KR101636477B1 - Human-like Knowledge Expansion and Verification System Using Structured Knowledge Information and Document Crwaling, Method, Recording Medium - Google Patents

Abstract

By expanding and verifying the knowledge information according to the present invention, it is possible to solve the reliability problem of knowledge information generated in the process of expanding the knowledge information, which has been continuously and conventionally proposed, to reduce the cost burden of expansion and verification, We can evaluate the value and reliability of pre-built knowledge information by using knowledge information in collection, goal search, and extraction, and verify consistency and reliability by comparing / comparing new knowledge information collected and extracted with existing knowledge information A document collecting unit, a knowledge information extracting unit, a knowledge information verifying unit, and a knowledge information updating unit for effectively expanding knowledge information.

Description

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a knowledge extension and verification system using structured knowledge information and document collection,

As a part of the "Symbolic Approach-Based Human Model Self-Learning Intelligence Source Technology Development" supported by the Future Creation Science Department, the present invention relates to the meaning and knowledge used in the natural language processing and artificial intelligence such as ontology, semantic network, semantic web, knowledge base, , And expanding and verifying knowledge automatically and semi-automatically using knowledge information used in intelligent processing.

Ontologies, semantic networks, semantic webs, and knowledge bases are widely used as a way for computers to understand human needs and intentions. However, it is difficult to build and verify on a large scale due to high reliance on expert knowledge in construction and verification.

Recently, ontologies have been expanded to integrate ontologies constructed in different domains, and validation methods of ontologies constructed in this process have also been proposed. We also extend ontologies and knowledge bases by creating RDF triples based on large web documents.

The ontology is generated by the conventional generation method through the process of recognition, generation, evaluation, and recording. Traditional methods rely heavily on a lot of things. The ontology generation through heuristics has the advantage of generating an ontology that is rich in rich vocabulary and formal meaning.

However, there are disadvantages of consuming a lot of time and money because it relies heavily on many parts. The heuristic-dependent part is an obstacle to the reduction of time and cost. To overcome these drawbacks, proposed methods include domain ontology generation method using UML (XML), domain ontology generation method using formal concept analysis, data mining technique (ID3, AOI, association rules mining, clustering) And ontology generation methods using tools (Protege 2000, etc.).

However, the domain ontology generation method using UML depends on heuristic much in the process of analysis and design, and it consumes much time and money. The method using formal concept analysis relies on heuristics in much of the process of searching for valid keywords in documents and the data mining technique is used for applying data mining technique

It has a disadvantage of consuming a lot of time and money in selecting the data. The method of using the tool reduces the time and cost consumption through the tool which depends on the heuristic, but it basically consumes the time and the cost because it heavily depends on the heuristic, which helps the traditional generation method. There is a problem in that the verification of effectiveness can not be performed.

Semantic Web technology refers to the next generation intelligent web where computers can understand the meaning of information resources and logically reasoning. Like the current computer, it is a web that a computer understands, not a web that a person uses a mouse or a keyboard to find and understand desired information. It is an intelligent web that allows machines to communicate with each other by expressing them in a new language that can be understood by the computer instead of the current web, which is designed to be easy for people to read and interpret.

The Semantic Web principle is to transform the semantics of information resources into a language that can be understood by computers, unlike current Web documents that are focused on natural language so that people can understand them. This allows the computer to interpret the meaning of the information resource, and the machines can exchange information with each other and handle the tasks themselves.

Research related to semantic web is mainly composed of ontology technology based on Resource Description Framework (RDF) and Topic Map technology based on International Organization for Standardization (ISO). The former is a technology that allows the web to understand and process the meaning of information by assigning metadata, which is a language that describes resources such as resource (subject) / attribute (predicate) / attribute value (object) The latter is a technology that supports the distributed management of information and knowledge using the XML Topic Maps (XTM) language, which is a standard technology language based on ISO's XML (eXtensible Markup Language). It has a dual structure of knowledge layer and information layer. When realized, computers can process information automatically, maximizing productivity and efficiency of information system. It can be applied to various fields such as system integration (SI), intelligent robot system, medical informatization, and so on.

Prior art documents include a method disclosed in Japanese Laid-Open Patent Publication No. 2010-0003084 (an apparatus and method for extracting an ontology partial graph, a search user query and an ontology semantic matching apparatus using the same and a method thereof) A method for generating an ontology, and a device thereof). However, the prior art documents mainly focus on expansion and verification of knowledge information, and there is a lot of noise that is not intended for extended knowledge information And thus it is difficult to effectively utilize the expanded knowledge information.

The extension and verification of knowledge information including existing ontology tended to concentrate mainly on expansion. In addition, there is a lot of information (Noise) that does not fit with the intension / extension intention to extended knowledge information because extension and verification are not done together. This makes it difficult to utilize the expanded knowledge information.

The object of the present invention is to verify / extend knowledge information for computation processing in a similar manner to human knowledge verification / expansion by combining previously constructed knowledge information with a document collector (Crwaler).

In order to achieve the object of the present invention, a knowledge extension and verification system using knowledge information and document collection according to the present invention,

 A query generation unit (100) for converting the generated knowledge information into a query for a document search capable of collecting related documents for updating and verifying knowledge;

A document collection unit 200 for collecting / storing the returned result documents from the target domain for the query sentence generated by the query generation unit 100 and analyzing / evaluating the appropriateness of the collected documents in the target domain;

From the collected target documents, the structural framework and rules of prebuilt knowledge information, the meaning of the prebuilt knowledge information (vocabulary, relation, structure and used data value included in the relevant knowledge information), contents related to prebuilt knowledge information A knowledge information extracting unit 300 for extracting new knowledge information based on complementary or incompatible contents);

When the extracted knowledge information is compared with the already constructed knowledge information, if the previously constructed knowledge information is the same as the extracted knowledge information, the necessity of deleting the extracted knowledge information is compared with the extracted knowledge information In the case of containing less information, it is necessary to update the existing knowledge information. If the existing knowledge information includes more information than the extracted knowledge information, it is necessary to verify the necessity of supplementing the extracted knowledge information. (400);

And a knowledge information updating unit (500) for performing the process of deleting extracted knowledge information, updating basic knowledge information, or supplementing extracted knowledge information according to the verification result of the knowledge information verification unit (400) .

In order to accomplish the object of the present invention, the knowledge extension and verification method using knowledge information and document collection according to the present invention,

A. converting the constructed knowledge information into a query word of a domain for document collection;

B. collecting the document in the target domain with the translated query and performing analysis / evaluation on the collected document;

C. From the collected documents using the constructed knowledge information, the structural framework of the knowledge information, the meaning of the knowledge information (vocabulary, relation, structure and used data value included in the knowledge information) Extracting new knowledge information with reference to contents;

D. comparing and verifying extracted knowledge information with established knowledge information;

E. Deleting / updating / supplementing the previously constructed knowledge information and newly extracted knowledge information;

F. Repeat steps A to E if there is knowledge information to be deleted / updated / supplemented again after deleting / updating / supplementing in step E, and terminating the step if not.

The step of extracting new knowledge information according to the present invention includes a step (C-1) of extracting items structurally similar to the knowledge information constructed from the collected documents, a step of obtaining items semantically similar to the knowledge information constructed from the collected documents (C-3) extracting items that are incompatible with the knowledge information constructed from the collected documents, extracting items complementary to the knowledge information constructed from the collected documents (C-4 ).

The step of fusing the extracted knowledge information according to the present invention with the pre-established knowledge information includes a step (D-1) of deleting the extracted knowledge information when the pre-established knowledge information and the extracted knowledge information are the same, (D-2) updating the already-constructed knowledge information when the knowledge information includes less information than the extracted knowledge information; if the previously constructed knowledge information includes more information than the extracted knowledge information, And a step (D-3) of complementing and processing the information.

In addition, the above method is provided as a computer-readable recording medium that is programmed.

By expanding and verifying the knowledge information according to the present invention, it is possible to solve the reliability problem of knowledge information generated in the process of expanding the knowledge information, which has been continuously and conventionally proposed, to reduce the cost burden of expansion and verification, We can evaluate the value and reliability of pre-built knowledge information by using knowledge information in collection, goal search, and extraction, and verify consistency and reliability by comparing / comparing new knowledge information collected and extracted with existing knowledge information And thus the expanded knowledge information can be effectively used.

1 illustrates a knowledge extension and verification system using structured knowledge information and document collection according to the present invention;
2 is a diagram illustrating a knowledge extension and verification method using structured knowledge information and document collection according to the present invention.
3 is a diagram showing a process of extracting new knowledge information

The present invention is intended to provide reliability in expanding knowledge information by extending established knowledge information and verifying extended knowledge information, and the present invention will be described in brief as follows. The method includes the steps of converting previously constructed knowledge information into a query for document collection, collecting / storing the returned result documents for the query, extracting new knowledge information based on the constructed stone information from the collected documents, The information is compared with the previously constructed knowledge information to update the entire knowledge information, and the above process is continued until there is no longer updated knowledge information.

The present invention relates to a system and method for continuously accumulating, verifying, and expanding knowledge through the above-described procedures, and is based on a principle that humans acquire new knowledge based on current knowledge and supplement and expand existing knowledge.

First, a knowledge extension and verification system using structured knowledge information and document collection for pre-established knowledge information, which is one embodiment of the present invention, will be described in detail with reference to FIG.

1, the knowledge extension and verification system using structured knowledge information and document collection according to the present invention includes a query generation unit 100, a document collection unit 200, a knowledge information extraction unit 300, a knowledge information verification unit 400 And a knowledge information updating unit 500.

The query generation unit 100 analyzes the constructed knowledge information and generates a query for a document search that can collect related documents for updating and verifying knowledge. That is, the query generation unit of the present invention converts document contents of the target domain into a query form for document search. This method is similar to the query used by Naver to return the result of a search for a keyword passed by the user in the following search service. However, the query generated by the query generator in the present invention is used to collect the target document, whereas the search service of the general portal is used to respond to the query.

This collection method is a popular collection method such as Google's deep crawl.

The collection of initial documents, especially web documents, was a search space for the address space of the web page where the documents to be collected were stored. The collection of documents is the process of copying the documents stored in a given address space and the contents of the documents are not considered. However, as the document exchanging and providing system through the web is developed, the documents are stored in a separate DB instead of the address space of the web page, and the web documents are reworked according to the user's request or service policy. In this case, a web document can be acquired only when a request is transmitted to the interface providing the web service. Google's deep crawl is a way to obtain a web document that is not stored in the address space in this way.

The query generation unit in the present invention has the same purpose as the query generation that is generated for document collection. The document collection generates a query according to the address space used as the collection target or the service to be acquired.

For example, to acquire a document from a specific information providing service provider (meaning a target domain) for a used car, a keyword is generated based on a car name, a price, a production year, and a manufacturer as keywords. Therefore, it is a key technology to identify the appropriate services and determine appropriate query structures and keywords.

In the present invention, a document containing contents that can be extended and verified is collected. Therefore, since the query expression for document acquisition is generated from the pre-constructed knowledge information, the address of the target domain that services the document (for example, the web domain name such as http: //books.google.com/tech) It has nothing to do with configuration.

Even if they have the same keyword and structure, this is only valid if they are obtained from knowledge information. In other words, it should be noted that the query expression is composed not from the service structure of the used car service company but from the knowledge information that has been constructed for the used car.

For example, even if you search for the keyword "GM car", you can create different quality rituals depending on whether the "GM car" is related to the knowledge information "used car ticket" or "movie transformer" The evaluation results are also different.

Another example of query generation according to the present invention will be described. "Obama is the 21st President of the United States." Knowledge-based keywords for knowledge information include knowledge expressions such as "Obama (person: name), the United States (country: name), the President And can generate the following search query word from this.

serch keyword = (Obama & (type: name or type: person))

serch keyword = (United States & (type: name or type: country))

serch keyword = ((Obama & (type: name or type: person)) & (President) or (US & President)

The format of the search query form may be somewhat different according to the target domain (the address space of the collection target, the service provider, etc.) as the collection target. Document collection through query generation is described in Madhavan et. It is similar to query generation using template introduced in tal (2007, 2008). However, there is a difference in that the query generation is not based on the data type of the collection target but is based on the pre-built knowledge information, that is, the content of the data.

The document collection unit 200 collects / stores the result documents returned from each target domain, and analyzes / evaluates the appropriateness of the documents collected in the target domain, for the query sentence generated by the query generation unit 100.

The result returned from each target domain by the query term generated by the query generation unit 100 are documents having addresses on the network. Thus, collecting a document means collecting a typical web document.

The collection of documents generally means that the documents returned by copying and querying the documents obtained by the address space used for web collection are stored in a certain repository. At this time, the document to be stored includes the document acquisition path, subject query (eg, Obama), and metadata about the address space of the knowledge information used to generate the query (eg, US president, first black president, etc.) do.

In addition, the document collecting unit evaluates the appropriateness of the acquired document, and the information stored together with the document is used for the appropriateness evaluation of the document.

The evaluation of the relevance of a document is the filtering of unwanted documents from the query. For example, among the many documents collected about the query "US President Obama," there is no direct connection with Obama, It is an appropriateness evaluation that evaluates and filters appropriateness for these unwanted documents. Appropriateness assessment means that information that is stored with the collected documents is used to exclude documents that are not needed, that is, inappropriate documents.

For example, a query term is generated by the method described in the description section of the query generation section with respect to knowledge information called "President Obama ", and a large number of documents are collected from the target domain with respect to the query term. Among the collected documents, there is a document containing the knowledge information of President Obama of the United States, but it is the center of China 's science and technology, and it excludes such document. In other words, the document on the collected Chinese science and technology includes the knowledge information of the US President Obama, but it is not about the President Obama but about the Chinese science and technology. Obama is not a proper collection document for a query.

The function of evaluating the contents of the document uses a natural language processing technique, and the natural language processing technology is a known technology, so a detailed description is omitted.

The knowledge information extraction unit (300) extracts new knowledge information from the collected documents.

In extracting knowledge information, meta data included in the document and the basic construction information used in acquisition of the document are used.

In existing ontologies, knowledge bases, and semantic webs, new knowledge information instances are created according to the structural frameworks and rules of each knowledge knowledge.

However, in the present invention, as shown in FIG. 3, not only the structural framework and rules of the basic building information used for acquiring a document, but also the meaning (vocabulary, relation, structure and value of data used in the corresponding knowledge information) The new knowledge information is extracted not only by all the contents of the document but also by the contents related to the applied knowledge information. That is, the present invention is characterized in that new knowledge information is extracted by reflecting the meaning and contents of knowledge information in addition to the conventional structural framework and rules for extracting new knowledge information.

The new knowledge information extraction process applied to the present invention differs from the conventional knowledge information extraction process in that the human being acquires knowledge does not knowledge all the data that are contacted and that the analysis result and the obtained data are different depending on the existing knowledge applied when interpreting the data It is characterized by using points.

Hereinafter, a process of extracting new knowledge information based on the structure, meaning and contents of knowledge information constructed in the knowledge information extracting unit of the present invention will be described.

The knowledge information extraction unit 300 of the present invention extracts items structurally similar to the constructed knowledge information from the collected documents to extract new knowledge information.For example, The structure of the knowledge information that has been constructed becomes the Y-shape of X. A similar structural item is "the 21st President of the United States" and the structure is X (the number indicating the number of units) versus Y. Accordingly, the knowledge information extraction unit 300 extracts knowledge information having a structure of "President of the United States (numerical expression) vs. President ".

The knowledge information extracting unit 300 of the present invention extracts items that are semantically similar to the constructed knowledge information from the collected documents. For example, there may be a king, a prime minister, a commentary, etc., which have a similar meaning to the president. Therefore, "King of America". "Prime Minister of America", "Annotation of America", and so on. Likewise, "the President of Korea" using the name of a country with semantic system information similar to the United States. And extracts knowledge information including "President of France ".

In addition, the knowledge information extraction unit 300 of the present invention extracts new knowledge information from the content aspect of the constructed knowledge information. For example, extract items that are incompatible with the previously constructed knowledge information from the collected documents. For example, if the established knowledge information is "the most expensive smartphone", the item "cheapest smartphone" is extracted as the contradictory item, and the complementary items are "the most advanced smartphone", " Phone ". As described above, the knowledge information extracting unit of the present invention extracts new knowledge information in consideration of the structure, meaning, and content aspects of the knowledge information constructed from the highly related documents.

Next, the knowledge information verifying unit 400 verifies the previously constructed knowledge information and the newly extracted knowledge information. The verification of knowledge information has long been relied on by experts in the field or by experts on knowledge information. According to a survey by Hlomani Hlomani and Deborah Stacey (2014), the verification of knowledge information such as the ontology is based on a methodology-based method (Gold standard-based evaluation), an application or task-based evaluation , User-based evaluation, and data-driven evaluation are used.

The method proposed by the present invention is also included in the data-driven based on the collected document. Hlomani Hlomani and Deborah Stacey (2014) cite I. Nonaka and R. Toyama (2005), but the data driven method is not always consistent and not explicit because the domain knowledge is dynamic. And the interpretation of the data.

In the present invention, such a problem is solved by the two methods (the method used in the query generating unit and the knowledge information extracting unit of the present invention) proposed above. That is, the first is a function of the query generation unit 100 that increases the association between the collected document (i.e., data) and the target knowledge by using the previously constructed knowledge as a query, and the second is the function of the query generation unit 100, It is a method of analyzing data from the viewpoint of target knowledge using the constructed knowledge information and the metadata of the target documents.

In addition, the knowledge information updating unit 500, which will be described later, accepts the characteristics of the dynamic domain knowledge by adding new knowledge by comparing / collating the established knowledge information with the extracted knowledge information, supplementing the deficient knowledge, It complements knowledge. This process is repeated and reliability is assessed by the weights given each time the verification is performed. That is, the reliability of the knowledge information having a low update rate is maintained, and the reliability of the knowledge information having a high update rate is low.

The verification of the knowledge information performed by the knowledge information verifying unit 400 according to the present invention is performed in the following manner according to the object and type.

<When the established knowledge information and the newly extracted knowledge information are the same>

The newly extracted knowledge information, which is completely or consistently related to the constructed knowledge information and structure and contents, is unnecessary information of extension or verification of knowledge information, so that newly extracted knowledge information is deleted. This increases the reliability of the pre-established knowledge information.

For example, if the established knowledge information is "Obama is the 21st President of the United States" and the newly extracted knowledge information is "Obama is the President of the United States" and "Obama is the 21st President of the United States" Obama is the 21st President of the United States "is information consistent with the existing knowledge information, and" Obama is the President of the United States " The knowledge information "Obama is the president of the United States" and "Obama is the 21st president of the United States" are deleted as knowledge information that does not need to be extended to new knowledge information.

<When the pre-established knowledge information contains less information than newly extracted knowledge information>

If the pre-built knowledge information is insufficient than the extracted knowledge information (when the newly extracted knowledge information has a relation and attribute with more data (nodes) than the pre-established knowledge information), the pre-built knowledge information is newly extracted The knowledge information should be updated with the knowledge information. In this case, the reliability of the pre-established knowledge information is lowered, and the pre-established knowledge information is updated by the knowledge information update unit to be described later.

For example, the preliminary knowledge information is "Obama is the President of the United States" and the newly extracted knowledge information is "Obama is the black president of the United States" or "Obama, the 21st president of the United States, "Obama is the president of the United States," he said. There is insufficient information about whether it is black or white. Therefore, there is a need for renewal in terms of structure, meaning, and contents of the existing knowledge information "Obama is the President of the United States", and updated information is updated through a knowledge information update unit to be described later. In other words, the knowledge of the pre-built knowledge, "Obama is the President of the United States," reflects the information on how many presidents, blacks or whites are, and is updated with new knowledge.

<When the established knowledge information contains more information than the newly extracted knowledge information>

The fact that the constructed knowledge information contains more information than the extracted knowledge information means that the collected document does not have sufficient knowledge or the extracted knowledge information is incomplete. For example, if the pre-built information is information that "Obama is a US president from the Democratic Party" and the extracted information is "Lincoln is the President of the United States," the extracted information includes knowledge of Lincoln's Party . In this case, since the extracted knowledge information is incomplete, the necessity of supplement is indicated, and the reliability is set to zero.

As described above, the knowledge information verifying unit 400 of the present invention displays a deletion or a supplement to the newly extracted knowledge information, performs a function of displaying an update on the previously constructed knowledge information, To the generating unit.

Next, the knowledge information update unit 500 of the present invention performs processing of knowledge information according to an indication such as deletion, update, or supplement given in the verification result of the knowledge information verification unit 400. [

The deleted knowledge information includes the same structure, meaning, and content as the previously constructed knowledge information, and is deleted or moved to a separate storage space.

The updated knowledge information is attached to the necessary information (node, relationship, etc. to be added / corrected and its value (there may be no value)), changes the status indication to complement, and transmits it to the query generation unit. The knowledge information to be updated is not removed from the knowledge information list for document collection until the reliability of the knowledge information verifier is increased.

For example, if the established knowledge information is "Obama is the President of the United States" and the newly extracted knowledge information is "Obama is the 21st President of the United States" or "Obama the 21st President is the American black president" In the case of the construction information, "Obama is the President of the United States of America" is the knowledge information to be renewed, and the information required for renewal is the number of presidents, blacks or whites. At this time, the knowledge information renewal unit of the present invention changes the supplementary information to "Obama is the President of the United States," which is the basic construction information, and then transmits the related information (information indicating how many presidents or black or white people) 100 and transmits the information to the query generation unit 100. The query generation unit 100 generates a query term including the number of presidents and black or white in the query term structure and the content and the document collection unit 200 uses the generated query term After performing the described document collection and analysis / evaluation, the extraction and verification process shown in FIG. 1 is repeated.

The case of knowledge information supplemented is also described.

When the newly-extracted information is "Lincoln is the President of the United States", the newly extracted knowledge information is incomplete information and needs to be supplemented , And the knowledge information update unit transfers the information necessary for the supplement (information about the Lincoln's party) to the query generation unit. The query generation unit that has received the information generates a new query language (for example, a query having a structure of a person-party-country-title) containing the subject. After the query term is generated, the collection, extraction, and verification processes are repeated according to FIG.

Hereinafter, a knowledge extension and verification method using knowledge information and document collection, which is another embodiment of the present invention, will be described with reference to FIG.

As shown in FIG. 2, the knowledge extension and verification method using knowledge information and document collection, which is one embodiment of the present invention,

A. converting the constructed knowledge information into a query of domain for document collection;

B. collecting the document in the target domain with the converted query, and performing analysis / evaluation on the collected document;

C. From the collected documents using the constructed knowledge information, the structural framework of the knowledge information, the meaning of the knowledge information (vocabulary, relation, structure and used data value included in the knowledge information) Extracting new knowledge information with reference to contents;

D. comparing and verifying extracted knowledge information with established knowledge information;

E. Deleting / updating / supplementing the previously constructed knowledge information and newly extracted knowledge information;

F. Repeat steps A to E if there is knowledge information to be deleted / updated / supplemented again after deleting / updating / supplementing in step E, and terminating the step if not.

Step A: The step of converting the previously constructed knowledge information into a query word for document collection will be described.

Obama is the President of the United States, which is a knowledge base that has been constructed. In this case, the query form can be as follows.

serch keyword = (Obama & (type: name or type: person))

serch keyword = (United States & (type: name or type: country))

serch keyword = ((Obama & (type: name or type: person)) & (President) or (US & President)

Step B: The step of collecting the document in the target domain with the converted query and performing the analysis / evaluation on the collected document will be described.

The document is collected from the target domain (all portal or service site where document collection is possible) using the converted query. Inappropriate documents that are not included in the collected documents are excluded from the collection through analysis / evaluation. In this case, inappropriate documents are filtered using known natural language processing techniques.

For example, suppose that there is a document that Obama is the word of the collected documents, but the document is related to Chinese science and technology, which is far from "Obama is the President of the United States". .

At this time, documents related to Chinese science and technology are excluded from the collection subject by using known natural language processing techniques.

Step C: The step of extracting new knowledge information will be described.

In extracting new knowledge information for collected documents excluding inappropriate documents, semantic and related contents are reflected as extraction factors in addition to the structure of knowledge information that has been constructed.

"Obama is the president of the United States." Structurally similar knowledge is "Lincoln is the first president of the United States," and "Obama is the American black president."

Next, the semantic similarity between "Obama is the president of the United States" and "Obama is the King of America" and "Obama is the commentator of the United States" and so on.

"Obama is a popular president of the United States,""Obama is a president of Africa,""Obama is a US president," and "Obama is a president of the United States. It's a white president of "

Step D: The step of comparing and verifying the extracted knowledge information with the established knowledge information will be described.

<When the established knowledge information and the newly extracted knowledge information are the same>

The newly extracted knowledge information, which is completely matched with or included in the previously constructed knowledge information, is information that does not require expansion or verification of the knowledge information, so that the newly extracted knowledge information is deleted. This increases the reliability of the pre-established knowledge information.

For example, if the pre-built knowledge is "Obama is the 21st President of the United States" and the newly extracted knowledge information is "Obama is the President of the United States" or "Obama is the 21st President of the United States" (Ie, "Obama is the President of the United States," or "Obama is the President of the United States"), President Obama is the 21st President of the United States. ") It is deleted because it is meaningless knowledge information that can not expand the established knowledge information (" Obama is the 21st President of the United States ").

<When the pre-established knowledge information contains less information than newly extracted knowledge information>

If the pre-built knowledge information is insufficient than the extracted knowledge information (when the newly extracted knowledge information has a relation and attribute with more data (nodes) than the pre-established knowledge information), the pre-built knowledge information is newly extracted The knowledge information should be updated with the knowledge information. In this case, the reliability of the pre-established knowledge information is lowered.

For example, the established knowledge is "Obama is the President of the United States," and the newly extracted knowledge is "Obama is the 21st President of the United States," or "Obama, the 21st President, Obama is the president of the United States. There is insufficient information about whether it is black or white. Therefore, there is a need to update the structure, meaning, and content of Obama's "Obama is the President of the United States," which is the base knowledge of knowledge. , And information indicating whether the person is black or white is updated and displayed as knowledge information to be reflected.

<When the established knowledge information contains more information than the newly extracted knowledge information>

The fact that the constructed knowledge information contains more information than the extracted knowledge information means that the collected document does not have sufficient knowledge or that the extracted knowledge information is incomplete. For example, if the pre-built information is information that "Obama is a US president from the Democratic Party" and the extracted information is "Lincoln is the President of the United States," the extracted information includes knowledge of Lincoln's Party . In this case, since the extracted knowledge information is incomplete, it indicates the necessity of complement and the reliability is set to zero.

As described above, the step of verifying the extracted knowledge information with the preliminarily constructed knowledge information may include the step of displaying the deletion or the supplement to the newly extracted knowledge information or the function of displaying the renewed knowledge information to be.

Step E, F: Deleting / updating / supplementing the existing knowledge information and newly extracted knowledge information, and A / A information if there is knowledge information to be deleted / updated / supplemented after deletion / The step of repeating the step E and terminating the step is described below.

In step D, the deleted knowledge information includes the same structure, meaning, and content as the previously constructed knowledge information and is deleted or moved to a separate storage space.

In step D, the knowledge information that is updated is attached to the necessary information (node, relationship, etc., to be added / modified and its value (there may be no value)), do.

For example, if the pre-built knowledge is "Obama is the President of the United States" and the newly extracted knowledge is "Obama is the 21st President of the United States" or "Obama, the 21st President, Obama is the President of the United States, and the information needed for renewal is the president, black, or white. At this time, we repeat the steps A to E after the change of the existing construction information "Obama is the President of the United States" and the related information to be supplemented (information indicating how many presidents, black or white people) are reflected.

In other words, a new query term including "president", black or white is created in step A of the previously constructed knowledge information that "Obama is the President of the United States", and then the collection / extraction / verification steps are repeated.

The case of knowledge information supplemented in step D will be described.

When the newly-extracted information is "Lincoln is the President of the United States", the newly extracted knowledge information is incomplete information and needs to be supplemented And return to Step A with the information necessary for supplementation (information about Lincoln's party). In step A, a new query (eg, a query with a structure of person-party-country-title) containing the subject is created, and then the collection / extraction / verification process is repeated.

If there is no knowledge information to be deleted / updated / supplemented, the knowledge extension and verification method using knowledge information and document collection is terminated.

The knowledge extension and verification method using the knowledge information and the document collection according to the present invention can be performed automatically by computerized programming, and the program is provided in the form of a recording medium readable by a computer.

100: query generation unit
200: Document collecting section
300: knowledge information extracting unit
400: knowledge information verification unit
500: Knowledge information update section

Claims (5)

A knowledge extension and verification system using structured knowledge information and document collection for extending and verifying established knowledge information,
A query generation unit (100) for converting the generated knowledge information into a query for a document search capable of collecting related documents for updating and verifying knowledge;
A document collection unit 200 for collecting / storing the returned result documents from the target domain for the query sentence generated by the query generation unit 100 and analyzing / evaluating the appropriateness of the collected documents in the target domain;
From the collected target documents, the structural framework and rules of prebuilt knowledge information, the meaning of the prebuilt knowledge information (vocabulary, relation, structure and used data value included in the relevant knowledge information), contents related to prebuilt knowledge information A knowledge information extracting unit 300 for extracting new knowledge information based on complementary or incompatible contents);
When the extracted knowledge information is compared with the already constructed knowledge information, if the previously constructed knowledge information is the same as the extracted knowledge information, the necessity of deleting the extracted knowledge information is compared with the extracted knowledge information In the case of containing less information, it is necessary to update the existing knowledge information. If the existing knowledge information includes more information than the extracted knowledge information, it is necessary to verify the necessity of supplementing the extracted knowledge information. (400);
And a knowledge information update unit 500 for performing the process of deleting extracted knowledge information, updating basic knowledge information, or supplementing extracted knowledge information according to the verification result of the knowledge information verification unit 400
The knowledge information extraction unit 300 extracts items structurally similar to the knowledge information constructed from the collected documents, extracts items that are semantically similar to the knowledge information constructed from the collected documents, Extracting new knowledge information using at least one of extracting items that are in conflict with the knowledge information, extracting items that are complementary to the knowledge information constructed from the collected documents, and extracting new knowledge information. Knowledge Extension and Verification System. A. converting the constructed knowledge information into a query word of a domain for document collection;
B. collecting the document in the target domain with the translated query and performing analysis / evaluation on the collected document;
C. From the collected documents using the constructed knowledge information, the structural framework of the knowledge information, the meaning of the knowledge information (vocabulary, relation, structure and used data value included in the knowledge information) Extracting new knowledge information with reference to contents;
D. comparing and verifying extracted knowledge information with established knowledge information;
E. Deleting / updating / supplementing the previously constructed knowledge information and newly extracted knowledge information;
F. Repeating steps A to E if there is knowledge information to be deleted / updated / supplemented again after deleting / updating / supplementing processing in step E,
(C-1) extracting items structurally similar to the knowledge information constructed from the collected documents, (C-2) extracting items semantically similar to the knowledge information constructed from the collected documents, A step (C-3) of extracting an item contrary to the knowledge information constructed from the collected document, and a step (C-4) of extracting items complementary to the knowledge information constructed from the collected document Knowledge Extension and Verification Method Using Knowledge Information and Document Collection.
delete 3. The method of claim 2,
The step (D) includes the steps of (D-1) deleting the extracted knowledge information when the constructed knowledge information is the same as the extracted knowledge information, (D-2) renewing and displaying the constructed knowledge information, and (D-3) supplementing and displaying the extracted knowledge information when the constructed knowledge information includes more information than the extracted knowledge information Knowledge Extension and Verification Method Using Knowledge Information and Document Collection. A computer-readable recording medium having recorded thereon a program for executing a knowledge extension and verification method using knowledge information and document collection according to claim 2 or claim 4.

Images