JP5046170B2 - SEARCH SYSTEM, SEARCH METHOD, REPORT SYSTEM, REPORT METHOD, AND PROGRAM - Google Patents

Description

  The present invention relates to a search system, a search method, a report system, a report method, and a program. In particular, the present invention relates to a search system, a search method, a report system, a report method, and a program that search document data from a plurality of document data and report that the number of document data having a specific concept is increasing.

  Conventionally, as a search system that searches document data that includes the contents specified by the input search sentence from multiple document data, the search intention is reflected even if the search sentence itself is not completely included in the document data. Search systems that search for appropriate documents are being studied. Such a search system, for example, in a product manufacturer, creates a call log database in which an inquiry about a product sent to a call center and an answer to the inquiry are converted into text as document data, and the answer to the inquiry is made using this database. Can be utilized as a basic technology of a support system for appropriately performing (see Non-Patent Document 5).

  As an example of such a search system, a system that considers ambiguity when extracting a keyword of an independent word from a search sentence or document data used for search has been proposed (see Non-Patent Documents 1, 2, and 6). . In order to perform a search more accurately, a keyword incorporating a meaning expressed by an attached word has been proposed (see Non-Patent Document 5). Furthermore, not only whether or not a keyword is included in a search sentence or document data, but also taking into account dependency between words has been proposed (see Non-Patent Document 4, Patent Documents 1 and 2). As a system for outputting an answer to a question sentence, a system for learning based on an example of a correct answer to a question has been proposed (see Non-Patent Document 3).

  It is also important for companies to establish relationships of trust with customers and further improve product quality and customer support. For this reason, it is desirable for companies to detect problems in products and services at an early stage, and it is expected that call logs of call centers will be used as means for finding problems.

Non-patent document 7 has been proposed as a method of detecting a problem from information sequentially accumulated in this way. In addition, as an example of such a method, a system has been proposed in which a problem is detected by determining a portion of a document stream where a document input interval related to a specific keyword is small (see Non-Patent Document 8). . Further, in this determination, the number of writing per time is taken into consideration (see Non-Patent Document 9), or a warning is issued when the number of appearances of a specific topic exceeds a threshold (see Non-Patent Document 10). ), And those that detect that the frequency of keywords has increased and extract a sudden topic (see Non-Patent Document 11) have been proposed. Moreover, what performs a predictive analysis using the case of the known malfunction in a product etc. is proposed (refer nonpatent literature 12).
[Prior art documents]
[Patent Document 1] Japanese Patent Application Laid-Open No. 11-259524 [Patent Document 2] Japanese Patent 3266586 [Non-Patent Document 1] JUSTSYSTEM, “What is ConceptBase Technology”, [online], July 30, 2003, JUSTSYSTEM, [Search June 30, 2004], Internet <URL: http://www.justsystem.co.jp/km/whats/search_q_104.html>
[Non-Patent Document 2] NRI, “About Services (NRI Cyber Patent)”, [online], [Search June 30, 2004], Internet <URL: http://www.patent.ne.jp/01gaiyo /s-point/06.html>
[Non-patent document 3] Sasaki et al., “Learning-type question answering system SAIQA-II using SVM”, IPSJ Journal, Vol. 45, No 02, 2004 [Non-patent document 4] Matsumura et al., “Word Evaluation of Information Retrieval Method Using Dependency Relationship between Information Processing, ”IPSJ Journal, Vol. 41, No. SIG01-003, 2000 [Non-Patent Document 5] T. Nasukawa and T. Nagano,“ Text analysis and knowledge mining system ", IBM Systems Journal, Vol. 40, No. 4, 2001 [Non-Patent Document 6] Autonomy," Conceptual Search ", [online], [Search June 30, 2004], Internet < URL: http://www.autonomy.com/c/content/Products/IDOL/f/Conceptual_Search>
[Non-Patent Document 7] T. Fawcett and F. Provost, "Activity monitoring: Noticing interesting changes in behavior.", In Proc. Fifth ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pages 53--62, 1999 [ Non-Patent Document 8] Jon Kleinberg, "Bursty and hierarchical structure in streams", In Proc. The 8th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 2002 [Non-patent document 9] Fujiaki Yasuaki, Minamino Yasuyuki, Suzuki Yasuhiro, Okumura Manabu, "Discovery of burst in document stream", Information Processing Society of Japan, 2004- NL-160, p.85-92
[Non-Patent Document 10] Kenji Yamanishi, “Text Mining and NLP Business”, [online], NEC, [Search June 30, 2004], Internet <URL: http://it.jeita.or.jp/ eltech / committee / knowledge / PDF / 2003 / Yamanishi.pdf
[Non-Patent Document 11] Nomura Research Institute, “What is True Teller?”, [Online], [Search June 30, 2004], Internet <URL: http://www.trueteller.net/about/ index.shtml>
[Non-Patent Document 12] JUSTSYSTEM, “Alize”, [online], [searched on June 30, 2004], Internet <URL: http://www.justsystem.co.jp/km/ssm>

  In such a field, it is desired that the staff who receives the call inputs the contents of the inquiry as a search sentence and efficiently searches the document data according to the search intention.

  In a search system that considers ambiguity in keyword extraction, when only an independent word is targeted as a keyword, for example, “hard disk” and “recognition” are extracted from a search sentence “does not recognize hard disk”. As a result, the search intention “not recognized” is lost, and even document data “recognized” is searched.

  In addition, when an ancillary word is considered as a keyword, “hard disk” and “recognition [denial]” are extracted from a search sentence “does not recognize hard disk”, and the search intention “does not recognize” is reflected. However, because the search is performed based on whether or not the specified keyword appears in the document, document data “CD-ROM cannot be recognized but hard disk is recognized” is searched.

  In addition, even when taking into account dependency between words, it is difficult to match various expression formats that express search intent, for example, “Unable to recognize hard disk” and “Invisible hard disk”. It is. Because, even if each word is expanded in the range of synonyms and the search sentence is semantically analyzed, expressions (word combinations) used only in a specific situation such as “I can't see the hard disk” cannot be properly distinguished. It is.

  Further, when call center call logs are used as means for finding problems, it is difficult to isolate problems in word-based processing because there are few words expressing individual problems. Further, it is impossible to determine what kind of problem is caused by the keyword for which the increase has been reported. In addition, for new products, the number of calls for all problems tends to increase, but it is difficult to find a specific problem early in such a situation.

  Accordingly, an object of the present invention is to provide a search system, a search method, a report system, a report method, and a program that can solve the above-described problems. This object is achieved by a combination of features described in the independent claims. The dependent claims define further advantageous specific examples of the present invention.

  According to a first aspect of the present invention, there is provided a search system for searching document data including content specified by an input search sentence from a plurality of document data, the document database storing the plurality of document data, Based on a concept database that stores a plurality of predetermined concepts and a keyword included in each of the document data by using a hierarchical structure in which another concept including the one concept is an upper hierarchy of the one concept. A document data concept extracting unit that extracts the document concept that is the concept corresponding to the document data, and a search sentence concept that is the concept corresponding to the search sentence based on a keyword included in the search sentence The search sentence concept extraction unit, and the search sentence concept of each of the plurality of document data is a concept of an upper hierarchy or a lower hierarchy of the document concept A search system comprising: a concept search unit that searches for document data; and a search result output unit that outputs the document data searched by the concept search unit as the document data including content specified by the search sentence; A search method, a program, and a recording medium relating to the search system are provided.

  According to the second aspect of the present invention, there is provided a reporting system in which a plurality of document data is sequentially input, the document database storing the input document data sequentially, and other concepts including one concept. A concept database that stores a plurality of predetermined concepts, and a document concept that is the concept corresponding to the document data, based on a keyword included in each of the document data. A document data concept extraction unit that extracts the document data, a concept ratio calculation unit that calculates a ratio of the number of document data corresponding to each concept to the number of document data in the document database, and each of the concepts A relative frequency calculation unit that calculates a relative frequency indicating the magnitude of the ratio calculated by the concept ratio calculation unit with respect to a corresponding reference ratio; Of a plurality of concepts, a frequent concept selection unit that selects the concept having a relative frequency equal to or higher than a predetermined threshold, the first concept selected by the frequent concept selection unit, and the first A priority concept selection unit that selects one of the second concept and the second concept in the upper hierarchy of the concept based on the relative frequency of the first concept and the second concept; and the first concept or the first concept A reporting system including a notifying unit for notifying a user that the relative frequency of the concept selected by the priority concept selecting unit among the two concepts is high, and a reporting method and program related to the reporting system And a recording medium.

  The above summary of the invention does not enumerate all the necessary features of the present invention, and sub-combinations of these feature groups can also be the invention.

  According to the present invention, it is possible to search document data by appropriately reflecting the contents of a search sentence and to appropriately detect occurrence of a problem from sequentially added document data.

  Hereinafter, the present invention will be described through embodiments of the invention. However, the following embodiments do not limit the invention according to the scope of claims, and all combinations of features described in the embodiments are included. It is not necessarily essential for the solution of the invention.

  FIG. 1 shows a configuration of a search system 10 according to an embodiment of the present invention. The search system 10 appropriately searches document data including contents specified by a search sentence input by a user or the like from a plurality of document data. In the present embodiment, for example, the search system 10 searches document data in which an inquiry about a product sent to a call center and a response to the inquiry are converted into text.

  The search system 10 includes a document DB 100, a concept DB 105 (concept database), a product DB 106 (product database), a component DB 107 (component database), a dictionary DB 110 (dictionary database), and a synonym DB 115 (synonym database). , Document data normalization unit 120, concept extraction rule DB 125, document data concept extraction unit 130, search index DB 135 (search index database), search sentence normalization part 140, search sentence concept extraction part 145, concept A search unit 150, a concept selection support unit 155, and a search result output unit 160 are provided.

  The document DB 100 stores a plurality of document data. In the present embodiment, the document DB 100 stores, for each of a plurality of defects related to a product, document data including the contents of the problem, for example, contents of an inquiry from a user of the product and an answer to the inquiry. This document data is sequentially accumulated every time an inquiry and response are made.

  The concept DB 105 stores a plurality of predetermined concepts in a hierarchical structure in which another concept including the one concept is an upper hierarchy of the one concept. Here, the concept is individual information that is defined in advance by the manufacturer or user of the search system 10 and systematically classifies the semantic content of the text handled by the search system 10. The concept DB 105 according to the present embodiment stores a plurality of concepts that specify a plurality of defects about a product.

  The product DB 106 stores an inclusion relationship of product names of a plurality of products in a hierarchical structure. The component element DB 107 stores the inclusion relationship of each component element of the product in a hierarchical structure. The dictionary DB 110 stores a dictionary in which word parts of speech and regular expressions are described. The synonym DB 115 stores a correspondence between a predetermined phrase and a keyword that is a synonym of the phrase.

  The document data normalization unit 120 uses the dictionary DB 110 to perform morphological analysis and syntax analysis of each document data stored in the document DB 100. Further, the document data normalization unit 120 normalizes the document data by using the synonym DB 115 to replace a phrase included in each document data with a keyword that is a synonym of the phrase.

  The concept extraction rule DB 125 stores a concept extraction rule including a set of one or more keywords and a concept indicating the meaning content of the one or more keywords. The document data concept extraction unit 130 extracts a document concept that is a concept corresponding to the document data based on a keyword included in each document data. The document data concept extraction unit 130 according to the present embodiment applies the concept extraction rules stored in the concept extraction rule DB 125 to one or more keywords included in the document data, and the concepts corresponding to the one or more keywords. Convert to The search index DB 135 stores, for each document data, a correspondence between the document concept of the document data extracted by the document data concept extraction unit 130 and the document data.

  Similar to the document data normalization unit 120, the search sentence normalization unit 140 performs morphological analysis and syntax analysis of the search sentence and normalizes the search sentence. Similar to the document data normalization unit 120, the search sentence concept extraction unit 145 extracts a search sentence concept that is a concept corresponding to the search sentence based on a keyword included in the search sentence.

  The concept search unit 150 uses the search index DB 135 to search for document data having a document concept corresponding to the search statement concept among a plurality of document data. At this time, the concept search unit 150 searches for document data in which the search sentence concept is an upper hierarchy or lower hierarchy concept of the document concept among each of the plurality of document data. The concept selection support unit 155 supports high-level conceptualization and / or low-level conceptualization of a search sentence concept based on an instruction from a user of the search system 10.

  The search result output unit 160 outputs the document data searched by the concept search unit 150 as document data including contents specified by the search sentence.

  According to the search system 10 described above, concept inclusion relationships are systematically defined by a predetermined hierarchical structure, and document data is searched in consideration of the search statement concept and document concept inclusion relationship. Can do. As a result, it is possible to perform a search that appropriately maps the search text and document data to the concept and appropriately reflects the contents of the search text.

FIG. 2 shows an example of a concept hierarchy of defects stored in the concept DB 105 according to the embodiment of the present invention.
The concept DB 105 according to the present embodiment stores, as an example, a hierarchical structure in which a concept for specifying a certain thing or component is set as an upper hierarchy, and a concept for explaining the state or the like of the thing or component is set as a lower hierarchy.

  The concept hierarchy of defects is a systematic summary of each concept that identifies each defect that can occur in a product, with a hierarchical structure in which other concepts that semantically include one concept are the upper hierarchy of one concept. It is. For example, “hardware” which is a node in the hierarchical structure of FIG. 2 is a concept of meaning content “hardware has a defect” and is represented as “/ fault / hardware”. “Hard disk” is a concept of the meaning content “a hard disk (which is a part) has a defect” and is represented as “/ failure / hardware / hard disk”.

  In the present embodiment, the concept of the lower hierarchy is semantically included in the concept of the upper hierarchy of the concept. For example, “/ defect / hardware / hard disk”, which is a concept indicating that “a hardware (part of) hard disk has a defect”, is a concept indicating “defect in hardware” “/ defect. / Hardware ”and has an inclusive relationship in which the concept of the upper hierarchy is also established when the concept of the lower hierarchy is established. Similarly, “/ defect / hardware / display / flicker”, which is a concept indicating that “the display of the hardware (which is a part) flickers”, This is a form of “/ defect / hardware / display”, which is a concept indicating “defect”, and is included in the superordinate concept.

  Further, in the present embodiment, the concept DB 105 has a concept (eg ““ defect / hardware ”) indicating the status of the malfunction of the component in a lower hierarchy of the concept (eg,“ defect / hardware ”) indicating that the product component is defective. A plurality of concepts are stored in a hierarchical structure provided with a component (for example, “failure / hardware / hard disk”) indicating a component of a part of the component (for example, “failure / hardware / noise”).

FIG. 3 shows an example of a conceptual hierarchy of products stored in the product DB 106 according to the embodiment of the present invention.
The product concept hierarchy is a systematic collection of concept inclusions indicating product names of a plurality of products in a hierarchical structure. In the present embodiment, similarly to the concept hierarchy of defects, a hierarchical structure is used in which another concept including one concept is an upper hierarchy of the one concept. For example, “/ product / PC product / notebook / series A” in FIG. 3 is “/ product / PC product / notebook / series A / model A1” and “/ product / PC product / notebook / series A / model A2”. It is the concept which shows the generic product name which includes.

  The concept hierarchy of the product shown above may be stored in the concept DB 105 as a hierarchy structure different from the hierarchy structure regarding defects. In this case, the concept DB 105 stores each of the plurality of concepts as a node of any of a plurality of different hierarchical structures (first hierarchical structure, second hierarchical structure,...).

FIG. 4 shows an example of the conceptual hierarchy of the components stored in the component DB 107 according to the embodiment of the present invention.
The concept hierarchy of the component is a systematic summary of the inclusion relationship of each concept indicating each component constituting the product by a hierarchical structure. In the present embodiment, similarly to the concept hierarchy of defects, a hierarchical structure is used in which another concept including one concept is an upper hierarchy of the one concept. For example, “component / hardware” in FIG. 4 includes “component / hardware / hard disk”, “component / hardware / CPU”, “component / hardware / CD drive”, and “component / This is a concept indicating components including “hardware / keyboard”.

  The concept hierarchies of the constituent elements described above may be stored in the concept DB 105 as a hierarchical structure different from the hierarchical structure regarding the defects.

FIG. 5 shows an operation flow of the search system 10 according to the embodiment of the present invention.
First, the document data normalization unit 120 uses the dictionary DB 110 to perform text analysis such as morphological analysis and structural analysis of each document data stored in the document DB 100 (S500). Next, the document data normalization unit 120 normalizes the document data by replacing each word / phrase extracted as a result of the text analysis with a keyword corresponding to the original text of the word / phrase (S505). At this time, the document data normalization unit 120 performs normalization using this concept when a concept uniquely determined corresponding to the word is recorded in the dictionary.

  Next, the document data concept extraction unit 130 extracts the concept of the document data (document concept) based on the keyword included in each document data (S510). When the concept DB 105 stores a plurality of concept hierarchical structures, the document data concept extraction unit 130 extracts document concepts belonging to the hierarchical structure for each hierarchical structure corresponding to the document data. That is, for example, when the concept DB 105 stores each concept as a node of the first hierarchical structure or the second hierarchical structure, the document data concept extraction unit 130 corresponds to each document data in the first hierarchical structure. The first document concept that belongs to and the second document concept that belongs to the second hierarchical structure may be extracted. Further, the document data concept extraction unit 130 generates a document concept indicating a product name described in the document data and / or a document concept indicating a component described in the document data based on a keyword included in the document data. Further extraction may be performed. Then, the document data concept extraction unit 130 creates a search index in which the document concept of the document data is added to information for identifying each document data, and stores the search index in the search index DB 135 (S515).

  When a search sentence is input (S520), the search sentence normalization unit 140 performs text analysis of the search sentence in the same manner as the document data normalization unit 120 (S525). Next, the search sentence normalization unit 140 normalizes the search sentence by replacing a phrase included in the search sentence with a keyword that is a synonym of the phrase (S530).

  Next, the search sentence concept extraction unit 145 extracts the search sentence concept from the search sentence in the same manner as the document data concept extraction unit 130 (S535). In the present embodiment, the search sentence concept extraction unit 145 extracts a search sentence concept corresponding to the search sentence for searching for defects in the product input by the user in S520. In addition, the search sentence concept extraction unit 145 performs a search sentence concept indicating a component described in the search sentence and / or a search sentence indicating a product name described in the search sentence based on a keyword included in the search sentence. Concepts may be further extracted.

  Next, the concept retrieval unit 150 performs document data concept retrieval based on the retrieval sentence concept extracted from the retrieval sentence and the document concept extracted from the document data (S540). More specifically, the concept search unit 150 searches the document data for the search sentence when the document concept is the same as the search sentence concept or when the document concept is a concept of the upper hierarchy or lower hierarchy of the search sentence concept. Is selected as document data corresponding to. At this time, the concept search unit 150 corresponds to the document concept when the search sentence concept becomes a concept of an upper hierarchy or a lower hierarchy of the document concept stored in the search index DB 135 before the search sentence is input. Document data is output as a search result. As a result, the concept search unit 150 can perform a search based on the document concept that has already been extracted, and can perform a search at a higher speed than when the document concept is extracted from each document data each time the search is performed. .

  Next, the search result output unit 160 outputs the document data searched by the concept search unit 150 as a search result (S545). The search result output unit 160 according to the present embodiment outputs the document data searched by the concept search unit 150 as document data indicating the content of the defect of the product input by the user.

  Then, when the search system 10 next receives an input of a search sentence, the search system 10 advances the process to S520. When new document data is added to the document DB 100, the search system 10 advances the process to S500, extracts the document concept from the document data, and stores it in the search index DB 135.

  According to the search system 10 described above, concept inclusion relationships are systematically defined by a predetermined hierarchical structure, and document data is searched in consideration of the search statement concept and document concept inclusion relationship. Can do. As a result, it is possible to perform a search that appropriately maps the search text and document data to the concept and appropriately reflects the contents of the search text. When such functions are used to create a database of inquiries and answers about products and use them in response to new inquiries, it is desirable to accurately define a limited number of concepts so that they can be searched accurately. Is particularly effective.

FIG. 6 shows an example of normalization rules stored in the synonym DB 115 according to the embodiment of the present invention.
The synonym DB 115 stores the normalization rule illustrated in FIG. 6 in order to normalize the search sentence and the document data. The normalization rule of FIG. 6 replaces phrases such as “turn off” or “turn off” in the expressions “turn off power” or “turn off power” with the keyword “shut down” which is a synonym for these. It is a rule for normalization. When the document data normalization unit 120 and the search statement normalization unit 140 detect an expression in which the subject is “power” and the predicate is “cut” or “drop” as a result of the syntax analysis of the document data or the search statement, Replace predicate with “block”. In addition, the document data normalization unit 120 and the search sentence normalization unit 140 not only replace words but also replace “anger” with “anger”, or change “furious” to “very happen”. Expression normalization such as replacement may be performed based on a normalization rule. Then, the document data concept extraction unit 130 extracts the document concept from the normalized document data, and the search sentence concept extraction unit 145 extracts the search sentence concept from the normalized search sentence.

  In this way, by synchronizing synonyms in advance at the phrase level prior to concept search, the search accuracy of document data corresponding to a search sentence can be increased.

FIG. 7 shows an example of the concept extraction rules stored in the concept extraction rule DB 125 according to the embodiment of the present invention.
The concept extraction rule DB 125 stores the concept extraction rules illustrated in FIG. 7 in order to predefine concepts to be extracted from search sentences and document data. Here, the concept extraction rule is a concept that indicates one or more keywords in the syntax tree based on the sentence syntax (dependency relationship, etc.) obtained by text analysis, and indicates the semantic content of the one or more keywords. It is a rule for converting to. In FIG. 7, the concept is based on the keywords “hard disk” and “recognition” extracted from the sentence “cannot recognize hard disk” and the dependency relationship “reject” (hitei = “1”) of “recognition”. Rules for extracting “/ defect / hardware / hard disk” are defined.

  When one or more keywords included in any one of the concept extraction rules stored in the concept extraction rule DB 125 are included in the document data, the document data concept extraction unit 130 converts the concept included in the concept extraction rule into the document concept. Extract as Similarly, the search sentence concept extraction unit 145, when one or more keywords included in any concept extraction rule stored in the concept extraction rule DB 125 are included in the search sentence, the concept included in the concept extraction rule. Are extracted as a search sentence concept.

In the above, concept extraction rules are defined using not only keywords, dependency, and attributes but also generic terms such as “problem” and “request”, and the document data concept extraction unit 130 and search are performed based on these rules. The concept may be extracted by the sentence concept extraction unit 145. That is, for example, the document data concept extraction unit 130 and the search sentence concept extraction unit 145 may extract the concept “/ defect / hardware / hard disk” from the “hard disk problem”.
The concept extraction rule DB 125 may further store not only the concept hierarchy for defects but also the concept extraction rules for the concept hierarchy of products and components.

  Through the above processing, the document data concept extraction unit 130 and the search sentence concept extraction unit 145 extract “failure / hardware / hard disk”, “product / hardware” from the sentence “notebook series A / model A1 does not recognize hard disk”. Three concepts of “PC product / note / series A / model A1” and “component / hardware / hard disk” can be extracted.

  Thus, according to the search system 10 according to the present embodiment, corresponding concepts can be defined in advance according to one or a plurality of keywords and their dependency relationships. Thereby, natural language sentences can be appropriately converted into a systematic concept according to the application field of the search system 10.

  FIG. 8 shows a configuration of the concept search unit 150 according to the embodiment of the present invention. The concept search unit 150 includes an identical concept output unit 800, a higher concept acquisition unit 810, a generalized concept output unit 820, a lower concept acquisition unit 830, and a specialized concept output unit 840.

  The same concept output unit 800 outputs the document data to the search result output unit 160 as a search result when the search statement concept matches the document concept and the search statement concept is not specialized. When the search sentence concept does not match the document concept, the higher concept acquisition unit 810 acquires a search sentence superordinate concept that is a concept of a higher hierarchy of the search sentence concept. The generalized concept output unit 820 outputs the document data as a search result when the search sentence superordinate concept matches the document concept. The lower concept acquisition unit 830 replaces the search sentence concept with the search sentence subordinate concept when the same document data can be searched even if the search sentence concept is replaced with the search sentence subordinate concept that is a concept of the lower hierarchy. The specialized concept output unit 840 outputs document data whose search sentence subordinate concept matches the document concept as a search result.

FIG. 9 shows an operation flow of the concept search unit 150 according to the embodiment of the present invention.
First, the concept search unit 150 receives one or more search sentence concepts extracted from the search sentence. The concept search unit 150 receives one or more document concepts extracted from the document data for each document data. If the search statement concept matches the document concept (S900: Yes), the same concept output unit 800 advances the process to S940. As a result, the same concept output unit 800 outputs the document data as a search result to the search result output unit 160 on the condition that the search sentence concept cannot be replaced with the search sentence subordinate concept (S945: No) (S910). Here, when a plurality of search sentence concepts and a plurality of document concepts are extracted corresponding to a plurality of hierarchical structures, the same concept output unit 800 indicates that all of the search sentence concepts are the same as any one of the document concepts. In some cases, the document data is output as a search result. For example, if three search sentence concepts of “defect / hardware / hard disk”, “product / PC product / notebook / series A / model A1”, and “component / hardware / hard disk” are extracted, the same concept The output unit 800 outputs document data including all these three concepts as a document concept as a search result.

  On the other hand, if the search sentence concept does not match the document concept (S900: No), the superordinate concept acquisition unit 810 acquires the search sentence superordinate concept (S920). Here, when a plurality of search sentence concepts (for example, the first search sentence concept and the second search sentence concept) are extracted, the higher-level concept acquisition unit 810 performs the first search sentence concept and the second search sentence concept. Are not the same as the first document concept and the second document concept, respectively, the first search sentence superordinate concept in the upper hierarchy of the first search sentence concept and the first hierarchy in the upper hierarchy of the second search sentence concept. 2. The search sentence superordinate concept of 2 is acquired.

  The higher-level concept acquisition unit 810 according to the present embodiment, when there is a search statement concept that belongs to a failure concept hierarchy and indicates a failure state of the component or a failure state of the component, Is acquired as one of the search sentence superordinate concepts. Further, when there is a search sentence concept belonging to the concept hierarchy of the constituent element, a concept that is a higher hierarchy of the search sentence concept indicating the constituent element is acquired as one of the search sentence superordinate concepts. Further, when there is a search sentence concept belonging to the product concept hierarchy, a concept that is a higher hierarchy of the search sentence concept indicating the product name is acquired as one of the search sentence concepts.

  For example, when three search sentence concepts of “defect / hardware / hard disk”, “product / PC product / notebook / series A / model A1”, and “component / hardware / hard disk” are extracted, the superordinate concept The acquisition unit 810 has three search statement superordinate concepts of “defect / hardware”, “product / PC product / note / series A”, and “component / hardware” as a concept DB 105, a product DB 106, and a component DB 107. Get from. As a result, when the search sentence is generalized for one layer using the search sentence superordinate concept, the search sentence is superposed into the following three.

(1) Set of first search sentence superordinate concept “defect / hardware”, search sentence concept “product / PC product / notebook / series A / model A1”, and search sentence concept “component / hardware / hard disk” That is, for example, the document data concept extraction unit 130 extracts a document concept indicating that one component is defective based on a keyword included in the document data, and the search sentence concept extraction unit 145 When a search concept indicating that there is a defect in a part of the one component is extracted based on a keyword included in the sentence, the higher concept acquisition unit 810 is a concept of a higher hierarchy of the search sentence concept. A concept indicating that one component has a defect is acquired as a search sentence superordinate concept.

  Similarly, the document data concept extraction unit 130 extracts a document concept indicating that one component is defective based on a keyword included in the document data, and the search sentence concept extraction unit 145 includes the search sentence. When a search sentence concept indicating a failure state of one constituent element is extracted based on the keyword to be searched, the higher-level concept acquisition unit 810 has a defect in one constituent element that is a concept of a higher hierarchy of the search sentence concept. A concept indicating this is acquired as a search sentence superordinate concept. As a result, the search result output unit 160 can output, as a search result, document data having a document concept that matches the search sentence superordinate concept and indicates that one component has a defect.

(2) From the combination of search sentence concept “defect / hardware / hard disk”, second search sentence superordinate concept “product / PC product / note / series A”, and search sentence concept “component / hardware / hard disk” That is, for example, the document data concept extraction unit 130 extracts a document concept indicating one product name based on a keyword included in the document data, and the search sentence concept extraction unit 145 selects a keyword included in the search sentence. Based on this, when the search sentence concept indicating the product name in the lower hierarchy of the product name is extracted, the upper concept acquisition unit 810 acquires the concept corresponding to the product name in the upper hierarchy of the search sentence concept as the search sentence upper concept. .

(3) Search sentence concept “defect / hardware / hard disk”, search sentence concept “product / PC product / notebook / series A / model A1”, and third search sentence superordinate concept “component / hardware”
That is, for example, the document data concept extracting unit 130 extracts a document concept indicating one component based on a keyword included in the document data, and the search sentence concept extracting unit 145 is based on the keyword included in the search sentence. When a search sentence concept indicating some constituent elements of the constituent element is extracted, the upper concept acquisition unit 810 acquires a concept corresponding to a constituent element of an upper hierarchy of the search sentence concept as a search sentence upper concept.

  In the above, the higher concept acquisition unit 810 may acquire a concept that is positioned in a higher hierarchy for a plurality of hierarchies with respect to the search sentence concept as a search sentence superordinate concept. In this case, the superordinate concept acquisition unit 810 sequentially replaces the search sentence concept with the concept of the upper hierarchy until the search sentence superordinate concept matches the document concept, and the search sentence superordinate concept of any hierarchy matches the document concept. In such a case (S930: Yes), it may be determined to use the search sentence superordinate concept. In addition, when a combination of a plurality of types of search sentence concepts and / or search sentence superordinate concepts is acquired corresponding to the search sentence, the generalized concept output unit 820 displays an appropriate search sentence concept and / or search sentence superordinate concept. A combination is selected (S935).

  In this processing, the generalized concept output unit 820 selects a search sentence superordinate concept that can search document data having a higher amount of information. That is, for example, the generalization concept output unit 820 includes the first document data in which the first search sentence superordinate concept and the first document concept, and the second search sentence concept and the second document concept are the same. When the number is smaller than the number of second document data in which the first search sentence concept and the first document concept and the second search sentence superordinate concept and the second document concept are the same, the first Document data is output as a search result. As a result, the generalization concept output unit 820 can more appropriately select and output document data to be searched.

Next, the lower concept acquisition unit 830 acquires a search sentence subordinate concept that is a search sentence concept in S900 or a concept in a lower hierarchy of the search sentence concept obtained in S935 (S940). Here, if there is no search sentence subordinate concept of the search sentence concept obtained in S935 (S945: No), the generalized concept output unit 820 has document data having a document concept that matches the search sentence superordinate concept (S930). Reference) is output to the search result output unit 160 as a search result (S910).
Here, in the case where a plurality of search sentence superordinate concepts are acquired, when at least one search sentence concept is set as a high-order search sentence superordinate concept, all search sentence concepts or search sentence concepts replaced with search sentence superordinate concepts Document data having a document concept that matches the superordinate concept is output as a search result. That is, for example, when the first search sentence superordinate concept and the second search sentence superordinate concept are acquired, the generalized concept output unit 820 displays at least one of the first search sentence concept and the second search sentence concept. In the case of a higher-level concept, document data having a document concept that matches the first search sentence concept and a document concept that matches the second search sentence concept is output as a search result.

  On the other hand, when a search sentence subordinate concept exists (S945: Yes), the subordinate concept acquisition unit 830 searches the search sentence in which all document data having the same document concept as the search sentence concept is a concept in a lower hierarchy of the search sentence concept. On condition that the document concept is the same as the subordinate concept (S950: Yes), the search sentence concept is replaced with the search sentence subordinate concept, and the process proceeds to S940 (S950: Yes). Then, the lower-level concept acquisition unit 830 performs the processing of S940 and S945 again to further specialize the search sentence concept.

  In this way, the lower-level concept acquisition unit 830 sequentially replaces the search sentence concept with the lower-level concept until the above-described condition is not satisfied (S950: No). Thereby, the lower concept acquisition unit 830 can select a concept located in a lower hierarchy for a plurality of hierarchies with respect to the search sentence concept as a search sentence lower concept. Therefore, the lower-level concept acquisition unit 830 can select an appropriate search sentence concept according to the document concept of each document data stored in the search index DB 135.

  Then, the specialized concept output unit 840 outputs document data (S950: No) whose search sentence subordinate concept matches the document concept as a search result (S910).

  According to the concept search unit 150 described above, the search target document data can be appropriately searched by converting the search sentence concept into a higher concept or a lower concept according to the search result.

  As a result of the processing described above, if all of the obtained search sentence concepts do not have a lower hierarchy concept, the search result output unit 160 selects a document concept that matches the search sentence concept in S545 of FIG. A list of document data that the user has is displayed. On the other hand, if any one of the search sentence concepts has two or more subordinate concepts, the concept selection support unit 155 determines the number of document data to be searched when the search sentence concept is each of two or more subordinate concepts. May be displayed to the user, and any subordinate concept may be selected by the user. Instead, the search result output unit 160 may select any subordinate concept based on the number of document data to be searched, for example, based on a criterion such as reducing entropy.

FIG. 10 shows an example of generalization / specialization by the concept search unit 150 according to the embodiment of the present invention.
When there is no document data having a document concept that matches the search sentence concept, the higher-level concept acquisition unit 810 generalizes the search sentence concept as shown in S920 of FIG. For example, if the search sentence concept is “... / Series A / model A3” in the figure, the same concept output unit 800 will have a document having the same document concept as the search sentence concept “. No data can be extracted. Therefore, the higher-level concept acquisition unit 810 replaces the search sentence concept “.. ./Series A / model A3” with the higher-level search sentence higher level concept “... / Series A” and generalizes it.

  By this generalization, five pieces of document data having the same document concept as the search sentence superordinate concept can be extracted. Here, in the case of the figure, the number of document data corresponding to “.. ./Series A” is the same as the number of document data corresponding to “. It can be seen that all document data having the same document concept as the search sentence concept (that is, the search sentence superordinate concept) has the same document concept as the search sentence subordinate concept “... / Series A / model A2”. As described above, when all of the document data having the same document concept as the search sentence concept has the same document concept as the search sentence subordinate concept that is a concept of the lower hierarchy of the search sentence concept, the lower concept obtaining unit 830 Then, the search sentence concept is replaced with the search sentence subordinate concept and specialized. Thereby, the lower level concept acquisition unit 830 can uniquely specialize the search sentence concept.

  When a plurality of search sentence concepts are extracted, the lower-level concept acquisition unit 830 searches for the same document data if one or more search sentence concepts are set as lower-level concepts and can search for the same document data. You may specialize by replacing the sentence concept with a subordinate concept.

FIG. 11 shows an example of a display screen 1100 of the search system 10 according to the embodiment of the present invention.
The display screen 1100 includes a search text input screen 1110, a concept operation screen 1130, and a search result output screen 1160. The search text input screen 1110 is a screen for allowing a user of the search system 10 to input a search text. The search sentence normalization unit 140 causes the product model name and the search sentence to be input using the search sentence input screen 1110, and receives a search start instruction from the search button. Further, when generalization or specialization is performed by the concept search unit 150, the search sentence input screen 1110 displays that fact, for example, “Series A model A3 has been generalized to series A”.

  The concept operation screen 1130 displays each search sentence concept extracted from the search sentence based on an instruction from the concept selection support unit 155. When a plurality of search statement concepts are extracted, the concept selection support unit 155 displays the relationship (AND condition, OR condition) of these search statement concepts on the concept operation screen 1130. The concept selection support unit 155 may display the number (frequency) of document data having the same document concept as each search sentence concept. When the concept selection support unit 155 receives an instruction to delete the search sentence concept with the search sentence concept deletion button, the concept selection support unit 155 removes the search sentence concept from the search condition. Further, when an instruction to replace the search sentence concept with a higher-order search sentence superordinate concept is received by the superordinate concept button, the search sentence concept is replaced with the search sentence superordinate concept. The concept selection support unit 155 displays the search sentence concept via the concept operation screen 1130 and receives an operation on each search sentence concept, thereby supporting higher-level concept and / or lower-level conception of the search sentence concept.

  The search result output screen 1160 displays search results output by the search result output unit 160.

As described above, according to the search system 10, it is possible to search document data by appropriately reflecting the contents of the search sentence based on the concept organized by the hierarchical structure. The user of the search system 10 can efficiently perform the search process via the display screen 1100.
The search system 10 described above can be used as a technical information search system that stores various technical information as document data and performs a search based on a search sentence, in addition to being used for searching for inquiries and answers about products. . For example, the search system 10 stores information about various drugs as document data, and is used as a system for searching document data that conceptually matches a search sentence such as “a protein that amplifies cancer cells”. May be.

  FIG. 12 shows the configuration of the reporting system 20 according to the embodiment of the present invention. The reporting system 20 extracts each document concept of sequentially input document data, and notifies the user when the frequency of a specific document concept exceeds a predetermined value. As an example, the reporting system 20 according to the present embodiment extracts a document concept indicating a failure from document data obtained by texting an inquiry about a product sent to a call center, and the frequency of a specific document concept is equal to or higher than a predetermined value. In such a case, the product is notified that the trouble is frequently occurring.

  The reporting system 20 includes a document DB 100, a concept DB 105, a product DB 106, a component DB 107, a dictionary DB 110, a synonym DB 115, a document data normalization unit 120, a concept extraction rule DB 125, and a document data concept extraction unit 130. A search index DB 135, a concept ratio calculation unit 1200, a relative frequency calculation unit 1210, a frequent concept selection unit 1220, a priority concept selection unit 1230, a reference frequency calculation unit 1240, and a notification unit 1250. Here, the document DB 100, concept DB 105, product DB 106, component DB 107, dictionary DB 110, synonym DB 115, document data normalization unit 120, concept extraction rule DB 125, document data concept extraction unit 130, and search index DB 135 are shown in FIG. In order to adopt substantially the same function and configuration as the members having the same reference numerals, the description is omitted except for the following differences.

  The document DB 100 sequentially stores input document data. The document DB 100 according to the present embodiment stores document data indicating the contents of defects of the products for each of the plurality of products. The concept DB 105 stores a plurality of concepts that specify a plurality of defects for a product in the hierarchical structure illustrated in FIG.

  The concept ratio calculation unit 1200 uses the document concept stored in the search index DB 135 to calculate the ratio of the number of document data corresponding to each concept to the number of document data in the document DB 100. The concept ratio calculation unit 1200 includes an all product concept ratio calculation unit 1203 and a specific product concept ratio calculation unit 1206. The total product concept ratio calculation unit 1203 calculates the ratio of the number of document data corresponding to each concept to the number of document data for a plurality of products to be compared. For example, when the number of document data stored in the document DB 100 is 1000 and the number of document data having the document concept “/ defect / hardware / hard disk” is 35, the ratio is 3.5% (35/1000). It becomes.

  The specific product concept ratio calculation unit 1206 has a ratio of the number of document data for the product corresponding to each concept with respect to the number of document data for at least one product for which it is reported that many defects have occurred. Is calculated. For example, the number of document data stored in the document DB 100 for the product “/ product / PC product / notebook / series A / model A2” is 100, and the document data having the document concept “/ defect / hardware / hard disk” is stored. When the number is 10, the ratio is 10% (10/100).

  The relative frequency calculation unit 1210 calculates a relative frequency indicating the magnitude of the ratio calculated by the specific product concept ratio calculation unit 1206 in the concept ratio calculation unit 1200 with respect to the reference ratio corresponding to each concept. The relative frequency calculation unit 1210 according to the present embodiment uses the ratio calculated by the all product concept ratio calculation unit 1203 as a reference ratio, and the specific product concept ratio calculation unit 1206 with respect to the ratio calculated by the all product concept ratio calculation unit 1203. The relative frequency indicating the magnitude of the ratio calculated by is calculated. That is, in the case of the above example, the relative frequency corresponding to the document concept “/ defect / hardware / hard disk” for the product “/ product / PC product / notebook / series A / model A2” is about 2.9 (10% /3.5%).

  The frequent concept selection unit 1220 selects a concept having a relative frequency equal to or higher than a predetermined threshold among a plurality of concepts. The priority concept selection unit 1230 selects one of the first concept selected by the frequent concept selection unit 1220 and the second concept in the upper hierarchy of the first concept as a relative of the first concept and the second concept. Select based on frequency. Thus, the priority concept selection unit 1230 appropriately selects the concept hierarchy to be reported from the concepts selected by the frequent concept selection unit 1220 from the upper hierarchy or the lower hierarchy.

  The reference frequency calculation unit 1240 calculates a reference frequency for selecting which concept to report. The notification unit 1250 notifies the user that the relative frequency of the concept selected by the priority concept selection unit 1230 among the first concept or the second concept is high.

  According to the reporting system 20 shown above, when document data corresponding to a specific concept is frequently input, the user selects that the concept hierarchy is appropriately selected and the concept is frequently generated. Can be reported to. As a result, the document DB 100 that sequentially registers inquiries to the call center as document data can be used to detect and report at an early stage that a specific defect frequently occurs for a certain product.

  The reporting system 20 shown above may be provided as a part of the search system 10 shown in FIG. That is, for example, the search system 10 shown in FIG. 1 includes the concept ratio calculation unit 1200, the relative frequency calculation unit 1210, the frequent concept selection unit 1220, the priority concept selection unit 1230, the reference frequency calculation unit 1240, and the like shown in FIG. A notification unit 1250 may be further provided.

  FIG. 13 shows an operation flow of the reporting system 20 according to the embodiment of the present invention. In FIG. 13, the steps denoted by the same step numbers as in FIG. 5 perform substantially the same operations as in FIG.

  First, every time document data is input, the reporting system 20 performs text analysis (S500), normalization (S505), concept extraction (S510), and search index creation (S515) of the document data.

Next, the concept ratio calculation unit 1200 calculates the ratio of the number of document data corresponding to each concept or set of concepts to the number of document data in the document DB 100 (S1340). More specifically, the all-product concept ratio calculation unit 1203 calculates the ratio R _all for _all products by the following equation (1), and the specific product concept ratio calculation unit 1206 is a target for reporting defects. The ratio R for the product to be calculated is calculated by the following formula (2).

R _all = # (A _all ∩X) / #A _all (1)
R = # (A∩X) / #A (2)
Here, A _all indicates all products, A indicates a product for which a defect is to be reported, X indicates a concept or a set of concepts corresponding to the defect, and #C indicates the number of document data corresponding to the concept C. Here, all the products are a plurality of products that contribute to a reference ratio that is a reference for the relative frequency calculation unit 1210 to calculate the relative frequency. The reporting system 20 may use a plurality of products corresponding to a hierarchy in which the tendency of a failure to occur is substantially the same in the conceptual hierarchy of the product illustrated in FIG. 3 as a product that contributes to the reference ratio. That is, for example, the reporting system 20 uses “series A”, “series B”, “series C”, etc., which are located in the lower hierarchy of “product / PC product / notebook” in FIG. May be.

  When the concept DB 105 stores each of a plurality of concepts as a node having a plurality of hierarchical structures such as a defect concept hierarchy, a product concept hierarchy, and a component concept hierarchy, a document data concept extraction unit 130 may extract a plurality of document concepts belonging to each of a plurality of hierarchical structures corresponding to the document data (S510). For example, when the concept DB 105 stores each of a plurality of concepts as a node of the first hierarchical structure or the second hierarchical structure, the document data concept extraction unit 130 corresponds to each document data, The first document concept belonging to the hierarchical structure and the second document concept belonging to the second hierarchical structure may be extracted.

In this case, the concept ratio calculation unit 1200 has a first ratio R1 of the number of document data corresponding to the concept of the first hierarchical structure with respect to the number of document data in the document DB 100 for each of all products and target products. documents corresponding to a combination of _all and R1, and a second ratio R2 _all and R2 of the number of document data corresponding to the concept of the second hierarchical structure, and a concept of the first hierarchical structure and a concept of the second hierarchical structure The third ratios R1 & 2 _all and R1 & 2 of the number of data are respectively calculated.

Next, the relative frequency calculation unit 1210 has a relative frequency RR (= R) indicating the magnitude of the ratio calculated by the specific product concept ratio calculation unit 1206 in the concept ratio calculation unit 1200 with respect to the reference ratio corresponding to each concept. / R _all ) is calculated (S1350). The relative frequency calculation unit 1210 according to the present embodiment uses the ratio calculated by the all product concept ratio calculation unit 1203 as a reference ratio, and is calculated by the specific product concept ratio calculation unit 1206 for the product to be reported as a defect. A relative frequency is calculated indicating how large the ratio is relative to the average ratio for all products.

Here, when a set of the first document concept and the second document concept is extracted corresponding to the document data, the relative frequency calculation unit 1210 calculates the reference ratio R1 _all corresponding to the concept X1 of the first hierarchical structure. The first relative frequency RR1 (= R1 / R1 _all ) indicating the magnitude of the first ratio R1 with respect to the second and the second ratio R2 relative to the reference ratio R2 _all corresponding to the concept X2 of the second hierarchical structure 2 Relative frequency RR2 (= R2 / R2 _all ) and the third ratio R1 & 2 relative to the reference ratio R1 & 2 _all corresponding to the concept X1 & 2 corresponding to the combination of the concept of the first hierarchical structure and the concept of the second hierarchical structure A third relative frequency RR1 & 2 (= R1 & 2 / R1 & 2 _all ) indicating the above is calculated.

  Here, the relative frequency calculation unit 1210 may correct the relative frequency of the concept for the product to be smaller when the number of document data for the product corresponding to the concept is small than when the document data is large. More specifically, the relative frequency calculation unit 1210 may use the minimum value of the confidence interval in the interval estimation with the reliability coefficient of 80% as the relative frequency. Thereby, the relative frequency calculation part 1210 can avoid reporting a malfunction in the state where it is difficult to recognize that the number of samples is small and the malfunction occurs frequently.

  Next, the frequent concept selection unit 1220 selects a concept having a relative frequency equal to or higher than a predetermined threshold among a plurality of concepts (S1360). More specifically, the frequent concept selection unit 1220 selects a concept whose relative frequency for at least one product is equal to or higher than a predetermined threshold among a plurality of concepts. When a plurality of sets of the first hierarchical structure concept and the second hierarchical structure concept are extracted corresponding to the document data, the frequent concept selection unit 1220 includes the first hierarchical structure concept and the first hierarchical structure concept. Among the two combinations of the hierarchical structure concepts, a first hierarchical structure concept and a second hierarchical structure concept set whose relative frequency is equal to or higher than a threshold value is selected.

  Next, when the relative frequency of the document data corresponding to the combination of two or more document concepts is equal to or higher than the threshold, the reference frequency calculation unit 1240 reports the combination of these document concepts, or A frequency serving as a reference for selecting whether to report a combination of document concepts as a higher concept or a lower concept is calculated (S1370).

More specifically, the reference frequency calculation unit 1240 uses a reference frequency for determining whether to report a set of the first document concept and the second document concept or to report as a first document concept, A calculated value RR1 & 2 _base of a relative frequency when the first document concept and the second document concept are independent events is calculated. The reference frequency calculation unit 1240 calculates the reference frequency RR1 & 2 _{base based} on the relative frequency RR1 for the first document concept and the relative frequency RR2 for the second document concept by the following equation (3).
RR1 & 2 _base = RR1 × RR2 × (# (X1∩A _all ) × # (X2∩A _all )) / (# (X1∩X2∩A _all ) × # A _all ) (3)

  Next, the priority concept selection unit 1230 selects one of the concept selected by the frequent concept selection unit 1220 and the concept of the upper hierarchy of the concept based on the relative frequency of the concept and the concept of the upper hierarchy of the concept. Select (S1380). The priority concept selection unit 1230 may select and use the concept of the higher hierarchy from the concepts selected by the frequent concept selection unit 1220. By the processing of S1380, the priority concept selection unit 1230 selects whether to report the concept or a concept of a higher hierarchy of the concept when the relative frequency of the specific concept is equal to or higher than the threshold value. .

  More specifically, a concept X consisting of a set of one or more document concepts {X1, X2, ..., Xn} and a set of one or more document concepts {Y1, Y2, ..., Ym} Which of the more detailed concepts Y to report than X is selected as shown in (1) or (2) below. Here, concept Y is more detailed than concept X means that for any Xi, any Yj is the same or lower-level concept as Xi in a certain concept hierarchy. Concept X is more general than concept Y It is also expressed.

(1) When X = {X1, X2,..., Xn}, Y = {Y1, Y2,..., Yn}, and all Xk are the same or higher-level concepts as Yk in a certain concept hierarchy. The concept X is “/ defect / hardware / input device / pointing device” (= {X1}) and the concept Y is “/ defect / hardware / input device / pointing device / mouse” (= {Y1}).

In this case, the priority concept selection unit 1230 selects the concept Y (= {Y1, Y2,..., Yn}) selected by the frequent concept selection unit 1220 and the concept X (= {X1, X2,. Xn}) satisfies the following formula (4).
(Relative frequency of Y)> α × (Relative frequency of X) (4)
However, (alpha) is a predetermined ratio and takes the value of about 1.5-2, for example.

  Then, the prioritized concept selection unit 1230 selects the concept Y when the relative frequency of the concept Y is greater than a predetermined ratio α as compared with the relative frequency of the concept X, and is not greater than the predetermined ratio α. If you choose concept X. Thereby, the priority concept selection unit 1230 reports the concept Y when the relative frequency of the concept X in the upper hierarchy is low and the relative frequency of the concept Y is high, and when the relative frequency of the concept X in the upper hierarchy is high to some extent Concept Y can be reported together as Concept X.

(2) X = {X1, X2, ..., Xn}, Y = {Y1, Y2, ..., Yn, ..., Ym}, and all Xk (k = 1, ..., n) are in a certain concept hierarchy. For example, the concept X is “/ component / hardware / fan” (= {X1}) and the concept Y is “/ component / hardware / fan” (= = Y1) && “/ Problem / Hardware / Noise” (= Y2). “&&” indicates that the concept Y is an AND condition of the concepts Y1 and Y2.

In this case, the priority concept selection unit 1230 determines whether the concept Y selected by the frequent concept selection unit 1220 and the concept X of the higher hierarchy satisfy the following formula (5). However, Z is a set of concepts of differences between X and Y (= {Yn + 1,..., Ym}) in the above example when the difference between the upper and lower layers in the same concept hierarchy is not considered. }).
(Relative frequency of Y)> α × (Relative frequency of Y when X and Z are independent events (RR1 & 2 _base )) (5)

Then, the priority concept selection unit 1230 compares the relative frequency RR1 & 2 of the concept Y with a predetermined ratio α compared to the calculated value RR1 & 2 _base of the relative frequency of the concept Y when the concept X and the concept Z are independent events. If it is larger than the above, the concept Y that is a combination of the concept X and the concept Z is selected, and if it is not larger than the predetermined ratio α, the concept X is selected. Accordingly, the priority concept selection unit 1230 can report the concept Y when the concept Y is highly likely to be caused by the combination of the concept X and the concept Z.

  Note that the priority concept selection unit 1230 may obtain the concept X of a higher hierarchy by applying both (1) and (2) to the concept Y.

  Next, the notifying unit 1250 indicates that the user of the search system 10 has frequently reported that at least one product among the concepts Y or X corresponds to the concept selected by the priority concept selecting unit 1230. (S1390). Here, in the case of (2), the notification unit 1250 indicates that the concept Y, which is a combination of the concept X and the concept Z, selected by the priority concept selection unit 1230, or the relative frequency of the concept X is high. Notify the user.

  Then, when new document data is added to the document DB 100, the reporting system 20 advances the process to S500 (S1395). Instead, the reporting system 20 may perform the above processing every predetermined period such as one week.

  According to the reporting system 20 shown above, each document concept of sequentially input document data is extracted, and the concept selected by the frequent concept selection unit 1220 and the priority concept selection unit 1230 is notified to identify When the frequency of the document concept or the set of document concepts exceeds a predetermined value, the user can be notified of this. Thereby, for example, according to the number of inquiries about the product sent to the call center, it is possible to notify the malfunction of the product at an early stage.

  FIG. 14 shows an example of a hardware configuration of a computer 1900 according to this embodiment. A computer 1900 according to this embodiment is connected to a CPU peripheral unit having a CPU 2000, a RAM 2020, a graphic controller 2075, and a display device 2080 that are connected to each other by a host controller 2082, and to the host controller 2082 by an input / output controller 2084. Input / output unit having communication interface 2030, hard disk drive 2040, and CD-ROM drive 2060, and legacy input / output unit having ROM 2010, flexible disk drive 2050, and input / output chip 2070 connected to input / output controller 2084 With.

  The host controller 2082 connects the RAM 2020 to the CPU 2000 and the graphic controller 2075 that access the RAM 2020 at a high transfer rate. The CPU 2000 operates based on programs stored in the ROM 2010 and the RAM 2020 and controls each unit. The graphic controller 2075 acquires image data generated by the CPU 2000 or the like on a frame buffer provided in the RAM 2020 and displays it on the display device 2080. Instead of this, the graphic controller 2075 may include a frame buffer for storing image data generated by the CPU 2000 or the like.

  The input / output controller 2084 connects the host controller 2082 to the communication interface 2030, the hard disk drive 2040, and the CD-ROM drive 2060, which are relatively high-speed input / output devices. The communication interface 2030 communicates with other devices via a network. The hard disk drive 2040 stores programs and data used by the CPU 2000 in the computer 1900. The CD-ROM drive 2060 reads a program or data from the CD-ROM 2095 and provides it to the hard disk drive 2040 via the RAM 2020.

  The input / output controller 2084 is connected to the ROM 2010, the flexible disk drive 2050, and the relatively low-speed input / output device of the input / output chip 2070. The ROM 2010 stores a boot program that the computer 1900 executes at startup, a program that depends on the hardware of the computer 1900, and the like. The flexible disk drive 2050 reads a program or data from the flexible disk 2090 and provides it to the hard disk drive 2040 via the RAM 2020. The input / output chip 2070 connects various input / output devices via a flexible disk drive 2050 and, for example, a parallel port, a serial port, a keyboard port, a mouse port, and the like.

  A program provided to the hard disk drive 2040 via the RAM 2020 is stored in a recording medium such as the flexible disk 2090, the CD-ROM 2095, or an IC card and provided by the user. The program is read from the recording medium, installed in the hard disk drive 2040 in the computer 1900 via the RAM 2020, and executed by the CPU 2000.

  The search program installed in the computer 1900 and causing the computer 1900 to function as the search system 10 includes a document DB management module that manages the document DB 100, a concept DB management module that manages the concept DB 105, and a product DB management module that manages the product DB 106. A component DB management module that manages the component DB 107, a dictionary DB module that manages the dictionary DB 110, a synonym DB module that manages the synonym DB 115, a document data normalization module, and a concept that manages the concept extraction rule DB 125 An extraction rule DB module, a document data concept extraction module, a search index DB module for managing the search index DB 135, a search sentence normalization module, a search sentence concept extraction module, a concept check Comprising a module, a concept selection support module, and a search result output module. These programs or modules work on the CPU 2000 or the like to make the computer 1900, the document DB 100, the concept DB 105, the product DB 106, the component DB 107, the dictionary DB 110, the synonym DB 115, the document data normalization unit 120, Concept extraction rule DB 125, document data concept extraction section 130, search index DB 135, search sentence normalization section 140, search sentence concept extraction section 145, concept search section 150, concept selection support section 155, search results It functions as the output unit 160.

  The report program installed in the computer 1900 and causing the computer 1900 to function as the reporting system 20 includes a document DB management module that manages the document DB 100, a concept DB management module that manages the concept DB 105, and a product DB that manages the product DB 106. Manages a management module, a component DB management module that manages the component DB 107, a dictionary DB module that manages the dictionary DB 110, a synonym DB module that manages the synonym DB 115, a document data normalization module, and a concept extraction rule DB 125 Concept extraction rule DB module, document data concept extraction module, search index DB module for managing the search index DB 135, all product concept ratio calculation module, and specific product concept ratio calculation module Comprising a concept ratio calculating module having Yuru, the relative frequency calculating module, a frequent concept selection module, a priority concept selection module, a reference frequency calculation module, a notification module. These programs or modules work on the CPU 2000 or the like to make the computer 1900, the document DB 100, the concept DB 105, the product DB 106, the component DB 107, the dictionary DB 110, the synonym DB 115, the document data normalization unit 120, A concept extraction rule DB 125, a document data concept extraction unit 130, a search index DB 135, a concept ratio calculation unit 1200 having an all product concept ratio calculation unit 1203 and a specific product concept ratio calculation unit 1206, a relative frequency calculation unit 1210, The frequent concept selection unit 1220, the priority concept selection unit 1230, the reference frequency calculation unit 1240, and the notification unit 1250 are caused to function.

  The program or module shown above may be stored in an external storage medium. As the storage medium, in addition to the flexible disk 2090 and the CD-ROM 2095, an optical recording medium such as DVD or CD, a magneto-optical recording medium such as MO, a tape medium, a semiconductor memory such as an IC card, or the like can be used. Further, a storage device such as a hard disk or a RAM provided in a server system connected to a dedicated communication network or the Internet may be used as a recording medium, and the program may be provided to the computer 1900 via the network.

  As mentioned above, although this invention was demonstrated using embodiment, the technical scope of this invention is not limited to the range as described in the said embodiment. It will be apparent to those skilled in the art that various modifications or improvements can be added to the above-described embodiment. It is apparent from the scope of the claims that the embodiments added with such changes or improvements can be included in the technical scope of the present invention.

1 shows a configuration of a search system 10 according to an embodiment of the present invention. An example of the concept hierarchy of the malfunction which concept DB105 concerning the embodiment of the present invention memorizes is shown. An example of the concept hierarchy of the product which product DB106 concerning the embodiment of the present invention memorizes is shown. An example of the concept hierarchy of the component memorize | stored in component DB107 which concerns on embodiment of this invention is shown. The operation | movement flow of the search system 10 which concerns on embodiment of this invention is shown. An example of the normalization rule which the synonym DB115 which concerns on embodiment of this invention memorize | stores is shown. An example of the concept extraction rule which the concept extraction rule DB125 which concerns on embodiment of this invention memorize | stores is shown. The structure of the concept search part 150 which concerns on embodiment of this invention is shown. The operation | movement flow of the concept search part 150 which concerns on embodiment of this invention is shown. An example of generalization and specialization by the concept search unit 150 according to the embodiment of the present invention is shown. An example of the display screen 1100 of the search system 10 which concerns on embodiment of this invention is shown. 1 shows a configuration of a reporting system 20 according to an embodiment of the present invention. The operation | movement flow of the report system 20 which concerns on embodiment of this invention is shown. 2 shows an exemplary hardware configuration of a computer 1900 according to an embodiment of the present invention.

10 Search system 20 Reporting system 100 Document DB
105 Concept DB
106 Product DB
107 Component DB
110 Dictionary DB
115 Synonyms DB
120 Document data normalization unit 125 Concept extraction rule DB
130 Document Data Concept Extraction Unit 135 Search Index DB
140 Search sentence normalization part 145 Search sentence concept extraction part 150 Concept search part 155 Concept selection support part 160 Search result output part 800 Same concept output part 810 Upper concept acquisition part 820 Generalized concept output part 830 Lower concept acquisition part 840 Specialization Concept output unit 1100 Display screen 1110 Search sentence input screen 1130 Concept operation screen 1160 Search result output screen 1200 Concept ratio calculation unit 1203 All product concept ratio calculation unit 1206 Specific product concept ratio calculation unit 1210 Relative frequency calculation unit 1220 Multi-frequency concept selection unit 1230 Priority concept selection unit 1240 Reference frequency calculation unit 1250 Notification unit 1900 Computer 2000 CPU
2010 ROM
2020 RAM
2030 Communication interface 2040 Hard disk drive 2050 Flexible disk drive 2060 CD-ROM drive 2070 Input / output chip 2075 Graphic controller 2080 Display device 2082 Host controller 2084 Input / output controller 2090 Flexible disk 2095 CD-ROM

Claims (20)

A document database for sequentially storing input document data;
A concept database that stores a plurality of predetermined concepts by means of a hierarchical structure in which another concept including one concept is an upper hierarchy of the one concept;
A document data concept extraction unit that extracts a document concept that is the concept corresponding to the document data based on a keyword included in each of the document data;
A concept ratio calculation unit that calculates a ratio of the number of document data corresponding to each concept to the number of document data in the document database;
A relative frequency calculating unit that calculates a relative frequency indicating a magnitude of the ratio calculated by the concept ratio calculating unit with respect to a reference ratio corresponding to each of the concepts;
Among the plurality of concepts, a multi-frequency concept selecting unit that selects the concept having the relative frequency equal to or higher than a predetermined threshold;
One of the first concept selected by the frequent concept selection unit and the second concept in the upper hierarchy of the first concept is set as the relative frequency of the first concept and the second concept. A priority concept selection unit to select based on;
A reporting system comprising: a notifying unit for notifying a user that the relative frequency of the concept selected by the priority concept selecting unit from the first concept or the second concept is high. The document database stores, for each of a plurality of products, document data indicating contents of defects of the plurality of products,
The concept database stores the plurality of concepts for identifying a plurality of defects about a product,
The document data concept extraction unit extracts a document concept that is the concept corresponding to the document data based on a keyword included in each document data,
The concept ratio calculation unit
An overall product concept ratio calculating unit that calculates a ratio of the number of document data corresponding to each concept to the number of document data for the plurality of products;
A specific product concept ratio calculating unit that calculates a ratio of the number of document data for the product corresponding to each concept to the number of the document data for at least one product,
The relative frequency calculation unit calculates the relative frequency indicating the size of the ratio calculated by the specific product concept ratio calculation unit with respect to the ratio calculated by the all product concept ratio calculation unit,
The multi-frequency concept selection unit selects, from the plurality of concepts, the concept in which the relative frequency for the at least one product is equal to or higher than the predetermined threshold value,
The priority concept selection unit selects one of the first concept selected by the frequent concept selection unit and the second concept in a higher hierarchy of the first concept as the first concept and the first concept. Select based on the relative frequency of the two concepts,
The notification unit reports that the at least one product frequently has a defect corresponding to the concept selected by the priority concept selection unit from the first concept or the second concept. Notify system users
The reporting system according to claim 1 . The priority concept selection unit selects the first concept when the relative frequency of the first concept is greater than a predetermined ratio compared to the relative frequency of the second concept, and determines the predetermined concept. The reporting system according to claim 1 or 2 , wherein the second concept is selected when the ratio is not larger than a predetermined ratio. The concept database stores each of the plurality of concepts as a node of the first hierarchical structure or the second hierarchical structure,
The document data concept extraction unit extracts the first document concept belonging to the first hierarchical structure and the second document concept belonging to the second hierarchical structure corresponding to the document data,
The concept ratio calculation unit includes a first ratio of the number of the document data corresponding to the concept of the first hierarchical structure with respect to the number of the document data in the document database, and the second hierarchical structure. A second ratio of the number of document data corresponding to a concept, and a third ratio of the number of document data corresponding to a combination of the concept of the first hierarchical structure and the concept of the second hierarchical structure. Calculate
The relative frequency calculation unit includes a first relative frequency indicating a magnitude of the first ratio with respect to a reference ratio corresponding to the concept of the first hierarchical structure, and a reference corresponding to the concept of the second hierarchical structure. A second relative frequency indicating a magnitude of the second ratio with respect to a ratio, and a third ratio with respect to a reference ratio corresponding to a combination of the concept of the first hierarchical structure and the concept of the second hierarchical structure. Calculating the third relative frequency indicating the magnitude,
The frequent concept selection unit is configured such that the relative frequency of the combination of the concept of the first hierarchical structure and the concept of the second hierarchical structure is equal to or higher than the predetermined threshold value . Selecting a third concept of the second hierarchical structure and a fourth concept of the second hierarchical structure;
The reporting system is configured such that the third concept and the fourth concept are independent events based on the first relative frequency for the third concept and the second relative frequency for the fourth concept. A reference frequency calculation unit for calculating a calculated value of the third relative frequency in the case of
The priority concept selection unit is configured such that the third relative frequency is a ratio determined in advance in comparison with a calculated value of the third relative frequency when the third concept and the fourth concept are independent events. If greater than or equal to, select a combination of the third concept and the fourth concept; if not greater than the predetermined percentage, select the third concept;
The notification unit notifies the user that the combination of the third concept and the fourth concept selected by the priority concept selection unit or the relative frequency of the third concept is high.
The reporting system according to any one of claims 1 to 3 .   A search sentence concept extraction unit that extracts a search sentence concept that is the concept corresponding to the search sentence, based on a keyword included in the input search sentence;
  A concept search unit for searching for document data in which the search sentence concept is an upper hierarchy or lower hierarchy concept of the document concept among each of the plurality of document data;
  A search result output unit for outputting the document data searched by the concept search unit as the document data including content specified by the search sentence;
  The reporting system according to claim 1, further comprising:   A concept extraction rule database that stores a concept extraction rule that includes a set of one or more of the keywords and the concept indicating the semantic content of the one or more keywords;
  The search sentence concept extraction unit, when the one or more keywords included in any of the concept extraction rules are included in the search sentence, the concept included in the concept extraction rule as the search sentence concept Extract
  The reporting system according to claim 5.   The document data concept extraction unit extracts the concept included in the concept extraction rule as the document concept when the one or more keywords included in any of the concept extraction rules are included in the document data. Do
  The reporting system according to claim 6.   For each of the document data, further comprising a search index database that stores correspondence between the document concept of the document data extracted by the document data concept extraction unit and the document data,
  The concept search unit corresponds to a document concept when the search sentence concept is a concept of an upper hierarchy or a lower hierarchy of the document concept stored in the search index database before the search sentence is input. Output the document data as search results
  The reporting system according to any one of claims 5 to 7.   A synonym database that stores a correspondence between a predetermined phrase and the keyword that is a synonym of the phrase;
  A document data normalization unit that normalizes the document data by replacing the phrase included in each document data with the keyword that is a synonym of the phrase;
  A search sentence normalization unit that normalizes the search sentence by replacing the phrase included in the search sentence with the keyword that is a synonym of the phrase;
  Further comprising
  The document data concept extraction unit extracts the document concept from the normalized document data,
  The search sentence concept extraction unit extracts the search sentence concept from the normalized search sentence
  The reporting system according to any one of claims 5 to 8.   The concept search unit
  A superordinate concept acquisition unit that, when the search sentence concept does not match the document concept, acquires a search sentence superordinate concept that is a concept of an upper hierarchy of the search sentence concept;
  A generalized concept output unit that outputs the document data as a search result when the search sentence superordinate concept matches the document concept;
  The reporting system according to any one of claims 5 to 9.   The concept database stores each of the plurality of concepts as a node of the first hierarchical structure or the second hierarchical structure,
  The document data concept extraction unit extracts the first document concept belonging to the first hierarchical structure and the second document concept belonging to the second hierarchical structure corresponding to the document data,
  The search sentence concept extraction unit extracts the first search sentence concept belonging to the first hierarchical structure and the second search sentence concept belonging to the second hierarchical structure corresponding to the search sentence. ,
  The higher-level concept acquisition unit, when the first search sentence concept and the second search sentence concept are not the same as the first document concept and the second document concept, respectively. Obtaining the first search sentence superordinate concept in the upper hierarchy of the concept and the second search sentence superordinate concept in the upper hierarchy of the second search sentence concept;
  The generalization concept output unit includes the first document in which the first search sentence superordinate concept and the first document concept are the same as the second search sentence concept and the second document concept. The number of second document data in which the number of data is the same as the first search sentence concept and the first document concept, and the second search sentence superordinate concept and the second document concept. If smaller, output the first document data as a search result
  The reporting system according to claim 10.   The concept search unit
  When all of the document data having the same document concept as the search sentence concept has the same document concept as the search sentence subordinate concept that is a concept of a lower hierarchy of the search sentence concept, the search sentence concept is , A subordinate concept obtaining unit that replaces the search sentence subordinate concept,
  A specialized concept output unit for outputting the document data whose search sentence subordinate concept matches the document concept as a search result;
  The reporting system according to claim 5, comprising:   The concept database stores the plurality of concepts for identifying a plurality of defects about a product,
  The document database stores, for each of the plurality of defects, the document data indicating the contents of the plurality of defects,
  The search sentence concept extraction unit extracts the search sentence concept corresponding to the search sentence for searching for defects in the product, which is input by a user,
  The search result output unit outputs the document data searched by the concept search unit as the document data indicating the content of the defect of the product input by the user. Or reporting system.   The concept database stores the plurality of concepts according to a hierarchical structure in which the concept indicating the state of failure of the component is provided in a lower hierarchy of the concept indicating that the component of the product is defective.
  The document data concept extraction unit extracts the document concept indicating that one component is defective based on a keyword included in the document data,
  The search sentence concept extraction unit extracts the search sentence concept indicating a failure state of the one component based on a keyword included in the search sentence,
  The concept search unit
  An upper concept acquisition unit that acquires the concept indicating that there is a defect in the one component, which is the concept in the upper hierarchy of the search sentence concept, as a search sentence upper concept;
  A generalized concept output unit that outputs the document data having the document concept indicating that there is a defect in the one component that matches the search sentence superordinate concept, as a search result;
  Have
  The reporting system according to claim 13.   A component database for storing the inclusion relationship of each component of the product in a hierarchical structure;
  The document data concept extraction unit further extracts the document concept indicating the component described in the document data based on a keyword included in the document data,
  The search sentence concept extraction unit further extracts the search sentence concept indicating the components described in the search sentence based on a keyword included in the search sentence,
  The higher concept acquisition unit includes the concept that is a higher hierarchy of the first search sentence concept indicating that the component has a defect or a failure state of the component, and the second that indicates the component. Obtain the concept that is the upper hierarchy of the search sentence concept,
  The generalization concept output unit, when at least one of the first search sentence concept and the second search sentence concept is the higher-level concept, the document concept that matches the first search sentence concept And the document data having the document concept matching the second search sentence concept is output as a search result.
  The reporting system according to claim 14.   A product database for storing the inclusion relationship of the product names of the plurality of products in a hierarchical structure;
  The document data concept extraction unit further extracts the document concept indicating the product name described in the document data based on a keyword included in the document data,
  The search sentence concept extraction unit further extracts the search sentence concept indicating the product name described in the search sentence concept extraction unit based on a keyword included in the search sentence,
  The higher-level concept acquisition unit includes the concept that is an upper layer of the first search statement concept indicating that the component has a defect or a defect state of the component, and the second that indicates the product name. Obtain the concept that is the upper hierarchy of the search sentence concept,
  The generalization concept output unit, when at least one of the first search sentence concept and the second search sentence concept is the higher-level concept, the document concept that matches the first search sentence concept And the document data having the document concept matching the second search sentence concept is output as a search result.
  The reporting system according to claim 14 or 15. A reporting method in a reporting system in which a plurality of document data is sequentially input,
A document database storage stage for sequentially storing input document data;
A concept database storage stage for storing a plurality of predetermined concepts in a hierarchical structure in which another concept including the one concept is an upper hierarchy of the one concept;
A document data concept extraction step for extracting a document concept, which is the concept corresponding to the document data, based on a keyword included in each document data;
A concept ratio calculating step of calculating a ratio of the number of the document data corresponding to each of the concepts to the number of the document data stored in the document database storing step;
A relative frequency calculating step for calculating a relative frequency indicating a magnitude of the ratio calculated by the concept ratio calculating step with respect to a reference ratio corresponding to each of the concepts;
Of the plurality of concepts, a multi-frequency concept selection step of selecting the concept having the relative frequency equal to or higher than a predetermined threshold;
One of the first concept selected in the frequent concept selection stage and the second concept in the upper hierarchy of the first concept is set as the relative frequency of the first concept and the second concept. A priority concept selection stage to select based on;
A reporting method comprising: a notification step of notifying a user that the relative frequency of the concept selected in the priority concept selection step of the first concept or the second concept is high.   A search sentence concept extraction step of extracting a search sentence concept that is the concept corresponding to the search sentence based on a keyword included in the input search sentence;
  A concept search step of searching for the document data in which the search sentence concept is an upper hierarchy or lower hierarchy concept of the document concept among each of the plurality of document data;
  A search result output step of outputting the document data searched by the concept search unit as the document data including the content specified by the search sentence;
  The reporting method according to claim 17, further comprising: A program for a reporting system in which a plurality of document data is sequentially input,
The program uses the reporting system,
A document database for sequentially storing input document data;
A concept database that stores a plurality of predetermined concepts by means of a hierarchical structure in which another concept including one concept is an upper hierarchy of the one concept;
A document data concept extraction unit that extracts a document concept that is the concept corresponding to the document data based on a keyword included in each of the document data;
A concept ratio calculation unit that calculates a ratio of the number of document data corresponding to each concept to the number of document data in the document database;
A relative frequency calculating unit that calculates a relative frequency indicating a magnitude of the ratio calculated by the concept ratio calculating unit with respect to a reference ratio corresponding to each of the concepts;
Among the plurality of concepts, a multi-frequency concept selecting unit that selects the concept having the relative frequency equal to or higher than a predetermined threshold;
One of the first concept selected by the frequent concept selection unit and the second concept in the upper hierarchy of the first concept is set as the relative frequency of the first concept and the second concept. A priority concept selection unit to select based on;
A program that functions as a notifying unit for notifying a user that the relative frequency of the concept selected by the priority concept selecting unit of the first concept or the second concept is high.   The reporting system,
  A search sentence concept extraction unit that extracts a search sentence concept that is the concept corresponding to the search sentence, based on a keyword included in the input search sentence;
  A concept search unit for searching the document data in which the search statement concept is an upper hierarchy or lower hierarchy concept of the document concept among each of the plurality of document data;
  A search result output unit for outputting the document data searched by the concept search unit as the document data including content specified by the search sentence;
  The program according to claim 19, further functioning.

Images