KR102543703B1 - Knowledge extraction system for scientific technology papers - Google Patents

Abstract

The knowledge extraction system for science and technology papers according to the present invention includes a document management unit that collects a plurality of scientific and technical papers through a network, extracts data, classifies it according to a technical classification system, and stores it in a classified literature repository; a semi-structured knowledge extraction model for extracting semi-structured data as knowledge among data extracted from the plurality of science and technology papers stored in a classified literature repository with reference to the semi-structured knowledge extraction model; an unstructured knowledge extraction model for extracting unstructured data as knowledge among data extracted from the plurality of scientific and technical papers stored in a classified literature repository with reference to the unstructured knowledge extraction model; a structured knowledge extraction model for extracting structured data as knowledge from data extracted from the plurality of science and technology papers stored in a classified literature repository with reference to the structured knowledge extraction model; and a knowledge management unit that integrates the knowledge extracted from each of the semi-structured knowledge extraction unit, the unstructured knowledge extraction unit, and the structured knowledge extraction unit and stores them in a knowledge base.

Description

Knowledge extraction system for scientific technology papers}

The present invention relates to a knowledge extraction system, and more particularly to a knowledge extraction system for extracting knowledge from scientific and technical papers.

The present invention is derived from research conducted and supervised by Saltlux Co., Ltd. as part of the SW computing industry source technology development project (SW) of the Ministry of Science and ICT. [Research period: 2020.01.01.~2020.13.31., research management specialized institution: Information and Communication Technology Promotion Center, research project title: WiseKB: development of self-learning knowledge base and inference technology based on big data understanding, project identification number 1711103335, Detailed task number 2013-2-00109-008]

Science and technology papers contain academic achievements in the field of science and technology, and are a condensation of the efforts and wisdom of researchers in the field of science and technology, and are an important means of knowledge dissemination and academic exchange. made on the basis of

As the amount of scientific and technological papers published steadily increases and the information available through the Internet reaches a state of excess, the traditional information analysis method of obtaining and analyzing information from scientific and technological papers based on individual knowledge and experience is time-consuming, It has disadvantages such as biased information collection and analysis progress according to the viewpoint of the individual being analyzed.

The technical problem of the present invention relates to a knowledge extraction system for scientific and technical papers, which can efficiently extract knowledge from scientific and technical papers.

A knowledge extraction system for scientific and technological papers according to one aspect of the technical idea of the present invention for achieving the above technical problem collects a plurality of scientific and technological papers through a network, extracts data, classifies according to a technical classification system, a document management unit storing classified documents in a repository; a semi-structured knowledge extraction model for extracting semi-structured data as knowledge among data extracted from the plurality of science and technology papers stored in a classified literature repository with reference to the semi-structured knowledge extraction model; an unstructured knowledge extraction model for extracting unstructured data as knowledge among data extracted from the plurality of scientific and technical papers stored in a classified literature repository with reference to the unstructured knowledge extraction model; a structured knowledge extraction model for extracting structured data as knowledge from data extracted from the plurality of science and technology papers stored in a classified literature repository with reference to the structured knowledge extraction model; and a knowledge management unit that integrates the knowledge extracted from each of the semi-structured knowledge extraction unit, the unstructured knowledge extraction unit, and the structured knowledge extraction unit and stores them in a knowledge base.

The literature management unit, a literature collection module for collecting the plurality of science and technology papers; An optical character recognition (OCR) module for extracting text data from the plurality of scientific and technical papers when the plurality of scientific and technical papers collected by the document collection module are PDF (Portable Document Format) or image files; a literature classification model for classifying the plurality of scientific and technical papers by using text information extracted from the plurality of scientific and technical papers; and a classified literature repository for storing the plurality of scientific and technical papers classified in the literature classification model.

The OCR module may include a text extractor for extracting semi-structured data and unstructured data of the plurality of collected science and technology papers, the text information; and a table recognizer for recognizing and extracting table information, which is structured data of the plurality of collected scientific and technical papers.

The structured knowledge extraction unit may extract the table information, which is semi-structured data extracted from the table recognizer, as knowledge among data stored in the classified literature repository.

The structured knowledge extraction unit may include a header/value classification module for distinguishing a header cell from a value cell in the table information; a semantic analysis module that performs semantic analysis so that the header cell can be connected to knowledge; and a value refinement module that refines the value of the value cell so that it can be linked to knowledge.

The literature classification model performs a search for the plurality of scientific and technical papers collected using a predetermined keyword, classifies a scientific and technical paper corresponding to the predetermined keyword among the plurality of scientific and technical papers, and classifies the classified literature. The scientific and technical papers classified into a specific technical category among the plurality of scientific and technical papers may be stored in the classified literature repository by storing in a repository or by classifying the plurality of scientific and technical papers collected according to technical classification.

The semi-structured knowledge extraction unit extracts bibliographic information and identification numbers, which are semi-structured data, as knowledge from data stored in the classified literature repository, and the bibliographic information is a title, author, abstract, or keyword of a document, and the identification number may be an International Standard Serial Number (ISSN), a digital object identifier (DOI), or an International Standard Book Number (ISBN).

The unstructured knowledge extraction unit may include: an entity name recognition module recognizing entity name information from unstructured data among data stored in the classified document repository; a relationship extraction module for extracting a relationship between entity name information recognized by the entity name recognition module; and a sentence classification module for classifying the sentences from which the relationship between object name information is extracted in the relation extraction module according to a preset sentence classification system.

The knowledge management unit may include: a knowledge integration module for connecting and integrating knowledge extracted from each of the semi-structured knowledge extraction unit, the unstructured knowledge extraction unit, and the structured knowledge extraction unit; and a knowledge conversion module that stores the knowledge that is connected to each other and integrated into the knowledge base as a knowledge graph.

a collection literature repository for storing the plurality of scientific and technical articles collected; and an extraction model learning unit configured to perform learning on each of the semi-structured knowledge extraction model, the unstructured knowledge extraction model, and the structured knowledge extraction model, wherein the semi-structured knowledge extraction model is stored in the collected document repository. Learning may be performed in the extraction model learning unit using stored semi-structured data, and each of the unstructured knowledge extraction model and the structured knowledge extraction model uses unstructured data and structured data stored in the classified literature repository, respectively. Learning may be performed in the extraction model learning unit.

The knowledge extraction system for science and technology papers according to the present invention extracts knowledge through learned knowledge extraction models with both efficiency and accuracy for science and technology papers searched for specific keywords or scientific and technical papers classified into specific technical fields. Therefore, it is possible to quickly and accurately find the knowledge of science and technology papers.

1 is a schematic block diagram of a knowledge extraction system for scientific and technical papers according to an exemplary embodiment of the present invention.
2 is a schematic block diagram illustrating a method for learning a semi-structured knowledge extraction model in a knowledge extraction system for scientific and technical papers according to an exemplary embodiment of the present invention.
3 is a schematic block diagram illustrating a method of learning an unstructured knowledge extraction model and a structured knowledge extraction model in a knowledge extraction system for scientific and technical papers according to an exemplary embodiment of the present invention.
4 is a schematic block diagram illustrating a method of searching for knowledge by using a knowledge base constructed in a knowledge extraction system for scientific and technical papers according to an exemplary embodiment of the present invention.
5 is a schematic block diagram of a natural language understanding unit included in a knowledge extraction system for scientific and technical papers according to an exemplary embodiment of the present invention.
6 is a conceptual diagram illustrating a process of extracting semi-structured knowledge in a knowledge extraction system for scientific and technical papers according to an exemplary embodiment of the present invention.
7 is a conceptual diagram illustrating a process of extracting formal knowledge in a knowledge extraction system for scientific and technical papers according to an exemplary embodiment of the present invention.
8 is a conceptual diagram illustrating a knowledge base constructed by extracting from a knowledge extraction system for scientific and technical papers according to an exemplary embodiment of the present invention.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. The embodiments of the present invention are provided to more completely explain the present invention to those skilled in the art. Since the present invention can have various changes and various forms, specific embodiments will be illustrated in the drawings and described in detail. However, it should be understood that this is not intended to limit the present invention to the specific disclosed form, and includes all modifications, equivalents, and substitutes included in the spirit and scope of the present invention. In describing each figure, like reference numbers are used for like elements. In the accompanying drawings, the dimensions of the structures are shown enlarged or reduced than actual for clarity of the present invention.

Terms used in this application are only used to describe specific embodiments, and are not intended to limit the present invention. Expressions in the singular number include plural expressions unless the context clearly dictates otherwise. In this application, terms such as "comprise" or "have" are intended to designate that there is a feature, number, step, operation, component, part, or combination thereof described in the specification, but one or more other features It should be understood that it does not preclude the possibility of the presence or addition of numbers, steps, operations, components, parts, or combinations thereof.

Unless defined otherwise, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. Terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with the meaning in the context of the related art, and unless explicitly defined in this application, they are not interpreted in an ideal or excessively formal meaning. .

In the drawings and descriptions below, one block, for example, a component indicated or described as '~unit' or '~module' may be a hardware block or a software block. For example, each of the components may be an independent hardware block that exchanges signals with each other, or may be a software block executed on a single processor.

In order to fully understand the configuration and effects of the present invention, preferred embodiments of the present invention will be described with reference to the accompanying drawings.

1 is a schematic block diagram of a knowledge extraction system for scientific and technical papers according to an exemplary embodiment of the present invention.

Referring to FIG. 1, the knowledge extraction system (1, hereinafter referred to as the knowledge extraction system) for science and technology papers includes a document management unit 100, a semi-structured knowledge extraction unit 200, an unstructured knowledge extraction unit 300, and a structured knowledge extraction unit. 400, a knowledge management unit 500, and a knowledge base 600.

The document management unit 100 may include a document collection module 110 , an OCR module 120 , a collected document storage 130 , a document classification module 140 , and a classified document storage 150 . The literature collection module 110 may collect a plurality of scientific and technical papers (document, 40). The OCR module 120 may extract data from a plurality of collected scientific and technical papers 40 . The literature classification module 140 may classify the collected scientific and technical papers 40 according to a pre-stored technical classification system and store them in the classified literature repository 150 .

The literature collection module 110 may collect a plurality of scientific and technical papers 40 through the network 20 . The network 20 includes wired Internet service, local area network (LAN), wide area network (WAN), intranet, wireless Internet service, mobile computing service, wireless data communication service, wireless Internet access service, satellite communication service, wireless LAN, Bluetooth, etc. It may include all that can send and receive data through wired/wireless. When the network 20 is connected to a smartphone or tablet, the network 20 may be a wireless data communication service such as 3G, 4G, or 5G, a wireless LAN such as Wi-Fi, or Bluetooth. The science and technology thesis 40 may be a dissertation, a conference thesis, or an academic journal paper in the field of science and technology, but may also be a document in the field of science and technology, such as a dissertation and a report. In this specification, the scientific and technical papers 40 collected by the knowledge extraction system 1 may be referred to as 'documents'. The scientific and technical paper 40 may include text information, table information, and image information. The image information may include a photo, picture, or graph and a caption therefor.

The Optical Character Recognition (OCR) module 120 may include a text extractor 122 and a table recognizer 124 . The OCR module 120 may extract the collected scientific and technical papers 40 by substituting them into text information when the collected scientific and technical papers 40 are PDF (Portable Document Format) or image files.

The text extractor 122 may extract the text information of the science and technology paper 40 . The table recognizer 124 may recognize and extract the table information of the science and technology paper 40 . In some embodiments, the text extractor 122 may further extract texts of captions for image information of the scientific and technical paper 40 and link information in which images are stored. The text extractor 122 may extract semi-structured data and unstructured data from the science and technology paper 40 . The semi-structured data may be bibliographic information and identification numbers, and the unstructured data may be body contents. The table recognizer 124 may extract structured data from the table information of the scientific and technical paper 40 .

The table recognizer 124 may recognize table information such as texts included in the table information of the scientific technical paper 40 and cell positions of each text in the table. The OCR module 120, when the collected scientific and technical papers 40 are composed of recognizable text, does not perform OCR recognition and extracts the text information and the table information through the text extractor 122 and the table recognizer 124. can be extracted and recognized.

The OCR module 120 may store the text information extracted from the collected scientific and technical papers 40 and the recognized table information in the collected literature repository 130 . Images of the image information included in the collected scientific and technical papers 40 may be stored in the collected literature repository 130 .

The literature classification module 140 may classify the collected scientific and technical papers 40 by using the text information extracted by the OCR module 120 from the collected scientific and technical papers 40 .

In some embodiments, the literature classification module 140 performs a search for the collected scientific and technical papers 40 using a predetermined keyword, and classifies and classifies the scientific and technical papers 40 corresponding to the predetermined keyword. may be stored in the literature repository 150.

In some other embodiments, the literature classification module 140 may classify the collected science and technology papers 40 according to technical classifications such as the National Science and Technology Standard Classification System and store them in the classified literature storage 150 for each classification. For example, the literature classification module 140 may classify the collected scientific and technical papers 40 into major categories, middle categories, and small categories. The major categories may be nature, life, artifact, human science and technology, and non-scientific technology. The middle classification may be mathematics, physics, chemistry, and earth science (Earth/atmosphere/marine/astronomy) if the major classification is nature, and may be life science, agriculture, forestry, fishery, and health care if the major classification is life, and the major classification may be In the case of artifacts, they may be machinery, materials, chemical engineering, electricity/electronics, information/communication, energy/resources, nuclear power, environment, and construction/transportation, and if the major categories are human science and technology, they may be neuroscience and cognitive/emotional science. there is. If the main category is life and the middle category is life science, molecular and cell biology, genetics/genetic engineering, developmental biology|development/neurobiology, immunology/physiology, taxonomy/ecology/environmental biology, biochemistry/structural biology, convergence bio, biology It can be engineering, industrial bio, bioprocess/device, biorisk and other life sciences. The literature classification module 140 may not classify science and technology papers 40 whose main classification is classified as non-science and technology into middle and small classifications.

The classified literature repository 150 may store the collected scientific and technical papers 40 for each category, but unless otherwise specified in the present specification, the classified literature repository 150 uses keywords among the collected scientific and technical papers 40. It may mean that only the science and technology papers 40 searched using the search or science and technology papers 40 classified into a specific technology classification are stored, and the collection literature repository 130 stores all of the collected science and technology papers 40. It can mean what you are doing.

The semi-structured knowledge extractor 200 may extract semi-structured data, bibliographic information and an identification number, as knowledge from data stored in the classified literature repository 150 . The semi-structured knowledge extraction unit 200 may extract bibliographic information and an identification number from data stored in the classified literature repository 150 by referring to the semi-structured knowledge extraction model 250 . The semi-structured knowledge extraction unit 200 may include a bibliographic information extraction module 210 and an identification number extraction module 220. The bibliographic information extraction module 210 may extract the title, author, summary, keywords, etc. of the document from the structured semi-structured data extracted by the text extractor 122 . The identification number extraction module 220 may extract an identification number or the like. The identification number may be, for example, an International Standard Serial Number (ISSN), a digital object identifier (DOI), and an International Standard Book Number (ISBN).

The unstructured knowledge extraction unit 300 may extract unstructured data as knowledge from among data stored in the classified literature repository 150 . The unstructured knowledge extraction unit 300 may refer to the unstructured knowledge extraction model 350 and extract unstructured data as knowledge from among data stored in the classified literature repository 150 . The unstructured knowledge extraction unit 300 may include an object name recognition module 310 , a relationship extraction module 320 , and a sentence classification module 330 . The entity name recognition module 310 may recognize and obtain entity name information from data stored in the classified literature repository 150 to determine candidate sentences to be knowledge extracted. The relationship between the entity name information recognized by the relationship extraction module 320 and the entity name recognition module 310 may be extracted, and the sentence classification module 330 converts sentences from which the relationship between entity name information is extracted into a preset sentence. They can be classified according to a classification system. For example, the sentence classification module 330 may classify sentences from which relationships between object name information is extracted according to a predetermined sentence classification system based on an experiment method and an experiment result.

The structured knowledge extractor 400 may extract structured data, which is table information recognized by the table recognizer 124 of the OCR module 120 and stored in the classified document repository 150, as knowledge. The structured knowledge extraction unit 400 may refer to the structured knowledge extraction module 450 to extract table information as knowledge. The structured knowledge extraction unit 400 may include a header/value classification module 410, a semantic analysis module 420, and a value refinement module 430. The header/value classification module 410 may distinguish header cells and value cells from table information. The semantic analysis module 420 may perform semantic analysis so that the header cell may be connected to knowledge. The value refinement module 430 may refine the value of the value cell so that it can be linked to knowledge.

The knowledge extracted from each of the unstructured knowledge extraction unit 300, the structured knowledge extraction unit 400, and the knowledge management unit 500 may be structured data having a table structure or a tree structure. Structured data that can be provided by each of the semi-structured knowledge extraction unit 200, the unstructured knowledge extraction unit 300, and the structured knowledge extraction unit 400, for example, RDB (Relation DateBase), CSV (Comma-Separated Variables), XML (eXtensible Markup Language), JSON (JavaScript Object Notation), etc., but is not limited thereto.

The knowledge management unit 500 may integrate knowledge extracted from each of the semi-structured knowledge extraction unit 200, the unstructured knowledge extraction unit 300, and the structured knowledge extraction unit 400. The knowledge management unit 500 may include a knowledge integration module 510 and a knowledge conversion module 520 . The knowledge integration module 510 may connect knowledge extracted from each of the semi-structured knowledge extraction unit 200, the unstructured knowledge extraction unit 300, and the structured knowledge extraction unit 400 to each other. The knowledge conversion module 520 may convert knowledge integrated by connecting to each other in the knowledge integration module 510 to be stored in the knowledge base 600 . In some embodiments, the knowledge transformation module 520 may store knowledge in the knowledge base 600 as a knowledge graph.

The knowledge graph refers to a semantic graph that is a graph representing a semantic graph between entities so that search results can be improved by accumulating semantic information from various sources. Semantic information means that a relationship between resources is expressed in some kind of information. That is, the knowledge graph means that data is stored in the form of a graph by collecting semantic information.

A knowledge graph can be built using, for example, a Resource Description Framework (RDF). RDF is a model, language, and grammar for describing (explaining) properties, characteristics, and relationships of all resources (web pages, images, videos, etc.) with a URI (Uniform Resource Identifier), and is a graph-based data model. RDF expresses the relationship between two resources, a subject and an object, and a property or predicate describes the characteristics of this relationship. A sentence with a subject-property-object structure is called a triple, and the relationship may have a direction.

A knowledge graph is a subject-property-object relationship between one entity and another entity, for example, using one entity as the subject and another entity as the object. It is represented as a triple of structure, and another object is used as a subject and another object as an object. What is represented by a triple can consist of repeated multi-order relationship triples.

The knowledge extraction system 1 may further include a question response management unit 700 and a user interface (UI) 900. When the knowledge extraction system 1 further includes a question answering management unit 700 and a user interface (UI, 900), the knowledge extraction system 1 for science and technology papers may function as a science and technology question answering system.

The user interface 900 may receive a query in a natural language form from the user 10 through the network 20 through the user interface 900 and transmit a response. The user interface 900 may provide an interface for accessing the knowledge extraction system 1, that is, the science and technology question answering system, through a terminal used by the user 10. The user 10 may transmit a query to the knowledge extraction system 1, that is, the scientific and technological question answering system, through the user interface 900, and the knowledge extraction system 1, that is, the scientific and technological query, through the user interface 900. Responses to queries provided by the response system may be received.

The question response management unit 700 interprets a query made by the user 10 through the user interface 900 with reference to the natural language understanding unit 750 and generates a response to the query obtained through the knowledge base 600 in natural language. It can be provided to the user 10 through the user interface 900.

The knowledge extraction system 1 may further include an extraction model learning unit 800 . The extraction model learning unit 800 may perform learning on the semi-structured knowledge extraction model 250, the unstructured knowledge extraction model 350, and the structured knowledge extraction model 450. The extraction model learning unit 800 may include an input module 810 and an output module 820 . The input module 810 may receive data stored in the collected literature repository 130 or the classified literature repository 150, and the output module 820 may perform deep learning using the data input from the input module 810. Using one result, learning is performed on the semi-structured knowledge extraction model 250, the unstructured knowledge extraction model 350, or the structured knowledge extraction model 450, and the pre-learned semi-structured knowledge extraction model 250, the unstructured The knowledge extraction model 350 or the structured knowledge extraction model 450 may be updated.

In the extraction model learning unit 800, the input module 810 receives data stored in the collected literature repository 130 and performs deep learning, and the output module 820 uses the result to generate the semi-structured knowledge extraction model 250 ) can be learned and updated. The semi-structured knowledge extraction model 250 may be learned and updated by deep learning performed by the extraction model learning unit 800 using semi-structured data stored in the collected literature repository 130 .

In the extraction model learning unit 800, the input module 810 receives data stored in the classified literature repository 150 and performs deep learning, and the output module 820 uses the result to generate an unstructured knowledge extraction model 350. You can learn about and update. The unstructured knowledge extraction model 350 may be learned and updated by deep learning performed by the extraction model learning unit 800 using unstructured data stored in the classified literature repository 150 .

In the extraction model learning unit 800, the input module 810 receives the data stored in the classified literature repository 150 and performs deep learning, and the output module 820 uses the result to generate a structured knowledge extraction model 450. You can learn about and update. The structured knowledge extraction model 450 is learned by deep learning performed by the extraction model learning unit 800 using the structured data extracted from the table information of the science and technology paper 40 stored in the classified literature repository 150 and can be updated.

The scientific and technical paper 40 may have different terms, that is, entity names, depending on the technical classification. Therefore, the text content and table information of the science and technology paper 40 may have many differences according to the technical classification. However, bibliographic information and identification numbers, which are semi-structured data of the science and technology thesis 40, may have substantially similar formats regardless of technical classification.

Therefore, the extraction model learning unit 800 of the knowledge extraction system 1 according to the present invention performs learning on the semi-structured knowledge extraction model 250 for extracting knowledge from semi-structured data such as bibliographic information and identification numbers. The input data for performing learning on the unstructured knowledge extraction model 350 and the structured knowledge extraction model 450 for extracting knowledge with unstructured data and structured data such as text content and table information are different. can be selected

That is, the extraction model learning unit 800 occupies a relatively small amount in the science and technology paper 40, and semi-structured data such as bibliographic information and identification numbers that have a substantially similar format regardless of technology classification are irrelevant to technology classification. It is possible to learn the semi-structured knowledge extraction model 250 by receiving input from the collected literature repository 130 where all scientific and technical papers 40 that have been collected are stored, and occupy a relatively large amount in the scientific and technical papers 40 and classify the technology. Unstructured data and structured data, such as text content and table information, which have a lot of difference depending on the scientific and technical papers (40) searched using keywords among the collected scientific and technical papers (40) or scientific and technological papers classified into one technical classification Only (40) may be received from the stored classified literature repository 150 to learn the unstructured knowledge extraction model 350 and the structured knowledge extraction model 450.

Therefore, the extraction model learning unit 800 can perform learning on the semi-structured knowledge extraction model 250, the unstructured knowledge extraction model 350, and the structured knowledge extraction model 450 with both efficiency and accuracy.

The knowledge extraction system 1 for science and technology papers according to the present invention has efficiency and accuracy for science and technology papers 40 searched for specific keywords or science and technology papers 40 classified into specific technical fields, and learns Since knowledge is extracted through knowledge extraction models, it is possible to quickly and accurately find the knowledge of the science and technology paper 40 .

2 is a schematic block diagram illustrating a method for learning a semi-structured knowledge extraction model in a knowledge extraction system for scientific and technical papers according to an exemplary embodiment of the present invention.

Referring to FIG. 2 , the knowledge extraction system 1 may include a document management unit 100, a semi-structured knowledge extraction unit 200, a semi-structured knowledge extraction model 250, and an extraction model learning unit 800. .

In the extraction model learning unit 800, the input module 810 receives data stored in the collected literature repository 130 and performs deep learning, and the output module 820 uses the result to generate the semi-structured knowledge extraction model 250 ) can be learned and updated. The semi-structured knowledge extraction model 250 may be learned and updated by deep learning performed by the extraction model learning unit 800 using semi-structured data stored in the collected literature repository 130, and the semi-structured knowledge extraction unit 200 refers to the semi-structured knowledge extraction model 250 to extract bibliographic information and identification numbers, which are semi-structured data, from data stored in the classified literature repository 150 as knowledge.

3 is a schematic block diagram illustrating a method of learning an unstructured knowledge extraction model and a structured knowledge extraction model in a knowledge extraction system for scientific and technical papers according to an exemplary embodiment of the present invention.

Referring to FIG. 3, the knowledge extraction system 1 includes a document management unit 100, an unstructured knowledge extraction unit 300, an unstructured knowledge extraction model 350, a structured knowledge extraction unit 400, and a structured knowledge extraction model 450. , and an extraction model learning unit 800.

In the extraction model learning unit 800, the input module 810 receives data stored in the classified literature repository 150 and performs deep learning, and the output module 820 uses the result to generate an unstructured knowledge extraction model 350. You can learn about and update. The unstructured knowledge extraction model 350 may be learned and updated by deep learning performed by the extraction model learning unit 800 using unstructured data stored in the classified literature repository 150, and the unstructured knowledge extraction unit 300 may extract unstructured data as knowledge among data stored in the classified literature repository 150 by referring to the unstructured knowledge extraction model 350 .

In addition, in the extraction model learning unit 800, the input module 810 receives data stored in the classified literature storage 150 to perform deep learning, and the output module 820 uses the result to generate a structured knowledge extraction model ( 450) can be performed and updated. The structured knowledge extraction model 450 can be learned and updated by deep learning performed by the extraction model learning unit 800 using unstructured data stored in the classified literature repository 150, and the structured knowledge extraction unit 400 may extract table information, which is structured data, as knowledge from among data stored in the classified literature repository 150 by referring to the unstructured knowledge extraction model 450 .

4 is a schematic block diagram illustrating a method of searching for knowledge by using a knowledge base constructed in a knowledge extraction system for scientific and technical papers according to an exemplary embodiment of the present invention.

Referring to FIG. 4 , the knowledge extraction system 1 functioning as a science and technology question answering system may include a user interface 900, a question answering management unit 700, a natural language understanding unit 750, and a knowledge base 600. can

The user interface 900 may receive a query in a natural language form from the user 10 through the network 20 through the user interface 900 and transmit a response. The user interface 900 may provide an interface for accessing the knowledge extraction system 1, that is, the science and technology question answering system, through a terminal used by the user 10. The user 10 may transmit a query to the knowledge extraction system 1, that is, the scientific and technological question answering system, through the user interface 900, and the knowledge extraction system 1, that is, the scientific and technological query, through the user interface 900. Responses to queries provided by the response system may be received.

The question response management unit 700 interprets a query made by the user 10 through the user interface 900 with reference to the natural language understanding unit 750 and generates a response to the query obtained through the knowledge base 600 in natural language. It can be provided to the user 10 through the user interface 900. The query response management unit 700 may include a query reception module 710 , a knowledge search module 720 , and a response generation module 730 . The query reception module 710 may receive a query made by the user 10 through the user interface 900 and interpret it with reference to the natural language understanding unit 750 . The knowledge search module 720 may send the query interpreted by the query receiving module 710 to the knowledge base 600 and receive a result thereof. The response generation module 730 may generate the result received from the query search module 720 as a response to a query in natural language, and provide the result to the user 10 through the user interface 900 .

5 is a schematic block diagram of a natural language understanding unit included in a knowledge extraction system for scientific and technical papers according to an exemplary embodiment of the present invention.

4 and 5 together, the natural language understanding unit 750 includes a morpheme analysis unit 751, a syntax analysis unit 752, an object name analysis unit 753, a filtering analysis unit 754, an intention classification unit ( 755), a domain analysis unit 756, and a semantic role labeling unit (SRL, 757). The morpheme analyzer 751 may separate a sentence of the query in the form of a natural language of the user 10 into morpheme units. The syntax analysis unit 752 and entity name analysis unit 753 may perform syntax analysis and entity name analysis on sentence entities separated by morpheme units, respectively. The filtering analyzer 754 may generate a simplified sentence by removing unnecessary features from sentence objects. The intention classification unit 755 and the domain analysis unit 756 may perform intention classification and domain analysis of a query to which a semantic role is assigned based on the simplified sentence generated by the filtering analysis unit 754 . The semantic role labeling unit 757 may assign (label) a semantic role to a sentence entity.

6 is a conceptual diagram illustrating a process of extracting semi-structured knowledge in a knowledge extraction system for scientific and technical papers according to an exemplary embodiment of the present invention.

Referring to FIGS. 1 and 6 together, the knowledge extraction system 1 may extract semi-structured knowledge from a document that is a scientific and technical paper 40 . Semi-structured knowledge may be, for example, knowledge extracted from semi-structured data such as bibliographic information and identification numbers. In FIG. 6, DOI and ISSN, which are identification information, are also shown as part of bibliographic information for convenience of illustration.

Specifically, when the scientific and technical papers 40 collected by the literature collection module 110 are composed of PDF or image files, the text extractor 122 of the OCR module 120 extracts semi-structured data that is bibliographic information of the literature. The semi-structured knowledge extraction unit 200 may extract the semi-structured knowledge by referring to the semi-structured knowledge extraction model 250 . Semi-structured knowledge includes bibliographic information such as year of publication (Year), title (Title), brief description (Description), author (Creator.), keywords (Keyword), and publication (Published), digital object identifier (DOI), international standards It may have a property for an identification number such as a serial publication number (ISSN) and a value for each of them.

7 is a conceptual diagram illustrating a process of extracting formal knowledge in a knowledge extraction system for scientific and technical papers according to an exemplary embodiment of the present invention.

Referring to FIGS. 1 and 7 together, the knowledge extraction system 1 may extract formal knowledge from a document that is a science and technology paper 40 . The structured knowledge may be, for example, table information that is structured data.

Specifically, when the scientific and technical papers 40 collected by the literature collection module 110 include tables, the table recognizer 124 of the OCR module 120 may recognize tables of documents and extract table information. In addition, the structured knowledge extraction unit 300 may extract table information as structured knowledge by referring to the structured knowledge extraction model 350 .

One table may contain various types of information, and the header/value classification module 410 of the structured knowledge extraction unit 400 may distinguish between header cells and value cells in the table information, and the semantic analysis module 420 may separate the header cells and value cells from the table information. Semantic analysis can be performed so that the cell can be linked to knowledge, and the value refinement module 430 can refine the value of the value cell so that it can be linked to knowledge.

Through this, the knowledge extraction system 1 can extract, for example, group information, director, symptom information, experiment information, etc. from one table as structured knowledge to which meaning is assigned.

8 is a conceptual diagram illustrating a knowledge base constructed by extracting from a knowledge extraction system for scientific and technical papers according to an exemplary embodiment of the present invention.

Referring to FIGS. 1 and 8 together, the knowledge extraction system 1 includes a semi-structured knowledge extraction unit 200, an unstructured knowledge extraction unit 300, and a structured knowledge extraction unit 400 from the collected scientific and technical papers 40. The semi-structured knowledge, unstructured knowledge, and structured knowledge extracted through ) are connected and integrated in the knowledge integration module 510 of the knowledge management unit 500, and the knowledge conversion module 520 connects the integrated knowledge to the knowledge base. (600). The knowledge conversion module 520 may store knowledge in the knowledge base 600 as a knowledge graph 602 .

The knowledge graph 602 uses, for example, a research paper as a subject, a paper ID and a paper title as an object, and each value is a property It can be expressed as a triple of subject-property-object structure used as . In addition, you can use Research paper as the subject and Experiment as the object, and Experiment Result, Treatment, and Experiment group as the subject. Group) and control group can be used as objects.

In this way, with reference to the knowledge base 600 that stores the knowledge graph 602 constructed for the scientific and technical papers 40 searched for a specific keyword or the scientific and technical papers 40 classified in a specific technical field, the user ( With respect to the query of 10), the query response management unit 700 may generate and provide a response to the user 10 .

In the above, the present invention has been described in detail with preferred embodiments, but the present invention is not limited to the above embodiments, and various modifications and changes are made by those skilled in the art within the technical spirit and scope of the present invention. this is possible

1: knowledge extraction system, 10: user, 20: network, 40: literature, scientific and technical papers, 100: literature management unit, 200: semi-structured knowledge extraction unit, 250: semi-structured knowledge extraction model, 300: unstructured knowledge extraction unit, 350: unstructured knowledge extraction model, 400: structured knowledge extraction unit, 450: structured knowledge extraction model, 500: knowledge management unit, 600: knowledge base, 700: query response management unit, 800: extraction model learning unit, 900: user interface

Claims (10)

A literature collection module for collecting a plurality of science and technology papers through a network;
a collection literature storage for storing the plurality of scientific and technical papers collected in the literature collection module;
a literature classification module extracting data from the plurality of science and technology papers collected in the literature collection module and classifying them according to a technology classification system;
a classified literature storage for storing the plurality of science and technology papers classified according to the technology classification system in the document classification module by classification;
a semi-structured knowledge extraction model for extracting semi-structured data as knowledge among data extracted from the plurality of science and technology papers stored in the classified literature repository with reference to the semi-structured knowledge extraction model;
an unstructured knowledge extraction model for extracting unstructured data as knowledge from among data extracted from the plurality of science and technology papers stored in the classified literature repository with reference to the unstructured knowledge extraction model;
a structured knowledge extraction model for extracting structured data as knowledge from data extracted from the plurality of science and technology papers stored in the classified literature repository with reference to the structured knowledge extraction model;
a knowledge management unit integrating the knowledge extracted from each of the semi-structured knowledge extraction unit, the unstructured knowledge extraction unit, and the structured knowledge extraction unit and storing them in a knowledge base; and
An extraction model learning unit configured to perform learning on each of the semi-structured knowledge extraction model, the unstructured knowledge extraction model, and the structured knowledge extraction model;
The semi-structured knowledge extraction model is learned in the extraction model learning unit using the semi-structured data of all the plurality of scientific and technical papers stored in the collected literature repository,
Each of the unstructured knowledge extraction model and the structured knowledge extraction model corresponds to a predetermined keyword among the plurality of scientific and technical papers stored in the classified document repository or is an unstructured form of only scientific and technological papers of one technical classification classified according to the technical classification system. A knowledge extraction system in which learning is performed in the extraction model learning unit using each of data and structured data. According to claim 1,
An optical character recognition (OCR) module for extracting text data from the plurality of scientific and technical papers when the plurality of scientific and technical papers collected by the document collection module are PDF (Portable Document Format) or image files;
a literature classification model for classifying the plurality of scientific and technical papers by using text information extracted from the plurality of scientific and technical papers; and
The knowledge extraction system further comprising a; classified literature storage for storing the plurality of scientific and technical papers classified in the literature classification model. According to claim 2,
The OCR module,
a text extractor for extracting semi-structured data and unstructured data of the plurality of scientific and technical papers collected; and
A knowledge extraction system comprising a table recognizer for recognizing and extracting table information, which is the structured data of the plurality of scientific and technical papers collected. According to claim 3,
The formal knowledge extraction unit,
The knowledge extraction system, characterized in that for extracting the table information, which is semi-structured data extracted from the table recognizer, as knowledge among data stored in the classified literature repository. According to claim 4,
The formal knowledge extraction unit,
a header/value classification module for distinguishing header cells from value cells in the table information;
a semantic analysis module that performs semantic analysis so that the header cell can be connected to knowledge; and
A knowledge extraction system comprising a value refinement module for refining the value of the value cell so that it can be linked to knowledge. According to claim 2,
The document classification model,
Performing a search for the plurality of scientific and technical papers collected using the predetermined keyword, classifying scientific and technical papers corresponding to the predetermined keyword among the plurality of scientific and technical papers, and storing them in the classified literature repository;
The knowledge extraction system, characterized in that for classifying the plurality of scientific and technical papers collected according to the technical classification, and storing scientific and technical papers classified into a specific technical classification among the plurality of scientific and technical papers in the classified literature repository. According to claim 1,
The semi-structured knowledge extraction unit,
From the data stored in the classified literature repository, bibliographic information and identification numbers, which are semi-structured data, are extracted as knowledge,
The bibliographic information is a title, author, abstract, or keyword of a document, and the identification number is an International Standard Serial Number (ISSN), a digital object identifier (DOI), or an international standard book. Knowledge extraction system, characterized in that the number (International Standard Book Number, ISBN). According to claim 1,
The atypical knowledge extraction unit,
a entity name recognition module recognizing entity name information from unstructured data among data stored in the classified literature repository;
a relationship extraction module for extracting a relationship between entity name information recognized by the entity name recognition module; and
A knowledge extraction system comprising: a sentence classification module for classifying the sentences from which the relationship between entity name information is extracted in the relation extraction module according to a preset sentence classification system. According to claim 1,
The knowledge management department,
a knowledge integration module for connecting and integrating knowledge extracted from each of the semi-structured knowledge extraction unit, the unstructured knowledge extraction unit, and the structured knowledge extraction unit; and
A knowledge extraction system comprising: a knowledge conversion module that stores the knowledge that is connected to each other and integrated into the knowledge base as a knowledge graph. According to claim 3,
The text extractor further extracts link information in which texts and images of captions for the collected image information of the plurality of scientific and technical papers are stored,
The knowledge extraction system, characterized in that the collected literature repository stores images of the image information included in the plurality of scientific and technical papers collected.

Images