KR101061075B1 - How to Create Contour Maps for Research Trend Analysis - Google Patents

Abstract

A contour map generation method is disclosed. The method of generating a contour map according to the present invention includes a first step of extracting keywords using a text mining technique for a plurality of research materials stored in a database, a plurality of research data, and a vector space for the extracted keywords. The second step of generating, the third step of generating a two-dimensional coordinates by analyzing the correlation of a plurality of research data and keywords by using a correlation analysis (correspondence analysis) method for the generated vector space, the two-dimensional coordinates nuclear A fourth step of converting a height map into a contour map of three-dimensional coordinates indicated by height, and a fifth step of detecting peak positions in the contour map using a local maxima detection method. And a sixth step of applying a k-nearest neighbor detection technique to match the keyword to the detected peak position and to display the keyword.

Contour map, two-dimensional coordinates, three-dimensional coordinates, research trends, keywords

Description

How to create contour map for research trend analysis {Method for creating contour map for research trend analysis}

The present invention relates to a method for generating contour maps for research trend analysis, and more particularly, to a method for generating contour maps used for research trend analysis on a plurality of research data.

The characteristic of the knowledge society based on the Internet is the oversupply of information called information flood. However, it is becoming increasingly difficult to efficiently collect and analyze necessary information among a myriad of rapidly changing information.

Recently, the development of science and technology is very fast, and the research field is diverse and changing, so the related information is increasing more rapidly than ever before, and researchers are spending more and more time investing in preliminary research to determine the research field. to be. Therefore, researchers need to analyze research trends in order to prevent duplication of research and to improve efficiency.

Existing researchers analyzed research trends using patent map analysis and paper map analysis. The results of the analysis are used as important indicators in determining the research field.

A patent map classifies bibliographic matters (patents, patent numbers, etc.) of patents and technical contents of patents (claims, detailed description of the invention, drawings, etc.) and tabulates various information with the data. In addition, a paper map is created so that research information can be easily analyzed with respect to paper information similar to a patent map.

As described above, the patent map and the paper map are for foreign patents or SCI (Science Citation Intex) papers. Therefore, it is difficult to analyze research trends using patent maps or thesis maps because of the characteristics of Korean medicine, which are less occupied by foreign patents and SCI papers than other fields. Therefore, there is a need for a suitable method for analyzing the trends of oriental medical research using a large number of research data related to oriental medicine.

SUMMARY OF THE INVENTION The present invention has been made to solve the above problems, and an object of the present invention is to visually provide research trends by analyzing a plurality of research data and generating a contour map of three-dimensional coordinates in which the research data are displayed according to the degree of research activation. It is about how to create a contour map.

A contour map generation method according to an embodiment of the present invention for achieving the above object is the first step of extracting a keyword using a text mining technique for a plurality of research data stored in a database, the plurality of studies A second step of generating a vector space for the data, the extracted keywords, and analyzing the correlations between the plurality of research data and the keywords by using a correlation analysis technique for the generated vector spaces. A third step of generating two-dimensional coordinates; a fourth step of converting the two-dimensional coordinates into a contour map of the three-dimensional coordinates indicated by height using a kernel density estimation technique; local maxima detection The fifth step of detecting the peak position in the contour map using the detection technique and the k-nearest neighbor detection technique are applied. And a sixth step of displaying by matching keywords to group the detected peak position.

In this case, the first step is a first process of extracting the relevant text from the plurality of research materials, using the Korean dictionary for Chinese and Western terms, the English dictionary and the Chinese dictionary to display the related text in English and Chinese characters. A second process of generating a keyword by converting to Korean, a third process of deleting related texts written in English and Hanja that are not converted into Korean, and a fourth process of updating a hash map by analyzing the converted keyword in Korean; Include.

The second step may include a first step of receiving a search keyword, a second step of searching for a study data including the search keyword in the updated hash map and a keyword extracted from the study data, and in the second step A third step of generating the searched keywords in order of high frequency, and generating a vector for the keyword from the top to the nth and the research data; and detecting the frequency of the keywords in the vector and exceeding a preset threshold. And storing the frequency of the keyword in the first matrix, and storing the frequency of the keyword in the second matrix if it does not exceed the preset threshold.

In the third step, the generated first matrix is dimensionally reduced to generate the two-dimensional coordinates for the research data and a keyword exceeding the preset threshold.

On the other hand, the contour map generation method according to the present invention, by analyzing the peak height matched with the keyword, the seventh step of classifying the study data and the keyword into three stages of introduction, growth and maturation It may further include.

The seventh step may include the first step of calculating the height of the highest peak in the contour map of the three-dimensional coordinates by dividing the height into three equal parts, and introducing the height of the third divided contour map in the order of the lowest height. And a third process of classifying the maturation stage into a maturation stage, a third process of classifying and displaying the research data and the keyword matching the peaks according to the introduction stage, the growth stage, and the maturation stage.

On the other hand, the keyword is preferably made of the keywords described in the title of the research data, the abstract of the research data and the index items for each of the plurality of research data.

In addition, the contour map is preferably displayed by changing the color according to the height.

According to the present invention, by analyzing a plurality of research data to generate a contour map of the three-dimensional coordinates marked with the keyword, it is possible to analyze the research trend through the height of the contour of the three-dimensional coordinates. Therefore, by providing the research trend visually using the contour map, it can be usefully used to enable researchers to prevent the duplication of the research field and improve efficiency in determining the research field.

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

1 is a flowchart illustrating a contour map generation method according to an embodiment of the present invention. 2 is a flowchart illustrating a process of extracting a keyword using a text mining technique according to an embodiment of the present invention, and FIG. 3 is a vector space for research data and keywords according to an embodiment of the present invention. 4 is a flowchart illustrating a research trend analysis and display process according to an embodiment of the present invention. FIG. 5 is a diagram illustrating a vector space generated according to the process illustrated in FIG. 3. to be.

Referring to FIG. 1, in the method of generating a contour map according to an exemplary embodiment, keywords are extracted using text mining techniques for a plurality of research data stored in a database (S110). This process will be described in more detail with reference to FIG.

Referring to FIG. 2, first, relevant texts are extracted from a plurality of research data stored in a large database (eg, an oasis database) (S111). In this case, the plurality of research materials refer to all data related to a predetermined research, and specifically, may include a research paper, a research report, a research document, and the like.

In addition, the related text is text related to the contents of each of a plurality of research materials, and may be a noun (including both a single noun and a compound noun) included in the title, abstract, and index of the study material.

Extraction of related text is done in the following way. Specifically, since many originals include PDF files in which the original text itself is an image file, related texts can be extracted from the title, abstract, and index of the research material. First, the title, abstract, and index of the research data can be extracted, and the relevant text can be extracted by removing adjectives, surveys, verbs, etc. and leaving only nouns.

Then, using a dictionary to translate the relevant text written in Chinese characters and English to Korean (S112). Specifically, even though the research data is related to oriental medicine, there are many notations of English and Chinese characters, and a process of translating English and Chinese characters included in the research data into Korean is needed. In this case, a large database may be provided with a dictionary in addition to a plurality of research data. Also, the dictionary may be a 27,836 word Korean terminology dictionary composed of Korean, Chinese, and English, and a 2,144 word term dictionary composed of Korean and English.

In the process of performing step S112, if the related text written in Chinese characters and English is a word (term) not translated into Korean (S113), the related text is deleted (S114). In addition, when the Korean translation is completed, the hash map (Hashmap <Key, Value>) is updated using the translated keyword.

Specifically, the hash map is composed of "Key" and "Value". In the present invention, "Key" may be a keyword, and "Value" represents a frequency (cumulative number) in which the keyword is detected. This process may be performed through S115, S116 and S117.

Next, it is checked whether the keyword translated into Korean has been detected (S115), and if it is detected previously, the value for the corresponding keyword is increased to "1" (S116). For example, when the corresponding keyword is "ginseng", when "ginseng" has been previously detected, "1" may be added to update the hash map to <ginseng, 2>. In addition, if "ginseng" has been previously detected twice, "1" may be added to update the hash map to <ginseng, 3>.

If it is confirmed that the keyword is detected first instead of being detected in the past (S115), the value for the keyword is set to "1" (S117). That is, when the keyword is detected for the first time, the hash map is updated to the <keyword, 1>.

This process of S111 to S117 can be performed for all the research data stored in the database. In addition, when the keyword extraction by the text mining technique is completed through the process of S111 to S117, step S120 shown in FIG.

That is, a plurality of research data and a vector space for extracted keywords are generated (S120). This will be described in detail with reference to FIG. 3.

First, when a search keyword is input (S121), the search data including the search keyword and the keywords extracted from the study data in the updated hash map is searched (S122). Specifically, if a keyword for a field to be searched from a researcher, that is, a search keyword is input, search for all research data including the search keyword in the updated hash map, and search for all keywords extracted from the searched research data. .

Thereafter, the keywords searched in step S121 are listed in order of high frequency, and a vector for the keyword from the top to the nth keyword and the research data is generated (S123). In this case, the research data are listed in rows, and the keywords up to the nth column are listed in columns, indicating the frequency of keywords for each research data. The vector generated in this way is shown in FIG.

Referring to FIG. 5, first, only frequencies up to the nth frequency of the keywords are searched. And, research data are research data 1, research data 2, research data 3,… in row units. , M as the research data, and the extracted keywords are columnar units of keyword 1, keyword 2, keyword 3,... Create a vector by listing them with the keyword n. In addition, the value of keyword 1 in research data 1, that is, the frequency C ₁₁ in which keyword 1 appears in research data 1 is displayed. In this way, it is confirmed how many times the keywords 1 to n are detected in the study data 2 to the study data m to generate a vector as shown in FIG.

Meanwhile, after the step S123, when the vector is generated, if the frequency of each keyword in the vector exceeds a preset threshold (S124), the frequency of the research data and the corresponding keyword is stored in the first matrix (S125), and if not exceeded. In the study data and the frequency of the keyword is stored in the second matrix (S126).

For example, as shown in FIG. 5, study materials are listed in row units, first and second matrices each of corresponding keywords are arranged in column units, and the frequency of keywords extracted from each study data is set at a predetermined threshold. If the value is exceeded, the frequency of the keyword is stored in the research data and the position of the keyword in the first matrix. Otherwise, it is stored in the second matrix. The first matrix in which keywords exceeding a predetermined threshold value are stored may be viewed as a main matrix, the second matrix may be viewed as a sub-matrix, and both the first matrix and the second matrix may be vector spaces.

On the other hand, if a vector space is generated through step S120 (specifically, steps S121 to S126 shown in FIG. 3), a correlation analysis technique is applied to the vector space to search for a plurality of research data and keywords. The correlation is analyzed to generate two-dimensional coordinates (S130).

When a plurality of research data and correlations for each keyword are analyzed and displayed on the coordinates, the dimension is determined according to the number of each keyword. For example, when 50 keywords exist in a vector space, 50-dimensional coordinates may be generated. Therefore, the correlation between study data and keywords and the correlation between research data and research data are applied by applying a correlation analysis technique to a vector space where the number of rows is m (= p) and the number of columns is n (= q). Analyze the relationship, correlation between keywords and keywords, and generate low-dimensional coordinates, that is, two-dimensional coordinates, for research data and keywords.

In generating two-dimensional coordinates, principal coordinate analysis or multi-dimensional scaling can be used. However, these techniques can only analyze correlations between research data and research data. It is impossible to analyze the correlation between keywords. Therefore, for more accurate correlation analysis, it is preferable to use a correlation analysis technique that enables correlation between research data and keywords and correlation analysis between keywords and keywords.

2D coordinate generation using the correlation analysis technique may be performed through the following process.

에 행 백터 a₁, a₂, a₃, …, a_p를 사영하면

이며, a_i의 정사영들의 제곱합이 최대일 때, 차원축소에 의한 손실이 최소가 된다. 따라서, 제1 축을 결정하는 행렬식은 아래의 수학식 1과 같이 나타낼 수 있다. Unit vector from q-1 dimensional simplex S ^q

In row vectors a ₁ , a ₂ , a ₃ ,. , if you project a _p

When the sum of squares of the orthogonal projections of a _i is maximum, the loss due to dimension reduction is minimal. Therefore, the determinant for determining the first axis may be expressed as Equation 1 below.

이고,

이다. 따라서,

행렬 G를

로 정의할 때, 수학식 1은 아래의 수학식 2로 나타날 수 있다. In Equation 1

ego,

to be. therefore,

Matrix G

When defined as, Equation 1 may be represented by Equation 2 below.

의 최대 고유값(eigenvalue)과 일치하며 모든 고유값의 합은 아래의 수학식 3에서와 같이 카이제곱 통계값과 비례한다. Equation 2 is

The sum of the eigenvalues of and the sum of all eigenvalues is proportional to the chi-square statistic as in Equation 3 below.

에 대한 고유값

를 주관성(principal inertia),

를

에 대한 비정칙값(singular value)이라고 한다. 따라서, 고유벡터(eigenvector)

로 결정된 선형 부분 공간(linear subspace)에서 행 a₁, …, a_p는 각각 좌표값

로 표현되며 열 e₁, …, e_q는

로 표현된다. 이 같은 행렬의 값은 0 이 아닌 고유값

와 대응하는 고유벡터

마다 가능하지만 효과적인 차원축소에서는 처음

의 축만 활용한다. 따라서, s=2인 경우에는 아래의 수학식 4에서와 같이 행과 열이 2차원 공간으로 나타낸다. In equation (3)

Eigenvalues for

Subjective (principal inertia),

To

This is called the singular value for. Thus, eigenvectors

Rows a ₁ ,... In a linear subspace determined by. , a _p are the coordinate values

Represented by the columns e ₁ ,. , e _q is

It is expressed as The values of such a matrix are nonzero eigenvalues.

Eigenvectors corresponding to

Every time, but for the first time effective scaling

Use only the axis of. Therefore, when s = 2, rows and columns are represented in a two-dimensional space as shown in Equation 4 below.

행 표본의 표현 :

Representation of the row sample:

Representation of the ten categories:

를 설명하므로, s축 차원 근사도를 고유값의 누적비율인

로 정의할 수 있다. k is the total chi-square variation

Since the s-axis approximation is the cumulative ratio of the eigenvalues

Can be defined as

의 감소 패턴을 고려하여 결정될 수 있다. 평면 공간에 행과 열을 표현하는 것이 보기 편하기 때문에 s=2가 선호될 수 있다. 수학식 4에서 행 표본의 표현을 행의 주좌 표(principal coordinates)라 하고, 열 범주의 표현을 표준화좌표(standard coordinate)라고 한다. The number of dimensions (s) is the eigenvalues as in the principal component analysis

It may be determined in consideration of the reduction pattern of. S = 2 may be preferred because it is easier to represent rows and columns in planar space. In Equation 4, the representation of the row sample is called principal coordinates of the row, and the representation of the column category is called standard coordinate.

On the other hand, when the two-dimensional coordinates are generated through the step S130, the two-dimensional coordinates are converted into a contour map of the three-dimensional coordinates indicated by the height by applying the kernel density estimation (kernel density estimation, parzen window method) (S140).

The two-dimensional coordinates generated in step S130 are contours generated according to the correlation between the research data and the keywords. Specifically, research data and keywords having similar correlations are displayed in a dense manner. By applying the nuclear density estimation technique to such two-dimensional coordinates, the contour height can be displayed to generate a contour map of three-dimensional coordinates.

Nuclear density estimation is a method of estimating the probability density function of a random variable. It is a method of estimating the total population when several pieces of data are given. In the present invention, the population is all research data corresponding to "search keywords", and some data may be research data searched by "search keywords". For example, in a number of research data stored in a database, "ginseng" may be research data searched by a search keyword. Therefore, by applying the probability density function to the research data searched by "ginseng" and displaying it on the contour line, it is possible to generate three-dimensional coordinates that can analyze the research trends for all research data corresponding to "ginseng". Will be.

In other words, for the research data with high probability density function, the contour height is high, and for the research data with low probability density function, the contour height is low. In this case, the height indication in the contour may vary in color at each contour height. In detail, the color may be gradually lightened from the lowest contour to the highest contour. Alternatively, the colors may be displayed differently depending on the height.

를 만족하는 경우, 그

를 국소 최대값이라고 한다. 즉, 주어진 소정의 ε(>0)에 대하여

을 만족하는 모든 x는

를 만족해야한다. 이 경우, 등고선 맵이 3차원 좌표이므로, 국소 최대값을 검출하기 위해 x 좌표뿐만 아니라, y좌표로 검출해야한다. 따라서,

를 만족하는 모든 (x, y) 좌표를 검출해야한다. 이 같은 방식으로 검출된 모든 (x, y) 좌표가 3차원 좌표의 등고선 맵에서의 봉우리 위치가 될 수 있다. When the two-dimensional coordinates are converted into the contour map of the three-dimensional coordinates, the peak position is detected by applying a local maxima detection technique to the contour map of the three-dimensional coordinates (S150). In this case, a point x on the contour map of three-dimensional coordinates

If you meet that,

Is called the local maximum. That is, for a given ε (> 0)

All x satisfy

Should be satisfied. In this case, since the contour map is three-dimensional coordinates, not only the x coordinate but also the y coordinate must be detected in order to detect the local maximum value. therefore,

Should detect all (x, y) coordinates that satisfy All (x, y) coordinates detected in this way can be peak positions in the contour map of three-dimensional coordinates.

On the other hand, by applying a k-nearest neighbor detection (k-nearest neighbor detection) technique to match the keyword to the detected peak position (S160). Specifically, the position of the neighboring keyword at each detected peak position is calculated to determine the similarity according to the distance between the peak position and the calculated keyword position. Then, only the kth keyword up to the peak position can be displayed and displayed in the order of high similarity.

를 산출할 수 있다. Similarity according to distance between peak location and location of calculated research data using Equation 5 below

Can be calculated.

In Equation 5, X and Y are peak positions each composed of k keywords, and x _i and y _i are respectively the i th keyword position.

는 대상 봉우리 위치에서의 가중치로, 대상 봉우리 위치와 키워드들 간의 거리가 될 수 있다. 또한,

는 대상 i번째 키워드 위치(x_i, y_i)에 대한 표준 좌표 변환이다.In Equation 5,

Is a weight at the target peak position, and may be a distance between the target peak position and the keywords. Also,

Is the standard coordinate transformation for the target i th keyword position (x _i , y _i ).

On the other hand, when the contour map of the matched keyword is generated, the peak height matched by the keyword is analyzed, and the keyword is classified into three stages of an introduction stage, a growth stage, and a maturation stage (S170). A detailed description of this step uses FIG. 4.

Referring to FIG. 4, first, the height of each peak is calculated from the contour map of the three-dimensional coordinates (S171). At this time, the peak height divides the distance from the bottom surface (or the lowest peak) to the highest peak height by three in the contour map.

Thereafter, the peak height is divided into three and classified into an introduction stage, a growth stage, and a maturation stage (S172). In this case, the triangular peaks are sorted in order of introduction, growth and maturity in order of lowest height.

Thereafter, it is checked which stage corresponds to the peak where each keyword is matched (S173), and each keyword is classified into an introduction stage, a growth stage, and a maturation stage (S174). For example, the peak matched by the keyword "ginseng" corresponds to the stage of introduction, growth, and maturation. As a result, when "ginseng" is analyzed to correspond to the growth stage, the research data including the keyword "ginseng" is classified. This process can be applied to all keywords matched to the 3D contour map.

As a result, on the screen, what keywords are present in the introduction stage, growth stage, and maturity stage, and the number of research data is displayed. Therefore, it is possible to know the field that is currently being actively researched and the blank technology field.

The research step analysis method provided in the present invention is a medical field currently being studied through the process of steps S110 to S170 using a device such as a computer in a database (for example, an oasis database), and Research phases in the field can be labeled and labeled as introduction, growth and maturity. Therefore, not only the current research stage can be known, but also it can be very helpful in determining the future research direction.

6 is a diagram illustrating a contour map of three-dimensional coordinates according to an embodiment of the present invention, FIG. 7 is a diagram illustrating a contour map of three-dimensional coordinates with matching keywords, and FIG. 8 is a study step according to a keyword. It is a figure which shows the contour map of a three-dimensional coordinate analysis and display three stages of an introduction stage, a growth stage, and a maturation stage.

FIG. 6 is a diagram in which two-dimensional coordinates are converted into contour maps of three-dimensional coordinates by applying a nuclear density estimation technique, and different colors are displayed according to the height of the contour lines. The points displayed on the contour map in FIG. 5 may relate to keywords or research data.

In addition, FIG. 7 is a diagram of a state in which similarity between keywords is determined and matched on a contour map of three-dimensional coordinates, and keywords are matched and displayed at peak positions.

On the other hand, FIG. 8 is divided into three parts based on the highest peak height in the contour map, and is classified into the introduction stage, the growth stage, and the maturity stage from the lowest portion, and displays keywords matching each portion. That is, referring to FIG. 8, keywords are classified and displayed in each of a mature field, a growth field, and a blank field (that is, an introduction field), and the number of research data of each field is displayed on the left side. This labeling allows researchers to identify current research stages in the field of medicine by maturity, growth, and gaps, and to determine future research direction.

The method provided by the present invention is not only applicable to the medical field, but is applicable to all fields requiring analysis of the research stage.

While the above has been shown and described with respect to preferred embodiments of the invention, the invention is not limited to the specific embodiments described above, it is usually in the technical field to which the invention belongs without departing from the spirit of the invention claimed in the claims. Various modifications may be made by those skilled in the art, and these modifications should not be individually understood from the technical spirit or the prospect of the present invention.

1 is a flowchart illustrating a contour map generation method according to an embodiment of the present invention;

2 is a flowchart illustrating a process of extracting a keyword using a text mining technique according to an embodiment of the present invention;

3 is a flowchart illustrating a process of generating a vector space for research data and keywords according to an embodiment of the present invention;

4 is a flowchart illustrating a research trend analysis and display process according to an embodiment of the present invention;

5 is a diagram illustrating a vector space generated according to the process illustrated in FIG. 3;

6 is a view showing a contour map generated according to an embodiment of the present invention;

7 is a view showing a contour map with matching keywords according to an embodiment of the present invention;

8 is a diagram illustrating a contour map of analyzed and displayed research trends according to an embodiment of the present invention.

Claims (8)

Extracting keywords using a text mining technique for a plurality of research materials stored in a database; Generating a vector space for the plurality of research data and the extracted keywords; A third step of generating two-dimensional coordinates by analyzing correlations between the plurality of research data and the keyword by using a correlation analysis technique with respect to the generated vector space; A fourth step of converting the two-dimensional coordinates into a contour map of three-dimensional coordinates of which height is indicated by using a kernel density estimation technique; Detecting a peak position in the contour map using a local maxima detection technique; And and a sixth step of matching and displaying a keyword at the detected peak position by applying a k-nearest neighbor detection technique. The method of claim 1, The first step, Extracting relevant text from the plurality of research materials; A second process of generating a keyword by converting the related text written in English and Chinese characters into Korean using a Korean dictionary, an English dictionary, and a Chinese dictionary for Chinese and Western terms; A third process of deleting related texts written in English and Chinese characters that are not converted into Korean; And And a fourth process of updating a hash map by analyzing the keyword converted into Korean. The method of claim 2, The second step, A first step of receiving a search keyword; A second step of searching for research data including the search keyword and keywords extracted from the research data in the updated hash map; A third process of arranging keywords searched in the second process in order of high frequency, and generating a vector for the keyword from the top to the nth and the research data; And A fourth method of detecting a frequency of the keywords in the vector and storing the frequency of the keyword in the first matrix if the predetermined threshold is exceeded, and storing the frequency of the keyword in the second matrix if the preset threshold is not exceeded. Contour map generation method comprising the process. The method of claim 3, The third step, And dimensionally reducing the generated first and second matrices to generate the two-dimensional coordinates for the research data and the keyword exceeding the preset threshold. The method of claim 1, And a seventh step of analyzing the peak height matched with the keyword and classifying the research data and the keyword into three stages of introduction, growth, and maturation. The method of claim 5, The seventh step, A first step of calculating a height of the highest peak in the contour map of the three-dimensional coordinates and dividing it into three equal parts; A second process of classifying the heights of the third divided contour map into the introduction step, the growth step, and the maturation step in order of lowest height; And a third process of classifying and displaying the research data and the keyword matching the peaks according to the introduction stage, the growth stage, and the maturity stage. The method of claim 1, The keyword is Method for generating a contour map, characterized in that consisting of the keywords of the title of the research data, the abstract of the research data and the index items for each of the plurality of research data. The method of claim 1, And the contour map is displayed by changing colors according to heights.

Images