JP2005196668A - Query answering device and query answering system - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To obtain a proper answer without impairing processing efficiency. <P>SOLUTION: A query answering device comprises a searched document processing part 40 for analyzing a searched document, and creating a document index information for searching the searched document and a word information obtained by associating a document identifier, the attribute and an appearance position with an answer word for giving the answering to the query, and a search processing part 50 for acquiring a search word as a cornerstone of the query, and acquiring the searched document identifier fitted to the search word in reference to the word information, and an answer processing part 60 for extracting an answer word in reference to the word information on the basis of the document identifier, the search word and the kind of query, at a client 20 side. The document identifier and the search word from the search processing part at a client side are managed in corresponding to the query, and a session management part 1300 outputting the search word and the document identifier to the query extracting part is mounted at a server 30 side. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

  The present invention relates to a question answering apparatus and a question answering system for obtaining an answer to a question sentence from information held in advance.

Japanese Patent Application Laid-Open No. 2004-151867 discloses a question answering system that obtains an answer to a question sentence from information held in advance and responds.
According to the invention of Patent Document 1, a morpheme of a question sentence is analyzed, a question analysis process for obtaining a search word and a question type that are the main contents of a question, and a document set (searched document) held in advance, A document search process for searching a document in which a search term is indicated, and in the document obtained by the search, a score is given to the word according to the question type based on the search word and the appearance frequency, and the answer is based on the score An answer extraction process for determining a word to be performed is performed, and an answer corresponding to the question content is acquired.

For example, a description will be given of a case in which a document with a search target document is held as “About fuel cell” and a text indicating “What is a battery using fuel?” Is input as a question sentence.
In the question analysis process, morphological analysis of the question sentence is performed, and words separated by this analysis are acquired. That is, in the question analysis process, morphemes are analyzed as “What is the battery that uses fuel?” And words such as “fuel” and “battery” are acquired.

  The document search process uses each acquired word as a search word, and checks whether the search word is included in the search target sentence. In other words, in the document search process, the searched sentence is morphologically analyzed as “about fuel cell”. As a result, the “fuel cell” in the searched document is divided into “fuel” and “battery”, and the searched word corresponding to the question sentence is obtained by matching the divided word with the word acquired as the search word. Identified.

In the answer extraction process, a score is assigned to each word of the specified document to be searched based on the question type and the search word / appearance frequency, and the answer candidate word of the question sentence is extracted based on the given score.
Japanese Patent Laid-Open No. 2002-132911

  By the way, in the searched document shown as “About fuel cell”, the morphological analysis of “About fuel cell” is better than the morphological analysis of “Search about fuel cell”. Is appropriate. However, the answer to the question sentence obtained from the searched document should be “fuel cell”.

  In order to solve such a problem, it is conceivable to individually perform the morphological analysis method of the searched document in the document search processing and the morphological analysis method of the searched document in the answer extraction processing according to each processing. In the client-server model, in order to perform session management on the client side, it is necessary to perform data communication between the client and the server on the analysis result obtained by each morphological analysis method. Since this data communication increases in proportion to the number of documents and the length of the document, there is a risk that the processing efficiency may be reduced as the amount of data communication increases.

  Therefore, in view of the above-described problems, an object of the present invention is to provide a question answering apparatus and a question answering system that can obtain an appropriate answer without reducing processing efficiency.

The present invention adopts the following configuration in order to solve the above points.
<Configuration 1>
The question answering device of the present invention receives a search target document, analyzes a search target document processing unit that analyzes a morpheme of the search target document, acquires a search term that is a key point of the question content from the question sentence, Response processing unit including a search processing unit that acquires a document that conforms to the above and a response processing unit that assigns a score to each word of the acquired document and extracts a word that is an answer to the question sentence based on the score In the apparatus, the searched document processing unit includes: a first document analyzing unit that generates document index information in which each word obtained by analyzing a morpheme of an accepted searched document is associated with a document identifier of the searched document; Based on the words obtained by the first document analysis unit, words that can be answered in the question sentence are generated as appropriate, the generated words are replaced with the original words, and these words are used as answer words. For the answer word, the attribute and the searched document And a second document analysis unit that generates word information that associates the document identifier of the searched document with the search processing unit, based on each word obtained by analyzing the morpheme of the question sentence A question analysis unit that appropriately generates words that can be answered in a question sentence, adds the generated words to words obtained by analyzing the morphemes of the question sentence, and outputs these words as search words; and a document index A document search unit that refers to information and obtains a document identifier of a document that matches the search term, the answer processing unit refers to the word information, and the document identifier from the document search unit and the question analysis unit And an answer extracting unit for extracting a word to be an answer to the question sentence from the word associated with the document identifier based on the search word from.

<Configuration 2>
The question analysis unit in the first configuration includes a question sentence morpheme analysis processing unit that performs the same morphological analysis as the morpheme analysis in the first document analysis unit on the question sentence, and each word acquired in the question sentence morpheme analysis process A search word post-processing unit that generates a compound word that can be an answer target of a question sentence and eliminates fluctuations in the document, and a compound word generated by the search word post-processing part is converted into the question sentence morphological analysis processing unit And a search word addition processing unit for adding to a word acquired in step (1).

<Configuration 3>
The second document analysis unit in the configuration 1 generates a compound word that can be an answer target of the question sentence based on the word obtained by the first document analysis unit, and also performs word post-processing for eliminating the shaking of the document And a word replacement processing unit that replaces the compound word generated by the word post-processing unit with the compound source word.

<Configuration 4>
A question answering system of the present invention is a system in which a network is provided between a client and a server, and the server accepts a searched document and analyzes a searched document processing unit that analyzes a morpheme of the searched document, A search processing unit that obtains a search word that is the key to the contents of a question from a sentence, acquires a document that matches the search word, and assigns a score to each word of the acquired document, and a question sentence based on the score In the question answering system, the client includes an input processing unit that extracts a question sentence from the user, and an output unit that outputs the answer to the question sentence. The searched document processing unit generates a document index information in which each word obtained by analyzing a morpheme of an accepted searched document is associated with a document identifier of the searched document; and First Based on the words obtained by the document analysis unit, the words that can be answered in the question sentence are generated as appropriate, the generated words are replaced with the original words, and these words are used as the answer words. A second document analysis unit that generates word information that associates the attribute, the appearance position in the searched document, and the document identifier of the searched document, and the search processing unit analyzes a morpheme of the question sentence Based on each word obtained, generate words that can be answered in the question sentence as appropriate, add the generated words to the words obtained by analyzing the morphemes of the question sentence, and output these words as search words And a document search unit that obtains a document identifier of a document that matches the search term by referring to the document index information, and the answer processing unit refers to the word information from the document search unit. Document identifier and quality An answer extraction unit that extracts a word as an answer to a question sentence from words associated with the document identifier based on a search word from the analysis unit, and the client includes the search word from the question analysis unit And a session management unit that manages the document identifier from the document search unit in association with a question sentence acquired by the input unit, and outputs the search term and the document identifier to the question extraction unit. And

<Configuration 5>
The session management unit according to Configuration 4 displays a search target document associated with a document identifier to be acquired, causes a user to select a desired search target document, and associates the search target document with the selected search target document. The document identifier is output to the answer extraction unit.

<Configuration 6>
The question analysis unit according to Configuration 4 acquires the question sentence by the question sentence morpheme analysis processing unit that performs the same morphological analysis as the morpheme analysis in the first document analysis unit, and the question sentence morpheme analysis process. Each word is compounded to generate a compound word that can be an answer target of the question sentence, and a search word post-processing unit that eliminates the fluctuation of the document, and the compound word generated by the search word post-processing part is analyzed by the question sentence morpheme analysis And a search word addition processing unit to be added to the words acquired by the processing unit.

<Configuration 7>
The second document analysis unit according to Configuration 4 generates a compound word that can be an answer target of a question sentence based on the word obtained by the first document analysis unit, and a word that eliminates shaking of the document A post-processing unit and a word replacement processing unit that replaces the compound word generated by the word post-processing unit with a compound source word are provided.

  According to the question answering device of the present invention, the morpheme analysis performed for acquiring the search word that is the key of the question sentence and the morpheme analysis of the searched document are performed in the same method, and the word obtained by the morphological analysis of the searched document Based on the above, a word that can be an answer to a question sentence is generated, an answer word obtained by replacing the generated word with the original word is referred to, and an answer word that matches the search word is extracted as an answer to the question sentence Therefore, the morphological analysis of the searched document is performed in the same way as the morphological analysis of the question sentence, and the morphological analysis for extracting the answer word using the analysis result is performed. Since there is no need to perform morpheme analysis by another method for extracting the response word from the searched document after performing the morphological analysis to acquire the searched document corresponding to Can get .

  Further, according to the question answering system of the present invention, the searched document is analyzed, the document index information for searching the searched document, and the document identifier, attribute, and appearance position for the answer word for answering the question sentence To-be-searched document processing unit for creating the word information that associates the search term, and the search word that is the key to the question sentence is acquired, and the document identifier of the searched document that matches the search word is acquired by referring to the word information The server includes an answer processing unit that extracts an answer word by referring to word information based on the document identifier, the search word, and the question type, and the document identifier and the search word from the server-side search processing unit. Is associated with a question sentence, and a session management unit that outputs the search word and document identifier to the question extraction unit is provided on the client side, thereby obtaining the document identifier of the searched document and extracting the answer word No The morpheme analysis method of each is performed, and the position of each search word in each searched document among the results of the search processing unit does not need to be communicated between the client and the server, that is, the document identifier and the search word Therefore, it is possible to obtain an appropriate answer efficiently without increasing the amount of data communication.

  Hereinafter, an example of a question answering system having a client / server configuration will be described in detail with reference to the drawings.

  As shown in FIG. 1, the question answering system 10 of the present invention includes a server group 20 and a client 30, and the server group 20 and the client 30 are connected to a network as a transmission path (not shown). Data communication is performed between the server group 20 and the client 30 via the network.

The server group 20 includes a searched document processing unit 40, a search processing unit 50, and an answer processing unit 60.
The searched document processing unit 40 is based on a document input unit 200 that receives the searched document 100, a first document analysis unit 300 that analyzes the received document, and an analysis result of the first document analysis unit 300. And a second analysis unit 400 that performs analysis for word editing.
The search processing unit 50 associates a word acquired by morpheme analysis in the first document analysis processing unit 300, an appearance frequency of the word, and an index (document identifier) for searching the search target document. A document index information accumulating unit 500 that holds as index information, a question analyzing unit 700 that analyzes a question sentence from the client 30 and obtains a word (a word group) that is a key of the question content as a search term, and a document index information accumulating unit A document search unit 800 that refers to the unit 500 and acquires a document identifier associated with a word corresponding to the search term is provided.

Based on the analysis result in the second document analysis unit 400, the response processing unit 60 obtains word information in which the document identifier of the search target document, an answer word to be described later, its attribute, and its appearance position are associated. A word information storage unit 600 that holds the information, and an answer extraction unit 900 that refers to the word information stored in the word information storage unit 600 and extracts an answer word as an answer to a question sentence based on a document identifier and a search word from a client. Prepare.
The client 30 manages communication with the server group 20 based on the input unit 1200 that receives the question text 1100, the question text received by the input unit 1200, the search word obtained by the question analysis unit 700, and the document identifier. A session management unit 1300 and an output unit 1400 for presenting an answer from the answer extraction unit 900 to the user are provided.

Next, each part will be described in detail.
The search target document 100 received by the document input unit 200 is, for example, a newspaper, a paper, a patent document, a document (document group) shown on a Web (World Wide Web) page, or the like. When the searched document 100 includes information (words) as an answer to the question sentence 1100, the question answering system 10 can obtain an accurate answer.

  The document input unit 200 reads a file designated as a search target document, such as a file in which the search target document 100 is held and managed on a disc. If the searched document 100 is indicated on a Web page or the like, the document input unit 200 accesses a specified URL (Uniform Resource Locator) by means such as HTTP (Hyper Text Transfer Protocol) to search the searched document 100. To get. In addition, when the searched document 100 is printed on a medium such as paper, the document input unit 200 acquires the print contents of the searched document 100 with the same function as an OCR (Optical Character Reader).

  The first document analysis unit 300 includes a morpheme analysis processing unit 310. The morpheme analysis processing unit 310 analyzes the morpheme of the searched document 100 acquired by the document input unit 200, and obtains words or words obtained by the analysis. Get a word with a specific part of speech. Each acquired word is a word that can be searched in the document search unit 800 described later, and a document identifier for identifying a plurality of documents (searched document 100) and an appearance frequency in the document are morphological analysis processing. Corresponding in section 310. This associated information is held in the document index information storage unit 500 as document index information.

  When the second analysis unit 400 receives the text analyzed by the first document analysis unit 300, the second analysis unit 400 generates a word that can be an appropriate answer as an answer of the question answering system 10 as necessary. The second analysis unit 400 associates a document identifier, an attribute, an appearance position in a document, and the like with a response word such as a generated word or a word obtained by analyzing a morpheme with the first document analysis unit 300. The associated information is held in the word information storage unit 600 as word information.

By the way, the second analysis unit 400 includes a word post-processing unit 410 and a word replacement processing unit 420. When the word post-processing unit 410 receives the document divided by the morpheme analysis processing unit 310, When it is appropriate to combine a plurality of words into one from the dependency or character type separation, the combined words are generated. For example, if each word is divided into “○ × Electricity”, “Industry”, and “Inc.” in the morphological analysis, “Inc.” is considered to be the prefix or suffix of the company name. Alternatively, words included in a range that can be a company name in the back are combined into one word, and words such as “◯ × Electric Industry Co., Ltd.” and “O Corporation” are generated.
In addition, the word post-processing unit 410 performs a normalization process to unify “computer” so as to eliminate a difference called “computer” and “computer”.

  The word replacement processing unit 420 replaces the word created by the word post-processing unit 410 with the word that is the creation source. That is, the word replacement processing unit 420 converts the words “Ox Electric”, “Industry”, and “Inc.” into the generated “Ox Electric Industry Co., Ltd.” or “Ox Electric Co., Ltd.” in the previous example. Execute the replacement process so as not to overlap. The word replacement processing unit 420 associates the replaced word and the word that does not need to be replaced as an answer word, associates the answer word with an attribute, an appearance position, and a document identifier, and uses this as word information. Generate. The generated word information is held in the word information storage unit 600.

  The question analysis unit 700 performs a process of extracting, as a search term, a word that should be the key of the question content from the question sentence 1100 received from the user by the input unit 1200. This search term is a word (word group) that is included in both the document index information in the document index information storage unit 500 and the word information in the word information storage unit 600. Incidentally, the question analysis unit 700 is mounted on the server group 20 side or the client 30 side. In this embodiment, an example in which the question analysis unit 700 is mounted on the server group 20 side will be described.

  The above-described question analysis unit 700 includes a question sentence morpheme analysis processing unit 710, a search word post-processing unit 720, and a search word addition processing unit 730. The question sentence morpheme analysis processing unit 710 analyzes the morpheme of the question sentence 1100 acquired by the input unit 1200 and acquires a word obtained by the analysis or a word of a specific part of speech. That is, the question sentence morpheme analysis processing unit 710 has the same function as the morpheme analysis processing unit 310 described above.

  When the query word post-processing unit 720 receives the question sentence divided in units of morphemes by the question sentence morpheme analysis processing unit 710, it is appropriate to combine a plurality of words into one due to dependency and character type separation. In this case, a single word is generated. That is, the search word post-processing unit 720 has the same function as the word post-processing unit 410 described above.

  The search word addition processing unit 730 performs processing for adding the word generated by the search word post-processing unit 720 to the word that is the creation source. For example, when a “fuel cell” is generated based on “fuel” and “battery” in the search word post-processing unit 720, the search word addition processing unit 730 adds “fuel cell” to “fuel” and “battery”. To do. The search word addition processing unit 730 outputs the word group of “fuel”, “battery”, and “fuel cell” to the document search unit 800 and the session management unit 1300 as search words.

  When the document search unit 800 acquires the search word from the search word addition processing unit 730, the document search unit 800 refers to the document index information in the document index information storage unit 500 and searches for a word corresponding to the search word. This search can be performed by a general technique based on the appearance frequency in general known documents, the appearance frequency in each sentence included in the document, the appearance frequency in the question sentence, and the like. When the document search unit 800 acquires a document identifier associated with the word obtained by the search, the document search unit 800 outputs the document identifier to the session management unit 1300.

  Since the question sentence morpheme analysis processing unit 710 of the question analysis unit 700 has the same function as the morpheme analysis processing unit 310, the word (search word) acquired from the question analysis unit 700 is the same as the morpheme analysis processing unit 310. May contain words obtained by analyzing morphemes. That is, the morpheme analysis processing unit 310 that analyzes the searched document 100 and the question sentence morpheme analysis processing unit 710 that analyzes the question sentence 1100 have the same morpheme analysis function. When the same word is included in 1100, the question analysis unit 700 outputs a search word including a word obtained by morphological analysis by the morpheme analysis processing unit 310.

  Further, since the word generated by the search word post-processing unit 720 is added by the search word addition processing unit 730, the search word output by the question analysis unit 700 is a word that is not output by the morphological analysis processing unit 310, that is, a document A word that is not included in the index information storage unit 500 is included. Therefore, such a word has an appearance frequency of 0 in the document index information of the document index information storage unit 500, and it is possible to use this appearance frequency to search for a document that does not affect the document search. it can. As a result, the document search unit 800 can perform a document search without being affected by the words added by the search word addition processing unit 730, and the answer extraction unit 900 described later added by the search word addition processing unit 730. It is possible to perform a search that includes words.

  The session management unit 1300 performs management in which the document identifier associated with the word obtained by the search by the document search unit 800, the search word from the question analysis unit 700, and the question sentence 1100 acquired by the input unit 1200 are associated with each other. Do. For example, when the document identifier obtained based on the search word in the document search unit 800 is a Web URL, the session management unit 1300 accesses the URL and displays the content on a display input unit (not shown). In order to allow the user to select a desired identifier, management of the question sentence 1100, the search word, and the document identifier is performed in order to receive input from the user.

  The session management unit 1300 is provided in the client 30, and a complicated logical operation such as a plurality of document identifiers and a plurality of search terms by a plurality of question sentences 1100 is performed by the client. There is no need to manage.

  When the answer extraction unit 900 obtains an identifier indicating a document from which an answer from the session management unit 1300 is extracted, a question type, and a search word, the answer extraction unit 900 refers to the word information storage unit 600 and attributes according to the question type. Are extracted from word information. The question type is, for example, a person name, a place name, or a technical name, and can be obtained by analyzing a conventionally known question sentence 1100. In addition to this method, the questioner may specify using the input terminal, or if the system answers only a predetermined question type (for example, only the name of the person), only the target may necessarily be answered. .

  In addition, the answer extraction unit 900 obtains the importance of the answer based on the positional relationship in the document between the search term from the session management unit 1300 and the answer candidate word, and gives an answer highly relevant to the question sentence 1100. Extract. By the way, the search word may include a word that is not included in the word information storage unit 600, that is, a word that is replaced by the replacement in the word replacement processing unit 420 is stored in the word information storage unit 600. Not accumulated. However, when an answer highly relevant to the question sentence 1100 is obtained using such a search word, it is not used for evaluation of the positional relationship with the answer candidate word. There is no effect.

  When the output unit 1400 receives the answer obtained by the answer extraction unit 900, the output unit 1400 considers convenience such as associating the answer with a question sentence managed by the session management unit 1300 and displaying it to the user.

Next, the operation of the question answering system 10 will be described in detail using specific examples of the searched document 100 and the question sentence 1100.
The search target document 100 will be described using two document examples D1 and D2, as shown in FIG. D1 says “Oxaro:“ Development of fuel cells ”Hydrogen and oxygen fuel cells are indicated as“ ... ”, and D2 is a battery that generates electricity using hydrogen and oxygen as fuel. × Developed by Taro (● Professor) ″ Is shown.

First, the operation of the question answering system 10 until D1 and D2 as the search target document 100 are held in the document index information storage unit 500 and the word information storage unit 600 will be described.
When the document D1 and the document D2 acquired by the document input unit 200 are sent to the morpheme analysis processing unit 310 of the first document analysis unit 300, the morpheme analysis processing unit 310 divides the document D1 and the document D2 into morpheme units. Taking document D1 as an example, document D1 is divided into “Taro XX:“ Development of fuel cell ”. Fuel cell with hydrogen and oxygen is divided into“ ... ”.

  At this time, as shown in FIG. 3, the morphological analysis processing unit 310 obtains the part of speech of the divided word and acquires the independent word. That is, the morphological analysis processing unit 310 acquires “◯ ×”, “Taro”, “fuel”, “battery”, “development”, “hydrogen”, and “oxygen”. In this embodiment, for the sake of convenience of explanation, an example in which only independent words are acquired will be described. However, the present invention is not limited to this.

  After that, the morphological analysis processing unit 310 obtains the appearance frequency of each acquired word in the document as the appearance frequency. That is, the appearance frequency of “fuel” and “battery” is twice, and the appearance frequency of other words is one. The morphological analysis processing unit 310 associates the appearance frequency with the word and also associates the document identifier “D1”, and then causes the document index information storage unit 500 to store the document identifier “D1” as the document index information. When the morphological analysis processing unit 310 performs the above-described processing for the document D2, the document index information is generated so that the document index information shown in FIG. Is output to the document index information storage unit 500.

  Further, the morpheme analysis processing unit 310 outputs the result of the morpheme analysis to the word post-processing unit 410 of the second analysis unit 400. When the word post-processing unit 410 acquires the word acquired by the morphological analysis processing unit 310, the word post-processing unit 410 combines the acquired words and generates an appropriate word as an answer. For example, the word post-processing unit 410 that acquired “Ox”, “Taro”, “Fuel”, “Battery”, “Development”, “Hydrogen”, and “Oxygen” in the analysis of the document D1 is based on, for example, the relationship of “last name + first name”. Then, the word “Ox Taro” is generated from “Ox” and “Taro”, and the word “Fuel cell” is generated from “Fuel” and “Battery”. That is, the word post-processing unit 410 creates a word as “a compound noun” by “noun” + “noun”.

  Further, the word post-processing unit 410 is based on the particle of “(Omitted) Oxygen fuel cell is ...” shown in the document D1, that is, a range where a plurality of words are combined. May be determined. In addition, in the case of noun + “battery”, the word post-processing unit 410 may hold knowledge indicating the type of battery in advance and use it to create a compound word.

In the word substitution processing unit 420, words such as “XX Taro” and “fuel cell” generated by the word post-processing unit 410 and other words “XX”, “Taro”, “fuel”, “battery”, “development”, “ When the original words “hydrogen” and “oxygen” are acquired, the original words are replaced with compound words. That is, the original words “Ox” and “Taro” are replaced with “Ox Taro”, and the original words “fuel” and “battery” are replaced with “fuel cell”. As a result, the word replacement processing unit 420 replaces the original word with a compound word, and acquires each word of “Taro Taro”, “fuel cell”, “development”, “hydrogen”, and “oxygen” as an answer word.
When the word replacement processing unit 420 performs the above-described processing for the document D2, as shown in FIG. 5, word information that associates the attribute with the answer word, the appearance position in the document, and the document identifier is converted into word information. The word information is output to the word information storage unit 600 to be stored in the storage unit 600.

Next, the operation of the question answering system 10 will be described until a “fuel cell developer” whose user asks a person's name is input as a question text at the input unit 1200 of the client 30 and an answer to the question text is obtained.
The question text input by the input unit 1200 of the client 30 is sent to the question analysis unit 700 and the session management unit 1300 of the client 30 via the network. At this time, the “question of person name” question type acquired by the input unit 1200 is also sent to the session management unit 1300.

  In the question analysis unit 700, first, the question sentence morpheme analysis processing unit 710 analyzes the morpheme of the question sentence. That is, the question sentence morphological analysis processing unit 710 divides the question sentence into “fuel cell developers” and obtains independent words as shown in FIG. This morpheme analysis is the same as the morpheme analysis method in the morpheme analysis processing unit 310 described above.

  The search word post-processing unit 720 that acquired the word obtained by the question sentence morphological analysis processing unit 710 generates a compound word based on the appropriately acquired word. The generation of the compound word is performed by the same method as the processing in the word post-processing unit 410 described above. That is, a “fuel cell” is generated from “fuel” and “battery”, and a “developer” is generated from “development” and “person”. And the search word post-processing part 720 performs the normalization process which eliminates the fluctuation | variation of a description.

  Thereafter, the search word addition processing unit 730 adds the word generated by the search word post-processing unit 720 to the word group acquired by the question sentence morphological analysis processing unit 710. That is, the search word addition processing unit 730 adds “fuel cell” and “developer” to “fuel”, “battery”, “development”, and “person”. The word group to which “fuel cell” and “developer” are added by the search word addition processing unit 730 is output to the document search unit 800 and the session management unit 1300 as a search word.

  The document search unit 800 that has acquired the search word from the question analysis unit 700 refers to the document index information storage unit 500 to check whether or not the word corresponding to the search word is included in the search information shown in FIG. If there is a word that matches the search word, the document identifier associated with the word is output to the session management unit 1300. That is, the document search unit 800 determines that the search terms “fuel”, “battery”, and “development” are included in the search information, acquires the document identifiers D1 and D2 associated with the words, The identifier is output to the session management unit 1300.

  The session management unit 1300 displays the question text together with the acquired document identifier on a display input device (not shown). The user selects a desired document identifier based on the display content. At this time, for example, when a URL or the like is associated with the document identifier, the session management unit 1300 accesses the URL and displays all or part of the contents shown on the Web page on the display input device. Thereby, the user can select a desired document identifier based on the displayed contents. Accordingly, the document identifier output by the session management unit 1300 is a document identifier arbitrarily selected by the user so as to include a word (answer word) for answering the question sentence.

  Thereafter, the document identifier selected by the user, the search term from the question analysis unit 700, and the question type are output from the session management unit 1300 to the answer extraction unit 900 via the network.

  By the way, in the session management in the conventional question answering system, in order to acquire a document identifier that matches the search term that is the key to the question sentence, the result of the morphological analysis is performed between the server and the client. That is, information on each word separated by morphological analysis is communicated between the server and the client. Further, after obtaining the document identifier, in order to extract the answer to the question sentence, the result of the morpheme analysis performed again on the searched document associated with the document identifier is transmitted between the server and the client. As a result, the longer the text of the search target document and the question sentence, the greater the amount of data communication and the lower the processing efficiency.

  On the other hand, in the session management of the present invention, it is not necessary to perform morpheme analysis for each document identifier acquisition process and answer word extraction process, and it is not necessary to acquire information of each word separated by morpheme analysis. That is, in the session management of the present invention, only the document identifier and the search word need be communicated, so the amount of data communication can be reduced compared to the conventional case.

  The answer extraction unit 900 that has acquired the document identifier, the search word, and the question type refers to the word information shown in FIG. 5 and, in the answer word associated with the document identifier, W11 and W12 “XX” corresponding to the question type. "Taro" is extracted as an answer. Further, the answer extraction unit 900 extracts “fuel cell”, “development”, “fuel”, and “battery” that match the search word from the answer word associated with the document identifier in the word information, that is, in FIG. R11, R12, R13, R15, R18, R19 and R20 are extracted. The answer extraction unit 900 assigns a score to each extracted answer word based on the attribute, the appearance position relationship in the document, the question type, and the like by a conventionally known method. Each answer word given the score is sent from the answer extraction unit 900 to the output unit 1400.

  The output unit 1400 that has acquired each answer word given a score displays the answer words in descending order of the score, for example. At this time, the output unit 1400 may display a question sentence and the like together with the answer word.

  As described above, according to the question answering system 10 of the present invention, both in the document D1 including the words “fuel cell” and in the document D2 including the individual words “fuel” and “cell”, The document search unit 800 can acquire a document identifier including a key word of the question sentence, that is, “fuel cell”, and the answer extraction unit 900 can obtain the question sentence based on the acquired document identifier, the search word, and the question type. It is possible to extract an appropriate answer word as an answer.

Next, a case will be described in which the question sentence is, for example, “developed by Professor XX × Taro” and the question type is a noun.
The question sentence is analyzed by the question sentence morphological analysis processing section 710 of the question analysis section 700 into “what was developed by Professor Taro X”, and as independent words “○ ×”, “Taro”, “Teacher”, “Development”, “Things” Is obtained. The search word post-processing unit 720 generates “○ × Taro” as a compound word based on “◯ ×” and “Taro”. When the generated compound word is acquired by the search word addition processing unit 730, the search word addition processing unit 730 reads from “XX”, “Taro”, “Teacher”, “Development”, “Things”, and “XX Taro”. A word group (search word) is generated.

  The document search unit 800 refers to the document index information storage unit 500 and acquires the document identifier associated with the word that matches the search word in the document search information shown in FIG. That is, the document search unit 800 acquires D1 and D2, and outputs the document identifier to the session management unit 1300. Having acquired the document identifier, the session management unit 1300 acquires the document identifier from the document search unit 800, acquires the search word from the question analysis unit 700, and acquires the question sentence and the question type from the input unit 1200. The search term and question type are output to the answer extraction unit 900.

  The answer extraction unit 900 refers to the word information storage unit 600 and extracts answer words based on the acquired document identifier, search word, and question type. That is, in the word information of the word information storage unit 600, the answer extraction unit 900 includes “fuel cell”, “hydrogen”, “oxygen”, “fuel”, “battery” in which the question type is a noun in the answer word associated with the document identifier. ", That is, R11, R13, R14, R15, R16, R17, R18 and R19 in FIG. 5 are extracted. Thereby, the answer extraction unit 900 can extract “fuel cell” in which “fuel” and “battery” are collected as an answer to the question sentence.

The answer extraction unit 900 assigns a score to each extracted answer word based on the attribute, the appearance position relationship in the document, the question type, and the like by a conventionally known method. It outputs to the output part 1400 as a reply of a sentence.
By the way, in this embodiment, since words having both parts of speech of nouns and verbs are often not appropriate as answers to question sentences, such answer words are excluded from extraction targets. In addition to excluding answer words that are not suitable as answers, the number of points assigned by the answer extraction unit 900 may be kept low.

  As described above, according to the question answering system 10 of the present invention, the document index information is created by the morphological analysis of the search target document 100 by the first document analysis unit 300, and the second analysis unit 400 combines it. A word is created and replaced with the compound word to create word information. According to the question answering system 10 of the present invention, the question analysis unit 700 generates a compound word by morphologically analyzing the question sentence 1100, and the document search unit 800 uses the document index based on the search word to which the compound word is added. A document identifier associated with an information word is acquired, and an answer word associated with the document identifier is extracted by an answer extraction unit 900 based on a question type or a search word. Even if it does not directly contain the word that is the key of the sentence, it is possible to obtain an identifier indicating the document that is a candidate for the answer to the question sentence, and an appropriate answer word based on the word information associated with the obtained document identifier Can be extracted.

  In the embodiment, the session management is performed on the client 30 side. However, the session management may be performed on the server group 20 side.

It is a block diagram of the question answering system of the present invention. It is a figure which shows the example of the text of a to-be-searched document. It is a figure which shows the morphological analysis example of a to-be-searched document. It is a figure which shows document index information. It is a figure which shows word information. It is a figure which shows the morphological analysis example of a question sentence.

Explanation of symbols

DESCRIPTION OF SYMBOLS 10 Question answering system 20 Server group 30 Client 40 Searched document process part 50 Search process part 60 Answer process part 100 Searched document 200 Document input part 300 1st document analysis part 310 Morphological analysis process part 400 2nd analysis part 410 Word post-processing unit 420 Word replacement processing unit 500 Document index information storage unit 600 Word information storage unit 700 Question analysis unit 710 Question sentence morpheme analysis processing unit 720 Search word post-processing unit 730 Search word addition processing unit 800 Document search unit 900 Answer extraction Part 1100 question sentence 1200 input part 1300 session management part 1400 output part

Claims (7)

A search target document processing unit that accepts a search target document and analyzes a morpheme of the search target document, and a search that acquires a search term that is the key to the contents of a question from a question sentence and acquires a document that matches the search term In a question answering device comprising a processing unit and an answer processing unit that assigns a score to each word of the acquired document and extracts a word that becomes an answer to a question sentence based on the score,
The searched document processing unit includes:
A first document analysis unit that generates document index information in which each word obtained by analyzing a morpheme of an accepted searched document is associated with a document identifier of the searched document;
Based on the words obtained by the first document analysis unit, words that can be answered in the question sentence are generated as appropriate, the generated words are replaced with the original words, and these words are used as answer words. A second document analysis unit that generates word information that associates a word with an attribute, an appearance position in the searched document, and a document identifier of the searched document;
The search processing unit
Based on each word obtained by analyzing the morpheme of the question sentence, a word that can be an answer target of the question sentence is appropriately generated, and the generated word is added to the word obtained by analyzing the morpheme of the question sentence. A question analysis unit that outputs words as search terms;
A document search unit that refers to document index information and obtains a document identifier of a document that matches the search term,
The answer processing department
An answer that refers to the word information and extracts a word that becomes an answer to a question sentence from words associated with the document identifier based on a document identifier from the document search unit and a search word from the question analysis unit A question answering apparatus comprising: an extraction unit. The question analysis unit, a question sentence morpheme analysis processing unit for performing the same morpheme analysis as the morpheme analysis in the first document analysis unit for the question sentence,
Compounding each word acquired in the question sentence morpheme analysis process, generating a compound word that can be an answer target of the question sentence, and a search word post-processing unit that eliminates shaking of the document,
The question answering device according to claim 1, further comprising: a search word addition processing unit that adds a compound word generated by the search word post-processing unit to a word acquired by the question sentence morphological analysis processing unit. The second document analysis unit generates a compound word that can be an answer target of the question sentence based on the word obtained by the first document analysis unit, and a word post-processing unit that eliminates the shaking of the document;
The question answering apparatus according to claim 1, further comprising: a word replacement processing unit that replaces the compound word generated by the word post-processing unit with a compound source word. A system in which a network is established between a client and a server,
The server receives a search target document, analyzes a search target document processing unit that analyzes a morpheme of the search target document, obtains a search term that is a key point of a question content from a question sentence, and obtains a document that matches the search term. A search processing unit for acquiring, and a response processing unit for assigning a score to each word of the acquired document and extracting a word that becomes an answer to the question sentence based on the score,
In the question answering system, the client includes an input unit that obtains a question sentence from a user, and an output unit that outputs an answer to the question sentence.
The searched document processing unit includes:
A first document analysis unit that generates document index information in which each word obtained by analyzing a morpheme of an accepted searched document is associated with a document identifier of the searched document;
Based on the words obtained by the first document analysis unit, words that can be answered in the question sentence are generated as appropriate, the generated words are replaced with the original words, and these words are used as answer words. A second document analysis unit that generates word information that associates a word with an attribute, an appearance position in the searched document, and a document identifier of the searched document;
The search processing unit
Based on each word obtained by analyzing the morpheme of the question sentence, a word that can be an answer target of the question sentence is appropriately generated, and the generated word is added to the word obtained by analyzing the morpheme of the question sentence. A question analysis unit that outputs words as search terms;
A document search unit that refers to document index information and obtains a document identifier of a document that matches the search term,
The answer processing department
An answer that refers to the word information and extracts a word that becomes an answer to a question sentence from words associated with the document identifier based on a document identifier from the document search unit and a search word from the question analysis unit An extraction unit,
The client manages a search word from the question analysis unit and a document identifier from the document search unit in association with a question sentence acquired by an input unit, and extracts the query word and the document identifier from the question A question answering system comprising: a session management unit that outputs to a unit.   The session management unit displays a search target document associated with a document identifier to be acquired, causes a user to select a desired search target document, and sets the document identifier associated with the selected search target document 5. The question answering system according to claim 4, wherein the question answering system outputs the answer to the answer extracting unit. The question analysis unit, a question sentence morpheme analysis processing unit for performing the same morpheme analysis as the morpheme analysis in the first document analysis unit for the question sentence,
Compounding each word acquired in the question sentence morpheme analysis process, generating a compound word that can be an answer target of the question sentence, and a search word post-processing unit that eliminates shaking of the document,
5. The question answering system according to claim 4, further comprising: a search word addition processing unit that adds a compound word generated by the search word post-processing unit to a word acquired by the question sentence morphological analysis processing unit. The second document analysis unit generates a compound word that can be an answer target of the question sentence based on the word obtained by the first document analysis unit, and a word post-processing unit that eliminates the shaking of the document;
5. The question answering system according to claim 4, further comprising: a word replacement processing unit that replaces the compound word generated by the word post-processing unit with a compound source word.

Images