JP4057962B2 - Question answering apparatus, question answering method and program - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a question answering apparatus, a question answering method, and a program for outputting an answer to a question inputted by a user.
[0002]
[Prior art]
As represented by Internet search engines, techniques for searching and ranking documents that match user search requests are widely used. However, the document search can satisfy search requests such as “I want to read newspaper articles about…” or “I want to see a web page about…”, but “Who is the president of XX?” Can't answer directly to questions such as "Whales are going to be extinct?" Since the document search only outputs a document or a passage in the document, the user must search for an answer from the output result.
[0003]
There is a question answering system that directly outputs an answer to a question like the latter. The question answering system does not output a document about XX company (for example, XX company's website) to a question such as "Who is XX company president?" A person's name is output, and an answer such as “Mt. Fuji is 3776 m” is output to a question such as “How tall is Mt. Fuji?”.
[0004]
Such a question answering system has recently attracted attention as a development form of research such as information retrieval and information extraction, and it has become possible to return a direct answer to a user's question to some extent. For example, as in Patent Document 1, a direct response to a certain degree to a user's question text and a rationale document that allows the user to confirm that the answer is an answer to the question text are output. The system is known. The basis document is, for example, a source document from which the system has extracted answers to questions. The user can know which document is the answer based on the basis document.
[0005]
[Patent Document 1]
JP 2002-132812 A
[0006]
[Problems to be solved by the invention]
There may be only one evidence document as described above, but generally there may be a plurality of documents. In the conventional question answering system, when there are a plurality of ground documents, all the ground documents are listed and presented, or representative ground documents are selected and presented. However, the user may want only an answer to a single question, but often desires a series of knowledge and information. In the latter case, if the rationale document relating to the answer of a certain question contains useful information (eg, related information such as answers to other questions that the user should do afterwards) in addition to the answer For example, the user can efficiently acquire information by browsing the rationale document (for example, without asking other questions).
[0007]
However, the conventional question answering system cannot provide a presentation function related to information other than the answer existing in the ground document. For example, depending on the document evaluation method that is based on the degree of matching with a search keyword included in a question sentence, which is generally used as an evaluation method for a ground document, how information other than the answer to the question sentence is displayed. It is difficult to evaluate whether useful information is included. In addition, for example, in the method of displaying the summary of the ground document, there are only ones in which the same summary is generated regardless of the content of the question, or a summary that is biased toward the original question sentence is generated.
[0008]
The present invention has been made in consideration of the above circumstances, and in presenting a response to a question sentence, a question answering apparatus, a question answering method, and a program capable of presenting in consideration of information included in the basis document. The purpose is to provide.
[0009]
[Means for Solving the Problems]
A question answering apparatus according to the present invention includes a first database that stores a plurality of documents, a character string that can be an answer candidate included in the document for each document stored in the first database, Based on the second database for storing category information indicating categories as character string answer candidates in association with each other, input means for inputting a question sentence in natural language, and the first sentence based on the question sentence. A search means for searching a database; an extraction means for extracting an answer character string as an answer to the question sentence from the document searched by the search means; and the search means by referring to the second database. Acquisition means for acquiring category information corresponding to a character string other than the answer character string included in the basis document that is the document including the answer character string among the retrieved documents , Characterized in that an output means for outputting the response information including the category information corresponding to the character string other than the answer string and the reply string.
[0010]
The present invention relating to the apparatus is also established as an invention relating to a method, and the present invention relating to a method is also established as an invention relating to an apparatus.
Further, the present invention relating to an apparatus or a method has a function for causing a computer to execute a procedure corresponding to the invention (or for causing a computer to function as a means corresponding to the invention, or for a computer to have a function corresponding to the invention. It is also established as a program (for realizing) and also as a computer-readable recording medium on which the program is recorded.
[0011]
According to the present invention, when an answer to a question sentence is presented, it is possible to present the information in consideration of information included in the basis document. Therefore, according to the present invention, an answer to the question can be obtained, and for example, what information is included in the basis document in addition to the answer character string can be easily grasped.
[0012]
DETAILED DESCRIPTION OF THE INVENTION
Hereinafter, embodiments of the invention will be described with reference to the drawings.
[0013]
FIG. 1 shows a configuration example of a question answering system according to an embodiment of the present invention.
[0014]
As shown in FIG. 1, the question answering system includes an input unit 1, a document search unit 2, an answer type determination unit 3, an answer candidate extraction unit 4, an answer generation unit 5, an evidence document information addition unit 6, and an output unit 7. It has.
[0015]
The question answering system may include a document database 11, an expression category database 12, an answer candidate database 13, a question pattern database 14, and a rational document information table 15.
[0016]
A configuration without the document database 11 and / or the answer candidate database 13 is also possible. For example, these databases 11 and 13 may be connected to each other via a network such as a LAN or the Internet and be searchable.
[0017]
If the answer candidate database 13 is not provided or the answer candidate database 13 is provided, but the contents of the answer candidates can be acquired from the outside via a network such as the Internet, the answer candidate extraction unit 4 and the expression category database 12 are provided. You do n’t have to.
[0018]
The outline of each part in FIG. 1 is as follows.
[0019]
The input unit 1 inputs a question sentence.
[0020]
The document search unit 2 searches the document database 11 based on the inputted question sentence, and scores the obtained document.
[0021]
The answer type determination unit 3 uses the question pattern database 14 to determine the answer type of the input question sentence.
[0022]
The answer candidate extraction unit 4 extracts an answer candidate character string from a search target document stored in the document database 11 using, for example, the expression category database 12 and assigns an answer category to the answer candidate character string. , Answer candidate information is generated and stored in the answer candidate database 13.
[0023]
The answer generation unit 5 receives the input question text, the search result obtained by the document search unit 2, the answer type obtained by the answer type determination unit 3, and the answer candidate information stored in the answer candidate database 13. Based on the above, response information is generated.
[0024]
The basis document information adding unit 6 uses the basis document information table 15 to add the basis document information related to the basis document to the response information generated by the response generation unit 5.
[0025]
The output unit 7 outputs the answer information to which the ground document information is added.
[0026]
Here, when this question answering system is realized using a computer, the input unit 1 and the output unit 7 correspond to a user interface. For example, an input device such as a keyboard, a mouse, and a microphone, a computer display, a speaker, and the like. It can be realized using an output device. Further, the document search unit 2, the response type determination unit 3, the response candidate extraction unit 4, the response generation unit 5, and the rational document information addition unit 6 can be realized by a program.
[0027]
Further, when the question answering system is realized as a client / server system, for example, the input unit 1 and the output unit 7 among the components are mounted on the client computer side, and other parts are mounted on the server computer side. .
[0028]
Hereinafter, the flow of processing of the question answering system will be described using a specific example.
[0029]
FIG. 2 shows an example of the document database 11. In this example, each document includes a document ID, a title, and text. XX is the name of a movie director, and XXX, YYY, ZZZ, and WWW are all titles of a movie work directed by XXX.
[0030]
Here, the answer candidate extraction unit 4 and the answer candidate database 13 will be described.
[0031]
First, when the answer candidate extraction unit 4 is used, the answer candidate extraction unit 4 generates answer candidate information based on a search target document registered in the document database 11 in advance, and stores the answer candidate information in the answer candidate database 13. Keep it. About the process of the answer candidate extraction part 4, you may use well-known techniques, such as what is called a proper noun extraction and ontology tagger. In answer candidate extraction, processing may be performed on the surface representation of the search target document, processing may be performed on the result of morphological analysis, or processing may be performed on the result of syntax and dependency analysis. May be.
[0032]
A specific example of the answer candidate extraction process is shown below. First, a pair of (information for detecting an answer candidate character string, answer category) is registered in the expression category database 12. For example, when “XX Director” is detected from the target document including the phrase “XXX Director” and an answer category “producer” is to be assigned, the expression category database 12 stores the specific category. The answer candidate character string (director, producer, producer) is registered. Then, the answer candidate extraction unit 4 searches for the search target document registered in the document database 11 (for example, the document with document ID = 0000000 in FIG. 2) and the expression registered in the expression category database 12 (for example, the above-described document). (XX director, producer) compared to XX director), extract the answer candidate character string (for example, XXX director), and extract the answer candidate character string to the document ID ( In this example, document ID = 000050) and answer category information corresponding to the extracted answer candidate character string (in this example, XXX director) (in this case, (XX director, producer) above) Is created, and answer candidate information including a document ID, answer candidate character string, and answer category information is generated and stored in the answer candidate database 13. Even in the case of this example, the form of the expression category database 12 can be various. For example, in the expression category database 12, a character string that should be included in the answer candidate character string (director, producer) is registered, and a character string “XXX director” including “director” is extracted from the target document. An answer category “producer” may be assigned thereto. Also, for example, (“*”, title) is registered in the expression category database 12 (where * indicates an arbitrary character string), and “XXX” ”enclosed in parentheses from the target document. A character string may be detected, and a portion “XXX” obtained by omitting ““ ”from the character string may be used as a candidate character string, and an answer category“ title ”may be assigned thereto. Is possible.
[0033]
FIG. 3 shows an example of the answer candidate database 13 in which the answer candidate information generated based on the documents stored in the document database 11 of FIG. 2 is registered.
[0034]
On the other hand, if it is necessary not to use the answer candidate extraction unit 4 and to acquire answer candidate information from the outside and register it in the answer candidate database 13, this work is performed.
[0035]
Further, when it is not necessary to obtain the answer candidate information from the outside and register it in the answer candidate database 13 without using the answer candidate extraction unit 4, the answer candidate database 13 may be accessed when necessary.
[0036]
FIG. 4 shows an example of the flow of processing from question text input to result output of the question answering system.
[0037]
The user inputs a question sentence by text or voice via the input unit 1.
[0038]
In addition, since it is possible to convert a question sentence into text data by a well-known voice recognition technique even when voice input is performed (in this case, for example, the input unit 1 may have a function of the conversion). Hereinafter, the case where the question sentence is obtained as text data will be described as an example.
[0039]
When receiving an input of a question sentence from the user, the input unit 1 sends the input question sentence to the document search unit 2, the answer type determination unit 3, and the answer generation unit 5 (step S1).
[0040]
Here, who is the director of “YYY”? ”Is entered.
[0041]
The document search unit 2 searches the search target document in the document database 11 based on the question sentence received from the input unit 1, obtains the document score of each obtained document, and ranks higher in the document score order. Select a document from a specified number (for example, a fixed number or a user-specified number as appropriate), and search results including the document ID and document score of the selected document (foundation document), The data is sent to the answer generation unit 5 (step S2). A known technique may be used for the processing of the document search unit 2.
[0042]
In Figure 5, “Who is the director of“ YYY ”? 2 shows an example of an output result when a document stored in the document database 11 of FIG. 2 is searched based on the question sentence “(In this example, the larger the document score value, the more suitable the question sentence is. ).
[0043]
On the other hand, before or after or in parallel with this search processing, the answer type determination unit 3 compares the question text received from the input unit 1 with the expressions registered in the question pattern database 14 to determine the answer type. The response type information including the determination result is sent to the response generation unit 5 (step S3). Note that a known technique may be used for the processing of the document type determination unit 3 (for example, see JP-A-2002-132812).
[0044]
FIG. 6 shows an example of the question pattern database 14. In this example, when “when” is included in the question sentence, it is determined that the answer type is “date and time”, and when “who” is included in the question sentence, the answer type is “ It is determined that it is “person name”, and when “where” is included in the question sentence, it is determined that the answer type is “location”.
[0045]
In Figure 7, “Who is the director of“ YYY ”? 6 shows an example of the result of determination of the answer type by the question pattern database 14 of FIG. 6. In this case, the answer type is determined to be “person name” by “who” in the question text. The
[0046]
Now, the response generation unit 5 expands the response type information sent from the response type determination unit 3 into one or a plurality of response category information based on a predetermined expansion rule, and a plurality of response category information obtained by this expansion. Using the answer category information and the document ID included in the search result sent from the document search unit 2 as a key, the answer candidate database 13 is searched to obtain answer candidate information (step S4). Further, in step S4, the answer generating unit 5 further merges the obtained answer candidate information having the same surface character string, and includes answer information including the answer character string, the answer category information, and the document ID of the basis document ( A question sentence and a document score may be included), and the response information is sent to the ground document information adding unit 6.
[0047]
The expansion rule is defined, for example, by associating response type information with one or a plurality of response category information to be expanded. For example, the response type information “location” is expanded to the response category information “country name”, “place name”, “hometown”, and the response type information “date” is expanded to the response category information “year” “date” “time”. The type information “person name” is a rule that expands to answer category information “person name”.
[0048]
In the above specific example, the answer type is determined to be “person name”. Therefore, if the development rule as in the above example is applied to this, the answer category information “person name” is developed.
[0049]
In the above specific example, the answer generation unit 5 receives the answer category information “person name” obtained by the above expansion and the document ID = 0000000, 00201 included in the search result sent from the document search unit 2 or The answer candidate database 13 is searched using 00560 as a key. At this time, the answer generation unit 5 has, for example, knowledge information indicating that the answer category information “person name” matches the answer category information “producer”. As a result, the answer category information “person name” is obtained. In addition to the answer candidate information possessed, answer candidate information having the answer category information “producer” is extracted.
[0050]
FIG. 8 shows an example of answer candidate information obtained by the answer generation unit 5 in the case of the above specific example. FIG. 9 shows an example of answer information sent from the answer generating unit 5 to the rational document information adding unit 6 in this case (the question text is omitted in FIG. 9).
[0051]
In the above, the response type information “person name” is expanded to the response category information “person name”, and in the search stage, not only the search based on the expanded answer category information “person name” but also the response category information “ The knowledge that “person name” and answer category information “producer” match was used. Instead of this, the present invention expands the response type information “person name” into the response category information “producer”, etc., and the search does not use the above rules, but based on the expanded response category information “producer”, etc. You may make it search for.
[0052]
Next, the basis document information adding unit 6 uses the basis of the answer information received from the answer generation unit 5, the answer candidate information held in the answer candidate database 13, and the document information held in the document database 11. Data relating to related information held in the document is generated and recorded in the basis document information table 15. Based on the recorded data and the response information received from the response generation unit 5, the basis document information adding unit 6 generates presentation information including the response information and the basis document information, and outputs this to the output unit 7. Send (step S5).
[0053]
Here, the processing of the ground document information adding unit 6 will be described in more detail.
[0054]
Upon receiving the response information from the response generation unit 5, the rational document information adding unit 6 searches the document database 11 using the document ID of each rational document included in the response information as a key, and acquires the title information of each rational document, Record in the rationale document information table 15. In addition, the answer candidate database 13 is searched using the document ID of each evidence document as a key, and the expression extracted from the question sentence (“YYY” in this specific example) or the answer character string (this book) in each answer candidate information In the case of a specific example, the answer category information related to the answer candidate information excluding the answer candidate information including the answer candidate character string that matches any one of “xxx director” is extracted, and this information is counted for each category information. The number is recorded in the basis document information table 15 together with the number.
[0055]
FIG. 10 shows an example of the ground document information table 15 in the case of this specific example.
[0056]
Finally, the output unit 7 outputs a question sentence, an answer thereto, and the basis document information based on the information received from the basis document information adding unit 6 (step S6).
[0057]
FIG. 11 shows an example of a display screen displayed from the output unit 7 in the case of this specific example. In this example, who is the director of the question ““ YYY ”? "And the answer" XXX Director "in response to this, the evidence document information is displayed. In the evidence document information of Fig. 11, three evidence documents are arranged in the order of their document scores. In addition, “other information” in each ground document information indicates the response category information in FIG. 10 and the number of documents to which the category is attached. ) Indicates that there is no “other information”. In document (2), “other information” includes answer category information “title”, “place of origin”, and “year”, three, one, and one respectively. In the document (3), “other information” indicates that there is one having answer category information “title”.
[0058]
Although FIG. 11 shows an example in which three ground documents are arranged in the order of their document scores, they may be arranged in order according to other criteria. For example, it may be arranged and presented in the order of evaluation points described later (in this case, for example, the documents are arranged in the order of document IDs 00100, 050,000, and 00560 (see FIG. 13)).
[0059]
In FIG. 11, for example, the content of the selected rationale document may be displayed by selecting the title information of the rationale document with the mouse or selecting the rank number of the rationale document with the keyboard. .
[0060]
In FIG. 11, only the title information and other information are displayed for all the ground documents. For example, the ground documents in the first rank (from the document score or the evaluation score) (or from the first rank) For the basis documents up to a predetermined order), the content of the basis document may be displayed instead of the title information and other information or together with the title information and other information.
[0061]
As described above, according to the present embodiment, the answer to the question is presented, and what information is included in the document that is the basis for obtaining the answer to the question. Since the information is presented, the user can easily grasp whether or not there is information that the user wants to know other than the question in the rationale document.
[0062]
In the following, another example of the processing of the rational document information adding unit 6 and the output method by the output unit 7 will be described.
[0063]
(First modification)
First, a first modification will be described.
[0064]
Upon receiving the response information from the response generation unit 5, the rational document information adding unit 6 searches the document database 11 using the document ID of each rational document included in the response information as a key, and acquires the title information of each rational document, Record in the rationale document information table 15. In addition, the answer candidate database 13 is searched using the document ID of each evidence document as a key, and the expression extracted from the question sentence (“YYY” in this specific example) or the answer character string (this book) in each answer candidate information In the case of a specific example, the response category information related to the answer candidate information excluding the answer candidate information including the answer candidate character string that matches any of “XX director” is extracted and recorded in the basis document information table 15. To do. The above is similar to the example already described (see FIG. 10).
[0065]
Next, the evidence document information adding unit 6 obtains an evaluation score for each evidence document based on the rules registered in the evidence document evaluation pattern database (not shown), and records this in the evidence document information table 15. To do.
[0066]
FIG. 12 shows an example of information registered in the ground document evaluation pattern database. The answer category (1) is answer category information related to the answer information (see FIG. 9), the answer category (2) is each answer category information in the evidence document information table 15 (see FIG. 11), and the score is This is a score added to the evaluation score in the case of the combination of the answer category (1) and the answer category (2). In the evaluation point calculation process, for example, a value obtained by adding all the points related to the corresponding rule among the rules in FIG. For example, when the answer category information related to the answer information to the question sentence is “producer” and the contents of the evidence document information table 15 are as shown in FIG. The content of the basis document information table 15 is as shown in FIG. 13, for example.
[0067]
In this case, the output unit 7 presents the evidence document information based on the evaluation score given to the evidence document.
[0068]
FIG. 14 shows an example of this case. In this example, only the first-order basis document having the highest evaluation score (in this example, the document with document ID = 00210 in FIG. 2) is presented, and the other documents have two. It only presents that it exists. For other documents, for example, the number of the other document is selected with the mouse, or the number is selected with the keyboard, and the content of the selected ground document is displayed. For example, when number 1 is selected, the contents of the ground document having the second-ranked evaluation score (in this example, the document with document ID = 0000000 in FIG. 2) is displayed.
[0069]
In the above description, the content of only the first-ranked ground document having the highest evaluation score (in this example, the document with document ID = 00210 in FIG. 2) is presented. You may make it show the content about the ground document to rank.
[0070]
In FIG. 14, no specific information is presented for the ground document other than the ground document for presenting the contents. For example, the title information and other information are displayed as shown in FIG. It may be. Of course, in this case as well, for example, the content of the selected evidence document may be displayed by selecting the title information of the evidence document with the mouse or selecting the rank number of the evidence document with the keyboard. Good.
[0071]
As described above, according to the present embodiment, it is determined that a lot of related information is included in the document that provides the answer to the question and is the basis for obtaining the answer to the question. Documents can be displayed with priority, so that the user can easily grasp whether or not there is information that he / she wants to know other than the question in the rationale document. Become. And, if the information you want to know is included in the evidence document, you can get the information you want to know just by browsing the evidence document without entering a new question and searching. .
[0072]
(Second modification)
Next, a second modification will be described.
[0073]
Upon receiving the response information from the response generation unit 5, the rational document information adding unit 6 searches the document database 11 using the document ID of each rational document included in the response information as a key, and acquires the title information of each rational document, Record in the rationale document information table 15. In addition, the answer candidate database 13 is searched using the document ID of each evidence document as a key, and the expression extracted from the question sentence (“YYY” in this specific example) or the answer character string (this book) in each answer candidate information In the case of a specific example, the response category information related to the answer candidate information excluding the answer candidate information including the answer candidate character string that matches any of “XX director” is extracted and recorded in the basis document information table 15. To do. The above is similar to the example already described (see FIG. 10).
[0074]
Next, the basis document information adding unit 6 classifies the basis documents based on the similarity of the response categories registered in the basis document information table 15 and reflects the classification result in the basis document information table 15 (for example, Information regarding the classification to which the document belongs is added to each document in the document document information table 15). As a classification method, a known technique may be used.
[0075]
For example, in the case document information table 15 shown in FIG. 10, document category 1 is a category to which a basis document (document ID = 00210, 0950) including an answer candidate related to the response category “title” belongs, and document category 2 is an answer. The classification to which the rational document (document ID = 00210) including the answer candidate related to the category “Birthplace” belongs, and the document classification 3 is the classification to which the rational document (document ID = 00210) including the response candidate related to the response category “year” belongs. In the rational document information table 15, information indicating document classification 1, document classification 2, and document classification 3 is added to the rational document with the document ID = 00210, and the document classification is included in the rational document with the document ID = 000050. 1 is added (the basis document with document ID = 00560 does not have information related to the classification or the classification to which it belongs. Information indicating it is added).
[0076]
In this case, the output unit 7 presents the evidence document information based on the information related to the classification given to the evidence document.
[0077]
FIG. 15 shows an example of this case. In this example, for each classification, the answer category related to the classification and the title of the rational document belonging to the classification are presented in descending order of the number of evidence documents belonging to the classification.
[0078]
In FIG. 15, in each document classification, the number of answer candidate information related to the answer category included in each evidence document may be presented. For example, in the ground document (document ID = 00210) for the title “YYY” of document classification 1, there are three pieces of answer candidate information having the answer category “title” (see FIG. 10), so “YYY (3)”. As described above, the number may be displayed beside the display of the title of the basis document of the document classification.
[0079]
Further, for example, in FIG. 15, for the first-ranking ground document (or the ground document from the first rank to a predetermined rank) (for example, document score or evaluation score), the contents of the ground document May be displayed.
[0080]
As described above, according to the present embodiment, an answer to a question is presented, and the basis document is classified based on related information in addition to the answer existing in the document that is the basis for obtaining the answer to the question. Therefore, the user can easily grasp whether or not there is information that he / she wants to know in addition to the question in the basis document. And, if the information you want to know is included in the evidence document, you can get the information you want to know just by browsing the evidence document without entering a new question and searching. .
[0081]
Note that the variations of the processing of the ground document information adding unit 6 and the output method of the output unit 7 described above can be combined as appropriate. Also, a plurality of methods may be prepared as the processing of the rational document information adding unit 6 and the output method by the output unit 7 so that the user can set which one to use.
[0082]
In the above description, the search target document is shown as plain text. However, a document tagged in advance such as an XML document can be similarly implemented. In this case, pre-tagged information can be used as the answer category information.
[0083]
In the above explanation, the response type and the response category are defined as one-to-many correspondence. However, when the response type and the response category have the same name, the response type and the response category are many-to-many correspondence. Or, even in the case of many-to-one correspondence, it can be similarly implemented.
[0084]
Also, in the above explanation, higher-level concept tags such as “person name” and “place name” are used as the answer category, but also when meta concepts such as “definition expression” and “means expression” are used as tags. Can be implemented as well.
[0085]
Further, in the above description, morphological analysis and syntax analysis means are not particularly explicitly used, but the same can be implemented when these means are used in each processing. In this case, it is also possible to assign a category attribute to the morphological analysis dictionary and perform category identification by syntax pattern matching.
[0086]
Each of the above functions can be realized even if it is described as software and processed by a computer having an appropriate mechanism.
The present embodiment can also be implemented as a program for causing a computer to execute predetermined means, causing a computer to function as predetermined means, or causing a computer to realize predetermined functions. In addition, the present invention can be implemented as a computer-readable recording medium on which the program is recorded.
[0087]
Note that the present invention is not limited to the above-described embodiment as it is, and can be embodied by modifying the constituent elements without departing from the scope of the invention in the implementation stage. In addition, various inventions can be formed by appropriately combining a plurality of components disclosed in the embodiment. For example, some components may be deleted from all the components shown in the embodiment. Furthermore, constituent elements over different embodiments may be appropriately combined.
[0088]
【The invention's effect】
According to the present invention, when an answer to a question sentence is presented, it is possible to present the information in consideration of information included in the basis document.
[Brief description of the drawings]
FIG. 1 is a diagram showing a configuration example of a question answering system according to an embodiment of the present invention.
FIG. 2 is a diagram showing an example of information in a document database
FIG. 3 is a diagram showing an example of information in an answer candidate database
FIG. 4 is a flowchart showing an example of a processing procedure of the question answering system according to the embodiment.
FIG. 5 is a diagram showing an example of search results
FIG. 6 is a diagram showing an example of information in a question pattern database
FIG. 7 is a diagram showing an example of an answer type determination result
FIG. 8 is a diagram showing an example of extracted answer candidate information
FIG. 9 is a diagram illustrating an example of response information including a response character string, a response category, and a document ID of a rational document.
FIG. 10 is a diagram showing an example of a rationale document information table
FIG. 11 is a diagram showing a display example of presentation information
FIG. 12 is a diagram showing an example of information in the basis document evaluation pattern database
FIG. 13 is a diagram showing another example of the basis document information table
FIG. 14 is a diagram showing a display example of presentation information
FIG. 15 is a diagram showing a display example of presentation information
[Explanation of symbols]
DESCRIPTION OF SYMBOLS 1 ... Input part, 2 ... Document search part, 3 ... Answer classification determination part, 4 ... Answer candidate extraction part, 5 ... Answer generation part, 6 ... Ground document information addition part, 7 ... Output part, 11 ... Document database, 12 ... expression category database, 13 ... answer candidate database, 14 ... question pattern database, 15 ... basis document information table

Claims (18)

A question answering apparatus that inputs a question sentence in a natural language and outputs an answer character string that is a character string serving as an answer to the question sentence,
A first database storing a plurality of documents;
For each document stored in the first database, and candidate character strings taken as to obtain candidates for the answer string from among the character strings included in the document, semantic content of the candidate character strings the answer candidate information including the category information indicating a category relating to a second database for each arbitrary number memorize,
And input means for inputting the question,
Search means for searching the first database based on the question sentence;
From the document retrieved by said retrieval means, extracting means for extracting the answers string,
Among previous SL retrieval means documents retrieved by, the answer string as target basis document respectively a free plain coarse document from all the answer candidate information about the evidence documents stored in the second database The acquisition means for excluding the answer candidate information including the candidate character string matching the answer character string, and for obtaining the category information included in the answer candidate information for the answer candidate information obtained as a result ;
The extraction means extracting said answer character string, and question-answering apparatus being characterized in that an output means for outputting the category information obtained by the obtaining unit. The acquisition unit is configured to select all of the answer candidates stored in the second database for the basis document for each of the basis documents that are documents including the answer character string from among the documents searched by the search unit. From the information, answer candidate information including a candidate character string that matches either the character string included in the question sentence or the answer character string is excluded, and the answer candidate information obtained as a result is included in the answer candidate information. 2. The question answering apparatus according to claim 1, wherein the category information included is acquired. And the output means, when performing the output, for each of the grounds document, and information indicating the rationale document, claim 1 in association with the category information obtained for the rationale document and outputs it or 2. The question answering apparatus according to 2. The output means further outputs, for each category information acquired for the basis document, information indicating the number of candidate character strings corresponding to the category information included in the basis document for performing the output. The question answering apparatus according to claim 3 , wherein The output means, when performing the output, when there are a plurality of the evidence documents, outputs the category information related to the evidence documents in order from the evidence documents highly evaluated by a predetermined evaluation method. The question answering device according to claim 3 or 4 . The output means classifies the basis document based on the category information acquired for the basis document when performing the output, information indicating the basis document belonging to the category for each category, and a category related to the category 4. The question answering apparatus according to claim 3 , wherein information is output. 3. The question according to claim 1, wherein the output unit outputs the contents of the basis document for the basis document that is most highly evaluated by a predetermined evaluation method when performing the output. 4. Response device. A first database storing a plurality of documents;
For each document stored in the first database, a character string that can be an answer candidate included in the document and category information indicating a category as an answer candidate of the character string are stored in association with each other. Database of
An input means for inputting a question sentence in natural language;
Search means for searching the first database based on the question sentence;
Extraction means for extracting an answer character string that is an answer to the question sentence from the document searched by the search means;
Referring to the second database, the first character string corresponding to each of the character strings other than the answer character string included in the ground document that is the document including the answer character string among the documents searched by the search means. Acquisition means for acquiring first category information and second category information corresponding to the answer character string;
About the basis document, an evaluation score is obtained based on a rule registered in a pattern database of a combination of the acquired first category information and the acquired second category information. An evaluation means for evaluating the evidence document;
A question answering apparatus comprising: output means for outputting the answer character string and the contents of a basis document having a predetermined number of documents selected based on an evaluation result by the evaluation means. The evaluation means refers to a third database that stores information indicating evaluation points determined for each combination of the first category information and the second category information when evaluating the basis document. A value obtained by summing up the evaluation points determined from each of the first category information acquired from the document and the evaluation points determined from the second category information is used as the evaluation value of the basis document. The question answering apparatus according to claim 8 . 9. The question answering apparatus according to claim 8 , wherein the output means outputs the content of the basis document that is most highly evaluated by the evaluation means. The extraction means obtains a condition to be satisfied by category information corresponding to an answer character string for the question sentence based on the question sentence,
The acquisition unit obtains category information satisfying the condition among the category information stored for the document searched by the search unit in the second database,
The extraction means uses the candidate character string stored in the answer candidate information in association with category information satisfying the condition in the second database as the answer character string. The question answering apparatus according to 2 or 8 . Extracting a character string to be answered candidate from the first database stored document, which was said candidate character string, in the candidate character string, and applying the category information, per the document, the candidate the question answering system according to the answer candidate information including the character string and the category information to claim 1, 2 or 8, further comprising a means for storing in said second database. In order to input a question sentence in a natural language and output an answer character string that is a character string that is an answer to the question sentence, a first database that stores a plurality of documents and a first database that is stored in the first database for each document, including a candidate character string retrieved as can be a candidate for the answer string from among the character strings included in the document, and a category information indicating a category related semantic content of the candidate character strings the answer candidate information, a second and a database, an input means, search means and the extraction means and the acquisition means and, question answering method of question and answer device and output means for memorize by any number,
It said input means, an input step for inputting said question,
A search step in which the search means searches a first database storing a plurality of documents based on the question sentence;
The extraction means, from the document retrieved by the retrieving step, an extraction step of extracting the answers string,
All the acquisition means, among pre-Symbol search documents retrieved by step, the answer string as target people grounds document each is free plain coarse document, stored for the grounds document to the second database The step of obtaining the category information included in the answer candidate information for the candidate answer information obtained as a result of excluding the answer candidate information including the candidate character string matching the answer character string from the answer candidate information When,
Question answering wherein said output means, and having an output step of outputting the reply character string extracted by the extraction step, and the category information obtained by the obtaining step. The acquisition unit is configured to select all of the answer candidates stored in the second database for the basis document for each of the basis documents that are documents including the answer character string from among the documents searched by the search unit. From the information, answer candidate information including a candidate character string that matches either the character string included in the question sentence or the answer character string is excluded, and the answer candidate information obtained as a result is included in the answer candidate information. 14. The question answering method according to claim 13, wherein category information included is acquired. A first database that stores a plurality of documents, a character string that can be an answer candidate included in the document for each document stored in the first database, and a category as an answer candidate for the character string A question answering method for a question answering apparatus comprising: a second database for storing category information to be associated with each other, an input means, a search means, an extraction means, an acquisition means, an evaluation means, and an output means. There,
The input means for inputting a question sentence in a natural language;
A search step in which the search means searches a first database storing a plurality of documents based on the question sentence;
An extracting step in which the extracting means extracts an answer character string that becomes an answer to the question sentence from the document searched in the searching step;
The acquisition means associates, for each document stored in the first database, a character string that can be an answer candidate included in the document and category information indicating a category as an answer candidate of the character string. Corresponding to each of the character strings other than the answer character string included in the ground document that is the document including the answer character string among the documents retrieved by the retrieval step. Obtaining first category information and second category information corresponding to the answer character string;
The evaluation means obtains an evaluation score for the basis document based on a rule registered in a pattern database of a combination of the acquired first category information and the acquired second category information, An evaluation step for evaluating the rationale document according to an evaluation point; and
The question answering method, characterized in that the output means includes an output step of outputting the answer character string and the contents of a basis document of a predetermined number of documents selected based on the evaluation result of the evaluation step. . In order to input a question sentence in a natural language and output an answer character string that is a character string that is an answer to the question sentence, a first database that stores a plurality of documents and a first database that is stored in the first database For each document, a candidate character string extracted as a candidate for the answer character string from among character strings included in the document, and category information indicating a category related to the semantic content of the candidate character string is included. A program for causing a computer to function as a question answering apparatus including a second database that stores answer candidate information in an arbitrary number ,
And the input function for entering the question,
A search function for searching a first database storing a plurality of documents based on the question sentence;
From the document retrieved by the search function, the extracting function of extracting the answers string,
Among pre Symbol search function documents retrieved by, the answer string as target basis document respectively a free plain coarse document from all the answer candidate information about the evidence documents stored in the second database , excluding the answer candidate information including a candidate character string that matches the answer string as an object the answer candidate information obtained as a result, an acquisition function of acquiring the category information included in the answer candidate information,
The extraction feature extracted the answer string, and a program for realizing an output function on a computer that outputs the category information that the acquirer has acquired. The acquisition function is for all the answer candidates stored in the second database for the basis document for each of the basis documents that are documents including the answer character string among the documents searched by the search function. From the information, answer candidate information including a candidate character string that matches either the character string included in the question sentence or the answer character string is excluded, and the answer candidate information obtained as a result is included in the answer candidate information. The program according to claim 16, wherein category information included is acquired. A program for causing a computer to function as a question answering device,
An input function for inputting questions in natural language;
A search function for searching a first database storing a plurality of documents based on the question sentence;
An extraction function for extracting an answer character string as an answer to the question sentence from the document searched by the search function;
For each document stored in the first database, a character string that can be an answer candidate included in the document and category information indicating a category as an answer candidate of the character string are stored in association with each other. The first category corresponding to each of the character strings other than the answer character string included in the ground document that is the document containing the answer character string among the documents searched by the search function An acquisition function for acquiring information and second category information corresponding to the answer character string;
About the basis document, an evaluation score is obtained based on a rule registered in a pattern database of a combination of the acquired first category information and the acquired second category information. An evaluation function to evaluate the rationale document;
A program for causing a computer to realize the output character string and an output function for outputting the contents of a document with a predetermined number of documents selected based on the evaluation result by the evaluation function.

Images