degree of similarity between arbitrary case sentences, and the process target sentence extraction process (step S1) Based on the obtained similarity, each case sentence is classified (clustered) to create cluster information and inter-cluster relation information for each field, and the created information is stored in the case database 50. Next, the case cluster editing process (step S
In 4), information of clusters stored in the case database 50 and relation information between clusters are input, and the hierarchical structure of the case clusters is edited as a problem solving tree.

In FIG. 2, the similar clustering process (step S3) is configured to call the similar sentence matching unit 47 in the case accumulation / search apparatus shown in FIG. It may be configured as a module (a set of steps) called from the similar clustering process (step S3) and realized by a program.

Next, the outline of the operation of the case accumulation function, which is another function of the case accumulation / retrieval apparatus according to the first embodiment of the present invention, will be described with reference to FIG.

FIG. 24 is a flowchart showing a flow of processing of the case search function in the case storage / search apparatus according to the first embodiment of the present invention. In FIG. 24, FIG.
The same or corresponding parts are denoted by the same reference numerals and description thereof is omitted.

In FIG. 24, the search sentence input processing (step S41) is executed in the search sentence input unit 46 in the configuration of the case accumulation / search apparatus shown in FIG. The similar case search process (step S42) is executed in the similar case search unit 52 in the configuration of the case accumulation / search device shown in FIG. The search result display processing (step S43) is executed in the search result display unit 53 in the configuration of the case accumulation / search apparatus shown in FIG.

First, search sentence input processing (step S41)
In, a search sentence for searching for a case stored in the case database 50 is input. Next, in the sentence analysis process (step S2), a search sentence input process (step S4)
A morphological analysis and a syntax analysis are performed on the search sentence input in 1) to generate a structure of the search sentence (specifically, a dependency structure described later). Next, in a similar case search process (step S42), a sentence analysis process (step S2)
Sentence structure and case database 5 generated by
The similar sentence matching unit 47 is called for the structure of each case sentence stored in 0 and the similarity between the search sentence and each case sentence is obtained. Based on the obtained similarity, the search sentence input processing ( A similar case sentence similar to the search sentence input in step S41) is searched from the case database 50. Next, in a search result display process (step S43), the similar case sentence obtained by the similar case search process (step S42) and the contents of the case cluster stored in the case database 50 are displayed.

In FIG. 24, the similar case search process (step S42) is configured to call the similar sentence matching unit 47 in the case storage / search device shown in FIG. May be configured as a module (set of steps) called from the similar case search process (step S42), and may be implemented by a program.

Hereinafter, the operation of the case accumulation function in the case accumulation / retrieval device according to the first embodiment of the present invention will be described in detail with reference to the flowchart shown in FIG. .

First, the sentence to be processed is extracted (step S
1) will be described. In step S1, a digitized document file such as an inquiry record to be subjected to case accumulation processing is input, and a case sentence to be processed is extracted from the input digitized document file.

FIG. 3 is a diagram showing an example of the structure of a digitized document file. In FIG. 3, reference numeral 54 denotes one digitized document file to be subjected to the case accumulation processing; 55 and 56 each exist in the digitized document file 54; a document having a plurality of fields; A fixed field (as an attribute) whose contents are represented by a number or a simple character string among the fields to be described. 58 is described by a plurality of sentences expressed in a natural language among the fields existing in the document 55. An atypical field (as an attribute), 59 is a document start tag indicating the beginning delimiter of the document 55, and 60 is a document 5
5 is an end-of-document tag indicating the end of the end of the document.

Reference numeral 61 denotes a document field start tag indicating a head delimiter of each field (hereinafter, referred to as “document field”) in the document 55 irrespective of the fixed field and the non-standard field. This is a document field end tag indicating the end delimiter of the document field in the document 55.

The digitized document file 5 shown in FIG.
In FIG. 4, reference numerals 61 and 62 specifically indicate a document field start tag for the document field “customer name” and a document field end tag for the document field “customer name”.

As shown in FIG. 3, a plurality of documents may be stored in one digitized document file 54. In the example of the digitized document file shown in FIG. 3, the document start tag 59 “<DOC>” and the document end tag 60 </ DO
C> represents a document break.

One document is composed of a plurality of document fields. For example, as shown in FIG. 3, in the document 55, the description content is a fixed form such as a document field such as "customer name", "customer telephone", "model", and "subject". Field 57, which can be represented by a simple character string, or a document field such as "question" or "answer", the description content is atypical and is described by a plurality of sentences expressed in a natural language. And an atypical field 58.

In the example of the digitized document file shown in FIG. 3, one document field has a document field start tag 61 surrounded by symbols “<” and “>”, and a symbol “<” and a symbol “<”. A character string including “>”, and a character string obtained by removing “<” and the symbol “>” from the document field start tag 61 is a field name. On the other hand, the sentence field end tag 62 is a character string that is surrounded by the symbol “<//” and the symbol “>” and includes the symbol “<//” and the symbol “>”. The character string excluding </ and> is the field name.

FIG. 4 is a flowchart showing the flow of the processing target sentence extraction process (step S1) in the flowchart shown in FIG. The processing target sentence extraction process (step S1) when the digitized document file 54 as shown in FIG. 3 is input will be described with reference to the flowchart shown in FIG.

First, until the document start tag 59 "<DOC>" is found (step S6), the digitized document file 54
Are sequentially read from the beginning (step S5).
If the document start tag 59 is not found by the time the end of the digitized document file 54 is detected (denoted by “EOF” in the flowchart shown in FIG. 4), the process ends. If the document start tag 59 is found, the process proceeds to step S7, where the document end tag 60 “</ D
Until "OC>" is found (step S8), the character strings of the digitized document file 54 are sequentially read (step S7). If the document end tag 60 is not found before the end of the digitized document file 54 is detected, the process ends.
If the document end tag 60 is found, the process proceeds to step S9.

In step S9, the character string read in step S7 (here, the document start tag 59 and the document end tag 60 are read and are not included in the read character string) are empty. If not, the read character string is determined to be one document 55, cut out, and a unique case number is assigned to the document 55 in order to process the cut out document 55 as a case (step S9). .
Next, with respect to the document 55 cut out in step S9, the processing flow is as follows.
And the above processing is repeated.

In the following three cases (1) to (3), (a) a warning is issued to the user and the processing is continued.
(B) A warning is issued to the user, the process is temporarily interrupted, the process waits for the user's instruction, the process is performed according to the user's instruction, and then the process is continued. (C) A warning is issued to the user and the process ends. Take one of the following actions. (1) When there is data to be read in step S6 except for the document start tag 59 and the document end tag 60. (2) In step S8, the digitized document file 54
Is detected, and there is an unprocessed read character string in steps S9 and S10. (3) In step S9, the read character string is empty.

FIG. 5 is a flow chart showing a process-target sentence extracting process for one document in the flowchart shown in FIG.
It is a flowchart which shows the flow of 0). Next, the processing target sentence extraction processing (step S10) for one document will be described using the flowchart shown in FIG.

In the processing target sentence extraction processing for one document (step S10), a character string for one document in the digitized document file 54 is received as an input. First, a character string surrounded by a symbol "<" and a symbol ">" is input until a character string next to the symbol "<" is not the symbol "/", that is, a document field start tag 61 is found (step S12). The character strings for one document are sequentially read from the beginning (step S11). The end of the character string for one input document is detected (“EOD” in the flowchart shown in FIG. 5).
If the document field start tag 61 is not found before the process is completed, the process is terminated.

If the document field start tag 61 is found, the process proceeds to step S13, and the character string surrounded by the symbol "<//" and the symbol ">", that is, until the document field end tag 62 is found (step S14). ) The input character strings for one document are sequentially read (step S13). If the document field end tag 62 is not found before the end of the input character string of one document is detected, the process is terminated. If the document field end tag 62 is found, the process proceeds to step S15.

In step S15, step S7
If the character string read in (i.e., the document field start tag 61 and the document field end tag 62 are stored separately and not included in the read character string) is not empty, the read character string The column is determined to be the contents of one field and cut out, and the field name is extracted from the document field start tag 61 and / or the document field end tag 62 to specify the field.

FIG. 6 is a diagram showing an example of document field information in which information about a document field such as a fixed field or an irregular field is described in advance. In FIG. 6, reference numeral 63 denotes document field information, 64 denotes a "field name" column in which the field names of all fields handled in this case accumulation / search method are registered, and 65 denotes a field name registered in the "field name" column 64 In the “Typical / Atypical” column 66 in which the attribute of a field is defined as a standard field or an atypical field is registered in advance, and 66 can be set for each field registered in the “field name” column 64. An "option" column in which option information is registered, 67 is a "category" attribute set to classify a set of case data largely, and 68 is an identification of each case when displaying a list of case data. This is a “title” attribute used to perform the operation.

The field name extraction process is performed by extracting the character string obtained by removing the symbols “<” and “>” from the document field start tag 61 and the symbols “<//” and “>” from the document field end tag 62. If the removed character string is identical (coincident) and not empty, the character string is determined to be a field name and taken out.

Next, the field specification processing refers to the document field information 63 shown in FIG. 6, and if the field name extracted in the field name extraction processing is registered in the “field name” column 64 , “Formal / Atypical” column 6 corresponding to the registered field name
The attribute value of No. 5 is extracted, and it is specified whether the field is a fixed field or a non-fixed field.

Next, when the extracted attribute value is "fixed pattern", the process proceeds to step S16, and the process for the fixed pattern field is executed. If the extracted attribute value is "irregular", the process proceeds to step S17 to execute processing on the irregular field.

Step S1 for processing a fixed field
In step 6, the contents of the fields extracted in step S15 are registered in the corresponding columns of the case database 50 as they are. In the example of the digitized document file 54 shown in FIG. 3, when the field is “<customer name>”, characters are added to the field contents field corresponding to the field where the field name field of the case database 50 is “customer name”. The column "Taro Yamada" is registered as it is.

In step S17 for processing an irregular field, the processing flow calls "irregular field processing" described later. In the example of the digitized document file 54 shown in FIG. 3, the "irregular field processing" is called for the "<question>" field and the "<answer>" field.

After the processing in step S16 or S17 is completed, the flow returns to step S11 to repeat the above processing.

In the following five cases (1) to (5), (a) a warning is issued to the user and the processing is continued.
(B) A warning is issued to the user, the process is temporarily interrupted, the process waits for the user's instruction, the process is performed according to the user's instruction, and then the process is continued. (C) A warning is issued to the user and the process ends. Take one of the following actions.

(1) When there is data to be read except for the document field start tag 61 and the document field end tag 62 in step S12. (2) In step S8, the digitized document file 54
Is detected, and there is an unprocessed read character string in step S16 or step S17. (3) In step S15, the read character string is empty. (4) If the field name extracted from the document field start tag 61 for the extracted character string (corresponding to one field) does not match the field name extracted from the same document field end tag 62 in step S15. (5) When the field name extracted in step S15 does not exist in the document field information 63 shown in FIG. The case where the extracted field name is empty is also included.

FIG. 7 is a flowchart showing the flow of the processing (step S17 in the flowchart shown in FIG. 5) for the irregular fields in the digitized document file 54 shown in FIG.

In FIG. 7, 69 is the case number set for each document, 70 is the field name of the atypical field to be processed, and 71 is the sentence extraction process (step S18) described below for each sentence. The extracted case sentence, 72 is a series of sentence numbers assigned to each of the case sentences 71, and 73 is a sentence tag assigning process described later (step S20).
, A sentence tag assigned to each of the case sentences 71,
74 is a series of sentence tag-by-sentence numbers assigned to each of the example sentences 71 (having the same sentence tag 73) for each sentence tag 73.

Next, the processing flow of the irregular field will be described with reference to the flowchart shown in FIG.

First, the case number 69 of the document to be processed (eg, the document 55 in FIG. 3) and the field name 70 of the atypical field to be processed (eg, the “<question>” field of the atypical field 58 in FIG. 3) are input. From among a plurality of sentences described in the atypical fields to be processed in the digitized document file 54,
One sentence is cut out based on a character string representing a bullet point such as a period ".", A middle point ".", And a chapter number (step S18).

In the example of the digitized document file 54 shown in FIG. 3, from the "<question>" field, a sentence "sound from the air conditioner." Here, the sentence extracted in the sentence extraction processing (step 18) is referred to as "case sentence" constituting the case data. Step S18
When the case sentence 71 can be extracted in the extracted case sentence 71, the case number 69 of the case to which the case sentence 71 belongs, the field name 70 of the field to which the case sentence 71 belongs, the field to be processed (for example, "<" Shown in FIG.
A sentence number 72 indicating the number of the sentence in the “question>” field) is assigned.

Next, it is checked whether or not a case sentence can be cut out (step S19). If the case sentence cannot be cut out, it is determined that all sentences in the processing target field have been extracted. It is confirmed that there is a document field end tag (for example, "</ question>" in FIG. 3) (step S21), the atypical field processing ends, and the atypical field processing in the flowchart shown in FIG. The process returns to step 17, which is the calling source of the field processing, and then proceeds to step S11. In step S21, if it cannot be confirmed that the document field end tag exists, a warning is issued.

On the other hand, if a case sentence can be cut out in step S19, the sentence sentence 71, which is a label indicating the type of the case sentence 71, is added to the cut-out case sentence 71.
73 is given (step S20).

FIG. 8 is a diagram showing an example of the sentence tag list. In FIG. 8, reference numeral 75 denotes a sentence tag list. In the sentence tag assigning process (step S20), for example, a sentence tag list table 75 shown in FIG. 8 is referred to as information for assigning the sentence tag 73.

As shown in FIG. 8, in the sentence tag list 75, a “condition” part which is a key expression in a sentence for each “field” and a “sentence tag” corresponding to the “condition” part are previously stored. Define it.

For example, in the example of the digitized document file 54 shown in FIG. 3, the first example sentence “sound from the air conditioner” in the <question> field corresponds to the sentence tag list shown in FIG. "Field" is "<question>"
In the “condition” part of which is described, the keywords “use in” and “gas” are described. However, since the keyword “gas” is present in the case sentence, the sentence tag of the case sentence is The “sentence tag” corresponding to the “condition” part of “ga suru” in the sentence tag list 75 is extracted and becomes the “symptom”.

Further, the digitized document file 5 shown in FIG.
In the example of FIG. 4, for the last case sentence in the <answer> field, “No countermeasure is required because it is not abnormal,” referring to the sentence tag list shown in FIG. 8, the “field” is changed to “<answer>”.
The "condition" part is "for", "it seems", "it seems"
The keywords "necessary", "unnecessary" are described. However, since the keyword "unnecessary" exists in the case sentence, the sentence tag of the case sentence is referred to as "unnecessary" in the sentence tag list table 75. The “statement tag” corresponding to the “condition” part is extracted and used as “measures”.

FIG. 8 shows a case where a sentence tag is specified by using a character string of a surface layer (characters included in a sentence) for the sake of simplicity. A syntax pattern and a modal expression such as negation and guesswork can be specified, and a sentence analysis process described later is performed on the extracted example sentence, and then a sentence tag is determined by associating with the sentence tag list 75. You may.

FIG. 9 shows a process of extracting a target sentence (step S).
FIG. 4 is a diagram showing an example of “case” generated from the document 55 of the digitized document file 54 shown in FIG. 3 by 1).
In FIG. 9, reference numeral 76 denotes a case. In FIG. 9, FIG.
The same or corresponding parts are denoted by the same reference numerals and description thereof is omitted.

In the example shown in FIG. 9, case number 69 (“0001” in the example shown in FIG. 9) is assigned as case 76.

Further, each of the document fields “<customer name>”, “<customer telephone>”, “<
For the “subject>” and “<model>”, as the case 76, the field names “customer name”, “customer number”, “subject”, “model” and the contents of the standard fields corresponding to the field names “Taro Yamada”, “Taro Yamada” 0 □□□-△ ▽ -OOOO ”,“ refrigerant sound of air conditioner ”, and“ air conditioner ”are generated.

The irregular field 58 shown in FIG.
For each of the fields <Question> and <Answer>
As a case 76, "Field sound 70 such as" Question "and" Answer ", and corresponding to each field name 70," Sound from the air conditioner. "
"1", "2", "3" for each case sentence 71 belonging to the field.
A series of sentence numbers 72 such as "4" (the above <question> field), "1" and "2" (the above <answer> field), "symptom", "environment", "symptom", "symptom" (the above <question> field) " Sentence tags 73 such as "cause" and "measures" (the above <answer> field) are generated.

The sentence tag 73 "symptom" is given to the sentence numbers 1, 3, 4 belonging to the <question> field.
Is different from the example sentence 71, and the other example sentences 71 are given different sentence tags 73.
“1,” “1”, “2”, “3”, “1”, and “1” for example sentences 71 of 2, 3, 4 (or more <question field>) and 1 and 2 (or more <answer> field) Sentence number 74 for each sentence tag.

FIG. 9 shows a case where the sentence extraction process (step S18 in the flowchart shown in FIG. 7) and the sentence tag attaching process (step S20) are all successful. There is a case where segmentation of a case sentence is erroneous due to delimitation of a sentence or ambiguity of a sentence tag attaching condition. In this case, a display function is provided in the processing target sentence extraction unit 45 in the case accumulation / search device shown in FIG. 1, and the case is presented (displayed) to the user in a format as shown in FIG. By adopting a configuration that accepts edit / correction inputs, manual edit / correction becomes possible.

Thus, the sentence extraction process (step S1) in the flowchart shown in FIG. 2 is completed, and then the sentence analysis process (step S2) is executed.

In the sentence analysis processing (step S2),
Morphological analysis and syntax analysis are performed on a set of processing target sentences (case sentences) obtained in the processing target extraction process (step S1) to generate a sentence structure.

FIG. 10 is a flowchart showing the flow of the sentence analysis process (step S2) in the flowchart shown in FIG. In FIG. 10, reference numeral 77 denotes an analysis word dictionary used in morphological analysis (step S22) described later. In FIG. 10, the same or corresponding parts as those in FIG. Next, a sentence analysis process (step S
The flow of 2) will be described.

As shown in FIG. 10, the sentence analysis process (step S2) includes a morphological analysis process (step S22) and a syntax analysis process (step S23). The morphological analysis process (step S22) refers to the analysis word dictionary 77, and the syntax analysis step S23 refers to the area ontology 44. Alternatively, the analysis word dictionary 77 may be included in the area ontology 44, and the morphological analysis processing (step S22) may refer to the area ontology 44.

FIG. 11 is a flow chart showing the flow of the morphological analysis processing (step S22) in the flow chart shown in FIG. 10, and the morphological analysis section in the sentence analysis section 47 in the case storage / search apparatus shown in FIG. FIG. 2 is a configuration diagram showing a configuration (not shown).

In FIG. 11, reference numeral 78 denotes a morphological analysis unit which exists in the sentence analysis unit 47 of the case storage / search apparatus shown in FIG. 1 and performs morphological analysis, and 79 denotes an auxiliary word used when performing morphological analysis. An attached word dictionary stored in advance, 80 is an attached word connection table stored in the attached word dictionary 79 in which connection relations of attached words stored in the attached word dictionary 79 are described in advance, and 81 is a processing target of the morphological analysis process (step 22). Input sentence,
Reference numeral 82 denotes a result of the morphological analysis of the input sentence 81 during the analysis,
Reference numeral 83 denotes a morphological analysis result obtained by executing the morphological analysis processing (step 22), and reference numerals 84, 85, and 86 denote structures of the analysis word dictionary 77, which are "heading" and "part of speech", respectively.
Information, a "semantic symbol". In FIG. 11, FIG.
The same or corresponding parts as 0 are denoted by the same reference numerals and description thereof will be omitted.

Next, the operation of the morphological analysis unit 78 and the flow of the morphological analysis process (step S32) will be described using the configuration diagram and flowchart of the morphological analysis unit 78 shown in FIG.

The morphological analysis section 78 executes a morphological analysis process (step S22). In the morphological analysis process (step S22), the analysis word dictionary 77, the attached word dictionary 7
9. Perform morphological analysis with reference to the attached word connection table 80,
The input sentence 81 is divided into morpheme columns. The morphological analysis method, the auxiliary word dictionary 79, and the auxiliary word connection table 80 have been described in detail in many documents, and thus description thereof is omitted here. In the following description, as shown in the analysis result 82 in FIG. 11, the delimiters of the morphemes are abbreviated as "/".

The analysis result 82 shown in FIG. 11 is based on the information described in the analysis word dictionary 77 (headline 84, part of speech information 85, meaning symbol 86 which is conceptual information indicating the type of word) from the independent word part. Can be referred to. At this time, pointer information for referring to the analysis word dictionary 77 may be held in the analysis in-progress result 82, or it takes time to refer to the analysis word dictionary 77 due to the presence of the analysis word dictionary 77 in the secondary storage device. In this case, the information of the analysis word dictionary 77 may be copied on the primary storage device. In the following description, as shown in FIG. 11, the meaning symbol 86 is represented by enclosing a word between a symbol “<” and a symbol “>”.

FIG. 12 is a flow chart showing the flow of the syntax analysis processing (step S23) in the flow chart shown in FIG. 10, and also included in the sentence analysis section 45 in the case accumulation / analysis apparatus shown in FIG. FIG. 3 is a configuration diagram illustrating a configuration of a syntax analysis unit (not shown).

In FIG. 12, reference numeral 87 denotes a syntactic analysis unit which exists in the sentence analysis unit 45 and performs syntactic analysis; 88, a grammatical rule stored in advance for use in the dependency processing (step S25); 83, the phrase structure generated by the phrase structure generation process (step S24);
Is a dependency structure generated by the dependency analysis process (step S25) using the phrase structure 89 as an input.

Next, the operation of the syntax analysis unit and the flow of the morphological analysis process (step S23) will be described using the configuration diagram and flowchart of the syntax analysis unit 87 shown in FIG.

The syntactic analysis unit 87 includes a phrase structure generation unit (not shown) that generates a phrase structure 89, which is a basic unit for performing dependency analysis, and a dependency analysis unit that performs dependency analysis with reference to the grammar rules 88. (Not shown).

The syntax analysis processing shown in FIG. 10 (step S
23), first, a phrase structure generation process (step S2)
In 4), the morphological analysis output result 83 shown in FIG.
To generate a phrase structure 89, which is a basic unit for performing dependency analysis. The phrase structure 89 is composed of only one or more independent word morphemes or one or more independent word morphemes and one or more adjunct word morphemes connected to the independent word morpheme.

FIG. 13 is a diagram showing an example of the phrase structure 89. In FIG. 13, reference numeral 91 denotes a dependency attribute which is an attribute of a dependency portion, 92 denotes a reception attribute which is an attribute of a dependency portion, and 93 denotes a pointer to at least one independent word morpheme information constituting the phrase. Is the independent word information, and 94 is auxiliary word information storing pointers to zero or more auxiliary word morpheme information.

In the dependency analysis process (step S25), the dependency analysis of the phrase structure 89 is performed in accordance with the grammar rule 88, and the dependency structure 90 is generated as a result of the syntax analysis. Since the dependency analysis method is described in many documents and the like, a detailed description is omitted here. In general, a number of ambiguities are generated during the dependency analysis, but the area ontology 44 is appropriately referred to in order to resolve the ambiguity. The grammar rule 88 may be embedded in a program for the dependency analysis process (step S25).

Thus, the sentence analysis processing (step S2) in the flowchart shown in FIG. 2 is completed. Next, a sentence to be processed (case sentence) is called while calling a similar sentence matching unit 48 for matching similar sentences with reference to the area ontology 44.
Sentence clustering processing for classifying (clustering) a set of the syntax analysis results 90 for each similar sentence (step S
Execute 3).

FIG. 14 is a diagram showing an example of a problem solving tree 95 composed of clusters including a set of similar case sentences generated from the digitized document file 54 shown in FIG.

In FIG. 14, 95 is a problem solving tree, 96
Is the cluster labeled "Sound of interest",
Reference numeral 97 denotes a cluster labeled “main body lamp not lit”, 98 denotes a cluster labeled “main body display lamp blinking”, 99 denotes a cluster labeled “input timer operation”, and 100 denotes the corresponding cluster. The number of cases (sentences) associated with the cluster, 101 is a confirmation check box used at the time of search, 102 is a display label of the cluster set for cluster display, 103 is the cluster located at the end of the problem solving tree 95 In this case, the end is displayed so that the cluster is the end.

A problem solving tree 95 composed of clusters including a set of similar sentences as shown in FIG. 14 is constructed, and each cluster of the created problem solving tree 95 has a set of similar case sentences and is shown in FIG. As described above, each case sentence 71 is stored in the case database 50 shown in FIG. 1 in a format having a pointer to the case 76. Before describing the flow of the similar sentence clustering process (step S3) in the flowchart shown in FIG. 2, first, the similar sentence clustering process (step S3)
The operation of the similar sentence matching unit 48 called from 3) to obtain the similarity between sentences will be described.

Calculation of similarity between sentences in the similar sentence matching unit 48 for matching similar sentences with reference to the area ontology 44 is performed by weighting for each layer of nodes in the sentence dependency structure and weighting by attributes in the nodes. (Process of making modal expressions such as negation and guesswork as calculation targets).

FIG. 15 is a diagram showing an example of a dependency structure of two sentences. In FIG. 15, reference numeral 104 denotes a dependency structure of a sentence So "sensor recognizes clogging", and reference numeral 105 denotes a dependency structure of sentence Si "sensor detects dew condensation".

Next, a method of calculating the similarity between sentences in the similar sentence matching unit 48 will be described with reference to FIG. The function of calculating the similarity between natural language sentences in the similar sentence matching unit 48 is Sim (A, B, D). Arguments A and B are shown in FIG.
0 is a dependency structure 90 (FIG. 12) which is a syntax analysis result obtained as a result of executing the syntax analysis process (step S23), and has a tree structure as shown in FIG.
The argument D is the level of detail of the matching when calculating the similarity between sentences. When calculating the similarity between the two tree-structured dependency structures 104 and 105 shown in FIG. This is a value indicating from which layer up to which layer is to be processed. Here, for the sake of simplicity, description will be made assuming that D = 2.

First, an initial similarity of 1.0 is given. A similarity of 1.0 means that the two input sentences have exactly the same meaning. In the following processing, the information of each node is compared while tracing the tree structure, and a “penalty” is given to a different part, and the difference is subtracted from 1.0. When the similarity becomes 0 or a predetermined value, the comparison target is regarded as dissimilar and the similarity calculation is terminated.

FIG. 16 is a diagram showing an example of a penalty calculation rule in the similarity calculation.

First, comparison between first level nodes is performed. Here, since the semantic symbols are equal in "<detection operation>" and the actual words are different, a penalty value of -0.01 is given according to rule R1 shown in FIG. Next, a comparison between the second level nodes is performed. At this time, no penalty is given to the left node “sensor <monitoring device>” because the information completely matches. The nodes on the right have different semantic symbols, so penalty-0.
Give 3. Thus, the similarity is 1.0-0.
01-0.3 is calculated to be 0.69.

For example, if the value D of the argument is "1"
Is given, only the first-level nodes are targeted for the similarity calculation, and the similarity is calculated as 1.0-0.01 to be 0.99. As described above, since the precision of the similarity calculation can be controlled by the value of D, flexible processing (similarity calculation) can be performed according to the search situation.

FIG. 17 is a diagram showing an example of knowledge about an area stored in the area ontology 44. In FIG. 17, reference numeral 106 denotes an IS-A indicating a relationship between a superordinate concept and a subordinate concept.
Relational knowledge 107 is HAS-A indicating a whole-part relation
Relation knowledge, 108 is case relation knowledge showing the relation between concepts, 1
09 is paraphrase knowledge that absorbs the difference when the difference in expression is large, and 110 is HAS related to “<timer display>”.
-A relationship knowledge, and at the same time, knowledge indicating a conflict that cannot occur. In FIG. 17, the same or corresponding parts as those in FIG. 2 are denoted by the same reference numerals, and description thereof will be omitted.

That is, the area ontology 44 stores the case
IS-A related knowledge 106, HAS-A related knowledge 107, and case related knowledge 10 depending on the area to be searched.
8. It is composed of the paraphrase knowledge 109 and the like, and it is also possible to describe knowledge that indicates a conflict. The IS-A related knowledge 106 and the HAS-A related knowledge 107 are usually called "thesaurus".

In addition, the dependency structure 104 shown in FIG.
When the rule R2 shown in FIG. 16 is applied to
When the meaning symbol of the word “monitor device” is “<main device>” according to the IS-A relation knowledge 106 shown in FIG. 2, the similarity of the sentence “the monitor device recognizes clogging” is 1.
0-0.1 is calculated to be 0.9.

On the other hand, the rule R5 shown in FIG. 16 is changed from the dependency structure 9 shown in FIG.
When applied to 0, the attribute of the node of the syntax analysis result includes a negation, so that the similarity with the dependency structure of the sentence “Timer enters at RC” according to rule R5 shown in FIG.
0.9 is calculated to be 0.1. If a predetermined value (threshold) for terminating the similarity calculation is set to, for example, 0.2, the two sentences are not similar, and the similarity calculation is discontinued at this time. .

Similarly, when a sentence includes a guess expression such as "-Would be" and the attribute of the node as a result of the syntax analysis includes guesswork, the rule R6 shown in FIG. 16 is applied and the penalty is- Given a similarity less 0.15.

Similarly, when the rule R3 shown in FIG. 16 is applied to the sentence “Does not accept the input of the timer button”, the sentence “HAS-A related knowledge 107 shown in FIG. 17 is used. The degree of similarity with "Remote controller input is not accepted" is calculated to be 1.0-0.2 and is relatively high at 0.8.

In the HAS-A relation knowledge 107 of the area ontology 44, information that is contrary to the above can be described. FIG.
This will be described using the example of the HAS-A related knowledge 110 regarding “<timer display>” shown in FIG.

The “<timer display>” has states of “none”, “off timer”, and “on timer”, but the HAS-A related knowledge 110 regarding “<timer display>” shown in FIG. , "None" and "Off timer" or "On timer"
Is not displayed at the same time, indicating that it is contrary. According to the HAS-A-related knowledge 110 regarding this “<timer display>”, when the sentence “no timer display” and the sentence “timer display has an input timer” in the case sentence, the case Database 5
0 can be used to store information that a cluster including each sentence is a conflict.

As shown in FIG. 17, the area ontology 44
In the paraphrase knowledge 109, expressions of words having the same meaning can be described. In the example shown in FIG.
No. 09 describes paraphrase knowledge 109 that “<breaker> flies” and “<breaker> falls” have the same meaning, and this paraphrase knowledge 10 is used when similarity is calculated.
By referring to No. 9, the similarity between the sentence “breaker flies” and the sentence “breaker falls” can be handled as 1.0 (semantically coincident).

FIG. 18 is a flowchart showing the flow of the similar sentence clustering process (step S3) in the flowchart shown in FIG. In FIG. 18, reference numeral 111 denotes a case number 69, a field name 70, a case sentence 71, a sentence number 72, a sentence tag 73, a sentence number 74 for each sentence tag,
Example sentence data composed of the dependency structure 90, 112
Is a set of case sentence data composed of a plurality of case sentence data 111. Note that, in FIG. 18, the same or corresponding parts as those in FIGS. 7 and 12 are denoted by the same reference numerals, and description thereof will be omitted.

Next, the flow of the similar sentence clustering process (step S3) in the flowchart shown in FIG. 2 will be described with reference to FIG. The similar sentence clustering process (step S3) includes a set 112 of case sentence data.
As input parameters.

First, the field name 70 to be classified (clustered), the type of the sentence tag 73, and the maximum layer number of the cluster layer are specified (step S26).
Next, assuming that the current layer is “1”, the clustering process (step S27) is called with the set of case sentences meeting the specified condition, the maximum number of layers (layer), and the current layer “1” as input parameters. The processing flow of the clustering processing (step S27) will be described later. Next, the problem solving tree 95 shown in FIG. 14, which is the processing result of the clustering processing (step S27), is stored in the case database 50.

FIG. 19 is a flowchart showing the flow of the clustering process (step S27) shown in FIG. In FIG. 19, reference numeral 113 denotes an execution result of a clustering sub-process (step S30) for performing actual clustering described later.

Clustering process (step S27)
First, when the set of case sentences, the maximum number of layers, and the current layer i are received as parameters, the current layer i is compared with the maximum layer number layer (step S29),
If the current layer i is larger than the maximum layer number layer, the process ends. On the other hand, the current hierarchy i is the maximum hierarchy number l
In the case of ayer or less, the clustering sub-process (step S30) is called for the set of case sentences, and the actual clustering process is performed. The processing flow of the clustering sub-process (step S30) will be described later. As the execution result 113 of the clustering sub-process (step S30), the number k of clusters and k clusters {C1,
.., Ck} are obtained.

Next, j is a counter provided inside the clustering process (step S27), and the value of the counter j is set to an initial value 1 (step S31). Counter j
Is compared with the value of the number of clusters k (step S32),
If the value of the counter j is larger than the number of clusters k, the process ends. On the other hand, when the value of the counter j is equal to or smaller than the number k of clusters, the clustering process (step S37) is recursively called using the set Cj of the case sentences, the maximum hierarchy layer, and the current hierarchy (i + 1) as input parameters. next,
The value of the counter j is incremented by 1 and the process returns to step S32, and the processing from step S32 is repeated.

FIG. 20 is a flowchart showing the flow of the clustering sub-process (step S30) in the flowchart shown in FIG. Next, referring to FIG. 20, a clustering sub-process for performing clustering using similarity calculation by the similar sentence matching unit 48 (step S
The flow of the process 30) will be described.

First, a threshold value th for determining that two sentences are similar is set (step S34). Here, the description will be made assuming that the value of the threshold th is 0.75. It is assumed that the threshold th can be changed according to the cluster creation status. Next, one cluster Ci (i = 1, 2,...,
n; n is the number of case sentences to be processed (n is the number of case sentences to be processed), each constituting an initial cluster consisting of one case sentence Si (step S3).
5). Subsequently, the similar sentence matching unit 48 is called to call Sim (S
i, Sj, D) is obtained, and the similarity Sim (C
(i, Cj) is Sim (Si, Sj, D), and an initial similarity table between the clusters is created. Where i = 1
2,..., N, j = i + 1, i + 2,.
The value of D is set as needed.

FIG. 21 shows that the case sentences are S1, S2,.
5, five sentences, C1 = {S1}, C2 = {S
It is a figure which shows the example of the similarity table which stored the similarity between clusters at the time of starting a clustering process from five clusters of 2}, ..., C5 = {S5}. In FIG.
114 is an initial similarity table between clusters, 115 is a similarity table when the first clustering process is performed, 11
Reference numeral 6 denotes a similarity table when the second clustering process is performed. Note that these similarity tables 114, 115,
Since 116 is a target matrix, values are stored only in the lower half.

Next, a pair Cu and Cv of the cluster having the maximum similarity is obtained from the initial similarity table 114 (step S37). In the similarity table 114 shown in FIG. 21, Cu = C4, Cv = C5, and Sim (Cu, Cv) =
Sim (C4, C5) = 0.99, and the similarity between clusters is the maximum.

Next, the maximum similarity Sim (Cu, Cv)
Is compared with the threshold th (step S38), and when the maximum similarity Sim (Cu, Cv) is larger than the threshold th, the clustering process is continued, and the maximum similarity Sim (C
If u, Cv) is equal to or smaller than the threshold th, the clustering process ends. In the example of the initial similarity table 114 shown in FIG. 21, the maximum similarity Sim (C4, C5) is 0.99,
Since the threshold th is 0.75, the clustering process is continued.

Next, in step S2406, the pair of clusters Cu and Cv is converted into one cluster Cu, v = ｛Su, S
v｝. When the first clustering process is performed on the example of the initial similarity table shown in FIG. 20, the cluster pair C4 and C5 is one cluster C4,5 = ｛S
4, S5}.

Next, the similarity Sim (Cu, v, Ci) between the newly generated cluster Cu, v and other clusters Ci by unifying the cluster pairs is represented by (Sim
The similarity is calculated using the formula (Ci, Cu) + Sim (Ci, Cv)) / 2 to reorganize the similarity table (step S40). In the example shown in FIG. 21, for the initial similarity table 114,
A similarity table 115 obtained when the first clustering process is performed is obtained.

Next, returning to step S37, step S37 is executed.
38. Similarity between clusters Sim (Cu, Cv) at 38
Is repeated until the value becomes equal to or smaller than the threshold th.

In the equation (the above equation) in step S40 shown in FIG. 20, the simple average value of the similarity between the two clusters Cu and Cv is calculated, but the weighted weight is calculated in consideration of the number of case sentences in the cluster. An average value or the like may be used.

In the example shown in FIG. 21, after the second repetition, the largest similarity table 116 in the similarity table 116 in the case where the second clustering process created by combining the clusters C1 and C2 is performed. Degree Sim (C1,2, C4,
5) becomes 0.73, which is 0.75 or less of the threshold th, so the clustering sub-process (step S30) is ended in step S38. At this time, in the example shown in FIG. 21, C1,2 = {S1, S2}, C3 = {S
Three clusters of {3}, C4,5 = {S4, S5} are generated.

Thus, the similar sentence clustering process (step S3) for performing the case sentence classification process in the flowchart shown in FIG. 2 is completed. Next, the case cluster editing process in the flowchart shown in FIG.
By executing 4), the cluster hierarchy of the case data stored in the case database 50 is edited and displayed as a problem solving tree 95 having a structure as shown in FIG.

The example of the problem solving tree 95 shown in FIG. 14 is shown in FIG. 6 among the case sentences 71 shown in FIG. 9 extracted in the processing target sentence extracting process (step S1) in the flowchart shown in FIG. Field 64 has a field name of "<
The case sentence 71 having the first, second, and third appearances having the “symptom” as the sentence tag 73 from the field “question>”, that is, the case sentence 71 having the sentence tag per minute number of 1, 2, and 3, It is created by performing clustering of similar sentences by specifying the number of hierarchical levels of the cluster 3 in the similar sentence clustering process (step S3) in the flowchart shown in FIG.

The problem solving tree 95 shown in FIG. 14 is created for each value of the “category” attribute 67 in the option 66 column corresponding to the field name in the field name 64 column shown in FIG. The display of each cluster in the case cluster editing unit 51 shown in FIG. 1 is displayed from top to bottom in the descending order of the number of case sentences included in the cluster. Each cluster is composed of a set of similar case sentences, and it is assumed that the display label 102 of each cluster is designated by an operation such as a mouse and information in the cluster can be referred to. Further, the case cluster editing unit 51 can set and edit the character string of the display label 102 of the cluster by the operation of the user.

FIG. 22 is a diagram showing an example of the case sentence data 111 stored in the case database 50. In FIG. 22, the same or corresponding parts as those in FIG.

The output of the processing target sentence extraction process (step S1) in the flowchart shown in FIG. 2, specifically, the case number 69, the field name, which is the output of the sentence tag attaching process (step S20) in the flowchart shown in FIG. A sentence analysis process (step S2) in the flowchart shown in FIG. 2 is described above the reference numeral 70, the case sentence 71, the sentence number 72, the sentence tag 73, and the sentence number 74 for each sentence tag. After the dependency structure 90 obtained by executing the reception analysis process (step S25) is added, the similar sentence clustering process (step S25) in the flowchart shown in FIG.
In 3), the cluster to which the case sentence data belongs is determined, the cluster number of the cluster is assigned, and stored in the case database 50. At this time, the case database 50 also stores information indicating the hierarchy between clusters, such as the problem solving tree 95 shown in FIG.

FIG. 23 shows a similar sentence clustering process (step S) in the flowchart shown in FIG.
FIG. 15 is a diagram showing an example of inter-cluster information other than the hierarchical relationship stored in the case database 50 in the process of storing the problem solving tree 95 (see FIG. 14) in the case database 50 in 3).

The inter-cluster information is, as shown in FIG.
A relationship between two clusters represented by cluster numbers is defined. Examples of relationship types are -1 (contradiction) and 1
(Similar). The conflicting relationship is created with reference to the conflicting relationship shown in the HAS-A relationship knowledge 110 regarding “<timer display>” in FIG.

The similarity relation is such that a representative sentence representing the cluster can be set in each cluster, and the similarity between the representative sentences is stored in the case database 50 as inter-cluster information.

The detailed description of the operation of the case accumulation function in the case accumulation / retrieval apparatus according to Embodiment 1 of the present invention has been completed.

Hereinafter, the operation of the case retrieval function in the case accumulation / retrieval apparatus according to the first embodiment of the present invention will be described in detail with reference to the flowchart as shown in FIG. .

First, a search sentence input process (step S41)
In, the user inputs a search sentence (a description of a new problem) for searching for a desired document. At this time, the input of the search sentence is performed using an input device such as a keyboard, a character recognition device, and a voice recognition device (not shown).

Next, in the sentence analysis process (step S2), the search sentence input by the search sentence input process (step S41) is analyzed.

In the sentence analysis process (step S2), the search sentence is processed in the order of the morphological analysis process (step S22) and the syntax analysis process (step S23) shown in FIG. Analyze and generate a dependency structure for the search sentence. The above processes are the same as those in the case construction process described above, and thus description thereof is omitted.

Next, a similar case search process (step S4)
The flow of the process 2) will be described. First, a primary search of the case database 50 is performed. When the capacity of the case database 50 is small, there is a method of performing the processing of the similar sentence matching unit 48 between the input search sentence and all the case sentences, but in general, the case database 50 has a large capacity. Since the case sentences are stored, if the input case sentence is compared with all the case sentences, a problem occurs in the processing time. Therefore, an index based on the syntactic element of the structure or the semantic symbol of the syntactic element is provided for the dependency structure of the case sentence, and after performing the primary search processing using the index, the similar sentence matching unit Forty-eight processes are performed, and the case sentences to be processed are narrowed down in advance.

Next, in the similar case search process (step S42) in the flowchart shown in FIG. 24, the similar sentence matching section 48 is activated to input the dependency analysis result of each case sentence in the primary search result. The similarity between the search sentence and the dependency analysis result is obtained, and the similar sentence is collated based on the obtained similarity. In the similar sentence matching process, similar sentence matching is performed with reference to the area ontology 44 shown in FIG.

Next, the flow of the similar case search process (step S42) in the flowchart shown in FIG. 24 will be described with reference to the flowchart shown in FIG.

First, initialization setting for loop processing is performed (step 44). Next, the loop processing from step S45 to step S47 is executed for each case sentence of the primary search result.

First, in step S46, the ith case sentence Si of the primary search result (case sentence Si shown in FIG. 15)
Dependency structure 10 for "sensor detects condensation"
5) and the input search sentence So (dependency structure 10 for the case sentence So “sensor recognizes clogging” shown in FIG. 15)
4) and the similarity calculation Sim (Si, So,
Perform D). The value of D is set as needed as described above.

Finally, the search results are sorted in order of similarity (step S48), and the sorted results are output to the search result display section 53 shown in FIG. 1 (step S43).

According to the case storage / search method of the present invention, even if the search sentence and the case sentence completely match at the level of the keyword included in the case sentence, the meaning of the sentence between the search sentence and the case sentence If the contents are different, the similarity decreases. Also,
Even if the keywords included in the search sentence and the case sentence are different, the similarity between the search sentence and the case sentence having the same semantic content is increased by performing similar sentence matching using the area ontology 44. .

In the conventional keyword-based case accumulation / search, since the detailed processing described above is not performed, an unnecessary case sentence as a search garbage at the time of case search such as a negative sentence is output at a higher position. However, according to the case accumulation / search method of the present invention, for example, by setting such that case sentences having a similarity of 0.5 or less are not displayed, unnecessary case sentences are not output as similar sentences. Can be.

As described above, at the time of case search, the input search sentence is divided into words in the morphological analysis process (step S22) shown in FIG. 10 in the sentence analysis process (step S2) shown in FIG. A dependency structure for the input search sentence is generated in the syntax analysis process (step S23) shown in FIG. 23, and a search using the dependency structure is performed in the similar case search process (step S42) shown in FIG. Similar sentence matching unit 4
8, the similarity between the input search sentence and the search result is calculated with reference to the area ontology 44, and the similarity is calculated in the search result display process (step S43) shown in FIG. 23 according to the calculated similarity. By outputting the search result to the search result display unit 53 shown in FIG. 1 according to the degree, a case search with less search dust can be realized for various natural language expressions.

The search result display method in the search result display processing (step S43) shown in FIG. 24 is such that a problem solving tree 95 as shown in FIG. The cluster including the case sentence may be highlighted. FIG. 14 includes, for example, an input sentence “unpleasant sound is heard” includes a case sentence “unusual sound is heard” that is most similar to the cluster 96 labeled “unwanted sound”. The example highlighted as a cluster is shown.

In the search result display unit 53 shown in FIG. 1, similarly to the case cluster editing unit 51, the problem solving tree 95 is displayed from the top to the bottom in the descending order of the number of case sentences included in the cluster. The user can easily refer to a case where the number of cases stored in is large. The user can designate each cluster, refer to the case corresponding to the case sentence included in the designated cluster, and display the case in a format as shown in FIG. One cluster includes a plurality of case sentences, each of which corresponds to another case. When a case list is displayed when referring to a case, a “title” attribute 68 described in an option 66 column corresponding to each field name in a field name 64 column shown in FIG. Just list them.

The case database 50 also stores the inter-cluster information shown in FIG. 23 in addition to the case sentence data 111 shown in FIG. Next, how to use the inter-cluster information will be described.

In the inter-cluster information, a reciprocal relation and a similar relation between two clusters represented by cluster numbers are defined. By using this inter-cluster information, for example,
In response to the input sentence "No timer display", if the cluster 97 "main unit lamp is not lit" is searched in the problem solving tree 95 shown in FIG. 14, the user confirms that the state of the cluster 97 is correct and checks When the column 101 is selected, the cluster 98, which is a similar cluster, “main body area display is not lit” is highlighted, or the display of the cluster 99, which is a contradictory cluster, “on timer operation” is not highlighted. Can be. As in the first embodiment, by storing in the case database 50 in advance the case data including the relationship between data at the time of case accumulation, the search efficiency can be improved in consideration of the relationship between data. Can be.

Further, the case accumulation / retrieval program obtained by realizing the above case accumulation / retrieval processing as a program is recorded on a computer-readable recording medium.
The case accumulation / search program recorded on the recording medium may be loaded into a computer such as a personal computer or a microprocessor and executed.

Further, the case accumulation / retrieval program comprises:
The program may be stored in a computer or a disk device connected to a network, and the program may be loaded via the network.

The case accumulation / retrieval program may exist in a computer connected via a network, and may access the computer to execute the case accumulation / retrieval program.

[0194]

According to the first case accumulation / retrieval apparatus of the present invention, even if the case accumulation / retrieval target is composed of a plurality of case sentences, an electronic document described in a natural language can be used. By storing case data obtained by cutting out and classifying each case sentence from the case database and comparing the input search sentence with the case sentence classified and stored in the case database, a complicated case is obtained. This also has an effect that the problem can be easily solved by the similarity search.

According to the first case accumulation method according to the present invention, even if the case accumulation / retrieval target includes a plurality of case sentences, each case is selected from the digitized document described in the natural language. By storing case data obtained by extracting and classifying case sentences in a case database, and collating the input search sentence with the case sentences classified and stored in the case database, even for complex cases, There is an effect that the problem can be easily solved by the similarity search.

According to the second case accumulating method of the present invention, in the processing target sentence extracting step, the case sentence to be subjected to the case accumulating process is added by assigning the type of the case sentence to the extracted case sentence. Therefore, in the classification processing, it is possible to perform the classification in which the type and description order of the sentence are reflected.

According to the third case accumulating method of the present invention, further, in the similar sentence clustering step, the number of hierarchical levels of the clusters that classify each case sentence and / or the case sentences are similar to each other. Since the threshold value of the similarity between the clusters to be used when judging that the case has been specified is specified, the number of layers or the number of By appropriately setting the threshold value of the similarity, it is possible to perform the clustering process finely and accurately, and to achieve an effect of increasing the efficiency of the clustering process.

According to the fourth case accumulation method of the present invention, further, in the area ontology, IS-A relation knowledge indicating a semantic upper-lower relation, and HAS indicating a semantic part-whole relation, -A Relationship knowledge, case relationship knowledge indicating the relationship between concepts, paraphrase knowledge that absorbs differences when the difference in expressions is large, and knowledge that is at the same time incompatible with each other are described. The description of knowledge depending on the target area becomes flexible, and detailed knowledge can be described, so that the accuracy of sentence analysis processing and similar sentence matching processing is improved, and furthermore, similar sentence clustering processing during case search And the accuracy of similar case search processing can be improved.

According to the fifth case accumulating method of the present invention, in the similar sentence matching step, each case sentence obtained by parsing each case sentence using the area ontology is referred to as (dependency). ) Since the similarity is obtained by comparing the semantic structure with the attribute of the syntactic element in the structure, it can handle various natural language expressions including the modal expressions (negation, guesswork, etc.) of the sentences constituting the case, Searching for cases that include sentences that are syntactically and / or semantically similar by making judgments in consideration of the detailed (fine-grained) similarities set in the nodes of the syntax tree of the syntax analysis result It has the effect of being able to do so.

According to the sixth case accumulating method of the present invention, further, in the similar sentence matching step, the degree of detail of matching used when obtaining the similarity between each case sentence is specified, so that the user can take the initiative. It is possible to provide a similarity criterion with the right, and it is also possible to finely provide a similarity criterion based on the “sentence tag” of the case sentence to be subjected to the case accumulation processing and its order.

According to the seventh case accumulation method of the present invention, further, in the sentence analyzing step, the structure of the case sentence is created as a tree structure, and in the similar sentence matching step, the similarity of each case sentence is compared. Since the degree of detail of the collation used to determine the degree is specified by the depth of the tree structure to be processed for the case sentence, there is an effect that the efficiency of the similar sentence collation processing can be improved.

According to the eighth case accumulation method according to the present invention, further, since a similarity relation or a contradiction relation is described as a relation between clusters in the case database, when a certain cluster is designated, special case calculation such as similarity calculation is performed. Since a cluster similar to the cluster can be searched without performing a complicated calculation, it is possible to improve the efficiency of case search processing such as similar case search processing and search result display processing.

According to the first case retrieval method according to the present invention, even if a case where the case accumulation / retrieval target includes a plurality of case sentences, each case is selected from the digitized document described in the natural language. By storing case data obtained by extracting and classifying case sentences in a case database, and collating the input search sentence with the case sentences classified and stored in the case database, even for complex cases, There is an effect that the problem can be easily solved by the similarity search.

According to the computer-readable recording medium on which the first case accumulation program according to the present invention is recorded, the case accumulation / retrieval target can be obtained by executing the case accumulation program recorded on the recording medium. For cases consisting of multiple case sentences, case data is extracted from each digitized document written in natural language and stored in the case database, and the case data is stored in the case database. By collating with a case sentence classified and stored in the case database, it is possible to easily solve a problem by similarity search even for a complicated case.

According to the computer-readable recording medium on which the first case search program according to the present invention is recorded, by executing the case search program recorded on the recording medium, the case accumulation / retrieval target can be obtained. For cases consisting of multiple case sentences, case data is extracted from each digitized document written in natural language and stored in the case database, and the case data is stored in the case database. By collating with a case sentence classified and stored in the case database, it is possible to easily solve a problem by similarity search even for a complicated case.

[Brief description of the drawings]

FIG. 1 is a configuration diagram showing a configuration of a case accumulation / search device according to a first embodiment of the present invention.

FIG. 2 is a flowchart showing a flow of processing of a case accumulation function in the case accumulation / retrieval apparatus according to the first embodiment of the present invention.

FIG. 3 is a diagram illustrating an example of the configuration of a digitized document file.

FIG. 4 is a flowchart showing a flow of a process target sentence extraction process (step S1).

FIG. 5 is a flowchart showing the flow of processing target sentence extraction processing (step S10) for one document.

FIG. 6 is a diagram illustrating an example of document field information in which information about a document field such as a fixed field or an irregular field is described in advance.

FIG. 7 is a flowchart showing a flow of processing (step S17 in the flowchart shown in FIG. 5) for an irregular field in the digitized document file 54;

FIG. 8 is a diagram showing an example of a sentence tag list.

9 is a diagram illustrating an example of a “case” generated from the document 55 of the digitized document file 54 illustrated in FIG. 3 by the processing target sentence extraction process (step S1).

FIG. 10 is a flowchart showing a flow of a sentence analysis process (step S2).

FIG. 11 is a flowchart showing a flow of a morphological analysis process (step S22), and is a configuration diagram showing a configuration of a morphological analysis unit (not shown) in a sentence analysis unit 47 in the case accumulation / search device.

12 is a flowchart showing a flow of a syntax analysis process (step S23) in the flowchart shown in FIG. 10, and a syntax analysis unit in a sentence analysis unit 45 in the case accumulation / analysis device shown in FIG. (Not shown)
FIG. 2 is a configuration diagram showing the configuration of FIG.

FIG. 13 is a diagram showing an example of a phrase structure 89.

14 is a problem solving tree 9 composed of a cluster including a set of similar case sentences generated from the digitized document file 54. FIG.
FIG. 9 is a diagram showing an example of No. 5;

FIG. 15 is a diagram illustrating an example of a dependency structure of two sentences.

FIG. 16 is a diagram showing an example of a penalty calculation rule in similarity calculation.

FIG. 17 is a diagram showing an example of knowledge about an area stored in an area ontology 44.

FIG. 18 is a similar sentence clustering process (step S3).
3 is a flowchart showing the flow of the processing of FIG.

FIG. 19 is a flowchart showing the flow of a clustering process (step S27).

FIG. 20 is a flowchart showing the flow of a clustering sub-process (step S30).

FIG. 21 is a diagram illustrating an example of a similarity table storing similarities between clusters when a case sentence is subjected to clustering processing.

FIG. 22 shows a process (step S28) for storing the problem solving tree 95 (see FIG. 14) in the flowchart shown in FIG. 18 in the case database 50 (see FIG. 2).
FIG. 7 is a diagram illustrating an example of inter-cluster information other than a hierarchical relationship stored in a case database 50;

FIG. 23 shows a process (step S28) for storing the problem solving tree 95 (see FIG. 14) in the case database 50 (see FIG. 2) in the flowchart shown in FIG.
FIG. 7 is a diagram illustrating an example of inter-cluster information other than a hierarchical relationship stored in a case database 50;

FIG. 24 is an illustration of case accumulation and storage according to the first embodiment of the present invention.
It is a flowchart which shows the flow of a process of the case search function in a search device.

FIG. 25 is a flowchart showing a flow of a similar case search process (step S42) in the flowchart shown in FIG. 24;

FIG. 26 is a configuration diagram illustrating a configuration of a similar case search device according to the related art 1.

FIG. 27 is a configuration diagram showing a configuration of a knowledge base device according to the conventional technology 2.

FIG. 28 is a configuration diagram showing a configuration of a document information search device according to a conventional technology 3.

FIG. 29 is a configuration diagram showing a configuration of an input analysis unit 28 in FIG. 28;

[Explanation of symbols]

 44 area ontology 45 processing target sentence extraction unit 46 search sentence input unit 47 sentence analysis unit 48 similar sentence matching sentence 49 similar sentence clustering unit 50 case database 51 case cluster editing unit 52 similar case search unit 53 search result display unit Step S1 processing target Sentence extraction processing step S2 Sentence analysis processing step S3 Similar sentence clustering processing step S4 Case cluster editing processing step S41 Search sentence input processing step S42 Similar case search processing step S43 Search result display processing step The same reference numerals indicate the same or corresponding parts.

 ──────────────────────────────────────────────────の Continuing on the front page (72) Inventor, Yoshiyuki Aikawa 2-3-2 Marunouchi, Chiyoda-ku, Tokyo Inside Mitsubishi Electric Corporation (72) Inventor Yamahiko Ito 2-3-2, Marunouchi, Chiyoda-ku, Tokyo Mitsui Electric Co., Ltd. F-term (reference) And a computer-readable recording medium recording a case accumulation program and a computer-readable recording medium recording a case retrieval program

Claims (12)

[Claims] 1. An area ontology in which knowledge relating to a term depending on an area to be stored and searched for and a relation between the terms is stored in advance, and the case accumulation is performed from an electronic document described in a natural language. A processing target sentence extracting unit that cuts out each case sentence to be searched; a search input unit that inputs a search sentence for searching for a case sentence desired by the user; The morphological analysis and the syntactic analysis are performed by referring to the knowledge stored in the area ontology with the input of each case sentence or the search sentence input by the search input means as input, and Sentence analysis means for creating a structure, and knowledge stored in the area ontology with the structure of each case sentence or the search sentence created by the sentence analysis means as input A similarity sentence matching unit that obtains a similarity between the case sentences or the similarity between each of the case sentences and the search sentence with reference to the reference sentence; and each of the cases obtained by the similar sentence matching unit. Similar sentence clustering means for classifying each of the case sentences based on the similarity between sentences to form a case cluster, and creating case data composed of the information of the case clusters and the relationship information between the case clusters A case database storing case data created by the similar sentence clustering unit; and the search sentence based on a similarity between each of the case sentences and the search sentence obtained by the similar sentence matching unit. Similar case search means for searching a similar case sentence similar to the search sentence input by the input means from the case data stored in the case database; The searched by similar case retrieval means a similar case statement to display the search result display unit and case storage and retrieval apparatus comprising the. 2. A processing target sentence extracting step of extracting each case sentence to be stored as an object from an electronic document described in a natural language, and each case extracted by the processing target extracting step A morphological analysis and a syntactic analysis are performed by referring to an area ontology in which a sentence is input and knowledge on a term depending on an area to be stored in the case and a relationship between the terms is stored in advance, and a structure of each of the case sentences is performed. A sentence analysis step of creating, and inputting the structure of each case sentence created by the sentence analysis step as input and referring to knowledge stored in the area ontology to obtain a similarity between the case sentences A sentence matching step, and classifying each of the case sentences based on the degree of similarity between the case sentences obtained in the similar sentence matching step. And a similar sentence clustering step of generating case data composed of the information of the case clusters and the relationship information between the case clusters and storing the same in a case database. . 3. The case storage method according to claim 2, wherein the processing target sentence extracting step assigns a case sentence type to the extracted case sentence. 4. The similar sentence clustering step includes the step of classifying each of the case sentences and the number of clusters constituting the case sentences and / or the number of clusters used when judging that the case sentences are similar to each other. The case accumulation method according to claim 2, wherein a threshold value of the similarity is designated. 5. The domain ontology describes IS-A relation knowledge describing a semantic upper-lower relation, HAS-A relation knowledge describing a semantic part-whole relation, and describes a relation between concepts. Claims characterized by describing at least one kind of knowledge out of case relation knowledge, paraphrase knowledge describing one meaning / concept / relation in a plurality of expressions, and knowledge describing conflicting relations that cannot occur at the same time. Item 2. The case accumulation method according to Item 2. 6. The similar sentence collating step includes collating a semantic structure between the case sentences based on an attribute of a syntactic element in a structure of each case sentence created in the sentence analyzing step. The case accumulation method according to claim 2, wherein the degree is obtained. 7. The case storing method according to claim 2, wherein the similar sentence matching step designates a level of detail of matching used when obtaining a similarity between the case sentences. 8. The sentence analyzing step creates the structure of the case sentence as a tree structure, and the similar sentence matching step sets a detail level of matching used when calculating the similarity between the case sentences. 9. The article example sentence is designated by a tree structure depth.
Case accumulation method described in. 9. The case storage method according to claim 2, wherein the case database describes a similarity relationship or a contradiction relationship as a relationship between the case clusters. 10. A search inputting step of inputting a search sentence for searching a case sentence desired by a user from a case database storing case data to be searched for a case, and inputting by the search inputting step. The structure of the search sentence is obtained by performing a morphological analysis and a syntax analysis by referring to an area ontology in which a search sentence is input and a knowledge on a term depending on an area to be searched for the case and a relationship between the terms is stored in advance. A sentence analysis step to be created, and a similar sentence matching for obtaining a similarity between each of the case sentences and the search sentence by referring to the area ontology with the structure of the search sentence created by the sentence analysis step as an input. And a search sentence input based on the similarity between each of the case sentences and the search sentence obtained in the similar sentence matching step. A similar case search step of searching a similar case sentence similar to the search sentence input in the step from the case data stored in the case database; and a search displaying the similar case sentence searched by the similar case search step. And a result display step. 11. A processing target sentence extraction procedure for extracting each case sentence to be stored as an object from an electronic document described in a natural language, and each case extracted by the processing target extraction procedure A morphological analysis and a syntactic analysis are performed by referring to an area ontology in which a sentence is input and knowledge on a term depending on an area to be stored in the case and a relationship between the terms is stored in advance, and a structure of each of the case sentences is performed. A sentence analysis procedure for creating the case sentence, and a similarity for obtaining a similarity between the case sentences by referring to the knowledge stored in the area ontology with the structure of each case sentence created by the sentence analysis procedure as an input. A sentence matching procedure, and classifying each of the case sentences based on the similarity between the case sentences obtained by the similar sentence matching procedure to form a case cluster; And a similar sentence clustering procedure for creating case data composed of data of the case data and the relationship information between the case clusters and storing the same in a case database. Possible storage medium. 12. A search input procedure for inputting a search sentence for searching for a case sentence desired by a user from a case database storing case data to be searched for a case, and a search input procedure input by the search input procedure. The structure of the search sentence is obtained by performing a morphological analysis and a syntax analysis by referring to an area ontology in which a search sentence is input and a knowledge on a term depending on an area to be searched for the case and a relationship between the terms is stored in advance. A sentence analysis procedure to be created, and a similarity between each of the case sentences and the search sentence with reference to the knowledge stored in the area ontology with the structure of the search sentence created by the sentence analysis procedure as input. A similar sentence matching procedure for calculating the similar sentence, and the search sentence input procedure based on the similarity between each of the case sentences and the search sentence obtained by the similar sentence matching procedure. Similar case search procedure for searching a similar case sentence similar to the input search sentence from the case data stored in the case database, and a search for displaying the similar case sentence searched by the similar case search procedure And a result display procedure. A computer-readable storage medium storing a case search program characterized by comprising a result display procedure.