KR20030006201A - Integrated Natural Language Question-Answering System for Automatic Retrieving of Homepage - Google Patents

Abstract

PURPOSE: An integrated answering system of natural language is provided to prepare a recommended question which is similar to a user's question by attaching the recommended question on each web page, searching the most similar questions to an inputted user's question, and suggesting the questions. CONSTITUTION: A morpheme analyzing unit(110) receives a natural language query of a user from a natural language query input unit(100) and analyzes the natural language query by one morpheme. A partial sentence structure analyzer(120) decides a mathematical expression relation between the analyzed morphemes. An integrated query indexer(200) stores a search-objected web document and a representative query of a web page in an index database. An integrated query finder(300) extracts a similar query list, a correct answer, and a document using a Lexico Syntactic pattern based on a morpheme and sentence structure analyzing result. An integrated result processing unit(400) integrates and manages the result of the integrated query finder(300).

Description

Integrated Natural Language Question-Answering System for Automatic Retrieving of Homepage}

The present invention relates to a natural language query-response system, and more particularly, to an integrated natural language query-response system in which general users can conveniently and accurately find information desired by a user regardless of the form of a query on a homepage by effectively processing a descriptive query. .

Conventional question-and-response retrieval systems have a disadvantage in that response time is very slow by extracting and scoring correct candidates and filtering out unnecessary information, and also effectively reflecting the context around the correct candidates due to time constraints. There was a problem that can not. To solve this problem, an example of a natural language query-response system that recommends a correct answer to a short-answer question in which a user's question can be answered shortly is described by the present applicant as a "natural language query using a real-time response index in a paragraph unit-". Patent application 2001-12071, entitled "Response Search System."

However, in such a natural-language query-response retrieval system, a noun phrase is used to answer a descriptive query that asks a 'how' or 'reason' that cannot be answered in a short answer form. Rather than being able to do this, each procedure should be explained descriptively. That is, there is a problem that cannot be answered with a noun phrase.

In addition, the query-response system that recommends similar questions about the descriptive query in the natural language search method performed by the conventional search site (for example, naver.com) classified all websites (homepages) and made a question. However, there is a problem that the similar question recommended for the input question is too comprehensive or different from the desired one, so that the user's query cannot be answered most similarly.

According to the present invention, each web page of a specific website is pre-posted a question that the page can answer, and a question that is recommended by searching for and presenting a question most similar to the input user's question is significantly different from the user's question. To provide a descriptive natural language query-response system that uses similar high-precision vertical techniques for searching specific data.

Another object of the present invention is to provide an integrated query-response system that can process a user's descriptive query and a short answer query together, and recommends similar queries by comparing a user's query with a pre-built representative query set.

Another object of the present invention is to provide a representative query construction system that allows administrators to create a query example expected through web surfing or query log analysis of an existing user and link it to a homepage.

1 is a block diagram of an integrated natural language query-response system according to the present invention;

2 is a structural diagram of an integrated query indexer according to the present invention;

3 is a structural diagram of an integrated query searcher according to the present invention;

4 is a flow diagram illustrating a descriptive query-response processing method in accordance with the present invention.

5 is a flow chart showing a short-answer question-response processing method according to the present invention.

FIG. 6 is a detailed flowchart of a paragraph classification step of determining a paragraph including a correct answer from a document having a correct answer candidate in FIG. 5 according to the present invention. FIG.

7 is a block diagram of a representative query building system according to the present invention;

8 is a flowchart illustrating a representative query construction method according to the present invention.

Explanation of symbols on the main parts of the drawings

110: stemmer 120: partial parser

200: integrated query indexer 201: descriptive query indexer

202: short answer query indexer 300: integrated query searcher

301: Descriptive Query Searcher 302: Short Answer Query Searcher

400: integrated result handler

Natural language query-response system of the present invention for achieving the above object comprises: a morpheme analyzer for analyzing the input of the natural language query of the user; A partial parser for determining a mathematical relationship between the morphemes; An integrated query searcher comprising a short-answer query searcher that recommends correct answers and documents for short-answer questions according to the morphological analysis and the syntax analysis result, and a descriptive query searcher that recommends a query most similar to a descriptive query; The required term of the virtual document is extracted by constructing a short-answer query indexer that extracts the correct candidates of the searched web document and assigns scores to the correct index database and the intention of the searched web page as a virtual document. An integrated query indexer composed of a descriptive query indexer that scores and stores the score in a descriptive query index database; And an integrated result processor for integrating the results of the integrated query searcher to recommend a similar query pre-built with the correct answer for the short answer query, and to recommend a similar query for the descriptive query.

Hereinafter, with reference to the accompanying drawings will be described in detail the present invention.

1 is a block diagram of an integrated natural language query-response system according to the present invention. In FIG. 1, a morphological analyzer 110 for receiving a user's natural language query from a natural language query inputter 100 and analyzing the morpheme with a partial syntax analyzer 120 for determining a mathematical relationship between the analyzed morphemes, a search target web document, and An integrated query indexer 200 which stores a representative query of a web page in an index database and extracting a list of similarities, correct answers, and documents using Lexico Syntactic patterns of the morpheme and parsing results. An integrated query searcher 300 and an integrated result processor 400 that integrates and processes the results of the integrated query searcher 300.

2 is a structural diagram of an integrated query indexer 200 according to the present invention.

In FIG. 2, the integrated query indexer 200 of the present invention comprises a virtual document builder 210 which constitutes a query set composed of queries having the same intention of a search target web page as one virtual document, and the virtual document is morphologically. Contents of the virtual document using the results of the morpheme analyzer 212 to analyze, the partial parser 214 to determine the mathematical relationship between the analyzed morphemes, and the results of the morphological analyzer 212 and the partial parser 214. Content-weighted extractor 216 for extracting a word, and weighting the content according to the term frequency (TF), inverse document frequency (IDF) and parts of speech extracted from the content-extractor Analyze stems of sentences in the descriptive query indexer 201 including a score assigner 218 and a descriptive index database 220 for storing the content word and the score. A morpheme analyzer 222, a partial parser 224 for determining a mathematical relationship between the analyzed morphemes, and a correct candidate candidate extractor 226 for extracting a correct candidate using a region dictionary and a correct answer type rule in the analyzed syntax. ), A paragraph separator 228 for determining a range of surrounding contexts of candidates extracted by the correct candidate extractor, and a degree of association between terms existing within the separated paragraphs and correct answer candidates, by calculating a degree of surrounding words. A short answer query indexer 202 including a score giver 230 for scoring a score, and a classification store 232 for classifying the types of the correct answer candidates and storing the correct word candidates in a correct answer index database 234. The integrated query indexer 200 is preferably operated in a server system of a web search site.

3 is a structural diagram of an integrated query finder 300 according to the present invention.

In FIG. 3, the integrated query searcher 300 of the present invention receives the results of analyzing the user's natural language query using the morpheme analyzer 110 and the partial parser 120 in FIG. 1, and Lexico Syntactic. Using the information of the essential term extractor 310 and the descriptive index database 220 to extract and classify the key words in the user's query into an essential term and an optional term using a pattern. Representative query builder 312 for extracting the representative query including the required term, similarity calculator 314 for calculating the similarity between the representative query extracted from the representative query builder and the user's query ranking The descriptive query searcher 301 including the representative query ranker 316 to be formed and the natural language query of the user are analyzed by the stemmer 110 and the partial parser 120. Extracts candidate correct answers using information from the correct answer index database 234 using the intention analyzer 318 for determining a user's query intent using a Lexico syntax pattern, and the similarity between the query and each correct candidate It consists of a short-answer query searcher 302 that includes a similarity calculator 320 to calculate the. The short answer query searcher 302 preferably extracts candidate answers using information from a general document searcher 322 and a document index database 326 that search a web document in response to a user's natural language query. And a similarity integrator 328 that integrates the similarity between the correct candidate and the query word to the similarity between the correct candidates. The integrated query searcher 300 as described above is preferably installed to operate on a server computer of a web search site.

4 and 5, the integrated query-response system will be described in more detail. As shown in FIG. 4, the descriptive query indexer 201 has the same intention of the searched web page in the virtual document composer 210. Consisting a query set into a single virtual document (for example, "I want to go back to school", "How do I get a back home?"). Stemming and parsing the paragraph of the virtual document through the morphological analyzer 212 and the partial parser 214 of the virtual document creation step (S2), and the virtual document created through the virtual document creator 210 (S3) ).

After performing the step S3, the content word extractor 216 performs a step (S4) for extracting the words that are the contents of the virtual document for each part of speech. When the content word is extracted in step S4, the weight is applied to the content word by calculating the product (TF * IDF) of the TF indicating the frequency of the content word usage in one document and the IDF representing the inverse of the number of content word usage in the various documents. After performing step S5, the step S6 of weighting is performed according to the parts of speech of each content word (for example, proper nouns give a higher weight than common nouns).

The weights given in the steps S5 and S6 and the scores assigned to the content words are stored together with the content words in the descriptive index database 220 (S7).

Meanwhile, the descriptive query searcher 301 illustrated in FIG. 4 performs language analysis on the user's natural language query received through the natural language query inputter 100 of FIG. 1 through the morpheme analyzer 110 and the partial syntax analyzer 120. The result of the analysis is performed by extracting the required term by classifying the required term and the selected term using the Lexico syntax pattern H in the essential term extractor 310. The essential term is the key word in the user's query. The term must appear in the document or sentence to be searched. The selection term is the word that helps to identify the user's intention.

The extraction of mandatory terms is determined by the syntactic elements and forms around the term. For example, in the sentence "Article about China", "China" is the subject of the query by the syntactic element of "about". It is extracted to the required term.

When the essential term and the selection term of the user query extracted in the step S8 are extracted, the representative queries of each home page pre-built in the descriptive index database 220 are recommended to measure and recommend the similarity with the user query. In step S9, a representative query set is configured. For each homepage, a typical query for the user to get information from the homepage (for example, a homepage corresponding to 'Bachelor Administration' contains a mandatory term 'return', such as "How do I get back to school?") Is built in advance.

When the representative query set composed of the representative queries of each homepage is configured in step S9, the similarity calculator 314 removes the representative queries that do not include the required term, and between the representative queries including the required term and the user's query. A step S10 of calculating the number of index terms shared in step S10 is performed. In the step S10, the representative query sharing a lot of index terms between the representative query and the user query gives a high similarity.

If the number of shared index terms among the representative queries recommended in step S10 is identical, a similarity is given according to a product of weights assigned to the user's query term and the index term of the representative query (S11). . In the step S11, the weight of the user's query term is determined according to syntactic importance, and is given from the high score to the low score in the order of the main word, the subject, and the object. If the weights are matched up, a high score is given to a representative query with a small number of index terms.

After performing the step S11, the same homepage, queries with the same intention to perform the step (S12) of merging into a representative query having the highest score (S12), the score given while performing the steps S10 to S12 Using the representative query ranker 316 performs the step (S13) of ranking each of the representative queries from a high score to a low score.

As illustrated in FIG. 5, the short-answer query indexer 202 may stem and partially parse a paragraph of a target document through the stemmer 222 and the partial parser 224 of FIG. 2 (S14). Perform After performing a series of linguistic processing, the correct candidate candidate extractor 226 includes an area dictionary (A) including a life dictionary, a place name dictionary, and an organization name dictionary containing information corresponding to the type of the correct answer, and a homepage address or an email address. Candidates to be recommended as correct answers are extracted using the correct answer type rule (B) including a regular grammar that can be recognized (S15). The types of correct answer candidates to be extracted are predefined, and the short-answer query searcher 302 can generate a response only for a user query that matches the defined type. I recommend you.

Once the correct candidate is extracted, paragraph separator 228 determines the range of surrounding sentences that may affect the candidate using markers indicating contextual linkage, such as substitute terms and lexical chain information (C). Perform (S16).

In addition, words that may affect the correct candidate in the paragraph are extracted, and the scorer 230 indicates information (D), positional distance difference (E), indicating whether or not each word is equal to the correct candidate. The score is assigned using the part-of-speech information F, the lexical chain information G, and the like (S18).

After performing step S18, a corresponding one of the correct answer index database 234 is selected according to the type of the correct answer currently extracted by the classification storage unit 232, and the peripheral words are stored (S20). The surrounding words become the keys of the correct answer index database 234, and the contents of the database also store information such as the position of the correct answer candidate in the document together with the scores assigned to each word.

The step S16 will be described in detail with reference to FIG. 6. First, if a document and a sentence having a correct answer candidate are selected and extracted (S141), it is determined whether up to and following how many sentences are the maximum paragraph sizes based on the correct candidate sentence (S142 and S143). When the size of the maximum paragraph is determined, the alternative word or vocabulary chain existing in the sentence is examined, and it is determined whether to include in the current paragraph (S144, S145). For example, if there is a lexical chain between the correct candidate candidate sentence and the previous sentence, or if the substitution can be observed, the previous sentence is included in the current paragraph. The paragraph determination method between the correct candidate sentence and the next sentence is performed in the same manner as the above method.

Meanwhile, the short-answer query searcher 302 illustrated in FIG. 5 performs a language analysis on the user's natural language query received through the natural language query inputter 100 of FIG. 1 through the morpheme analyzer 110 and the partial syntax analyzer 120. In step S21, the analysis result is determined by the user intention analyzer 318 using the Lexico syntax pattern H.

After performing step S21, the similarity calculator 32 selects one of the correct answer index databases 234 suitable for the intention of the query, and generates correct answer candidates using the user query term as a key (S22). The similarity calculation between the correct answer candidates and the user query word is calculated using the scores of the respective words stored in the correct answer index database 234. When the similarity calculation is finished, the step of recommending the correct answer by prioritizing is performed (S23).

After performing step S23, in addition to the step, a general document searcher 322 and a document similarity analyzer 324 for searching a general document search a web document in response to a user's natural language query, and a document index database By using the information of 326, the candidate correct answers are extracted, and the document-query similarity is separately obtained according to the general document retrieval method S24 for calculating the similarity between the query word and each correct candidate, and the similarity integrator 328 determines the similarity. Integrating the correct query query similarity according to the step S22 to the document-query similarity which is a general document search result, and then rerank the documents (S25).

The integrated result processor 400 presents the user with a ranked high-order similar query list output from the descriptive query searcher 301 for the descriptive natural language query, and the short answer query is output from the descriptive query searcher 301. A list of the ranked top plurality of correct answers outputted from the short-answer query searcher 302 together with the ranked top plurality of similar queries and the top plurality of documents including the correct answers or linked to the recommended similar query are presented. If the user selects a representative query that is most similar to his query among the suggested similar queries, the user is connected to the web page (that is, the answer page) connected to the representative query.

7 is a block diagram of a representative query building system.

The administrator surfs the Internet through a representative web browser for constructing a representative query that can access the Internet homepage. Queries are inputted to the representative query construction query set builder 540 to construct a query set, provided as an input of the descriptive query indexer 201, and after performing a process of the descriptive query indexer 201. The representative query is stored in the descriptive index database 220.

8 is a flowchart illustrating a flow of a system for easily constructing a representative query by a user / administrator.

The administrator creates a query set through the representative query construction query set generator 540 for the user's existing queries accumulated in the homepage surfing or natural language query-response system (S41), and the homepage is stored in the descriptive index database 220. It is determined whether to add an address, an intention, and a representative query (S42, S43, S44). If there is no intention to register the homepage, delete steps S47, S48, and S49 are performed. When the homepage is registered in the descriptive index database 220, a query set created by the representative query construction query set builder 540 is used. Is provided as an input of the descriptive query indexer 201 (S45), and stored in the descriptive index database 220 through the process of the descriptive query indexer 201 (S46).

In addition, since there are representative queries with different kinds of intentions in one page, the queries are classified and stored according to intents. For example, in order to find a page corresponding to 'Bachelor Administration', users may want to (1) "I want to go back to school", (2) "How do I get a back home?" Or (3) "How many times can I take a leave of absence?" You can query like this. Here, (1) and (2) indicate the same intention, and (3) indicate different intentions but have the same page. By assigning the same ID to a representative query having a different intention on such a same page, the descriptive query indexer 201 points to the same page and integrates it into a representative query having the same intention.

The integrated natural language query-response system according to the present invention as described above effectively processes descriptive queries that are difficult to process in the existing natural language query-response system so that general users can conveniently and accurately find the information they want regardless of the form of the query. Thus, by recommending a similar query pre-built with the correct answer for the user's query, the user can quickly and easily find the desired information.

In the above, the present invention has been shown and described with respect to certain preferred embodiments. However, the present invention is not limited to the above-described embodiments, and those skilled in the art to which the present invention pertains may vary without departing from the spirit of the technical idea of the present invention described in the claims below. It will be possible to carry out the change.

Claims (24)

In a natural language question-answer system, A morpheme analyzer for morphologically analyzing a user's natural language query; A partial parser for determining a mathematical relationship between the morphemes; A descriptive query indexer configured to extract an indispensable term of the virtual document by constructing an intention of a search target web page as a virtual document, assigning a score to the mandatory term, and storing it in a descriptive index database; And And a descriptive query searcher that ranks and recommends a representative query most similar to a user's descriptive natural language query using the stemmer, the parser result, and the index result of the descriptive query indexer. In a natural language question-answer system, A morpheme analyzer for morphologically analyzing a user's natural language query; A partial parser for determining a mathematical relationship between the morphemes; A short-answer query indexer that extracts the correct candidates for the searched web documents, assigns scores and stores them in the correct index database, and extracts the required terms of the virtual documents by constructing the intention of the searched web pages into virtual documents. An integrated query indexer composed of a descriptive query indexer that scores and stores the score in a descriptive index database; A short answer query searcher that recommends correct answers and documents for a short answer query using the stemmer, the parser result, and the index results of the integrated query indexer, and a descriptive query searcher that recommends the most similar query for a descriptive query. Integrated query finder; And And a unified result processor for integrating the results of the unified query searcher to recommend a pre-built similar query with a correct answer for the short answer query, and to recommend a similar query for the descriptive query. 3. The method of claim 1 or 2, wherein the descriptive query indexer comprises: a virtual document builder which composes queries having the same intention of the searched web page into one virtual document; A morpheme analyzer for morphologically analyzing the virtual document; A partial parser for determining a mathematical relationship between the analyzed morphemes; A content extractor for extracting content words of the virtual document using the stemmer and the partial parser results; A score giver for assigning a weight to content words by differentially assigning weights according to TF, IDF, and parts of speech extracted from the content word extractor in the virtual document; And And a descriptive index database for storing the content word. The method of claim 1 or 2, wherein the descriptive query searcher inputs a result of analyzing the user's natural language query using a morpheme analyzer and a partial parser, and uses a Lexic syntax pattern to become a key part of the user's query. An essential term extractor for classifying and extracting words into essential terms and selection terms; A representative query generator for extracting a representative query including a required term by using information of the descriptive query index database; A similarity calculator that calculates a similarity between the representative query extracted by the representative query builder and the user's query; And And a representative query ranker for ranking the results of the similarity calculator. 3. The system of claim 2, wherein the short answer query indexer comprises: a morpheme analyzer for morphologically analyzing the searched web document; A partial parser for determining a mathematical relationship between the analyzed morphemes; A correct candidate candidate extractor which extracts a correct answer candidate using an area dictionary and a correct answer type rule in the analyzed syntax; A paragraph separator for determining a range of surrounding context of candidates extracted by the correct candidate extractor; A score giver that scores the surrounding words by calculating a degree of association between the words existing in the divided paragraph and the correct answer candidates; And And a classification store for classifying the type of the correct answer candidate and storing the correct answer candidate in a correct answer index database. The apparatus of claim 2, wherein the short answer query searcher comprises: an intention analyzer configured to determine a natural language query intention using a Lexico syntax pattern as an input of a result obtained by analyzing a user's natural language query using a morpheme analyzer and a partial syntax analyzer; And And a similarity calculator for extracting candidate correct answers using information of the correct answer index database and calculating similarity between the query and each correct candidate. The apparatus of claim 6, further comprising: a general document searcher for searching a web document in response to the user's natural language query; A document similarity analyzer that extracts candidate correct answers using information in a document index database and calculates similarity between the query and each correct candidate; And And a similarity integrator for integrating the similarity between the query word and the correct answer candidate in the similarity between the document and the query word calculated in the document similarity analyzer. The method of claim 2, wherein the integrated result processor presents a user with a ranked high-number of similar query list output from the descriptive query searcher for a descriptive natural language query and ranks the output from the descriptive query searcher for a short answer query. A natural language query-response system, comprising: a ranked list of top-ranked correct answers output from the short-answer query searcher and a plurality of top-level documents linked to a recommended similar query, together with the top-number of similar queries. Morphologically analyzing the input natural language query of the user; A partial parsing step of determining the mathematical relationship between the morphemes; Storing the intention of the search target web page as a virtual document and storing the descriptive index database; Ranking and recommending the most similar representative query for the descriptive natural language query word; And And linking to a homepage connected to a representative query selected by the user among the recommended representative queries. Morphologically analyzing the user's natural language query; A partial parsing step of determining the mathematical relationship between the morphemes; Extracting correct candidates of the searched web document, assigning scores, and storing the correct candidates in a correct answer index database; Storing the intention of the search target web page as a virtual document and storing the descriptive index database; Recommending a correct answer and a document for the short answer natural language query; Ranking and recommending the most similar representative query for the descriptive natural language query word; Integrating and presenting a similar query list, correct answers, and documents to the user; And And linking to a homepage linked to a similar query selected by the user among the recommended similar query lists. The method of claim 9 or 10, wherein the storing in the descriptive index database comprises: Creating a query of the same intention of the search target web page into one virtual document; Morpheme and partial parsing of the virtual document to determine and extract a part-of-speech of a content word; And And assigning a score to the content word by differentially assigning weights according to TF * IDF and part-of-speech of the content word in the virtual document. The method of claim 9 or 10, wherein the step of ranking and recommending the most similar representative query for the descriptive natural language query word, Morphologically and parsing the natural language query of the user and extracting the core words of the user's query into essential terms and selection terms using a Lexicectic pattern; Extracting a representative query including the required term using the information stored in the descriptive index database; Calculating a similarity between the extracted representative query and the user's query; And Ranking the results of the similarity calculator. The method of claim 12, wherein the step of classifying and extracting a key word in the user's query into a required term and a selection term includes extracting a required term using syntactic elements and forms around the term. Natural language question-answer search method. The method of claim 12, wherein the calculating of the similarity between the extracted representative query and the user's query, Removing the representative query that does not include the required term; Calculating a number of index terms shared between the user's query and the representative queries including the required term; Multiplying the query term and the index term by weights when the number of shared index terms is identical; Assigning a high score to a representative query having a small number of index terms when the product matches the product of weights; And And selecting a representative query having the highest score among queries having the same intention and linked to the same homepage. 15. The natural language query of claim 14, wherein multiplying the weights of the query term and the index term when the number of shared index terms coincide comprises differentially assigning weights according to the parts of the query term. How to search for answers. The method of claim 10, wherein storing in the correct answer index database, Stemming and partial parsing the input document; Extracting a correct answer candidate from a language-analyzed document; Determining a range of sentences affecting the correct candidate; Assigning a score to words around the correct candidate; And Classifying correct candidates according to correct answer types and storing surrounding words in a correct answer index database. The method of claim 10, wherein recommending a correct answer and a document for the short answer natural language query word comprises: Identifying a user's intention using a Lexico syntax pattern based on the result of the morpheme and parsing of the user's natural language query; Selecting the correct answer index database corresponding to a user's intention and generating correct answer candidates using a user query word as a key; Recommending the correct answer by calculating and prioritizing the similarity between the correct answer candidates and the user query word; And And re-ranking the correct candidate by combining the correct candidate-query similarity with the document-query similarity obtained based on the document index database of a general information retriever. 17. The method of claim 16, wherein extracting the correct answer candidate comprises using a correct answer type rule including an area dictionary containing information corresponding to a correct answer type, and a regular grammar for recognizing a homepage address and an e-mail address. Natural language question-answer search method. The method of claim 16, wherein the determining the range of the sentence, Selecting and extracting a document and a sentence having a correct candidate; Determining how many sentences before and after the maximum paragraph size is based on the correct candidate sentences; And And determining the maximum paragraph size to determine whether to include the current paragraph or a vocabulary chain in the sentence to determine whether to include it in the current paragraph. 18. The method of claim 16, wherein the surrounding words stored in the correct answer index database are used as keys of the database, and the contents stored in the correct answer index database determine the position in the document of the correct answer candidate along with the scores assigned to the words around the correct candidate. Natural language query-response search method comprising a. A web browser for constructing a representative query that can access an Internet homepage; A query set generator for constructing a representative query for creating a representative query of a connected homepage while surfing the Internet; And A representative query building system including a descriptive query indexer for creating a home page intention as a virtual document, extracting a required term of the virtual document, assigning a score to the required term, and storing the score in the descriptive index database. 22. The representative query construction system as claimed in claim 21, wherein the representative query construction system stores a home page of representative queries worthwhile among user query logs accumulated in a natural language query-response system in the descriptive index database. Creating a representative query of a homepage accessed while surfing the Internet; Adding or deleting the address, intention, and representative query of the homepage; Indexing a representative query of the home page; And And storing the indexed representative query in a descriptive index database. 24. The method of claim 23, wherein the generating of the representative query further comprises creating a representative query as a representative query of the user's query log accumulated in the natural language query-response system.