JP2012208728A - Expert retrieval apparatus and expert retrieval method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To solve the problem that exerts could not be retrieved accurately for Web documents.SOLUTION: An expert retrieval apparatus comprises a document retrieval unit which acquires one or more documents relating to a topic from one or more Web server devices, an expert candidate extraction unit which extracts one or more expert candidates from the one or more documents, an opinion sentence extraction unit which extracts one or more opinion sentences which are sentences relating to opinions about the topic from the one or more documents, an opinion information acquisition unit which acquires opinion information from the one or more opinion sentences for each of one or more expert candidates, a relation degree acquisition unit which acquires for each of the one or more expert candidates a degree of relation between each of the one or more documents and the expert candidate, a score calculation unit which calculate a specialty score for the topic using the opinion information and the relation degree for each of the one or more expert candidates, an expert selection unit which acquires one or more experts using the score, and an expert output unit which outputs the one or more experts acquired by the expert selection unit. The expert retrieval apparatus can thereby retrieve experts accurately for Web documents.

Description

  The present invention relates to an expert search system that extracts one or more experts about a topic from a Web document.

  Conventionally, there has been a technique for searching for an expert for documents in an organization (see, for example, Non-Patent Document 1 and Non-Patent Document 2). In addition, there has been a technique for searching for experts in research papers (see, for example, Non-Patent Document 3). Common to these prior arts is that the degree of relevance to the topic is used as evidence of the expertise of the expert candidate for the document in which the expert candidate appears or the document for which the expert candidate is the author. is there. Institutional documents and research papers can be assumed that the document quality is above a certain level, and there is no problem in treating the degree of relevance to the topic of the document with which the expert candidate is involved as evidence of expertise.

Balog K, Azzopardi L, Rijke M de. Formal Models for Expert Finding in Enterprise Corpora. In: Proceedings of SIGIR’06 .; 2006: 43-50. Macdonald C, Ounis I. Voting for Candidates: Adapting Data Fusion Techniques for an Expert Search Task.In: Proceedings of CIKM’06 .; 2006: 387-396. Mimno D, Mccallum A. Expertise Modeling for Matching Papers with Reviewers. In: Proceedings of KDD’07 .; 2007: 500-509.

  However, when a document for searching for an expert is a document in an organization, a general user cannot use it. In addition, some topics are not research papers. Also, there are people who do not write research papers but are suitable for being called experts. Therefore, it is effective to extract experts from Web documents that are documents existing on the Web.

  However, when Web documents are targeted, not only documents that cause noise such as spam, but also many documents that do not necessarily have high quality such as bulletin board writing and blogs are included. Does not provide sufficient evidence, and causes a decrease in the accuracy of expert searches. In other words, in the conventional expert search system, it is not possible to search for experts with high accuracy for Web documents.

  Also, in the prior art, specialization candidates are treated as given. Expert candidates can be identified when targeting a specific organization or a specific research field, and such assumptions are valid. However, when an expert search is performed on an unspecified topic for a Web document, it is not realistic to obtain expert candidates in advance.

  The expert search device according to the first aspect of the present invention includes a reception unit that receives a topic, a document search unit that acquires one or more documents related to a topic from one or more Web server devices, and one or more documents. From the expert candidate extraction unit that extracts the above expert candidates, the opinion sentence extraction unit that extracts one or more opinion sentences that are sentences related to the topic opinion from one or more documents, and one or more opinion sentences, affirmation An opinion information acquisition unit that acquires, for each one or more expert candidates, opinion information including the number of positive opinion sentences that is the number of positive opinion sentences and the number of negative opinion sentences that are the number of negative opinion sentences; For each of one or more expert candidates, a score calculation unit that calculates the expert score for the topic using opinion information, and one or more experts from one or more expert candidates using the score Expert selection department to obtain and expert selection department An expert search device comprising an expert output unit for outputting one or more experts obtained.

  With this configuration, it is possible to search for a specialist with high accuracy for a Web document.

  In addition, the expert search device of the second invention has one or more expert candidates extracted by the expert candidate extraction unit and one or more documents acquired by the document search unit with respect to the first invention. Among these, for each document in which each expert candidate appears, a relevance degree acquiring unit that acquires the degree of association between the expert candidate and the document for each one or more expert candidates, and the score calculating unit, This is an expert search device that calculates a specialty score for a topic using opinion information and a degree of association for each of one or more expert candidates.

  With this configuration, it is possible to search for a specialist with high accuracy for a Web document.

  In the expert search device of the third invention, the expert output unit sorts and outputs one or more experts in descending order of scores in the first or second invention. Device.

  With such a configuration, it is possible to search for experts with high accuracy and appropriately present experts for Web documents.

  In the expert search device of the fourth aspect of the invention, the expert output unit obtains the expert selection unit instead of the output of one or more experts for any of the first to third inventions. The expert search device outputs one or more documents or one or more opinion sentences from which the scores of the one or more experts are calculated.

  With this configuration, it is possible to search for a specialist with high accuracy and present a document described by the specialist for a Web document.

  Further, the expert search device of the fifth aspect of the present invention provides one or more documents corresponding to some or all of the one or more experts with respect to any one of the first to third inventions, Alternatively, the expert search apparatus further includes a document output unit that outputs one or more pieces of access information that is information for accessing one or more documents.

  With this configuration, it is possible to search for a specialist with high accuracy and present a document described by the specialist for a Web document.

  According to the expert search device of the present invention, an expert can be searched with high accuracy for a Web document.

Conceptual diagram of expert search system 1 according to Embodiment 1 Block diagram of the expert search system 1 A flowchart for explaining the operation of the expert search device 13 Flowchart explaining the expert candidate extraction process Flow chart explaining the opinion information acquisition process Flowchart explaining the first relevance level acquisition process Flowchart explaining the second relevance level acquisition process Flow chart explaining the score calculation process The figure which shows the example of the same Web page Figure showing an example of the same opinion information Overview of the computer system Block diagram of the computer system

  Hereinafter, embodiments of an expert search system and the like will be described with reference to the drawings. In addition, since the component which attached | subjected the same code | symbol in embodiment performs the same operation | movement, description may be abbreviate | omitted again.

(Embodiment 1)
In the present embodiment, an expert search system that extracts and outputs one or more experts about a topic from a Web document will be described. Moreover, in this Embodiment, the expert search system 1 which sorts and outputs the expert in order of a score is demonstrated. In this embodiment, an expert search system that presents a highly reliable document written by an expert will be described.

  FIG. 1 is a conceptual diagram of an expert search system 1 in the present embodiment. The expert search system 1 includes one or more terminal devices 11, one or more Web server devices 12, and an expert search device 13. The terminal device 11 is a terminal used by a user for searching for an expert. The Web server device 12 is a so-called Web server device on the Web, and stores one or more documents. The expert search device 13 is a device that can search for experts. The terminal device 11, the Web server device 12, and the expert search device 13 can communicate via the Internet 4.

  FIG. 2 is a block diagram of the expert search system 1 in the present embodiment.

  The terminal device 11 configuring the expert search system 1 includes a terminal reception unit 110, a terminal transmission unit 111, a terminal reception unit 112, and a terminal output unit 113.

  The Web server device 12 includes a document storage unit 121 and a document transmission unit 122.

  The expert search device 13 includes a reception unit 130, a document search unit 131, an expert candidate extraction unit 132, an opinion sentence extraction unit 133, an opinion information acquisition unit 134, a relevance degree acquisition unit 135, a score calculation unit 136, and an expert selection unit. 137, an expert output unit 138, and a document output unit 139.

  The terminal reception unit 110 configuring the terminal device 11 receives topics from the user. A topic may be considered synonymous with a term. A topic is not limited to a word, but may be a phrase or a set of one or more words. The topic is a term or the like indicating a specialized area of an expert who wants to search. The topic input means may be anything such as a keyboard, mouse or menu screen. The terminal reception unit 110 can be realized by a device driver for input means such as a keyboard, control software for a menu screen, and the like.

  The terminal transmission unit 111 transmits the topic received by the terminal reception unit 110 to the expert search device 13.

  The terminal receiving unit 112 receives one or more experts, one or more documents, one or more opinion sentences, and the like from the expert search device 13.

  The terminal output unit 113 outputs one or more experts, one or more documents, or one or more opinion sentences received by the terminal receiving unit 112. Here, output refers to display on a display, projection using a projector, printing on a printer, sound output, transmission to an external device, storage in a recording medium, output to another processing device or other program, etc. It is a concept that includes delivery of processing results. The terminal output unit 113 may or may not include an output device such as a display or a speaker. The terminal output unit 113 may be realized by output device driver software, or output device driver software and an output device.

  The document storage unit 121 included in the Web server device 12 stores one or more documents. A document is a document on the Web. A document on the Web is called a Web document. Web documents include research papers, but also include CGM documents such as so-called blogs, miniblogs, and documents posted to SNS. The document does not have to be divided into one file. That is, the identification information of the document delimiter does not matter.

  The document transmission unit 122 transmits the document in the document storage unit 121 to the expert search device 13. Normally, the document transmission unit 122 transmits the document stored in the document storage unit 121 to the expert search device 13 in response to a request from the expert search device 13.

  The receiving unit 130 configuring the expert search device 13 receives a topic. Usually, the reception unit 130 receives a topic from the terminal device 11. However, when the expert search device 13 operates as a stand-alone, the reception unit 130 receives a topic from an input device such as a keyboard or a mouse. In other words, accepting here means accepting information input from an input device such as a keyboard, mouse, touch panel, receiving information transmitted via a wired or wireless communication line, optical disk, magnetic disk, semiconductor memory, etc. This is a concept including reception of information read from the recording medium.

  The document search unit 131 acquires one or more documents related to the topic from one or more Web server devices 12. Normally, the document search unit 131 receives a document from the Web server device 12. Here, the document search unit 131 may acquire all documents including a topic, may acquire a document including a topic at a location corresponding to a predetermined tag (for example, a title tag), You may acquire the document whose appearance frequency is more than a threshold value. In addition, there may be various algorithms for the document search unit 131 to acquire one or more documents related to a topic. Note that the document search unit 131 may have a so-called Web search function using topics as keys. In such a case, the document search unit 131 is a so-called Web search engine.

  The expert candidate extraction unit 132 extracts one or more expert candidates from one or more documents. The expert candidate may be a person name or an organization name. Further, the expert candidate extraction unit 132, for example, uses a unique expression extraction (NE) technique to give a unique name (person name / organization name) (for example, “http://www.sophia-it.com/content/ % E5% 9B% BA% E6% 9C% 89% E8% A1% A8% E7% 8F% BE% E6% 8A% BD% E5% 87% BA '' http://www.ntt.co.jp/ journal / 0806 / files / jn200806020.pdf "). More specifically, the expert candidate extraction unit 132 performs morphological analysis on one or more documents, divides each document into words, and then extracts them using a machine learning (sequence tagging) technique. Moreover, the expert candidate extraction part 132 may acquire a person's name, for example using a specific tag as a clue. For example, the expert candidate extraction unit 132 may extract the personal name “Yamada ○ Oo” from “Yamada ○ Oo” using, for example, the clue phrases “Mr.” and “Name”. In addition, the technique which acquires a person name and an organization name from a text is a well-known technique.

  The opinion sentence extraction unit 133 extracts one or more opinion sentences that are sentences related to the topic opinion from one or more documents. The opinion sentence extraction unit 133 extracts, for example, a sentence including a topic, a phrase for specifying an assertion, “is considered”, “considered”, and the like as an opinion sentence. In addition, the opinion sentence extraction unit 133, for example, “Tetsuji Nakagawa, Kentaro Inui and Sadao Kurohashi: Dependency Tree-based Sentiment Classification using CRFs with Hidden Variables, In Proceedings of Human Language Technologies: The 11th Annual Conference of the North American Chapter of Use the technology described in the Association for Computational Linguistics (HLT-NAACL 2010), June 2010. If the technique described in this paper is used, the opinion sentence extraction unit 133 performs syntax analysis on the sentence to be processed, and then the part of speech of words included in the sentence obtained as a result and the dependency relation between words. Using the clues, etc. as a clue, extract the places (phrases etc.) that express opinions based on the series tagging. Furthermore, the opinion sentence extraction unit 133 uses a conditional random field technique for a phrase or the like extracted as an opinion by using a dictionary related to polarity or a word dictionary that reverses the polar W property in addition to the dependency relationship. Outputs the polarity of the opinion sentence. Further, the opinion sentence extraction unit 133 outputs the opinion types such as merits, events, and events by the classifier based on the features from the result of the syntax analysis. In addition, the method of extracting the opinion sentence is not limited.

  The opinion information acquisition unit 134 acquires opinion information for each of one or more expert candidates from one or more opinion sentences. Opinion information is information about the opinion sentence acquired by the opinion sentence extraction unit 133. Opinion information is usually numerical information that can be acquired from one or more opinion sentences. The opinion information is usually statistical data related to one or more opinion sentences. Opinion information usually includes the number of positive opinion sentences, which is the number of positive opinion sentences, and the number of negative opinion sentences, which is the number of negative opinion sentences. The opinion information acquisition unit 134 holds, for example, an evaluation word DB that stores evaluation words such as “good”, “great”, “bad”, and “cannot be evaluated”. Judge whether it is a positive opinion sentence or a negative opinion sentence. And opinion information is acquired using this judgment result. The opinion information may include the number of positive opinion sentences and the number of negative opinion sentences, the total number of positive and negative opinions, the bias of opinions (the ratio of positive opinions to the total number of opinions), and the like. Opinion information may also include the opinion text or the degree of association between the document containing the opinion text and the topic, the domain type of the site where the document is stored (co.jp, go.jp, ac.jp, etc.) good.

  The degree-of-relevance acquisition unit 135 acquires the degree of association between the expert candidate and the document for each of one or more expert candidates. The degree-of-relevance acquisition unit 135 is, for example, a document in which each expert candidate appears among one or more expert candidates extracted by the expert candidate extraction unit 132 and one or more documents acquired by the document search unit 131. On the other hand, the degree of association between the expert candidate and the document is acquired for each of one or more expert candidates.

  More specifically, the relevance level acquisition unit 135 acquires the number (N) of expert candidates extracted by the expert candidate extraction unit 132 for one or more documents acquired by the document search unit 131. Then, the degree-of-relevance obtaining unit 135 calculates the degree of relevance of the expert candidate appearing in the document as “1 / N”.

  Note that the degree of association acquisition unit 135 may acquire the degree of association between a document and a topic. Then, the degree-of-association acquisition unit 135 may acquire the degree of association between the document in which the expert candidate appears and the topic for each expert candidate. That is to say, for example, the relevance degree acquisition unit 135 determines one or more documents corresponding to one or more expert candidates, and sets the relevance degree between the one or more documents and the topic for each one or more expert candidates. You may get to. In such a case, for example, the association degree acquisition unit 135 determines one or more documents corresponding to one or more expert candidates. Then, the degree-of-relevance acquisition unit 135 acquires, for each expert candidate, the degree of association between one or more documents corresponding to the expert candidate and the topic. Here, the relevance degree acquisition unit 135 acquires, for example, the appearance frequency of a topic in the document as the relevance degree between the love document and the topic. Further, the degree-of-relevance acquisition unit 135 may acquire the degree of association between a document and a topic using a cosine similarity using a vector space model or a method called Okapi BM25 (for example, “http: // gihyo. jp / dev / serial / 01 / search-engine / 0008 "," http://wpedia.goo.ne.jp/enwiki/Probabilistic_relevance_model_(BM25) "). That is, a technique for acquiring the degree of association between a document and a topic can be realized by a known technique.

  In addition, the degree of association acquisition unit 135 may acquire both the degree of association between the expert candidate and the document (first degree of association) and the degree of association between the document and the topic (second degree of association).

  The score calculation unit 136 calculates the expert score for the topic using the opinion information and the degree of association for each of one or more expert candidates. For example, the score calculation unit 136 calculates a larger score as the number of positive opinion sentences is larger and the degree of association is larger. The score calculation unit 136 holds information included in the opinion information and a calculation formula using the degree of relevance as a parameter, and uses the calculation formula to calculate the professional score for the topic for each expert candidate. To do. A specific example of the calculation formula will be described later.

  The expert selection unit 137 acquires one or more experts from one or more expert candidates using the score calculated by the score calculation unit 136. Here, the expert is a person name or an organization name. The expert selection part 137 acquires the expert candidate whose score is more than a threshold value as an expert, for example. Moreover, the expert selection part 137 may acquire the expert candidate to the top N rank of a score as an expert, for example. Furthermore, the expert selection part 137 may sort an expert or an expert candidate in order of a score, for example.

  The expert output unit 138 outputs one or more experts acquired by the expert selection unit 137. Moreover, it is preferable that the expert output unit 138 sorts and outputs one or more experts in descending order of score. Further, the expert output unit 138 replaces the output of one or more experts with one or more documents or one or more documents from which the scores of the one or more experts acquired by the expert selection unit 137 are calculated. May be output. Here, the output is normally transmission to the terminal device 11, but display on the display, projection using the projector, printing on the printer, sound output, storage on the recording medium, other processing devices and others It is also possible to pass the processing result to other programs.

  The document output unit 139 outputs one or more documents corresponding to some or all of one or more experts, or one or more access information that is information for accessing one or more documents. . When the expert output unit 138 outputs a document, access information, etc., the document output unit 139 is not necessary. Further, the document output unit 139 may rank and output documents written by the expert, access information, opinion sentences, etc. in the order of the expert's score. Here, the output is normally transmission to the terminal device 11, but display on the display, projection using the projector, printing on the printer, sound output, storage on the recording medium, other processing devices and others It is also possible to pass the processing result to other programs.

  The terminal transmission unit 111, the document transmission unit 122, the expert output unit 138, and the document output unit 139 are usually realized by wireless or wired communication means, but may be realized by broadcasting means.

  The terminal receiving unit 112 and the receiving unit 130 are usually realized by wireless or wired communication means, but may be realized by means for receiving broadcasts.

  The document search unit 131, expert candidate extraction unit 132, opinion sentence extraction unit 133, opinion information acquisition unit 134, relevance level acquisition unit 135, score calculation unit 136, and expert selection unit 137 are usually from an MPU, memory, or the like. Can be realized. The processing procedure of the document search unit 131 or the like is usually realized by software, and the software is recorded on a recording medium such as a ROM. However, it may be realized by hardware (dedicated circuit).

  Next, the operation of the expert search system 1 will be described. First, the operation of the terminal device 11 will be described. The terminal reception unit 110 of the terminal device 11 receives a topic from the user. Then, the terminal transmission unit 111 transmits the topic received by the terminal reception unit 110 to the expert search device 13. Next, in response to topic transmission, the terminal receiving unit 112 receives a topic expert, a document, an opinion sentence, or the like from the expert search device 13. Next, the terminal output unit 113 outputs an expert, a document, an opinion sentence, or the like received by the terminal receiving unit 112.

  Next, the operation of the Web server device 12 will be described. The document transmission unit 122 of the Web server device 12 normally transmits the document in the document storage unit 121 to the expert search device 13 in response to a request from the expert search device 13.

  Next, operation | movement of the expert search device 13 is demonstrated using the flowchart of FIG.

  (Step S301) The receiving unit 130 determines whether a topic has been received. If a topic is accepted, the process goes to step S302, and if no topic is accepted, the process returns to step S301.

  (Step S302) The document search unit 131 acquires one or more documents related to the topic received in step S301 from one or more Web server devices 12.

  (Step S303) The expert candidate extraction unit 132 extracts one or more expert candidates related to the topic received in step S301 from one or more documents acquired in step S302. This process is an expert candidate extraction process. The expert candidate extraction process will be described with reference to the flowchart of FIG.

  (Step S304) The opinion sentence extraction unit 133 extracts one or more opinion sentences of the topic received in step S301 from one or more documents acquired in step S302. And opinion information acquisition part 134 acquires opinion information using one or more acquired opinion sentences. This process is an opinion information acquisition process. The opinion information acquisition process will be described with reference to the flowchart of FIG.

  (Step S305) The relevance degree acquisition unit 135 determines one or more documents corresponding to each of the one or more expert candidates extracted in step S303, and sets the relevance degree between the one or more documents and the topic to 1 or more. Get for each expert candidate. This process is a relevance level acquisition process (may be referred to as a first relevance level acquisition process). Further, the degree of association acquired in the first degree of association acquisition process may be referred to as the first degree of association. The first relevance level acquisition process will be described with reference to the flowchart of FIG. Further, the relevance level acquisition unit 135 acquires the relevance level between the expert candidate and the document for each of one or more expert candidates. Such a process can also be said to be an association degree acquisition process (may be referred to as a second association degree acquisition process). The degree of association acquired in the second degree of association acquisition process may be referred to as the second degree of association. The second relevance level acquisition process will be described with reference to the flowchart of FIG.

  (Step S306) The score calculation unit 136 uses the opinion information acquired in step S304 and the relevance level (first relevance level or / and second relevance level) acquired in step S305, to generate one or more expert candidates. For each, calculate a specialty score for the topic. Such processing is score calculation processing. The score calculation process will be described with reference to the flowchart of FIG.

  (Step S307) The expert selection unit 137 acquires one or more experts from one or more expert candidates using the score calculated in step S306. Note that the expert selection unit 137 normally selects expert candidates whose scores are sufficiently high (scores equal to or higher than a threshold) as experts.

  (Step S308) The expert output unit 138 outputs one or more experts acquired in step S307.

  (Step S309) The document output unit 139 has one or more accesses corresponding to one or more documents corresponding to some or all of one or more experts, or information for accessing one or more documents. The information is output and the process returns to step S301.

  In the flowchart of FIG. 3, the process of step S309 may not be performed.

  Further, in the flowchart of FIG. 3, the process is ended by powering off or interruption for aborting the process.

  Next, the expert candidate extraction process of step S302 will be described using the flowchart of FIG.

  (Step S401) The expert candidate extraction unit 132 substitutes 1 for a counter i.

  (Step S402) The expert candidate extraction unit 132 determines whether or not the i-th document exists. If the i-th document exists, the process goes to step S403, and if it does not exist, the process returns to the upper process (step S304).

  (Step S403) The expert candidate extraction unit 132 performs morphological analysis on the i-th document.

  (Step S404) The expert candidate extraction unit 132 acquires one or more expert candidates using, for example, a machine learning (sequence tagging) technique using the result of the morphological analysis in step S403.

  (Step S405) The expert candidate extraction unit 132 substitutes 1 for the counter j.

  (Step S406) The expert candidate extraction unit 132 determines whether there is a j-th expert candidate. If the j-th expert candidate exists, go to step S407, and if not, go to step S411.

  (Step S407) The expert candidate extraction unit 132 determines whether or not the j-th expert candidate exists in a predetermined buffer. If it exists in the predetermined buffer, go to step S408, otherwise go to step S410.

  (Step S408) The expert candidate extraction unit 132 associates the j-th expert candidate with the document identifier of the i-th document, and at least temporarily accumulates them in a predetermined buffer.

  (Step S409) The expert candidate extraction unit 132 increments the counter j by 1, and returns to step S406.

  (Step S410) The expert candidate extraction unit 132 temporarily stores at least the document identifier of the i-th document in association with the j-th expert candidate already existing in the predetermined buffer.

  (Step S411) The expert candidate extraction unit 132 increments the counter i by 1, and returns to step S402.

  In the flowchart of FIG. 4, the algorithm that the expert candidate extraction unit 132 extracts expert candidates from each document does not matter.

  Next, the opinion information acquisition process in step S303 will be described with reference to the flowchart of FIG.

  (Step S501) The opinion sentence extraction unit 133 substitutes 1 for the counter i.

  (Step S502) The opinion sentence extraction unit 133 determines whether or not the i-th document exists among the documents acquired in step S302. If the i-th document exists, the process goes to step S503, and if the i-th document does not exist, the process goes to step S514.

  (Step S503) The opinion sentence extraction unit 133 substitutes 1 for the counter j.

  (Step S504) The opinion sentence extraction unit 133 determines whether or not the j-th sentence exists in the i-th document. If the jth sentence exists, go to step S505, and if the jth sentence does not exist, go to step S513. The sentence here may be a part of the sentence. In other words, the sentence may be a target unit for determining whether or not the sentence is an opinion sentence.

  (Step S505) The opinion sentence extraction unit 133 acquires the j-th sentence of the i-th document.

  (Step S506) The opinion sentence extraction unit 133 determines whether or not the sentence acquired in step S505 is an opinion sentence, and substitutes the determination result into the buffer. The opinion sentence extraction unit 133 parses the sentence, and uses the result to express an opinion based on the series tagging based on the part of speech of the word included in the sentence and the dependency relation of the word. Where it is and whether it is an opinion sentence or not.

  (Step S507) The opinion sentence extraction unit 133 goes to step S508 if the buffer value is information indicating that it is an opinion sentence, and goes to step S512 if it is information indicating that it is not an opinion sentence.

  (Step S508) The opinion information acquisition unit 134 determines whether the j-th sentence of the i-th document is a positive sentence, “positive”, a negative sentence or “negative”, and is “neutral” Etc.]. The opinion information acquisition unit 134 may calculate the polarity of the jth sentence. The opinion information acquisition unit 134 calculates the polarity of a sentence by a conditional random field technique using, for example, a dictionary regarding words, a dictionary related to polarity, or a word that reverses polarity. It is assumed that the opinion information acquisition unit 134 holds a dictionary related to polarity and a dictionary of words whose polarity is reversed. The polarity is information indicating the degree of affirmation and / or denial of a sentence.

  (Step S509) The opinion information acquisition unit 134 substitutes the determination result in step S508 for a flag (referred to as a posture flag), and adds the posture flag to the opinion sentence.

  (Step S510) The opinion information acquisition unit 134 acquires the degree of association between the j-th sentence of the i-th document and the topic, and assigns it to the opinion sentence.

  (Step S511) The opinion information acquisition unit 134 acquires the domain type of the i-th document and assigns it to the opinion sentence. The domain type is acquired by the document search unit 131 when the document search unit 131 acquires a document.

  (Step S512) The opinion sentence extraction unit 133 increments the counter j by 1, and returns to step S504.

  (Step S513) The opinion sentence extraction unit 133 increments the counter i by 1, and returns to step S502.

  (Step S514) The opinion information acquisition unit 134 substitutes 1 for the counter i.

  (Step S515) The opinion information acquisition unit 134 determines whether or not the i-th expert candidate exists. If the i-th expert candidate exists, the process goes to step S516, and if the i-th expert candidate does not exist, the process returns to the higher level process (step S305).

  (Step S516) The opinion information acquisition unit 134 acquires all document identifiers corresponding to the i-th expert candidate.

  (Step S517) The opinion information acquisition unit 134 acquires the number of sentences whose posture flag is “positive” in all documents corresponding to all document identifiers acquired in step S516. The number of such sentences is the number of positive sentences.

  (Step S518) The opinion information acquisition unit 134 acquires the number of sentences whose posture flag is “negative” in all the documents corresponding to all the document identifiers acquired in step S516. The number of such sentences is the number of negative sentences.

  (Step S519) The opinion information acquisition unit 134 acquires the total number of opinions in all the documents corresponding to all the document identifiers acquired in step S516. Here, the total number of opinions is usually the number of all opinion sentences.

  (Step S520) The opinion information acquisition unit 134 acquires the ratio of the number of positive sentences.

  (Step S521) The opinion information acquisition unit 134 increments the counter i by 1, and returns to step S515.

  In addition to the number of positive opinion sentences and negative opinion sentences acquired in the flowchart of FIG. 5, the total number of positive and negative opinions, the ratio of the number of positive sentences, the degree of association, the domain type, etc. It is an example.

  Next, the first relevance level acquisition process of step S305 will be described with reference to the flowchart of FIG.

  (Step S601) The degree-of-association acquisition unit 135 substitutes 1 for a counter i.

  (Step S602) The relevance degree acquiring unit 135 determines whether or not the i-th document exists. If the i-th document exists, the process goes to step S603, and if not, the process goes to step S606.

  (Step S603) The relevance level acquisition unit 135 acquires the relevance level between the i-th document and the topic.

  (Step S604) The relevance level acquisition unit 135 associates and stores the document identifier of the i-th document and the relevance level.

  (Step S605) The degree-of-association acquisition unit 135 increments the counter i by 1, and returns to step S602.

  (Step S606) The relevance degree acquiring unit 135 substitutes 1 for the counter i.

  (Step S607) The relevance degree acquisition unit 135 determines whether or not the i-th expert candidate exists. If the i-th expert candidate exists, the process goes to step S608, and if it does not exist, the process returns to the higher level process (step S306).

  (Step S608) The degree-of-association acquisition unit 135 acquires all document identifiers corresponding to the i-th expert candidate from the buffer.

  (Step S609) The relevance level acquisition unit 135 acquires one or more relevance levels corresponding to all the document identifiers acquired in step S608.

  (Step S610) The association degree acquisition unit 135 stores the association degree acquired in Step S609 in the buffer in association with the i-th expert candidate.

  (Step S611) The degree-of-association acquisition unit 135 increments the counter i by 1, and returns to step S502.

  Next, the second relevance level acquisition process of step S305 will be described with reference to the flowchart of FIG.

  (Step S701) The degree-of-association acquisition unit 135 substitutes 1 for a counter i.

  (Step S702) It is determined whether or not an i-th expert candidate exists. If the i-th expert candidate exists, the process goes to step S703, and if it does not exist, the process returns to the upper process (step S306).

  (Step S703) The relevance level acquisition unit 135 acquires all document identifiers corresponding to the i-th expert candidate from the buffer.

  (Step S704) The degree-of-association obtaining unit 135 substitutes 1 for a counter j.

  (Step S705) The degree-of-association acquisition unit 135 determines whether or not the j-th document identifier is present among the document identifiers acquired in step S703. If the jth document identifier exists, the process goes to step S706, and if not, the process goes to step S710.

  (Step S706) The degree-of-association acquisition unit 135 acquires the number of expert candidates corresponding to the j-th document identifier.

  (Step S707) The relevance degree acquisition unit 135 calculates the relevance degree between the document identified by the j th document identifier and the i th expert candidate using the number of expert candidates acquired in step S706. . The arithmetic expression for calculating the degree of association between the document and the expert candidate is a decreasing function with the number of expert candidates as a parameter.

  (Step S708) The association degree acquisition unit 135 temporarily stores the association degree calculated in step S707 in the buffer in association with the jth document identifier and the i th expert candidate.

  (Step S709) The association degree acquisition unit 135 increments the counter j by 1, and returns to step S705.

  (Step S710) The association degree acquisition unit 135 increments the counter i by 1, and returns to step S702.

  The score calculation process in step S306 will be described with reference to the flowchart of FIG.

  (Step S801) The score calculation unit 136 substitutes 1 for a counter i.

  (Step S802) The score calculation unit 136 determines whether or not the i-th expert candidate exists. If the i-th expert candidate exists, the process goes to step S803, and if it does not exist, the process returns to the host process (step S307).

  (Step S803) The score calculation unit 136 acquires opinion information corresponding to the i-th expert candidate.

  (Step S804) The score calculation unit 136 acquires the degree of association corresponding to the i-th expert candidate. The degree of association here is the first degree of association, the second degree of association, or the first degree of association and the second degree of association.

  (Step S805) The score calculation unit 136 calculates a score using the opinion information acquired in step S803 and the degree of association acquired in step S804.

  (Step S806) The score calculation unit 136 stores the score calculated in step S805 in the buffer in association with the i-th expert candidate.

  (Step S807) The score calculation unit 136 increments the counter i by 1, and returns to step S802.

  Hereinafter, a specific operation of the expert search system 1 in the present embodiment will be described.

  Now, it is assumed that the user inputs the topic “clear education” to the terminal device 11. Then, the terminal reception unit 110 of the terminal device 11 receives the topic “clear education”. Then, the terminal transmission unit 111 transmits the topic “clear education” received by the terminal reception unit 110 to the expert search device 13.

  Next, the reception unit 130 of the expert search device 13 receives the topic “clear education”. Then, the document search unit 131 performs a so-called Web search using the topic “clear education” as a key, and acquires one or more documents. Then, the document search unit 131 acquires two or more Web pages. Here, it is assumed that one Web page is, for example, the Web page shown in FIG.

  Next, the expert candidate extraction unit 132 extracts, for example, the expert candidate “Yo Yamada” related to the topic “clear education” from the document in FIG. 9. The expert candidate extraction unit 132 extracts the expert candidate “Yamada ○ o” using the technique of proper name recognition.

  Next, the opinion sentence extraction unit 133 extracts one or more sentences from the document of FIG. Then, the opinion sentence extraction unit 133 performs syntax analysis on the opinion sentence obtained by each sentence. The syntax analysis is performed by, for example, KNP (see http://nlp.kuee.kyoto-u.ac.jp/nl-resource/knp.html).

  Next, the opinion sentence extraction unit 133 determines whether each sentence is an opinion sentence, and substitutes the determination result into the buffer.

  Next, the opinion sentence extraction unit 133 acquires the type and polarity of each opinion sentence. Specifically, the opinion sentence extraction unit 133 acquires, for example, one or more features from the result of syntax analysis, gives the one or more features to a machine learning classifier, and acquires a type. Also, the opinion sentence extraction unit 133 performs syntax analysis on the sentence, for example, and acquires the polarity of the sentence in consideration of the polarity inversion word and the dependency relationship. The technology related to opinion sentence extraction, opinion sentence type, and polarity acquisition is described in the document `` Tetsuji Nakagawa, Kentaro Inui and Sadao Kurohashi: Dependency Tree-based Sentiment Classification using CRFs with Hidden Variables, In Proceedings of Human Language Technologies: The 11th Annual Conference of the North American Chapter of the Association for Computational Linguistics (HLT-NAACL 2010), June 2010. Note that opinion sentence extraction and polarity acquisition may be performed by methods other than those described above.

  And the opinion sentence extraction part 133 acquires the information of the opinion sentence shown in FIG. 10 in a buffer with respect to the sentence of FIG. Opinion sentence information is a set of records having “type”, “polarity”, and “opinion sentence”. Note that the sentence acquired by the opinion sentence extraction unit 133 may be a part of the sentence.

  Next, the opinion information acquisition unit 134 acquires the opinion information of the expert candidate “Yamada ○ o” and the opinion information for the document in FIG. 9 as follows. That is, the opinion information acquisition unit 134 acquires the number of opinion sentences “2” having a positive polarity in the table of FIG. 10 as the number of positive opinions. Further, the opinion information acquisition unit 134 acquires the number “1” of negative opinion sentences in the table of FIG. 10 as a negative opinion number. Furthermore, the opinion information acquisition unit 134 acquires the number of records “5” in the table of FIG. 10 as the number of opinion sentences.

Next, the relevance level acquisition unit 135 calculates the relevance level between the expert candidate and the document using Equation 1 below. It is assumed that the degree-of-association acquisition unit 135 holds information on the arithmetic expression of Formula 1 in advance.

  In Equation 1, aij is the degree of association between document i and expert candidate j. ni is the number of expert candidates extracted from the document i.

  That is, for example, when an expert candidate {Yo Yamada, Kazu Terada} is detected from the document in FIG. 9, the association degree acquisition unit 135 determines whether the expert candidate “Yo Yamada” and the document in FIG. The degree of association aij is ½. For example, when only the expert candidate {Yo Yamada} is detected from the document in FIG. 9, the relevance degree acquisition unit 135 determines the relevance degree aij between the expert candidate “Yo Yamada” and the document in FIG. 9. Becomes 1.

  Further, the relevance level acquisition unit 135 calculates a relevance level between each document and the topic “clear education”. For example, the degree-of-relevance acquisition unit 135 acquires the number of appearances “14” of the topic “clear education” in the document of FIG. Then, the relevance level acquisition unit 135 acquires the relevance level ri between the document of FIG. 9 and the topic “clear education” by an increasing function using the number of appearances as a parameter.

Next, it is assumed that the score calculation unit 136 calculates the score Ej of the expert candidate j using the following formula 2.

It is assumed that the quality score qi of the document i is calculated by the following mathematical formula 3.

  The feature types used in Equation 3 are a bias term, the number of positive opinions, the number of negative opinions, and the total number of opinions. In Equation 3, w is a model parameter. A model parameter is a parameter given to a feature type. When the model parameter is (a bias term parameter, a positive opinion number parameter, a negative opinion number parameter, a total opinion parameter), the model parameter w is, for example, w = (− 1.44, 0.06, 0). 390, 0.239). In Equation 3, xi is a set of features of document i. Here, xi is (bias term, number of positive opinions, number of negative opinions, total number of opinions). The bias term is a parameter independent of the document and determines the shape of the quality score function.

  Further, the score calculation unit 136 acquires the opinion information described above. The opinion information here is a set of features xi (1, 2, 1, 5) of the document i.

  Then, the score calculation unit 136 calculates “w · x” (see Equation 3) constituting the quality score qi of the document i as “wx = −1.44 + 0.12 + 0.390 + 1.195 = 0.265”. .

  Next, the score calculation unit 136 calculates the quality score qi of the document i as “qi = 0.666” according to Equation 3.

  Next, the score calculation unit 136 relates to the expert candidate “Yamada ○ Oo” according to Equation 2, the degree of relevance ri of the document i from which “Yamada ○ Oo” is extracted, and the quality score qi of the document i , The document i and the degree of association aij between the expert candidate “Yamada ○ o” are multiplied. Then, the score calculation unit 136 calculates, for each document, the product of the relevance level ri, the quality score qi, and the relevance level aij according to Equation 2, and calculates the sum of the calculated values as the expert candidate “Yamada ○ Calculated as “husband” score.

  The above processing is also performed on other expert candidates, and the score of each expert candidate is acquired.

  Next, the expert selection part 137 acquires one or more experts whose score is higher than a threshold value, for example, from one or more expert candidates using the calculated score. Then, the document output unit 139 sorts the experts in the order of the scores and transmits them to the terminal device 11.

  In addition, the document output unit 139 transmits the URL of one or more documents corresponding to each expert to the terminal device 11.

  Next, the terminal receiving unit 112 of the terminal device 11 receives from the expert search device 13 URLs of one or more experts and documents sorted in the order of score. Then, the terminal output unit 113 outputs the URL of one or more experts and documents received by the terminal receiving unit 112.

  As described above, according to the present embodiment, it is possible to search for an expert with high accuracy for a Web document.

  Also, according to the present embodiment, information written by an expert can be ranked and presented from a large number of information on the Web of the cobblestone chaos. This makes it possible to access more reliable information and enhance user convenience.

  Moreover, according to this Embodiment, it becomes possible to show a user the list of experts about a topic, or the information which the expert transmitted.

  Note that the processing in the present embodiment may be realized by software. Then, this software may be distributed by software download or the like. Further, this software may be recorded and distributed on a recording medium such as a CD-ROM. This also applies to other embodiments in this specification. Note that the software that implements the expert search device according to the present embodiment is the following program. That is, the program includes, for example, a computer from a reception unit that receives a topic, a document search unit that acquires one or more documents related to the topic from one or more Web server devices, and the one or more documents. An expert candidate extraction unit that extracts one or more expert candidates, an opinion sentence extraction unit that extracts one or more opinion sentences that are sentences related to the opinions of the topic from the one or more documents, and the one or more opinions Opinions for obtaining, from each sentence, opinion information including the number of positive opinion sentences, which is the number of positive opinion sentences, and the number of negative opinion sentences, which is the number of negative opinion sentences, for each of the one or more expert candidates An information acquisition unit, a score calculation unit that calculates the expert score for the topic using the opinion information for each of the one or more expert candidates, and the one or more specialists using the score House From the coenzyme, and experts selector which acquires one or more experts, a program to function as an expert output unit for outputting one or more experts the expert selection unit has acquired.

  FIG. 11 shows the external appearance of a computer that executes the program described in this specification to realize the expert search device or the like of the above-described embodiment. The above-described embodiments can be realized by computer hardware and a computer program executed thereon. FIG. 11 is an overview diagram of the computer system 340, and FIG. 12 is a block diagram of the computer system 340.

  In FIG. 11, a computer system 340 includes a computer 341 including an FD drive and a CD-ROM drive, a keyboard 342, a mouse 343, and a monitor 344.

  In FIG. 12, in addition to the FD drive 3411 and the CD-ROM drive 3412, the computer 341 stores an MPU 3413, a bus 3414 connected to the CD-ROM drive 3412 and the FD drive 3411, and a program such as a bootup program. A RAM 3416 for temporarily storing application program instructions and providing a temporary storage space; and a hard disk 3417 for storing application programs, system programs, and data. Although not shown here, the computer 341 may further include a network card that provides connection to the LAN.

  A program that causes the computer system 340 to execute the functions of the expert search device or the like of the above-described embodiment is stored in the CD-ROM 3501 or FD 3502, inserted into the CD-ROM drive 3412 or FD drive 3411, and further a hard disk 3417 may be transferred. Alternatively, the program may be transmitted to the computer 341 via a network (not shown) and stored in the hard disk 3417. The program is loaded into the RAM 3416 at the time of execution. The program may be loaded directly from the CD-ROM 3501, the FD 3502, or the network.

  The program does not necessarily include an operating system (OS), a third party program, or the like that causes the computer 341 to execute the functions of the expert search device of the above-described embodiment. The program only needs to include an instruction portion that calls an appropriate function (module) in a controlled manner and obtains a desired result. How the computer system 340 operates is well known and will not be described in detail.

  In the above program, in the step of transmitting information, the step of receiving information, etc., processing performed by hardware, for example, processing performed by a modem or an interface card in the transmitting step (only performed by hardware). Not included).

  Further, the computer that executes the program may be singular or plural. That is, centralized processing may be performed, or distributed processing may be performed.

  Further, in each of the above embodiments, it goes without saying that two or more communication units existing in one apparatus may be physically realized by one medium.

  In each of the above embodiments, each process (each function) may be realized by centralized processing by a single device (system), or by distributed processing by a plurality of devices. May be.

  The present invention is not limited to the above-described embodiments, and various modifications are possible, and it goes without saying that these are also included in the scope of the present invention.

  As described above, the expert search device according to the present invention has an effect that an expert can be searched with high accuracy for a Web document, and is useful as a Web search engine for experts.

DESCRIPTION OF SYMBOLS 1 Expert search system 11 Terminal apparatus 12 Web server apparatus 13 Expert search apparatus 110 Terminal reception part 111 Terminal transmission part 112 Terminal reception part 113 Terminal output part 121 Document storage part 122 Document transmission part 130 Reception part 131 Document search part 132 Specialization House candidate extraction unit 133 Opinion sentence extraction unit 134 Opinion information acquisition unit 135 Relevance degree acquisition unit 136 Score calculation unit 137 Expert selection unit 138 Expert output unit 139 Document output unit

Claims (6)

A receptionist that accepts topics,
A document search unit for acquiring one or more documents related to the topic from one or more Web server devices;
An expert candidate extraction unit that extracts one or more expert candidates from the one or more documents;
An opinion sentence extraction unit that extracts one or more opinion sentences that are sentences related to the opinion of the topic from the one or more documents;
An opinion information acquisition unit that acquires, from each of the one or more expert sentences, opinion information that is information about the opinion sentence for each of the one or more expert candidates;
For each of the one or more expert candidates, using the opinion information, a score calculation unit that calculates a professional score for the topic;
Using the score, an expert selection unit that obtains one or more experts from the one or more expert candidates;
An expert search device comprising: an expert output unit that outputs one or more experts acquired by the expert selection unit. For one or more expert candidates extracted by the expert candidate extraction unit and one or more documents acquired by the document search unit, for the document in which each expert candidate appears, A relevance level acquisition unit that acquires the relevance level for each of the one or more expert candidates,
The score calculation unit
The expert search device according to claim 1, wherein, for each of the one or more expert candidates, an expert score for the topic is calculated using the opinion information and the degree of association. The expert output unit
The expert search device according to claim 1, wherein one or more experts are sorted and output in descending order of the score. The expert output unit
The one or more documents or one or more opinion sentences from which the score of one or more experts acquired by the expert selection unit is calculated instead of the output of the one or more experts are output. The expert search device according to any one of claims 1 to 3. A document output unit that outputs one or more documents corresponding to some or all of the one or more experts, or one or more access information that is information for accessing the one or more documents; The expert search device according to any one of claims 1 to 3, further comprising: Expert search method realized by reception unit, document search unit, expert candidate extraction unit, opinion sentence extraction unit, opinion information acquisition unit, relevance level acquisition unit, score calculation unit, expert selection unit, and expert output unit Because
The reception unit receives a topic;
A document search step in which the document search unit acquires one or more documents related to the topic from one or more Web server devices;
The expert candidate extraction unit extracts one or more expert candidates from the one or more documents;
An opinion sentence extraction step in which the opinion sentence extraction unit extracts one or more opinion sentences that are sentences relating to the opinions of the topic from the one or more documents;
The opinion information acquisition unit includes, from the one or more opinion sentences, opinion information including the number of positive opinion sentences that is the number of positive opinion sentences and the number of negative opinion sentences that are the number of negative opinion sentences, An opinion information acquisition step to be acquired for each of one or more expert candidates;
The association degree acquisition unit determines one or more documents corresponding to the one or more expert candidates, and determines the association degree between the one or more documents and the topic for each of the one or more expert candidates. Relevance level acquisition step acquired in
A score calculating step in which the score calculation unit calculates a score of expertise for the topic using the opinion information and the degree of association for each of the one or more expert candidates;
The expert selection unit obtains one or more experts from the one or more expert candidates using the score; and
An expert search method comprising: an expert search step, wherein the expert output unit includes an expert output unit that outputs one or more experts acquired by the expert selection unit.