KR20200053334A - Method and System for the Researcher Map to Promote the Convergence Research - Google Patents

Abstract

The present invention relates to technology about a system which compares and analyzes research data such as papers and intellectual properties to analyze convergence research among researchers who belong to different research fields, and constructs a visualization map about the researchers related to the convergence research. According to the present invention, the system comprises: a data extraction unit extracting a researcher and a research abstract from the research data to construct a database; a morpheme analysis unit extracting a noun and an adjective from the research abstract to generate a word set; a topic modeling unit using a corpus to extract topic words, which become a subject of the research data, from the word set and grouping the related topic words to form a topic group; a user question unit receiving a search sentence from a user; a similarity measurement unit using the search sentence and the topic group to search for a key researcher and comparing the topic group extracted from the research data of the key researcher with the topic group extracted from other research data to search for a related researcher who is determined to perform a similar research to that of the key researcher; and a visualization unit forming and displaying a network in which the key researcher and related researchers searched in the similarity search unit are interconnected.

Description

Method and System for the Researcher Map to Promote the Convergence Research}

The present invention relates to a method and system for constructing a researcher map for facilitating convergence research, and more specifically, analyzing and analyzing convergence research possibilities between researchers with different research fields by comparing and analyzing research data such as papers and intellectual property rights. It is a technology for a system that builds a visualization map for researchers with research relationships.

In modern society, policies and institutions that emphasize groundbreaking changes such as convergence, complex, innovation, and creation are spreading. In terms of science and technology, convergence technology has become a universal phenomenon of modern science and technology innovation. Most products and services that have been successfully developed and commercialized are recognized as products of convergence technology, and all national science and technology innovation actors, such as corporations, research institutes, universities, and governments, are devoted to innovation in convergence technology. Globally, governments in each country are deeply aware of the importance of technology convergence research, and are promoting policies related to convergence research and development and establishment of departments to promote convergence research. In particular, corporations and universities are attracting attention as an innovator of convergence technology, and a lot of policy support is provided.

In order to build a relationship between researchers, the paper 'Analytical Study on the Relationship between Centralities of Research Networks and Research Performances' analyzed the association between centrality and research results in co-author, author citation, and author bibliography combination network. This is a study focused on the researcher's research performance analysis, and it is difficult to confirm whether convergence research can be conducted. In addition, the Stanford Natural Language Processing Group conducted network analysis based on key words in the fields related to corporate education and industrial education research to analyze HRD (Human Resource Development) research trends. Since the and links were generated to perform analysis on a limited field, it is not suitable for real-time convergence researcher network analysis and verification in various fields.

Registered Patent Publication No. 10-1426765

Accordingly, the present invention has been proposed to solve the above-mentioned problems, and the purpose of the present invention is to analyze and analyze research data such as thesis and intellectual property rights to analyze the possibility of convergence research between researchers with different research fields, It is to provide a system for constructing a visualization map for researchers with convergence research relationships.

In addition, another object of the present invention is to focus on the role of universities that can influence the development of convergence technology, and also the relationship between researchers who provide information for companies to conduct convergence industry-academic cooperation with universities. Is to build

In order to achieve the above object, the researcher map construction system for facilitating convergence research according to the technical idea of the present invention includes a data extraction unit that builds a database by extracting researchers and research outlines from research data, and nouns and A morpheme analysis unit that extracts adjectives to generate a word set, and a topic that forms topic groups by grouping related topic words by extracting topic words that are the subject of the research data from the word set using corpus A modeling unit, a user query unit that receives a search statement from a user, a search for a central researcher using the search statement and the topic group, and a topic group extracted from research data of the central researcher and a topic extracted from other research data By comparing the groups, it was determined to conduct similar studies to the above-mentioned central researchers. And similarity measurement unit for searching the researchers is, visible to the center researchers and researchers found in the similarity measurement unit configured to display the researchers networks connected to each other characterized in that it comprises a.

In addition, the data extraction unit extracts characters from the image by performing image processing when the research data is a document composed of images, and if the research data is an imaged paper, the starting coordinates and the abstract are terminated at the point where the abstract starts. Set the end coordinate at the point, replace the x-coordinate value of the end coordinate with the x-coordinate of the rightmost part of the green paragraph, and then extract the text contained within the rectangular range using both the start coordinate and the end coordinate as both end points. It can be characterized as.

In addition, the data extracting unit is characterized in that the space threshold width is pre-set to determine the space between words, and if the width of the space located on the left or right side of the word exceeds the space threshold width, the text line is determined to be terminated. Can be done with

)일 때, 상기

의 영역은

(

,

,

,

), 상기

의 시작 좌표

는 (

,

), 상기

의 종료 좌표

는 (

,

)가 되며,

의 조건을 만족할 때, i-1번째와 i번째 단어 간의 공백 너비

는

로 산출되는 것을 특징으로 할 수 있다.In addition, the data extraction unit is a word (W), the number of extracted words (n), the entire word (WA), the i-th extracted word (

), Above

The realm of

(

,

,

,

), remind

Starting coordinates

Is (

,

), remind

End coordinates of

Is (

,

),

The width of the space between the i-1th and ith words when the condition of

The

It can be characterized in that calculated.

In addition, the morpheme analysis unit reads the sentence included in the study outline in units of char, extracts the integer value of the character, and if the integer value is greater than 0x3131 and less than 0xD7A3, the study outline is Korean It can be characterized as judging that.

Also, if the research outline is Korean, the morphemes are analyzed using KOMORAN to extract common nouns, proper nouns, and Chinese characters words into a set of words, and if the research outline is English, use CoreNLP. The morphemes may be analyzed to extract words that are nouns and adjectives and constitute a set of words.

In addition, the topic modeling unit may be characterized by constructing a corpus using a word set and applying the corpus to a potential Dirichlet Allocation algorithm to generate a topic group.

In addition, the topic modeling unit randomly allocates one of K topic words to each study data, and exists in all study data (d), all words (w), and all words (w) included in each study data (d). For the topic word (t), the ratio p (t | d) of the topic word (t) in the word set (w) of each study data (d) is calculated, and the topic word (t) of all study data is allocated The calculated ratio p (w | t) may be calculated, and the topic word t may be newly selected according to the product of p (t | d) and p (w | t).

In addition, the topic modeling unit may be characterized in that when a researcher related to a search statement is searched, a topic group is extracted from research data including the researcher, and a search word-topic mapping is performed by connecting topic words matching the search statement. .

In addition, the topic modeling unit may be characterized by performing a similarity comparison by combining topic words of a topic group and keywords disclosed in a paper in order to solve a problem in which the meaning of the search statement may be ambiguous.

In addition, the similarity measurement unit searches a related researcher by calculating the similarity between a central researcher's topic group and another researcher's topic group using a jaccard algorithm, a SL (Scaled Levenshtein) algorithm, and a Soft TF / IDF algorithm. It can be characterized by.

회 수행하는 것을 특징으로 할 수 있다.In addition, the similarity measurement unit calculates the similarity of the researcher according to the number (N) of researchers.

It can be characterized by performing it once.

In addition, if the similarity value of the jacquard algorithm, the SL algorithm, and the Soft TF / IDF algorithm are all 1, the similarity measurement unit may determine that the researcher compared with the central researcher is a related researcher who performs a similar research.

In addition, the similarity measurement unit may be characterized in that the other researchers are excluded from the related researchers if the similarity value of the jacquard algorithm, SL algorithm, and Soft TF / IDF algorithm is 0 by comparing the central researcher with other researchers.

In addition, the similarity measurement unit compares the central researcher with other researchers and determines that there is no semantic similarity when the similarity value of two or more of the Jacquard algorithm, SL algorithm, and Soft TF / IDF algorithm is less than 0.5, and the other researcher It can be characterized as to be excluded.

In addition, the researcher network connects a figure in which a researcher's name is described, a figure and another figure, and a connection line whose thickness is determined according to the number of related topic words between the figure and the researcher corresponding to the other figure The visualization unit may include a visualization unit including the display unit, and display the topic word of the researcher on the outer circumferential surface of the figure when the figure is selected, and display the related topic word between researchers when the connection line is selected.

On the other hand, in order to achieve the above object, the method of constructing a researcher map for facilitating convergence research by the technical idea of the present invention includes the steps of the data extraction unit extracting the researcher and the research summary from the research data and constructing a database, and the morpheme analysis unit Generating a word set by extracting nouns and adjectives from the research summary, and a topic modeling unit extracts topic words as the subject of the research data from the word set using a corpus, and groups related topic words Forming a topic group, the user query unit receiving a search statement from the user, the similarity measuring unit searching the central researcher using the search statement and the topic group, and the similarity measuring unit the central researcher Topics extracted from research data and topics extracted from other research data Including comparing groups and searching for related researchers who are determined to perform similar research with the central researcher, and visualizing unit constructing and displaying a researcher network in which the central researcher and the related researchers are connected to each other. It is characterized by.

According to the researcher map construction method and system for facilitating convergence research according to the present invention,

First, the present invention can support the development of convergence technology by performing research similar to each other or by matching researchers using similar technologies to form a researcher network, thereby strengthening cooperation between researchers.

Second, even if the research data is an imaged document, the main research outline is extracted by automatically performing image processing, so analysis is possible without being limited to the format of the research data.

Third, it is possible to identify the spacing between words, line breaks, and the end of paragraphs by using the critical width of space during image processing, so that it is possible to accurately extract the required research outline.

Fourth, by identifying the language of the research data, it is possible to apply an optimal morpheme analysis system for each language, thereby obtaining an optimal word set for each language.

Fifth, by analyzing the word set using the LDA algorithm, it is possible to know the topic word and the frequency of each topic word for each study data, and this information can be used to easily search for research data related to researchers performing similar studies. There will be.

Sixth, when searching for related researchers, various approaches are attempted in judging similarity because the jacquard algorithm, SL algorithm, and Soft TF / IDF algorithm with different performance characteristics are used in combination, thereby allowing researchers and research data with similarities from various viewpoints. You will be able to navigate.

Seventh, the relationship between the central researcher and the related researchers is displayed in the researcher network, so that researchers who intuitively conduct similar research can be found, and if you click (select) the researcher, the topic words extracted from the researcher's research data are displayed. In addition, it becomes possible to easily grasp the research field and applied technology of the relevant researcher.

1 is a block diagram of a researcher map construction system for facilitating convergence research according to an embodiment of the present invention.
2 is an exemplary diagram showing that a rectangular range is set in the abstract area and the keyword area of the paper.
FIG. 3 is an exemplary diagram showing a process of matching a search word and a topic group, a topic group generated using a word set, and a word set extracted from the study outline.
FIG. 4 is a diagram showing an example of displaying information of a researcher searched by a search statement and information of research data displayed when each topic word or topic word of each researcher is selected.
FIG. 5 is an exemplary diagram showing a researcher network displayed when re-searching after selecting a topic of a specific researcher in FIG. 4.
FIG. 6 is an exemplary diagram showing a list of all topic words displayed when a key topic word is displayed around a shape when one of the researchers is clicked (selected) and displayed when a gray area around the shape is clicked.
7 is a flowchart illustrating a process of extracting a topic group from research data in a method of constructing a researcher map for facilitating convergence research according to an embodiment of the present invention.
8 is a flowchart illustrating a process from a search statement input to a researcher network in a method of constructing a researcher map for facilitating convergence research according to an embodiment of the present invention.
Figure 9 is a flow chart showing the detailed process of step S120.
10 is a flowchart showing the detailed process of step S140.
11 is a flow chart showing the detailed process of step S260.

A method and system for constructing a researcher map for facilitating convergence research according to embodiments of the present invention will be described in detail with reference to the accompanying drawings. The present invention can be variously changed and can have various forms, and specific embodiments will be illustrated in the drawings and described in detail in the text. However, this is not intended to limit the present invention to a specific disclosure form, and it should be understood that it includes all modifications, equivalents, and substitutes included in the spirit and scope of the present invention. In describing each drawing, similar reference numerals are used for similar components.

In addition, unless defined otherwise, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by a person skilled in the art to which the present invention pertains. Terms such as those defined in a commonly used dictionary should be interpreted as having meanings consistent with meanings in the context of related technologies, and should not be interpreted as ideal or excessively formal meanings unless explicitly defined in the present application. Does not.

The researcher map building system for facilitating convergence research according to an embodiment of the present invention may be implemented in which all components are included in one computer device, or components are distributed to multiple computer devices and the computer devices are organically connected. .

Referring to FIG. 1, a researcher map construction system for facilitating convergence research according to an embodiment of the present invention includes a data extraction unit that builds a database by extracting researchers and research outlines from research data, and nouns and adjectives in the research overview A morpheme analysis unit that extracts a word set to generate a word set, and a topic modeling that extracts a topic word as a subject of the research data from the word set using a corpus, and groups related topic words to form a topic group Wealth, a user query unit that receives a search statement from a user, a central researcher using the search statement and the topic group, and a topic group extracted from other research data and a topic group extracted from the research data of the central researcher By comparing the relevant researchers who are judged to conduct similar research to the central researcher Connection with the center researchers found in color similarity measurement unit, the similarity measurement unit for researchers characterized in that it includes a visualization display and to configure the network are connected to each other researchers.

In addition, an embodiment of the present invention further includes a researcher database in which the research data, the name of the researcher extracted from the research data, the research summary, and the topic group are configured and stored in one set.

Research materials include papers, research projects, patents, utility models, and copyrights. The research data can be carried out by inputting a file in a known format such as PDF, Ms word, and Hangul Hangul into this embodiment.

The researcher can be the author disclosed in the paper, the inventor disclosed in the patent or utility model.

The research outline includes the title of the paper, the name of the intellectual property right, the name of the research field, the principle used, laws, algorithms, and proper nouns such as phenomena.

The data extraction unit extracts information necessary for analysis from the research data. If the research material is a paper, the area corresponding to abstract and keywords is extracted from the paper.

When the research data is a document composed of images such as an image file or an imaged PDF file, the data extraction unit extracts characters from the image by performing image processing. If the article is an imaged document, set the starting coordinates at the starting point of the abstract and the ending coordinates at the ending point of the abstract, and replace the x-coordinate value of the ending coordinates with the x-coordinate of the rightmost part of the abstract paragraph. The text included in the rectangular range using both the starting coordinates and the ending coordinates as both end points is extracted (see FIG. 2).

The data extraction unit of this embodiment extracted text by analyzing the imaged paper PDF file using a PDFBox produced by Apache (Apache Software Foundation). If the word abstract or abstract is found in the extracted text, it is determined that the page contains the abstract, and a region of the abstract is set to extract the abstract content. Likewise, if a keyword or keywords exist in the extracted text, it is determined as a page containing keywords, and a rectangular area for extracting keyword words is set.

In order to set a rectangular area for abstract contents, the starting and ending coordinates (x2, y2) of the abstract are extracted using the PDFBox library. The starting coordinates of the abstract are the x and y coordinates of the upper left of the word on the page where 'Green' or 'Abstract' appears. The ending coordinates of the abstract are designated by the rightmost x-coordinate and the lower y-coordinate of the abstract content. In general, in the paper in image format, the abstract content is written in a plurality of text lines through line breaks, so the x-coordinate for the rightmost side of the abstract has the largest x-coordinate of the last word in the text line. The value is the x-coordinate of the end coordinate.

라 정의할 때,

의 영역은

(

,

,

,

),

의 시작 좌표

는 (

,

),

의 종료 좌표

는 (

,

)가 된다.

의 조건을 만족할 때, i-1번째와 i번째 단어 간의 공백 너비

는

로 산출된다. 이 값들 중 최대값

는 단어 사이의 공백으로 인식되는 공백 임계 너비로서, 1번째부터 n번째까지의 공백 SX 중 가장 큰 값으로 설정된다. 공백의 정의는 특정 픽셀의 RGB 값이 검은색(0x000000)보다 흰색(0xFFFFFF)에 가까운 경우를 의미한다. 이 공백 임계 너비를 이용하여 각 단어가 종료되는 x 좌표 이후의 공백이 다음 단어와 분리하기 위한 띄어쓰기인지, 줄바꿈을 위한 텍스트 라인의 끝 인지, 혹은 한 페이지의 우측 여백을 나타내는 것인지 판단할 수 있다. 어떤 단어의 우측 공백 너비가

보다 크면, 그 단어의 종료 좌표

(

,

)의

를 초록 내용의 최우측 x 좌표로 설정한다.In addition, the data extraction unit pre-sets a space threshold width to determine a space between words, and when the width of a space located on the left or right side of a word exceeds the space threshold width, it is determined that the corresponding text line is terminated. W is the word, n is the number of words extracted, WA is the whole word, and the i word is extracted.

When we define

The realm of

(

,

,

,

),

Starting coordinates

Is (

,

),

End coordinates of

Is (

,

).

The width of the space between the i-1th and ith words when the condition of

The

Is calculated as Maximum of these values

Is a space threshold width recognized as a space between words, and is set to the largest value of spaces SX from 1st to nth. The definition of blank means that the RGB value of a specific pixel is closer to white (0xFFFFFF) than black (0x000000). Using this space threshold width, it can be determined whether the space after the x coordinate where each word ends is a space to separate from the next word, the end of a line of text for line break, or the right margin of a page. . The width of the space to the right of a word

Greater than, the end coordinate of the word

(

,

)of

Is set to the rightmost x-coordinate of the abstract content.

는

으로 구할 수 있고, 이들 중 임계 높이 공백

를 구하여 하단의 공백들을 순차적으로 분석한다. 하단의 공백이

보다 크면, 초록이 종료되거나 문서 하단의 여백으로 판단하여 해당 단어의 종료 좌표

(

,

)의

를 초록의 최하단 y 좌표로 설정한다. 추출 된 x, y 좌표가 종료좌표로 설정되면, 시작좌표와 종료좌표를 각각 좌상측 꼭짓점과 우하측 꼭짓점으로 하는 사각형영역를 형성한 후 사각형 영역 내에 위치하는 텍스트에 대해 이미지 프로세싱을 실시한다. 추출된 텍스트는 초록으로 최종 결정된다.The method of obtaining the y-coordinate of the lower end of the abstract is similar to the method of obtaining the x-coordinate of the right end. Height with the word on the next line of each word

The

And the critical height of these spaces

To obtain and analyze the blanks at the bottom sequentially. The bottom space

If greater than this, the abstract ends or the end coordinate of the word is judged by the margin at the bottom of the document.

(

,

)of

Is set to the lowest y-coordinate of the green. When the extracted x and y coordinates are set as the end coordinates, a rectangular area is formed with the start and end coordinates as the upper left corner and the lower right corner, respectively, and image processing is performed on text located within the rectangular area. The extracted text is finally determined as abstract.

The keyword of the paper also extracts the start and end coordinates in the same way as the method of setting the green square area, and performs image processing by setting the square area to extract the words disclosed in the keyword.

After the line break, if the word that indicates the beginning of another paragraph, such as Introduction, Intro, or Research scope, appears, the content of the abstract is judged to have ended.

In addition, the data extraction unit removes unnecessary words and characters from the extracted abstract and keyword contents. Unnecessary words or characters include abstracts, abstracts, keywords, keywords, key words, etc., and also removes characters such as commas (',').

The morpheme analysis unit first classifies whether the text of the research outline is Korean or English. The morpheme analysis unit reads the sentences included in the research outline in units of char, and extracts the integer (int, character code) values of the characters. In general, English is used for research materials written in Korean, but Korean is rarely used for research materials written in English, so if the character code (integer) corresponding to each letter is greater than 0x3131 and less than 0xD7A3, research summary and research data Is judged to be Korean. It is judged that research data without a character code greater than 0x3131 and less than 0xD7A3 is in English.

When the language of the research outline is judged, the morpheme analysis unit analyzes the research summary using a corresponding language analysis method. In this embodiment, a well-known morpheme analysis library was used, and words necessary for analysis were collected based on the part-of-speech tagging information output after the morpheme analysis was completed.

Referring to FIG. 3, when the research outline is Korean, this example analyzed morphemes using KOMORAN of Shineware. Commoran performs morphological analysis based on 63 common nouns included in the user dictionary file and 307,435 unique titles analyzing wiki titles. Comoran performs morphological analysis to extract words with tag types of common noun (NNG), proper noun (NNP), and Chinese character (SH). In addition, if there is a hyphen (SS) at the end of any tag, it is determined that the word is separated by line breaks, and the word of the previous tag and the word of the subsequent tag are combined to make one tag.

If the study outline is in English, this example was analyzed for morphemes using CoreNLP (ver. 3.8) from Stanford NLP Group. CoreNLP performs morphological analysis to extract words with tag nouns (NN) and adjectives (JJ).

The common nouns, proper nouns, Chinese characters, nouns, and adjectives extracted from the morpheme analysis unit become the word set of the relevant research data. If the research material is a paper, the word set also includes words of keywords extracted from the data extraction unit.

The topic modeling unit individually clusters words included in the word set to form a topic group of a representative topic. The topic modeling unit uses a corpus, where the corpus is constructed using a set of words. A latent Dirichlet Allocation (LDA) algorithm is applied to the corpus.

The LDA algorithm extracts the subjects present in each document from a set of text documents consisting of a set of natural languages or words. That is, the LDA algorithm is an algorithm that derives a probability topic model through a generative probabilistic model from a set of text documents composed of natural languages, and is a probability model for what topics exist in each document.

Every document has a topic (topic), the documents are related to multiple topics, and the words appearing in the document are regarded as elements for forming the topics. In the LDA algorithm, words used in a document form a topic, and the topics are combined to model the document. In addition, hidden topics are derived using a method of randomizing the frequency of co-occurrences between words in a document. The LDA algorithm considers only the number of occurrences of a word, regardless of the order of words appearing in the document. Based on the distribution of the number of words per topic, the distribution of the number of words appearing in each document is analyzed and predicted what topics the document covers.

In this embodiment, the generated corpus is set as a document for LDA algorithm analysis, and a topic is extracted from the corpus through a topic modeling process. For example, assuming that an initial analysis of a project called JHotDraw produces a corpus consisting of the words 'Mycobacterium, tuberculosis-induced, granulocyte-macrophage, colony, stimulating, factor, ...', this corpus is an LDA algorithm. When topic modeling is performed using, words such as declining, macrophage, and cells are extracted together with variance values of 0.154, 0.74, and 0.065. The variance value can be regarded as the importance of each word in the whole corpus, and it is set as a list of main functions or features of the project by using words with values above a certain value.

In addition, the LDA algorithm is based on Gibbs sampling. Gibbs Sampling randomly assigns one of K topic words to each study. As a result, each document has a topic word and a distribution of the topic words. Since the distribution value of the topic word is an error value, an additional process is performed for improvement. Two calculations are performed on each study data (d), the word set (w) included in each study data (d), and the topic word (t) in the word set (w). First, the ratio p (t | d) of the topic word (t) among the word set (w) of the study data (d) is calculated. Second, the ratio p (w | t) to which the topic word (t) is allocated among all research data is calculated. Thereafter, the topic word t is newly selected according to the product of p (t | d) and p (w | t). According to this generative model, this can be regarded as the probability that the topic word (t) will generate a word set (w), so the topic word of each research data is set again according to the probability. That is, in this step, it is assumed that all the topic words are properly allocated in addition to the currently measured word, and the current word is updated by calculating the probability. When a stable state is achieved by repeating such a series of processes sufficiently, it is possible to check a topic word and its distribution in the document.

Topic modeling based on the LDA algorithm assumes that each study data contains one or more of k topic words. The topic group, the result of modeling, is a collection of arbitrary topic words. For example, assuming that the first group was extracted from the newspaper, the words Kim Gyeong-ho, violin, drum, and concert were extracted, and the second group extracted the words Dokdo, North Korea, summit, and nuclear. The first group becomes a topic group related to 'music', and the second group becomes a topic group related to 'politics'. That is, the topic modeling unit clusters words based on the distribution value to form a set of topic words representing a specific topic to form a topic group.

The Topic Modeling Department uses the MALLET Topic Modeling Toolkit library to perform topic modeling.

The topic modeling unit extracts a topic group from the research data included in the researcher when the researcher related to the search statement is searched, and performs search-to-topic mapping by connecting topic words matching the search statement.

In order to solve the problem that the meaning of the search statement composed of single or complex words may be ambiguous, similarity comparison is performed by combining topic words of the topic group and keywords disclosed in the article.

If there are words in the search statement among the words in the topic group, the topic group is connected to the words in the search statement. For example, if the search statement is 'GM-CSF, MEK1, Mycobacterium tuberculosis, P38 MAPK, PI3-K', the first topic group can be mapped to protein, k / mek, suggest, kinase, and induction. The second topic group can be mapped to kinase, mapk, treated, inmma, increase. The third topic group can be mapped to gm-csf, infection, factor, mediated, up-refulation. The fourth topic group can be mapped to bymtb, mma, mapk-associatedsignaling, mek, and thp. In the second topic group, mapk is mapped to P38 MAPK in the search statement, and in the third topic group, gm-csf is mapped to GM-CSF in the search statement. The search statement-topic map is completed through this mapping process (see FIG. 3).

The user query unit receives the search statement written by the user. The search statement can consist of a single word or a compound word. The input search statement is optimized as a word that can be used for search by performing part-of-speech tagging in the morpheme analysis unit.

The user query unit provides a configuration for inputting a search statement including a researcher's name or words on the monitor device. In this embodiment, a researcher input box for inputting a researcher's name and a keyword input box for inputting words are provided separately (see FIG. 4). When a search statement is input to the user query section, the visualization section configures and displays the researcher network on the monitor.

If a part of the name is entered in the researcher input box, the name of the researcher with the corresponding letter is displayed in the autocomplete list. The autocomplete list is displayed in the form of a layer just below the input box. If you select one of the researchers' names displayed on the autocomplete list, the researcher's school, department, e-mail, etc. may be additionally displayed.

In the keyword input box, a word to be searched can be entered.

In addition, the user query unit can select the research data to be searched among thesis, assignments, and intellectual property rights.

The search for related researchers in the similarity measurement unit is performed by calculating the similarity between the topic group of the central researcher and the topic group of other researchers using a jaccard algorithm, a scaled Levenshtein (SL) algorithm, and a Soft TF / IDF algorithm.

When the similarity values of the jacquard algorithm, SL algorithm, and Soft TF / IDF algorithm are all 1, it is determined that the researcher compared with the central researcher is a related researcher who performs similar research.

The similarity measuring unit performs two functions. First, search for the central researcher and words based on the search text. Second, the similarity of research between researchers is measured based on the topic group extracted from each researcher's research data.

The search for a central researcher and a word may be performed by performing a like search (a string search) in a database using a search statement in which morpheme analysis has been performed. The search results are generated as a list of researchers and a list of words, and the list is delivered to the visualization unit.

The similarity between researchers is measured based on the topic group created after the topic modeling is performed in the topic modeling unit. Jacquard algorithm, SL algorithm, and Soft TF / IDF algorithm are used to measure similarity.

Jacquard is a way to check the similarity between sets. Jacquard similarity J (A, B) is defined as the value of the intersection of two sets divided by the size of the two sets, and the relationship can be expressed by Equation 1.

[Equation 1]

는 수학식2와 같다.The SL algorithm is a correction of the result of the Levenshtein algorithm known as the edit distance algorithm. The SL algorithm calculates how many operations a and b need to perform on strings a and b to be the same. Here, the operations represent insert, delete, and replace. For two strings a, b, where | a | and | b | represent the lengths of the strings a, b, respectively.

Is the same as Equation 2.

[Equation 2]

의 값이 d이고, 문자열 a, b의 길이 중 큰 값을 n으로 둘 경우, 수학식3과 같다.The result of Equation 2 is output as an integer. Since the fluctuation range of the value is large according to the length of the word and the calculation amount, the SL algorithm corrects the result value to generate a value between 0 and 1. The correction formula

If the value of d is, and the larger value of the lengths of the strings a and b is n, it is as shown in Equation 3.

[Equation 3]

Soft TF / IDF algorithm is a weighting method considering partial match to TF / IDF. TF / IDF statistically indicates the importance of a particular word in a particular document when there is a document set consisting of multiple documents. TF represents the term frequency of words appearing in the document, and the higher the value, the more important words are considered. If the word is often used in other documents, it means that it is often used. This is called document frequency (DF). The inverse of this value is called the inverse document frequency (IDF), and the result of the TF / IDF algorithm is the product of TF and IDF. However, the TF / IDF algorithm does not consider typos, so even if the tokens of words are slightly different, they are recognized as different words. To correct this, the Soft TF / IDF algorithm performs both character- and token-based similarity measurements. To measure the token similarity of the Soft TF / IDF algorithm, string comparison algorithms such as Jaro and JaroWinkler and threshold values can be specified as input.

회 수행된다. 이때, 자카드 알고리즘 유사도는

, SL 알고리즘의 유사도는

, Soft TF/IDF 알고리즘의 유사도는

이다.To calculate the similarity, the topic groups of N researchers received from the topic modeling unit are sequentially compared. Comparison between researchers is total

Is performed once. At this time, the jacquard algorithm similarity

, The similarity of the SL algorithm is

, The similarity of Soft TF / IDF algorithm is

to be.

The first researcher (center researcher) and the second researcher are compared, and if any of the similarity results of the three algorithms shows 0, the second researcher is excluded from the related researcher. If all comparison values are 0, words in the topic group are compared. If 0 appears in this value, the first researcher and the second researcher judge that there is no similarity, and the second researcher is excluded from the related researcher.

+

+

= 3, 즉 세 알고리즘의 유사도 값이 모두 1이 나오면 제1연구자와 제2연구자의 유사도가 매우 높은 것으로 판단한다. 또한, 토픽그룹의 비교 후 일정 값 이상의 수치가 나오면 제2연구자를 관련연구자로 결정한다. 제2연구자가 관련연구자로 결정된 후 토픽그룹의 단어를 비교하여 유사 강도를 연산한다.When comparing similarity

+

+

= 3, that is, if all three algorithms have a similarity value of 1, it is determined that the similarity between the first and second researchers is very high. In addition, after comparing the topic group, if a value above a certain value appears, the second researcher is decided as a related researcher. After the second researcher is decided as a related researcher, the similar strength is calculated by comparing the words of the topic group.

For the other values, the similarity values of the three algorithms are compared and judged.

If the similarity value of two or more of the three algorithms is less than 0.5, the second researcher is excluded from the candidate for the related researcher. When the similarity value is less than 0.5, it is a word that is hard to be considered to have semantic similarity. As a result of the experiment, the similarity value of less than 0.5 was confirmed as the value output when only the same character was present.

On the other hand, if the similarity value of two or more of the three algorithms is 0.5 or more, the second researcher determines that there is a high probability of becoming a related researcher, and finally determines whether the researcher is related by comparing words of the topic group. When determined as a related researcher, similar strengths are calculated by comparing words in the topic group.

The related researcher's information is stored in the researcher database.

4 to 6, the user query unit and the visualization unit of this embodiment are implemented as web screens based on JSP and JQuery.

When a user inputs a search statement including a researcher's name or word in a function corresponding to a user query unit displayed on the monitor device, the visualization unit configures the researcher network to be displayed on the monitor.

For example, if a search is performed after a researcher's name is entered in the user query unit, the visualization unit causes major topics extracted from the researcher's thesis, assignment, and intellectual property rights to be output in order of frequency or constellation. When the word search is performed, the visualization unit displays the results of the researcher of the research data including the word and the related research data of the researcher. In the result list, an icon for checking the researcher network, related topics, and researcher information is also displayed.

When the network icon displayed with a specific topic is clicked in the search result list, the visualization unit displays a network screen connecting a researcher related to a specific researcher based on the topic. The network consists of similar intensity values calculated based on researchers' topics.

At the top of the screen, the researcher's name and representative topics that make up the network are displayed. In the upper right of the screen, there is a button to filter the area where the keyword is extracted by thesis, assignment, and intellectual property right, and you can limit and set the area of the output of the keyword according to selection and cancellation.

In the researcher network, shapes such as circles and polygons each represent the researcher. In the figure, the researcher's name is given. If you click (select) a shape, the topic word of the researcher is displayed on the outer peripheral surface of the shape. Due to the limitation of the displayed space, only a part of the topic words of the researcher is displayed according to the high frequency, and clicking the gray area around the figure displays additional topic words. To the right of the topic word, the frequency that appeared in the research data is displayed, and when the topic word is clicked, the title, keyword, author, and related topic list of the research data including the topic word are displayed.

The connecting line that connects researchers to form a network is determined in thickness according to the number of related topic words (similar strength) between researchers corresponding to one figure and another figure. The number of topic words related to the researcher is displayed in the connection line, and when the connection line is clicked (selected), a list of related topic words is displayed.

In the window on the left side of the screen, the entire topics of the researcher selected (clicked) by the user are displayed in order of frequency. If you enter a word in the text box at the bottom of the window, select the search button at the bottom of the text box, or check the checkboxes of some of the topics displayed in the window and select the search button, it will be based on the entered word or selected topic. A new researcher network is formed.

At the bottom of the screen, there is a department filter that can group and output the researchers' departments. If you click each department, the background of the same department is displayed in the same color.

Next, a method of constructing a researcher map for facilitating fusion research according to an embodiment of the present invention will be described.

7 and 8, a method for constructing a researcher map for facilitating convergence research according to an embodiment of the present invention includes a step of extracting a researcher and a research summary from a research material and constructing a database (S120). The morpheme analysis unit extracts nouns and adjectives from the research summary to generate a word set (S140), and the topic modeling unit extracts topic words that are the subject of the research data from the word set using a corpus, Grouping related topic words to form a topic group (S160), a user query unit receiving a search statement from a user (S220), and a similarity measurement unit searching for a central researcher using the search statement and the topic group Step S240, and the similarity measurement unit is extracted from the topic group and other research data extracted from the research data of the central researcher Searching for related researchers who are determined to perform similar research with the central researcher by comparing the pick group (S260), and visualizing unit constructing and displaying a researcher network in which the central researcher and the related researchers are connected to each other. (S280).

Referring to Figure 9, step S120 is specifically, the step of classifying what kind of document the research material is thesis, research project, intellectual property, etc. (S121), and identifying whether the classified document is an imaged document (S122) ).

If the article is imaged, a step of searching for words corresponding to 'green' and 'keyword' by performing image processing on the document (S126), and a rectangular range in the area where 'green' and 'keyword' are disclosed After setting, it includes extracting the text within the rectangular range (S127).

If the paper is in a state where text can be extracted, texts of 'green' and 'keyword' are extracted (S128).

The step S129 further includes removing unnecessary words or characters from the text extracted in step S127 or S128, that is, characters such as abstract, abstract, keyword, keywords, key words, and comma (',').

If the research data is not a thesis in step S121, a research summary is extracted from the research data in a corresponding way (S123).

Referring to Figure 10, step S140 is specifically, a step of classifying whether the study outline is Korean or English (S145) and, if the study outline is Korean, using a comoran, a general noun, a proper noun, and a Chinese character in the research outline Extracting, constructing the extracted word into a word set (S146), and if the research outline is English, extracting nouns and adjectives from the research overview using CoreNLP, and constructing the extracted word into a word set (S147).

Referring to FIG. 11, step S260 is specifically, calculating a similarity between a topic group of a central researcher and a topic group of another researcher using a jacquard algorithm, an SL algorithm, and a Soft TF / IDF algorithm (S264) and three algorithms. It includes the step of classifying the similarity value according to the preset classification (S265) and the step of selecting another researcher as a related researcher or excluding the related researcher according to the preset classification (S266).

In the step S266 of selecting another researcher as a related researcher or excluding the related researcher according to a preset classification, if the similarity value of the jacquard algorithm, the SL algorithm, and the Soft TF / IDF algorithm is 1, the other researcher is selected as a related researcher candidate. Step (S266a), if the similarity value of two or more of the jacquard algorithm, SL algorithm, and Soft TF / IDF algorithm is 0.5 or higher, selecting another researcher as a candidate for a related researcher (S266b), jacquard algorithm, SL algorithm, Soft TF / IDF If all algorithms have a similarity value of 0, excluding other researchers from related researchers (S266c), excluding Jacques algorithm, SL algorithm, and Soft TF / IDF algorithm if the similarity value is less than 0.5, excluding other researchers from related researchers. (S266d).

In S266a and S266b, another researcher selected as a candidate for the related researcher is selected as a related researcher when the similarity strength is higher than a predetermined criterion by comparing the central researcher and the topic words of the topic group (S268).

The preferred embodiments of the present invention have been described above, but the present invention can use various changes, modifications, and equivalents. It is clear that the present invention can be equally applied by appropriately modifying the above embodiments. Therefore, the above description is not intended to limit the scope of the present invention as defined by the following claims.

110: data extraction unit
120: morpheme analysis unit
130: topic modeling department
140: similarity measuring unit
150: user query unit
160: visualization unit
180: Researcher database

Claims (20)

A data extraction unit that builds a database by extracting researchers and research outlines from research data,
A morpheme analysis unit for generating a word set by extracting nouns and adjectives from the above study summary,
A topic modeling unit for extracting topic words that are the subject of the research data from the set of words using a corpus, and grouping related topic words to form a topic group;
A user query unit that receives a search statement from a user,
It is judged that a central researcher is searched using the search statement and the topic group, and a similar study to the central researcher is performed by comparing the topic group extracted from the researcher's research data and the topic group extracted from other research data. Similarity measurement unit to search for relevant researchers,
A researcher map construction system for facilitating convergence research, comprising a visualization unit configured to display a researcher network connected to a central researcher and related researchers searched in the similarity measurement unit. According to claim 1,
The data extraction unit extracts characters from an image by performing image processing when the research data is a document composed of images,
If the research data is an imaged paper, the starting coordinates are set at the point where the abstract starts, and the ending coordinates are set at the point where the abstract ends, and the x-coordinate value of the ending coordinates is replaced with the x-coordinate of the rightmost part of the abstract paragraph. A system for constructing a researcher map for facilitating convergence research, characterized by extracting text contained within a rectangular range using the start coordinate and the end coordinate as both end points. According to claim 2,
The data extracting unit determines that a space threshold width is pre-set to determine space between words, and when the width of a space located to the left or right of a word exceeds the space threshold width, the corresponding text line is determined to be terminated. Researcher map building system to promote convergence research. According to claim 3,
The data extraction unit is a word (W), the number of extracted words (n), the entire word (WA), the i-th extracted word (

)when,
remind

The realm of

(

,

,

,

), remind

Starting coordinates

Is (

,

), remind

End coordinates of

Is (

,

),

The width of the space between the i-1th and ith words when the condition of

The

Researcher map construction system for facilitating convergence research, characterized in that calculated by. According to claim 1,
The morpheme analysis unit reads the sentence included in the study outline in units of char, extracts the integer value of the character, and if the integer value is greater than 0x3131 and less than 0xD7A3, the study summary is determined to be Korean Researcher map construction system for facilitating convergence research, characterized in that. The method of claim 5,
If the research summary is Korean, the morpheme analysis unit analyzes the morphemes using KOMORAN, extracts words of common nouns, proper nouns, and Chinese characters and composes them into word sets.
If the research outline is English, a researcher map construction system for facilitating convergence research is characterized by analyzing morphemes using CoreNLP and extracting words that are nouns and adjectives into a set of words. According to claim 1,
The topic modeling unit constructs a corpus using a word set, and applies the corpus to a potential Dirichlet Allocation algorithm to generate a topic group, thereby creating a researcher map building system for facilitating convergence research. . The method of claim 7,
The topic modeling unit randomly allocates one of K topic words to each research data,
For each research data (d), all words (w) included in each research material (d), and topic words (t) existing in all words (w),
Calculate the ratio p (t | d) of the topic word (t) among the word set (w) of each study data (d),
Calculate the ratio p (w | t) assigned to the topic word (t) among all research data,
A researcher map building system for facilitating convergence research, characterized in that the topic word (t) is newly selected according to the product of p (t | d) and p (w | t). According to claim 1,
The topic modeling unit promotes convergence research, characterized in that when a researcher related to a search statement is searched, a topic group is extracted from the research data including the researcher, and a search word-topic mapping is performed by connecting topic words matching the search statement. For researcher map building system. The method of claim 9,
The topic modeling unit is a researcher map construction system for facilitating convergence research, characterized by performing a similarity comparison by combining topic words of a topic group and keywords disclosed in a paper in order to solve a problem in which the meaning of a search statement may be ambiguous. According to claim 1,
The similarity measurement unit searches a related researcher by calculating a similarity between a topic group of a central researcher and a topic group of another researcher using a jaccard algorithm, a scaled Levenshtein (SL) algorithm, and a Soft TF / IDF algorithm. Researcher map building system for facilitating convergence research. The method of claim 11,
The similarity measurement unit calculates the similarity of the researcher according to the number (N) of researchers.

Researcher map building system for facilitating convergence research, characterized in that it is performed once. The method of claim 11,
The similarity measurement unit is a researcher for facilitating convergence research, characterized in that if the similarity values of the jacquard algorithm, SL algorithm, and Soft TF / IDF algorithm are all 1, the researcher compared with the central researcher is determined to be a related researcher who performs similar research. Map building system. The method of claim 11,
The similarity measurement unit compares the central researcher with other researchers, and if any of the similarity values of the jacquard algorithm, the SL algorithm, and the Soft TF / IDF algorithm is 0, the other researcher is excluded from the related researcher to promote convergence research. Map building system. The method of claim 11,
The similarity measurement unit compares the central researcher with other researchers and determines that there is no semantic similarity when two or more similarity values of the Jacquard algorithm, SL algorithm, and Soft TF / IDF algorithm are less than 0.5, and excludes the other researchers from the related researchers. Researcher map construction system for facilitating convergence research, characterized in that. According to claim 1,
The researcher network includes a figure in which a researcher's name is described, a figure connecting a figure and another figure, and a connecting line in which the thickness is determined in correspondence to the number of related topic words between the figure and the researcher corresponding to the figure. It includes a visualization unit,
The visualization unit displays a researcher's topic word on the outer circumferential surface of the figure when the figure is selected, and displays a related topic word between researchers when the connection line is selected. A data extraction unit that builds a database by extracting researchers and research outlines from research data,
A morpheme analysis unit for generating a word set by extracting nouns and adjectives from the above study summary,
A topic modeling unit for extracting topic words that are the subject of the research data from the set of words using a corpus, and grouping related topic words to form a topic group;
A user query unit that receives a search statement from a user,
It is judged that a central researcher is searched using the search statement and the topic group, and a similar study to the central researcher is performed by comparing the topic group extracted from the researcher's research data and the topic group extracted from other research data. Similarity measurement unit to search for relevant researchers,
And a visualization unit configured to display a researcher network in which the central researcher and related researchers searched in the similarity measurement unit are connected to each other, and
The data extraction unit extracts characters from the image by performing image processing when the research data is a document composed of images, and if the research data is an imaged paper, the starting coordinates at the starting point of the abstract and the ending point of the abstract. Set the end coordinate, and the x-coordinate value of the end coordinate is replaced with the x-coordinate of the rightmost side of the green paragraph, and then the text included in the rectangular range using the start coordinate and the end coordinate as both end points is extracted.
The data extracting unit determines that a space threshold width is pre-set to determine space between words, and when the width of a space located on the left or right side of a word exceeds the space threshold width, it is determined that the corresponding text line is terminated.
The data extraction unit is a word (W), the number of extracted words (n), the entire word (WA), the i-th extracted word (

), Above

The realm of

(

,

,

,

), remind

Starting coordinates

Is (

,

), remind

End coordinates of

Is (

,

),

The width of the space between the i-1th and ith words when the condition of

The

Is calculated as
The morpheme analysis unit reads the sentence included in the study outline in units of char, extracts the integer value of the character, and if the integer value is greater than 0x3131 and less than 0xD7A3, the study summary is determined to be Korean And
The morpheme analysis unit analyzes morphemes using KOMORAN if the research outline is Korean, extracts words of common nouns, proper nouns, and Chinese characters into a set of words, and if the study outline is English, uses morphemes using CoreNLP. It analyzes and extracts words that are nouns and adjectives and composes them into word sets.
The topic modeling unit constructs a corpus using a word set, and applies the corpus to a latent Dirichlet Allocation algorithm to generate a topic group,
The topic modeling unit randomly allocates one of the K topic words to each study data, and the topics present in the whole word (w) and the whole word (w) included in each study data (d) and each study data (d). For word (t), the ratio p (t | d) of the topic word (t) in the word set (w) of each study data (d) is calculated, and the ratio of the topic word (t) allocated in all study data p (w | t) is calculated, and the topic word (t) is newly selected according to the product of p (t | d) and p (w | t),
When the researcher related to the search statement is searched, the topic modeling unit extracts a topic group from the research data including the researcher, and connects search terms to topic words by connecting the topic words matching the search statement,
The topic modeling unit performs a similarity comparison by combining topic words of a topic group and keywords disclosed in a paper, in order to solve a problem in which the meaning of the search statement may be ambiguous,
The similarity measurement unit searches a related researcher by calculating a similarity between a central researcher's topic group and another researcher's topic group using a jaccard algorithm, a SL (Scaled Levenshtein) algorithm, and a Soft TF / IDF algorithm.
The similarity measurement unit calculates the similarity of the researcher according to the number (N) of researchers.

Perform once,
The similarity measurement unit determines that if the similarity values of the jacquard algorithm, SL algorithm, and Soft TF / IDF algorithm are all 1, the researcher compared with the central researcher is determined to be a related researcher who performs similar research.
The similarity measurement unit compares the central researcher with other researchers, and if any of the similarity values of the jacquard algorithm, SL algorithm, and Soft TF / IDF algorithm is 0, the other researchers are excluded from the related researchers,
The similarity measurement unit compares the central researcher with other researchers and determines that there is no semantic similarity when two or more similarity values of the Jacquard algorithm, SL algorithm, and Soft TF / IDF algorithm are less than 0.5, and excludes the other researchers from related researchers. ,
The researcher network includes a figure in which a researcher's name is described, a figure connecting a figure and another figure, and a connecting line in which the thickness is determined in correspondence to the number of related topic words between the figure and the researcher corresponding to the figure. It includes a visualization unit,
The visualization unit displays a researcher's topic word on the outer circumferential surface of the figure when the figure is selected, and displays a related topic word between researchers when the connection line is selected. Includes a visualization unit including a figure in which a researcher's name is described, a connecting line that connects one figure to another figure, and determines the thickness corresponding to the number of related topic words between the figure and the researcher corresponding to the other figure and,
The visualization unit displays a researcher's topic word on the outer circumferential surface of the figure when the figure is selected, and displays a related topic word between researchers when the connection line is selected. The data extraction unit extracts the researcher and the research summary from the research data and builds a database.
The morpheme analysis unit extracts nouns and adjectives from the research outline to generate a word set,
A topic modeling unit extracting a topic word as a subject of the research data from the word set using a corpus, and forming a topic group by grouping related topic words;
The user query unit receives a search statement from the user,
A similarity measurement unit searching for a central researcher using the search statement and the topic group;
Searching for related researchers who are determined to perform similar research with the central researcher by comparing the similarity measurement unit by extracting the topic group extracted from the central researcher's research data and other research data;
Method of building a researcher map for facilitating convergence research, characterized in that the visualization unit comprises the step of constructing and displaying a researcher network in which the central researcher and the related researchers are connected to each other. The user query unit receives a search statement from the user,
The visualization unit connects a figure showing the name of the researcher searched for, and a figure and another figure, and monitors a connection line whose thickness is determined according to the number of related topic words between the figure and the researcher corresponding to the other figure A method of constructing a researcher map for facilitating convergence research, comprising the step of displaying in a.

Images