JPH113335A - Question and answer managing device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a question and answer managing device for easily and effectively utilizing a question content included in an existent document and an answering content to the question content. SOLUTION: Based on knowledge concerning an element constituting a question content part and knowledge concerning an element constituting an answer content part, a question and answer extracting means 2 specifies and extracts the question content part and the answer content part from the document and a question and answer list managing means 4 corresponds to the extracted question content part and answer content part to be stored to a data base 3. Then the means 4 receives a question from a user and retrieves an answer corresponding to the received question from the data base 3 to output. Consequently, the question content part and the answer content part to the question content part can be extracted from an existent document to allow the user to easily obtain an answer proper for the question by these extracted question content part and answer content part.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a question and answer management device for extracting and managing a question content portion and an answer content portion to the question content portion from a document.

[0002]

2. Description of the Related Art Conventionally, the contents of a question about a certain thing or matter and the answer to the question have been described in a document. 2. Description of the Related Art Disclosure is performed in a network environment such as the Internet and an intranet. In a document published in such a network environment, a pair of a frequently asked question and an answer to the question is often described as a question answer list called FAQ (Frequently Asked Questions).

An example of this FAQ is Sun Microsystems
http://www.sun.com/sunsoft/P in ms company homepage
roducts / Developer-products / WSReady / FAQ.html (hereinafter referred to as Reference 1), http://www.sun.com/smcc/sola
ris-migration / docs / faq / faq.html (hereinafter referred to as Reference 2), http://www.sun.com/tech/projects/spring/FA
Q.html and http://www.sun.com/tech/projects/spring/
faq / colormaphtml (hereinafter referred to as reference 3).

[0004] As a means for efficiently managing FAQs, http: //www.ncsa/uiuc.edu/SDG/IT94/Proceedings/
Auto-FAQ: an experiment in cybersp in Agnets / whitehead / whitehead.html (hereinafter referred to as Reference 4).
Ace leveraging is disclosed. This Auto-FA
Q manages the FAQ as a set of a description of the background of the question, a question, an answer, and an evaluation score calculated from the evaluation of the answer by the questioner who has obtained the answer. Then, when the user sends a question, the managing FA
An appropriate answer candidate is obtained from Q and returned to the user. The search mechanism in the Auto-FAQ obtains suitable answer text candidates from the degree of similarity between the user's question, the managed background description, and the question text. Also, Out
When there is no answer to the question sent from the user, the o-FAQ registers the question as a no-answer question, discloses it to the network environment, and accepts answers from experts widely existing in the Internet environment. I have.

Japanese Patent Application Laid-Open No. 1-112358 (hereinafter referred to as Document 5) discloses a document, a production rule of the form "if,...," And a knowledge grouping these. A method is disclosed that enables questions and answers to a document by managing the documents in association with each other. In this method, a question about the contents of a document is presented to an operator by instructing an inference operation on knowledge, and a question and answer is performed by the operator returning a response.

[0006]

Obtaining an answer to a certain question from a document requires reading and searching for the document, which is troublesome. In particular, when an answer to a question is obtained from documents which are scattered and published in a network environment, a considerable amount of time is required.

[0007] For example, a document published in the Internet or an intranet environment is an HTML (hyper text mark-
In this HTML, the description format of the FAQ is not formally defined (formally). For this reason, when searching for an FAQ that includes an answer to a question, a candidate document must be detected by using an information search technique such as a keyword search, and this information search technique often does not include an answer to a question. Only the candidate of the document including the noise document is obtained, and the requester himself has to individually access the candidate document and search for an answer to the question.

On the other hand, since the Auto-FAQ has both a semantic structure and a grammatical structure for a pair of a question and an answer, the answer to the question can be searched efficiently. However, the answers to be searched are limited to a small number of answers such as those managed by Auto-FAQ, and the semantic structure and the grammatical structure of the pair of the question and the answer are previously matched. There is a problem that it is necessary to keep it.

Further, according to the method described in Document 5, if the knowledge relating to the contents of a certain document is sufficiently described, the contents of the document can be grasped effectively and easily. However, it is difficult to accurately relate knowledge to many documents. In addition, it is difficult to accurately describe a large amount of knowledge without inconsistency by using production rules.

[0010] The present invention has been made in view of the above-described conventional circumstances, and provides a question-and-answer system that can easily and effectively use the contents of a question contained in an existing document and the contents of an answer to the question. It is intended to provide a management device.

[0011]

In order to achieve the above object, a question and answer management device according to the present invention extracts and manages a question content portion and an answer content portion corresponding to the question content portion from a document. A question and answer extracting means for identifying and extracting a question content part and an answer content part from a document, based on the knowledge about the elements making up the question content part and the knowledge about the elements making up the answer content part; Question and answer management means for storing the extracted question contents and the answer contents in a storage device in association with each other.

Here, the question content portion may be only a portion directly representing the question (hereinafter referred to as a question main portion), but further a portion representing the background of the question (hereinafter referred to as background information). It is preferable to combine them. In addition, the knowledge related to the elements constituting the question content portion includes knowledge that the question content portion includes elements such as symbols, phrases, and sentences in a specific format, and knowledge that the question content portion includes the answer content portion. And that it contains elements that have a specific document structure relationship. In addition, the knowledge related to the elements constituting the answer content part means that the answer content part includes elements such as symbols, phrases, sentences, etc. in a specific format, and the answer content part includes the question content part. And that it contains elements that have a specific document structure relationship.

In the above-mentioned question and answer management means, the question and answer extraction means is configured to perform the following based on the knowledge of the elements constituting the question content part and the knowledge of the elements constituting the answer content part.
The question content part and the answer content part are specified and extracted from the document, and the question and answer management means stores the extracted question content part and the answer content part in the storage device in association with each other. Therefore, a question content part and an answer content part corresponding to the question content part can be extracted from the existing document, and the user can easily obtain an appropriate answer to the question by the extracted question content part and answer content part. Can be.
Further, by extracting a question content portion and a question content portion corresponding to the question content portion from a large number of documents scattered and published in the network environment, a large number of question content portions and answer content portions can be obtained. Can easily obtain a more appropriate answer to the question.

[0014]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A question and answer management device according to one embodiment of the present invention will be described with reference to FIG. The question and answer management means includes a document search means 1 for searching for a document including an FAQ from a document existing in the Internet or intranet environment.
And a question / answer extracting means 2 for extracting a question content portion and an answer content portion from a document, a database 3 for holding the question content portion and the answer content portion, and associating the extracted question content portion with the answer content portion. And a question and answer list management means 4 for searching the database for answers to the questions and outputting the answers.

The document retrieving means 1 retrieves a document identifier (URL: uniform resource locato) from a document existing in an Internet or intranet environment described in HTML.
r) Or, search for a document that includes an FAQ equivalent in the title (contents enclosed by “/ title” in “Title”) or the body (contents enclosed by “body” and “/ body”). In addition, the phrase equivalent to FAQ includes FAQ, question and answer, question and answer, Frequently Asked Questio.
FA such as n, Answers, FAQs, Q & A, etc.
Q and its synonyms.

The question answer extracting means 2 uses the heuristic (empirical) knowledge about the FAQ for the document obtained by the document searching means 1 to generate "background information" representing the background of the question from the document. Then, a "question main part" directly representing a question and "answer contents" to the question are specified and extracted.

Here, for example, heuristic knowledge regarding FAQs includes the following. In this specification, Q, question, question, etc.
And its synonyms are called Q equivalent phrases, A, Answer, Answ
A such as er and its synonyms are called A equivalent phrases. (1) Document headings at the same level ("H1" and "/ H1"
In a text with "H2" and "/ H2" etc.), if the beginning of a paragraph starts with a Q-equivalent phrase and the beginning of the following paragraph starts with an A-equivalent phrase, a paragraph that starts with a Q-equivalent phrase Is the “question main part”, and the following paragraph is the “answer contents”. (2) In the body of a document having the same level heading,
If a question sentence (a sentence with the meaning of a question, for example, a sentence ending with “?” Or a sentence containing a phrase such as “why”), a line feed, multiple normal sentences, and a line feed, The question sentence is “question main part”, and a plurality of normal sentences are “answer contents”.

(3) In a body of a document having the same level heading, structures such as a reference to another document (a reference indicated by “a href = identifier”), a question sentence, and a line feed are repeatedly present, and If the body of the referenced document has the same level of heading as the question sentence, line feed, multiple normal sentences, and line feed, the question sentence in the referenced document is the "question main part". A plurality of ordinary sentences in the referenced document are “answer contents”. (4) Identifier (URL) of the document containing the sentence recognized as "question main part" and "answer content", and from the beginning of the document to the sentence recognized as "question main part" and "answer content" The content of the title, the content of the headline, etc. are "background information"
It is.

According to the knowledge of the above (1), the "question essential part" and the "answer contents" can be specified from a document such as the document 1, and according to the knowledge of the (2), the "question part" and the "answer content" can be specified from the document such as the document 2. According to the knowledge of (3), the "question main part" and "answer content" can be specified from a document such as Reference 3, and according to the knowledge of (4), reference 1 can be specified.
"Background information" can be specified from documents such as.

The above-mentioned question and answer extracting means 2 will be described in detail with reference to FIG. The question answer extraction means 2 includes a sentence segmentation unit 2a, a sentence type identification unit 2b, a sentence determination function management unit 2c, a sentence determination unit 2d, a candidate storage unit 2e, a candidate storage unit 2f, and a candidate clear unit. 2g and a registration unit 2h. The sentence extracting unit 2a sequentially extracts one sentence from the document retrieved by the document retrieval unit 1. In this embodiment, first, the identification name of the searched document is cut out. The sentence type identification unit 2b identifies whether the sentence extracted by the sentence extraction unit 2a is a title, a headline, or a body using an HTML tag.

The sentence judgment function management unit 2c manages a judgment function f, and the judgment function f receives, as inputs, a sentence to be judged, a type of the sentence to be judged, and a judgment result of the immediately preceding sentence. By comparing the received information with the heuristic knowledge, at the time when the judgment target sentence is read, the target sentence is a candidate for background information, a candidate for a question essential part, a candidate for an answer content, or any of the candidates other than the candidates. If the sentence to be determined is a candidate for background information, “background candidate” is returned. If the sentence is a candidate for a question main part, “question candidate” is returned. Returns "candidate", and if it is not any candidate, returns "non-candidate".

The sentence judging unit 2d calculates the sentence extracted by the sentence extracting unit 2a, the type of the sentence identified by the sentence type identifying unit 2b, and the judgment result of the judgment function f for the sentence immediately before the sentence. Input to the judgment function f and receive the judgment result. Then, according to the self state and the judgment result returned from the judgment function f, the candidate storage unit 2e, the candidate clear unit 2g,
An instruction is sent to any of the registration units 2h, and if necessary, the own state is changed. A detailed description of the function of the sentence determination unit 2d will be included in the operation description described later.

The candidate storage unit 2e stores the sentence in the candidate holding unit 2f as one of the background information candidate, the candidate for the question main part, and the answer content candidate in accordance with the instruction of the sentence determination unit 2d.
The candidate clearing unit 2g clears a sentence stored in the candidate holding unit 2f according to an instruction of the sentence determining unit 2d. As a result, a sentence that is determined not to satisfy heuristic knowledge by cutting out the following sentence from among the sentences stored as a candidate for background information, a candidate for a question main part, and a candidate for an answer content. Is deleted.

The registration unit 2h notifies the question answer list management unit 4 of the sentence stored in the candidate holding unit 2f in accordance with an instruction from the sentence determination unit 2d. As a result, a sentence that has been found to satisfy heuristic knowledge by cutting out the following sentence from among the sentences stored as background information candidates, question main part candidates, and answer content candidates. Is notified to the question and answer list management means 4 as "background information", "question main part", and "answer content".

The question / answer list management means 4 includes the “background information”, the “question main part”,
The "answer contents" are stored in the database 3 as question answer records as shown in FIG. 3, for example, and a group of a plurality of question answer records (question answer list) is managed. This question answer record includes information such as “background information”, “question main part”, and “answer contents”, as well as “evaluation score”, “background information”, and Information such as "copyright holder" indicating the owner of the "main part" and "answer contents" is also included. The question and answer list management unit 4 receives a question from the user, searches for a question and answer record including an answer corresponding to the received question, and returns the question and answer record to the user. A detailed description of the function of the question answer list management means 4 will be included in the operation description described later.

Next, the operation of the above-described question and answer management device will be described. First, the document retrieving means 1 retrieves a document including an FAQ from documents existing in the Internet or intranet environment, and the question answer extracting means 2 retrieves "background information", "questionnaire" for each document retrieved by the document retrieving means 1. An extraction process is performed to extract the “main part” and “answer contents”. In this extraction processing, the sentence segmentation unit 2a extracts 1 from the searched document.
The sentence collection identifying unit 2b identifies the type of the extracted sentence, and the sentence determination unit 2d determines the type of the extracted sentence, the type of the sentence, and the determination result by the determination function f for the immediately preceding sentence. Input to the function f and receive the judgment result of the sentence.

Next, the sentence determination unit 2d determines its own state,
In accordance with the received determination result, any one of the candidate storage unit 2e, the candidate clearing unit 2g, and the registration unit 2h is processed, and if necessary, a process of changing its own state is performed. This processing will be described in detail with reference to FIG. In the figure, the condition for transition from one state to another state is also indicated by an arrow, and the processing performed under the condition is indicated by () after the condition.
It is written in.

Here, the sentence judging section 2d is in an initial state before receiving the judgment result from the judgment function f for the sentence first cut out from the document. When the sentence determination unit 2d in the initial state receives the determination results “question candidate”, “answer candidate”, and “non-candidate”, it does nothing (step S1), and determines the determination result “background candidate”. Is received in the candidate storage unit 2e, the determined sentence is stored in the candidate storage unit 2f as a candidate for background information, and the own state is transited to the background stored state (step S2).

When the sentence judgment unit 2d in the background stored state receives the judgment result "background candidate", the sentence judged in the candidate storage unit 2e is sent to the candidate holding unit 2f. Stored as background information candidates (step S
3) If the judgment result “answer candidate” or “non-candidate” is received, nothing is performed (step S4), and if the judgment result “question candidate” is received, the judgment is stored in the candidate storage unit 2e. The prepared sentence is stored in the candidate holding unit 2f as a candidate for a question main part, and the own state is transited to a state where a question main part is stored (step S5).

When the sentence judging unit 2d in the stored state of the question main part receives the judgment result “background candidate” and “non-candidate”, the candidate clearing unit 2g sets the candidate clear part 2g as a candidate of the question main part. The stored sentence is deleted from the candidate holding unit 2f, and the own state is changed to the background stored state (step S6). When the determination result “question candidate” is received, the candidate storage unit 2e stores the relevant sentence in the candidate storage unit 2e. The determined sentence is stored in the candidate holding unit 2f as a candidate for a question main part (step S7). When the determination result “answer candidate” is received, the determined sentence is stored in the candidate storage unit 2e. The storage unit 2f causes the storage unit 2f to store the response content as a candidate, and shifts its own state to the response content stored state (step S
8).

When the sentence determination unit 2d in the answer content stored state receives the determination result "question candidate", the registration unit 2h sends the background information candidate and the question main part from the candidate holding unit 2f. And the candidate of the answer content are taken out and notified to the question / answer list management means 4, and the candidate storage unit 2e
Is stored in the candidate holding unit 2f as a candidate for the main part of the question, the state of the self is changed to the stored state of the main part of the question (step S9), and the judgment result “answer candidate” is received. In step S10, the sentence for which the determination is made is stored in the candidate storage unit 2e as a candidate for the answer content in the candidate storage unit 2f (step S10).

When the judgment result "background candidate" is received, the registration unit 2h retrieves a candidate for background information, a candidate for a main part of a question, and a candidate for an answer content from the candidate holding unit 2f to obtain a question answer list. The management unit 4 is notified and the candidate storage unit 2
e, the sentence that has been determined is stored in the candidate holding unit 2f as a candidate for background information, and its own state is transited to the background stored state (step S11). Then, the registration unit 2h retrieves the background information candidate, the candidate for the main part of the question, and the candidate for the answer content from the candidate holding unit 2f and notifies the question / answer list management unit 4 of the status. When the document is completed (step S12), the registration unit 2h causes the registration unit 2h to retrieve the background information candidate, the candidate for the main part of the question, and the candidate for the answer content from the candidate holding unit 2f, and to the question / answer list management unit 4. Let me know
The process ends (step S13). The initial state,
When the document is completed in the background stored state and the question main part stored state, the process ends without doing anything (steps S14 to S16).

By the above processing, the question answer extracting means 2
Is stored in the database 3 as a question and answer record in which the question and answer list management unit 4 associates the background information, the main part of the question, and the contents of the answer notified. In this manner, information such as background information, a main part of a question, and contents of an answer can be extracted from a plurality of documents existing in the Internet or an intranet environment and can be associated with each other.

Next, an example of a process of outputting an answer to a user's question using a question and answer list having a plurality of question and answer records created by the above process is shown in FIG.
This will be described with reference to FIG. Upon receiving the question from the user, the question and answer list management means 4 acquires one question and answer record from the question and answer list of the database 3 (step S).
21) The degree of coincidence B between the phrase included in the question received from the user and the phrase included in the background information is calculated (step S22), and the phrase included in the question received from the user and the main part of the question are calculated. The degree of coincidence Q with the included phrase is calculated (step S23), and the degree of coincidence A between the phrase included in the question received from the user and the phrase included in the answer content is calculated (step S24). The final matching score is calculated based on the calculated matching score B, matching score Q, and matching score A (step S25). In this embodiment, the matching degree Q
Are weighted to increase the relevance to the user's question.

Next, the question and answer list management means 4 compares the calculated final match degree with a preset threshold value (step S26), and sets a question answer record whose final match degree exceeds the threshold value as a search answer. For example, display and output are performed on a display device (not shown) such as a display (step S27). Then, the above-described processing (steps S21 to S27) is repeatedly performed on all other question and answer records in the question and answer list (step S28).
Therefore, the user can obtain an appropriate answer by an easy operation such as inputting a question.

The process of searching for an answer to a question using the question answer list is not limited to the above-described embodiment, and various generally known search processes can be used.

In the above-described embodiment, knowledge of the document structure relationship between the question content portion and the answer content portion is held.
By using the knowledge, the question content part and the answer content part corresponding to the question content part were appropriately specified and extracted from the document, but the knowledge of the relationship in the document structure was not retained In both cases, if the knowledge about the elements of the specific format included in each of the question content portion and the answer content portion is held, the question content portion and the answer content portion can be specified and extracted.

In the above-described embodiment, the HTML is used to extract the question contents and the answer contents. However, the present invention is not limited to this, and other structured documents may be used. Documents may be targeted. In addition,
If the knowledge of the elements that make up the question content part and the knowledge of the elements that make up the answer content part include knowledge of the relationship in the document structure, complex structures can be created by targeting structured documents. Since the document structure can be easily grasped without performing the analysis processing, appropriate question contents and answer contents can be efficiently extracted.

Each of the above-mentioned functional units is constituted by a processor executing a control program (prestored in a ROM, a RAM or the like). In the present invention, these functional units are constituted as independent hardware circuits. May be. In addition, the present invention allows the control program to be grasped as a computer-readable storage medium such as a floppy disk or a CD-ROM, and reads the control program from the storage medium and causes a processor to execute the control program. Such processing can also be executed.

[0040]

As described above, in the question and answer management apparatus according to the present invention, the contents of the question and the answer are determined based on the knowledge on the elements constituting the question and the knowledge on the elements constituting the answer. By extracting and associating the part and the answer content part, information associating the question content part with the answer content part can be created using existing documents, and by using the information, the user can appropriately answer the question. Answers can be easily obtained. Further, in the question and answer management device according to the present invention, a plurality of documents, for example,
Since the contents of the question and the answer are extracted from a plurality of documents scattered and published in the network environment and correspond to each other, it is possible to collect a large amount of information that associates the question and the answer. The user can obtain a more appropriate answer to the question by using the information.

[Brief description of the drawings]

FIG. 1 is a configuration diagram of a question and answer management device according to an embodiment of the present invention.

FIG. 2 is a configuration diagram of a question answer extraction unit according to one embodiment of the present invention.

FIG. 3 is a diagram showing a record format of a question answer record according to an embodiment of the present invention.

FIG. 4 is a diagram illustrating an operation of a sentence determination unit of the question answer extraction unit according to one embodiment of the present invention.

FIG. 5 is a flowchart of a search process by a question and answer list management unit according to one embodiment of the present invention.

[Explanation of symbols]

 1. Document search means 2. Question and answer extraction means 3. Database (storage means) 4. Question and answer list management means.

──────────────────────────────────────────────────続 き Continued on the front page (51) Int.Cl. ⁶ Identification code FIG06F 15/403 330C

Claims (4)

[Claims] A question and answer management device for extracting and managing a question content portion and an answer content portion corresponding to the question content portion from a document, wherein the knowledge and answer content portion relating to elements constituting the question content portion are configured. A question and answer extracting means for identifying and extracting a question content portion and an answer content portion from a document based on knowledge of the element, and a question answer storing the extracted question content portion and the answer content portion in a storage device in association with each other A question and answer management device, comprising: management means. 2. The question-and-answer management device according to claim 1, further comprising information for identifying a document including a question content portion and an answer content portion attached to a document in advance, or information included in a sentence. A search unit that searches a plurality of documents for a document that includes the question content portion and the answer content portion based on elements constituting the question content portion and the answer content portion; A question and answer management device characterized in that a question content portion and an answer content portion are specified and extracted from the obtained document. 3. The question-and-answer management device according to claim 2, wherein the question-and-answer management device is connected to a plurality of document holding devices that hold documents via a network. A question and answer management device characterized in that a document including a question content portion and an answer content portion is searched from documents of a plurality of document holding devices connected via the same. 4. The question and answer management device according to claim 1, wherein the question and answer management means receives a question from a user,
A question and answer management device characterized by retrieving and outputting answer contents corresponding to the question from a storage device.