KR20180099164A - Method and apparatus for analyzing technical document - Google Patents

Abstract

The present invention relates to a technical document analysis method and an apparatus therefor. The technical document analysis method comprises the following steps: a technical document collection unit collects a plurality of technical documents which is an analysis target from a technical document database; a keyword extraction unit extracts a plurality of keywords from the plurality of technical documents based on the number of appearances of at least one word included in each of the plurality of technical documents; and a technical document analysis unit analyzes the plurality of technical documents by inputting a plurality of keywords in at least one technical document analysis model previously generated based on machine learning.

Description

[0001] METHOD AND APPARATUS FOR ANALYZING TECHNICAL DOCUMENT [0002]

The present invention relates to a method and apparatus for analyzing a technical document based on machine learning.

Big data is spreading across various fields including social phenomena and various industrial fields. Big data is distinguished from general data in three aspects of capacity, diversity and speed.

Here, the capacity means a large size in the case of the big data, and diversity means that the big data includes various data such as numbers, letters, pictures, and moving pictures. In the case of the big data, .

At this time, the technical documents follow the characteristics of the big data described above, and the patent document which is an example of the technical document includes various descriptions about the invention including the title, abstract, inventor, filing date, Other examples of papers include various descriptions of the technology, including title, abstract, author, and open method.

As such, technical documents are characterized by big data, so technical documents have a large capacity, the data volume is rapidly increasing, various data are included, and analysis of these technical documents is costly and time consuming have.

Korean Registered Patent No. 10-0676693 (Jan. 25, 2007)

SUMMARY OF THE INVENTION An object of the present invention is to solve the problems described above, and it is an object of the present invention to extract a plurality of keywords from a plurality of technical documents based on the occurrence count of each of at least one word contained in each of a plurality of collected technical documents, And analyzing a plurality of technical documents by inputting a plurality of keywords into at least one technical document analysis model generated in advance based on the machine learning.

The problems to be solved by the present invention are not limited to the above-mentioned problem (s), and another problem (s) not mentioned can be clearly understood by those skilled in the art from the following description.

According to an aspect of the present invention, there is provided a method for analyzing a technical document, the method comprising: collecting a plurality of technical documents to be analyzed from a technical document database; Extracting a plurality of keywords from a plurality of descriptive documents based on the number of occurrences of each of at least one word contained in each of the plurality of descriptive documents, and analyzing at least one technical document analysis And analyzing the plurality of technical documents by inputting a plurality of keywords into the model.

For example, the step of extracting a plurality of keywords may include converting all text information into corpus information by removing a stopword from the entire text information included in each of the plurality of technical documents, Generating a word matrix for each descriptive document which is a matrix representing the number of occurrences of each of at least one word included in the information, and generating a word matrix for each descriptive word by using a predetermined number of occurrences of at least one word And selecting a word to extract a plurality of keywords.

According to one embodiment, the step of analyzing the plurality of technical documents may include the steps of: inputting a plurality of keywords into a technical document classification model generated in advance based on machine learning, and extracting a plurality of technical documents by bibliographic information .

For example, the step of classifying a plurality of technical documents by bibliographic information may include classifying each of a plurality of keywords included in each of the plurality of technical documents into a plurality of keywords, for each of a predetermined number of appearance frequency classes, Determining an appearance frequency rank of each of a plurality of keywords based on a degree of belonging indicating a degree of belonging to each of a plurality of keywords and a plurality of keywords, And classifying the plurality of technical documents by bibliographic information based on predetermined conditions for the relationship between the appearance frequency classes corresponding to each of the plurality of keywords.

According to one embodiment, the step of analyzing the plurality of technical documents includes the step of: the appearance frequency class estimating unit, by inputting a plurality of keywords into the keyword analysis model generated in advance based on the machine learning, .

For example, the step of estimating the appearance frequency class of a specific estimation target keyword may include estimating a plurality of keywords included in each of the plurality of technical documents, for each of a predetermined number of appearance frequency classes for classifying the plurality of keywords, Determining an appearance frequency rank of each of the plurality of keywords based on a degree of belonging indicating a degree to which the keyword belongs; inputting a plurality of keywords and an appearance frequency rank corresponding to each of the plurality of keywords into a keyword analysis model; And estimating an appearance frequency class of a specific estimation target keyword based on predetermined conditions for a relationship between appearance frequency classes corresponding to each of the plurality of keywords.

According to one embodiment, the step of analyzing the plurality of technical documents includes the steps of: the belonging technology group determination unit inputs a plurality of keywords to the belonging technology group determination model generated on the basis of the machine learning, And determining a technology group membership coefficient indicating the degree to which each of the plurality of technical documents belongs.

In order to achieve the above object, a technical document analysis apparatus according to an embodiment of the present invention includes a technical document collection unit for collecting a plurality of technical documents to be analyzed from a technical document database, A keyword extracting unit for extracting a plurality of keywords from a plurality of technical documents based on the number of appearance of each of the at least one word and a plurality of keywords in at least one technical document analysis model generated in advance based on the machine learning And a technical document analysis unit for analyzing a plurality of technical documents.

According to an embodiment of the present invention, a plurality of keywords are extracted from a plurality of technical documents based on the number of appearance of each of at least one word contained in each of a plurality of collected technical documents, Time and cost can be saved by analyzing a plurality of technical documents by inputting a plurality of keywords into at least one technical document analysis model generated in advance.

Further, according to the embodiment of the present invention, it is possible to classify the technical documents, to confirm the relationship between the keywords and the technical information, and to group the technical documents. Therefore, It can contribute to the formulation of research and development plans, and can also be applied to various technical management (MOT) issues.

1 is a block diagram for explaining a technical document analysis apparatus according to an embodiment of the present invention.
2 is a block diagram illustrating a technical document analysis unit in a technical document analysis apparatus according to an embodiment of the present invention.
3 is a flowchart for explaining a technical document analysis method according to an embodiment of the present invention.
4 is a flowchart for explaining a step of extracting a plurality of keywords in the technical document analysis method according to an embodiment of the present invention.
5 is a flowchart for explaining a first embodiment of analyzing a plurality of technical documents in the technical document analysis method according to an embodiment of the present invention.
6 is a flowchart for explaining a second embodiment of analyzing a plurality of technical documents in the technical document analysis method according to the embodiment of the present invention.
FIGS. 7 and 8 are diagrams for explaining a step of extracting a plurality of keywords in the technical document analysis method according to the embodiment of the present invention.
9 is a diagram for explaining a technical document classification model in a first embodiment of analyzing a plurality of technical documents of a technical document analysis method according to an embodiment of the present invention.
10A and 10B are diagrams for explaining a method of classifying a plurality of technical documents by bibliographic information in a first embodiment of analyzing a plurality of technical documents of a technical document analysis method according to an embodiment of the present invention .
11A, 11B and 11C illustrate a method for estimating the appearance frequency class of a specific estimation target keyword in the second embodiment of analyzing a plurality of technical documents of the technical document analysis method according to the embodiment of the present invention Fig.
12A and 12B are views for explaining a third embodiment of analyzing a plurality of technical documents in the technical document analysis method according to the embodiment of the present invention.
13 is a diagram for verifying the validity of a second embodiment of analyzing a plurality of technical documents of the technical document analysis method according to the embodiment of the present invention.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings in order to facilitate a person skilled in the art to easily carry out the technical idea of the present invention. . In the drawings, the same reference numerals are used to designate the same or similar components throughout the drawings. In the following description of the present invention, a detailed description of known functions and configurations incorporated herein will be omitted when it may make the subject matter of the present invention rather unclear.

Hereinafter, a method and apparatus for analyzing a technical document according to an embodiment of the present invention will be described in detail with reference to the accompanying drawings.

First, a technical document analysis apparatus 100 according to an embodiment of the present invention will be described with reference to FIG.

1 is a block diagram for explaining a technical document analysis apparatus according to an embodiment of the present invention.

1, the technical document analyzing apparatus 100 according to the embodiment of the present invention includes a technical document collecting unit 110, a keyword extracting unit 120, and a technical document analyzing unit 130. As shown in FIG.

The technical document collection unit 110 collects a plurality of technical documents to be analyzed from the technical document database 1.

The keyword extracting unit 120 extracts a plurality of keywords from a plurality of technical documents based on the number of occurrences of each of at least one word contained in each of the plurality of technical documents.

The technical document analysis unit 130 analyzes a plurality of technical documents by inputting a plurality of keywords into at least one technical document analysis model previously generated based on the machine learning.

Referring to FIG. 2, the technical document analysis unit 130 will be described in the technical document analysis apparatus 100 according to the embodiment of the present invention.

2 is a block diagram illustrating a technical document analysis unit in a technical document analysis apparatus according to an embodiment of the present invention.

2, the technical document analysis unit 130 includes a technical document classifying unit 131, an appearance frequency rating estimating unit 133, and a belonging technology group determining unit 135. [

The technical document classifying section 131 classifies a plurality of technical documents into bibliographic information by inputting a plurality of keywords into a technical document classification model previously generated based on the machine learning.

The appearance frequency class estimating unit 133 inputs a plurality of keywords into a keyword analysis model generated in advance based on the machine learning, and estimates an appearance frequency class of a specific estimation target keyword.

The belonging skill group determining unit 135 receives a plurality of keywords in a belonging skill group decision model generated in advance on the basis of the machine learning, and inputs a plurality of keywords to a technology group indicating a degree to which each of the plurality of skill documents belongs to each of the predetermined number of skill groups The group membership coefficient is determined.

According to one embodiment, each of the technical document classification model, the keyword analysis model, and the belonging skill group decision model described above is a fuzzy learning based on a fuzzy rule based system (FRBS) A fuzzy regression model, and a fuzzy clustering model, which are generated on the basis of the fuzzy regression model, but the present invention is not limited thereto.

According to an embodiment of the present invention, a method and apparatus for analyzing a technical document can utilize the fuzzy rule-based system described above for analyzing a technical document, wherein the method and apparatus for analyzing the technical document, A fuzzy rule-based system may include fuzzy inference and modeling systems that utilize " if-then " rules.

Here, the " if-then " rule may mean a rule for " determining the value of C in the case where A is a and B is b & Model and the fuzzy clustering model, respectively.

According to an embodiment of the present invention, fuzzy learning for generating each of the technical document classification model, the keyword analysis model, and the belonging skill group decision model described above is performed by using an R data language, which is software for fuzzy learning, , But the present invention is not limited thereto.

A more detailed description of the technical document analyzing apparatus according to the embodiment of the present invention will be described below with reference to FIGS. 3 to 12B, and redundant description will be omitted.

Referring now to FIG. 3, a method for analyzing a technical document according to an embodiment of the present invention will be described.

3 is a flowchart for explaining a technical document analysis method according to an embodiment of the present invention.

3, the technical document analysis method according to the embodiment of the present invention includes a technical document collection step S310, a plurality of keywords extraction step S320, and a plurality of technical document analysis steps S330 ).

In step S310, the technical document collection unit 110 collects a plurality of technical documents to be analyzed from the technical document database 1.

According to one embodiment, a technical document may refer to various documents describing a description of a technology including a patent document, a paper, and the present invention is not limited to a specific technical document.

For example, the technical document database 1 may mean various databases for storing technical documents.

For example, when the technical document is a patent document, the technical document database 1 may refer to various patent document databases. In the case where the technical document is a paper, the technical document database 1 may mean various paper databases, The present invention is not limited to the specific technical document database 1.

For example, a plurality of technical documents including the patent documents and theses collected in step S310 are large in capacity, including letters, numerals, pictures, etc., and a large number of technical documents are disclosed in real time, The plurality of collected technical documents may be the same as the structure of the big data, but the present invention is not limited thereto.

For example, the patent document, which is an embodiment of the technical document, includes various data including the name of the invention, the summary, the filing date, the inventor, the claim, the drawing, the international patent classification code (IPC) Patent documents can be a kind of big data structure because they meet the characteristic diversity and meet the peculiar speed of Big Data because a large number of patents are published in the patent office every day around the world. It does not.

Step S320 may mean a step of extracting a plurality of keywords from a plurality of technical documents based on the number of appearance of each of the at least one words included in each of the plurality of technical documents have.

For example, in step S320, the keyword extracting unit 120 extracts at least one word constituting the entire text information included in each of the plurality of technical documents, extracts at least one of the plurality of technical documents, The keywords for each of the plurality of technical documents can be extracted by sequentially selecting a predetermined number of words.

Now, referring to FIGS. 4, 7, and 8 at the same time, step S320 will be continued.

4 is a flowchart for explaining a step of extracting a plurality of keywords in the technical document analysis method according to an embodiment of the present invention.

FIGS. 7 and 8 are diagrams for explaining a step of extracting a plurality of keywords in the technical document analysis method according to the embodiment of the present invention.

As shown in FIG. 4, step S320 includes converting the text information into corpus information (S321), generating a word matrix by description document (S323), and extracting a plurality of keywords (S235).

Step S321 may refer to a step of converting the entire text information into corpus information by removing a stopword from the entire text information included in each of the plurality of technical documents.

For example, the full text information may mean information including various texts including words, sentences, and paragraphs constituting each of the plurality of technical documents described above.

For example, an abbreviation can mean an unnecessary word or a sentence symbol to describe the technical characteristics of each of a plurality of technical documents, including numbers, spaces, punctuation, articles, prepositions, investigations, conjunctions, and pronouns.

For example, the corpus information may mean set information of words that are expected to be necessary for describing the technical characteristics of each of a plurality of technical documents by removing an abbreviation from the whole text information.

For example, if each of the plurality of technical documents is an English technical document and the full text information is " He likes an apple. &Quot;, the pronoun "He", article "an" and punctuation " As a result, the corpus information extracted in step S321 may mean " likes apple ".

For example, if each of the plurality of technical documents is a Korean technical document, if the full text information is " he likes apples ", pronouns " Can be regarded as an abbreviation, and as a result, the corpus information extracted in the step S321 may mean " I like apples. &Quot;

Step S323 may be a step of generating a word matrix for each descriptive document, which is a matrix representing the number of occurrences of each of at least one word included in the corpus information for each of a plurality of technical documents.

Now, referring to FIG. 7, a word matrix for each technical document generated in step S323 will be described.

As shown in FIG. 7, the word matrix of each descriptive document may mean a matrix representing the number of occurrences of words included in each descriptive document.

For example, as shown in FIG. 7, in the word matrix according to one embodiment, technical document 1 includes three words 1, two words 3, four words 4, two words 4, The technical document 2 includes four words 1, three words 2, three words 3, three words 4, and five words 5, and the technical document 3 includes two words 1 2 words, 2 words, 0 words, 3 words, 3 words, 4 words, and 5 words.

Now, with reference to FIG. 4, step S325 will be described.

Step S325 may be a step of selecting a predetermined number of words in order of increasing the number of occurrences of each of at least one word contained in the word matrix of each descriptive document to extract a plurality of keywords.

For example, when there are two predetermined number of words for extracting keywords from the word matrix of the technical document shown in FIG. 7, words 1 and 3 can be extracted as keywords in technical document 1, The words 1 and 3 can be extracted as keywords, and in the description 3, words 3 and 5 can be extracted as keywords.

For example, at step S325, at least one word having the number of occurrences of each of at least one word included in the word matrix of each descriptive document having an appearance frequency equal to or greater than a preset occurrence frequency threshold may be extracted as a keyword.

For example, through the steps S321, S323, and S325, the top four keywords in each of the 2,583 English patents related to 3D printing are extracted, and the results of summarizing the number of appearances of the keywords for all 2,583 patents are shown in FIG. 8 Respectively.

For example, as shown in FIG. 8, the keyword "powder" appeared 1,387 times in a total of 2,583 patents, and the average value thereof was 0.54 times. Among the 2,583 patents, the keyword "powder" The number of occurrences of the keyword " powder " is 16.

Now, with continued reference to FIG. 3, step S330 will be described.

Step S330 may refer to a step in which the technical document analysis unit 130 analyzes a plurality of technical documents by inputting a plurality of keywords into at least one technical document analysis model previously generated based on the machine learning.

For example, the technical document analysis model may refer to a fuzzy model generated based on fuzzy learning based on fuzzy rules.

Now, a fuzzy rule for a technical document analysis model according to an embodiment of the present invention will be described, but the present invention is not limited thereto.

The fuzzy rule for the technical document analysis model according to the embodiment of the present invention can be defined as the following Equation 1.

[Equation 1]

As in the above equation (1), technical documentation analysis model according to an embodiment of the present invention, a keyword ₁ (Keyword 1) yi A _1, and keyword ₂ (Keyword 2) is A ₂ and the keyword p (Keyword _p) is A _p , The output target (Target) may be a model previously learned so as to determine that the target is B.

Referring now to Figures 5, 9, 10A and 10B, a first embodiment of a step of analyzing a plurality of technical documents in a technical document analysis method according to an embodiment of the present invention will be described.

5 is a flowchart for explaining a first embodiment of analyzing a plurality of technical documents in the technical document analysis method according to an embodiment of the present invention.

9 is a diagram for explaining a technical document classification model in a first embodiment of analyzing a plurality of technical documents of a technical document analysis method according to an embodiment of the present invention.

10A and 10B are diagrams for explaining a method of classifying a plurality of technical documents by bibliographic information in a first embodiment of analyzing a plurality of technical documents of a technical document analysis method according to an embodiment of the present invention .

For example, in the step of analyzing a plurality of technical documents, the technical document classification unit 131 may input a plurality of keywords into a technical document classification model generated in advance based on the machine learning, (Not shown).

In this case, the step of classifying the plurality of technical documents by bibliographic information (not shown) may mean the first embodiment of analyzing a plurality of technical documents.

As shown in FIG. 5, the step of classifying a plurality of technical documents by bibliographic information includes not only an appearance frequency rating step S331 of each of a plurality of keywords, a plurality of keyword and appearance frequency classes, (S332) of inputting to the model and classifying the plurality of technical documents by bibliographic information (S333).

At this time, the technical document classification model may mean a kind of fuzzy classification model generated based on the fuzzy learning.

In step S331, a plurality of keywords included in each of the plurality of technical documents are classified into a plurality of keywords based on the degree of belonging to which each of the plurality of keywords belongs to each of the predetermined number of appearance frequency classes for classifying the keywords, May be a step of determining the frequency grade of appearance of < RTI ID = 0.0 >

For example, in order to input the number of occurrences of each of the plurality of keywords into the above-described technical document classification model, the number of occurrences of each of the plurality of keywords should be classified according to the occurrence frequency interval.

To this end, in step S331, the technical document classification unit 131 determines the degree of belonging to each appearance frequency class of the appearance frequency of each of a plurality of keywords through the belonging function for fuzzy learning, The appearance frequency class can be determined as the appearance frequency class of each of the plurality of keywords described above.

In this case, a specific method for determining the belonging speed using the belonging function of the technical document classification unit 131 can use a conventionally known fuzzy learning algorithm, and a detailed description thereof will be omitted.

For example, in step S331, the appearance frequency grade is classified into three grades of "small" representing a keyword having a small number of occurrences, "medium" representing a keyword having a normal occurrence frequency and "large" representing a keyword having a large number of occurrences , The step S331 will be described.

In the above example, it is assumed that the number of appearance of the specific keyword A is 20, the degree of belonging to the appearance frequency class "small" is 0.3, the degree of belonging to the appearance frequency class "medium" is 0.7, The frequency rank of appearance of the specific keyword A may be determined as " medium ", but the present invention is not limited thereto.

According to an embodiment, a method of classifying each of a plurality of keywords into an appearance frequency class based on the degree of belonging in step S331 may be performed based on a conventionally known fuzzy learning algorithm, and a detailed description thereof will be omitted.

According to another embodiment, in step S331, the technical document classification unit 131 may determine the appearance frequency rank of each of the plurality of keywords based on a plurality of thresholds set in advance.

As an example of another embodiment described above, in step S331, the technical document classification unit 131 determines the above-mentioned appearance frequency classes "small", "medium", "large" based on the first threshold value and the second threshold value The keyword having the appearance frequency less than or equal to the first threshold value is determined as the appearance frequency grade " small ", the keyword having the appearance frequency exceeding the first threshold value and less than or equal to the second threshold value is determined as the appearance frequency level Quot; medium ", and a keyword whose occurrence frequency exceeds the second threshold value may be determined as the occurrence frequency grade " large ".

Step S332 may refer to a step of inputting the appearance frequency class corresponding to each of the plurality of keywords and the plurality of keywords into the technical document classification model.

According to one embodiment, the technical document classification model may mean a kind of fuzzy classification model generated on the basis of fuzzy learning, and the technical document classification model may include a keyword 1 extracted from a specific technical document, (Keyword ₁ ) is A ₁ , the appearance frequency frequency of Keyword ₂ is A _2, and the appearance frequency frequency of keyword p (Keyword _p ) is A _p , the bibliographic information of the specific technical document is B And may be a model that has been learned beforehand so that it can be determined.

For example, technical document classification model, the keywords (Keywords) by frequency of appearance rating (Linguistic) a first parameter (1 ^st), a second parameter of the membership functions for each as shown in Figure 9. Variable (2 ^nd ), The third parameter (3 ^rd ), and the fourth parameter (4 ^th ), and the specific method thereof is the same as that of the known fuzzy classification algorithm, so a detailed description will be omitted.

Step S333 may refer to a step of classifying a plurality of technical documents into bibliographic information based on predetermined conditions for a relationship between a plurality of keywords and appearance frequency classes corresponding to each of the plurality of keywords.

According to an exemplary embodiment, predetermined conditions for a relationship between appearance frequency classes corresponding to a plurality of keywords and a plurality of keywords may be stored in advance in the technical document classification model described above. As shown in Equation 1, If the appearance frequency frequency of the keyword 1 (Keyword ₁ ) extracted from the specific technical document is A ₁ , the appearance frequency frequency of the keyword 2 (Keyword ₂ ) is A _2, and the occurrence frequency grade of the keyword p (Keyword _p ) is A _p The bibliographic information in the technical document may refer to an " if-then "

For example, if the technical document is a patent document, the bibliographic information may refer to various bibliographic references including the applicant country, the applicant, the inventor, and the like.

For example, if the technical document is a paper, the bibliographic information may refer to various bibliographies, including the name of the author, journal name, author's affiliation, and so on.

According to an embodiment, after step S333, the method may further include a step of calculating the number of appearances from the appearance frequency class of each of the plurality of keywords by defuzzification, but the present invention is not limited thereto.

Referring now to FIGS. 10A and 10B, the case of applying S332 and S333 to the applicant country, which is one embodiment of the bibliographic information, will be described.

As shown in FIG. 10A, in a specific patent document A, when the appearance frequency frequency of the keyword "powder" is "small", the appearance frequency frequency of the keyword "composit" is "small" Quot; small ", and the appearance frequency class of the keyword " melt " is " small ", and the appearance frequency class of each keyword and keyword is input to the technical document classification model, The country (Nation) can output "1 (China)".

If the above-described process is applied to each of a plurality of patent documents, as shown in FIG. 10B, it can be seen that each of the plurality of patent documents can be classified by the application country.

Now, with reference to Figs. 6, 11A, 11B, 11C, and 13, a second embodiment of analyzing a plurality of technical documents in the technical document analysis method according to an embodiment of the present invention will be described.

6 is a flowchart for explaining a second embodiment of analyzing a plurality of technical documents in the technical document analysis method according to the embodiment of the present invention.

11A, 11B and 11C illustrate a method for estimating the appearance frequency class of a specific estimation target keyword in the second embodiment of analyzing a plurality of technical documents of the technical document analysis method according to the embodiment of the present invention Fig.

13 is a diagram for verifying the validity of a second embodiment of analyzing a plurality of technical documents of the technical document analysis method according to the embodiment of the present invention.

For example, in the step of analyzing the plurality of technical documents, the appearance frequency class estimating unit 131 may input a plurality of keywords into a keyword analysis model generated in advance based on the machine learning, (Not shown).

At this time, the step of estimating the appearance frequency class of the specific estimation target keyword (not shown) may mean the second embodiment of analyzing a plurality of technical documents.

As shown in FIG. 6, the step of estimating an appearance frequency class of a specific estimation target keyword includes an appearance frequency rating determination step (S335) for each of a plurality of keywords, a keyword analysis (S336) of inputting to the model, and estimating an appearance frequency class of a specific estimation target keyword (S337).

At this time, the keyword analysis model may mean a kind of fuzzy regression model generated based on fuzzy learning.

In step S335, the plurality of keywords included in each of the plurality of description documents are classified into the plurality of keywords based on the degree of belonging indicating the degree to which each of the plurality of keywords belongs to each of the predetermined number of appearance frequency classes for classifying the plurality of keywords, May be determined by determining the frequency of appearance of the frequency of occurrence.

At this time, since the step S335 is the same as the step S331 of FIG. 5, duplicate explanation will be omitted.

Step S336 may be a step of inputting the appearance frequency class corresponding to each of the plurality of keywords and the plurality of keywords into the keyword analysis model.

According to one embodiment, the keyword analysis model may be a kind of fuzzy regression model generated based on the fuzzy learning, and the keyword analysis model may include a keyword 1 extracted from a specific technical document ₁₎ frequency of appearance rates are a ₁ and when the appearance frequency level of appearance frequency level of keywords ₂ (keyword 2) is a ₂ and the keyword p (keyword _p) of a _p appearance frequency level of the specific estimation target keyword of is that the B And may be a model that has been learned beforehand so that it can be determined.

At this time, the keyword analysis model can be defined as Equation 2 below.

&Quot; (2) "

In this case, in the keyword analysis model (Keyword ₁ , ..., Keyword _p ), an input vector corresponding to each of a plurality of keywords, a keyword _target as an output variable, and B _n as a fuzzy regression variable.

이라고 하면,

는 낮은 데이터 지점에 맞추어 계산한 Keyword_target의 하한을 의미할 수 있고,

은 높은 데이터 지점에 맞추어 계산한 Keyword_target의 상한을 의미할 수 있으며, 키워드 분석 모델은 상술한 퍼지 회귀 구간을 활용하여 생성될 수 있으며, 키워드 분석 모델은 가능성(possibilistic)과 최소제곱(least-squares) 방법을 사용하여 적합화 될 수 있다.In generating the keyword analysis model through the process of fuzzy regression modeling, the fuzzy regression interval is defined as the keyword _target ,

Quot;

Can mean the lower bound of the Keyword _target calculated for the lower data points,

May mean an upper limit of a Keyword _target calculated in accordance with the high data point, the keyword analysis model can be generated by utilizing the above-described fuzzy return section, the keyword analysis model likelihood (possibilistic) and least squares (least-squares ) Method. ≪ / RTI >

At this time, the specific method of generating the keyword analysis model using the fuzzy regression interval is the same as the known fuzzy regression algorithm, and thus a detailed description thereof will be omitted.

Step S337 may refer to a step of estimating an appearance frequency class of a specific estimation target keyword on the basis of a predetermined condition for a relation between appearance frequency classes corresponding to a plurality of keywords and a plurality of keywords, respectively.

According to an exemplary embodiment, the predetermined condition for the relationship between the appearance frequency classes corresponding to the plurality of keywords and the plurality of keywords may be stored in advance in the keyword analysis model described above. If the appearance frequency level of Keyword ₁ extracted from the technical document is A ₁ , the appearance frequency frequency of Keyword ₂ is A _2, and the occurrence frequency grade of Keyword _p is A _p , frequency of appearance rates of the target keyword (keyword _target) may mean a "if-then" rules that determine that B.

At this time, as illustrated in Figure 11a, a plurality of the input keyword (Keyword _1, ..., _p Keyword) may refer to a technique, it is possible to affect the specific estimation technique in the target keyword (Keyword _target).

Furthermore, the specific estimation target keyword (Keyword _target) as shown in Figure 11a may sense the target technology in a particular technical field, the inputted plurality keyword _{(Keyword 1, ..., Keyword p} ) is the above-described target It can mean a sub-technology that influences the development of technology.

Now, the steps S336 and S337 will be described with reference to Figs. 11B and 11C.

As shown in FIG. 11B, among the input keywords, the appearance frequency frequency of the keyword "powder" is "small", the appearance frequency frequency of the keyword "composit" is "medium" When the appearance frequency class of each keyword and keyword is input to the keyword analysis model in the case where the appearance frequency class of the keyword "melt" is "medium" and the appearance frequency class of the keyword " Can be estimated to be "small".

When the above process is applied to each of a plurality of patent documents, it is confirmed that the appearance frequency class of the specific estimation target keyword " printhead " belonging to each of the plurality of patent documents is estimated for each of the plurality of patent documents, as shown in Fig. 11C .

Now, the performance of the keyword analysis model will be described with reference to FIG.

As shown in FIG. 13, in order to evaluate the performance of the keyword analysis model, a mean square error (MSE) and a mean square root error (RMSE) between an actual value and a predicted value are calculated. Is superior.

At this time, the mean square error (MSE) of the keyword analysis model is 1.66 and the mean square root error (RMSE) is 1.29. To validate the model using the test data to construct the keyword analysis model, (70%) and 775 test data (30%). As shown in FIG. 13, the actual appearance frequency class of the specific estimation target keyword and the estimated appearance frequency class of the specific estimation target keyword It can be confirmed that they are almost identical.

A third embodiment of analyzing a plurality of technical documents in the technical document analysis method according to an embodiment of the present invention will now be described with reference to FIGS. 3, 12A and 12B simultaneously.

12A and 12B are views for explaining a third embodiment of analyzing a plurality of technical documents in the technical document analysis method according to the embodiment of the present invention.

For example, in the step of analyzing a plurality of technical documents, the belonging skill group determining unit 135 may input a plurality of keywords into a belonging skill group determining model previously generated based on the machine learning, (Not shown) for determining a technology group membership coefficient indicating the degree to which each of the plurality of technical documents belongs.

At this time, the step of determining the technology group membership coefficient (not shown) may mean the third embodiment of the step of analyzing a plurality of technical documents.

At this time, the belonging skill group decision model may mean a kind of fuzzy clustering model generated based on the fuzzy learning.

Now, a method of determining the technology group membership coefficient based on the belonging technology group decision model in the step of determining the technology group membership coefficient (not shown) will be described.

이라 할때, 0에서 1사이의 기술군 소속 계수를 갖는 퍼지 집합 F는 아래 수학식 3과 같이 정의될 수 있다.A collection of multiple technical documents

, A fuzzy set F having a technology group membership coefficient between 0 and 1 can be defined as Equation 3 below.

&Quot; (3) "

At this time, each of the plurality of technical documents is mapped to a value between 0 and 1, and the above-described fuzzy set is used to cluster the plurality of technical documents.

이라 할 때, 분할 행렬 D는 아래 수학식 4와 같이 정의될 수 있다.At this time, the technology group consists of a fuzzy set of a plurality of technical documents,

, The division matrix D can be defined as shown in Equation (4) below.

&Quot; (4) "

Where D is the partition matrix and d _ij is the technical group membership coefficient of the technical document p _i for the technical group c _j .

At this time, the technology group membership coefficient can be selected within a range satisfying the following equations (5) to (7).

&Quot; (5) "

In this case, Equation (5) means that the technology group membership coefficient (d _ij ) is 0 or more and 1 or less.

&Quot; (6) "

Equation (6) implies that the total of the technical group membership coefficients (d _ij ) of the specific technical document (p _i ) determined for each technology group (c ₁ , c ₂ , ... c _m ) is 1.

&Quot; (7) "

Equation (7) indicates that the total of the technical group membership coefficients (d _ij ) of all technical documents (p ₁ , p ₂ , ... p _n ) determined for the specific technical group (c _j ) is greater than 0 and less than n do.

For example, the technology group membership coefficient may be determined to be inversely proportional to a distance between a point corresponding to each of the plurality of technical documents and a point corresponding to the center of each of the predetermined number of technology groups on a predetermined technology group graph model.

At this time, the more specific operation in which the belonging skill group decision model determines the technology group belonging coefficient is the same as the conventionally known fuzzy clustering model, and a detailed description thereof will be omitted.

As shown in FIG. 12A, when a technology group belonging coefficient for each of a predetermined number of technology groups of a plurality of technical documents is determined by the output of the belonging technology group determination model, a plurality of technical documents can be clustered.

In order to measure the clustering result of the plurality of technical documents, the sum of squares (SSE) as shown in Equation (8) below can be applied.

&Quot; (8) "

는 i번째 기술 문서(p_i)와 j번째 기술군(c_j) 간의 거리를 의미할 수 있다.At this time,

May refer to the distance between the i th descriptive document (p _i ) and the j th descriptive group (c _j ).

For example, when a plurality of technical documents are clustered based on a belonging technology group decision model, the specific technical document may belong to a technology group membership coefficient for several clusters.

According to one embodiment, when a plurality of keywords included in a specific technical document are input, the belonging skill group decision model is set to a learning process for determining a technical group belonging coefficient indicating the degree to which a specific technical document belongs to each of a predetermined number of technical groups, Which is a kind of fuzzy clustering model generated based on the learned fuzzy learning.

According to one embodiment, the belonging skill group determination model may determine the belonging coefficients of a plurality of keywords included in each of the plurality of descriptive documents, in addition to the above-described operations of determining the belonging coefficients of the plurality of descriptive documents.

12B shows a result obtained by multiplying the probability of belonging to a technology group by 100 by a factor of 100 for each of a total of ten keywords. As shown in Fig. 12B, the keyword " powder " The probability of belonging to the technology group 2 (cluster2) is 0.29, and the probability of belonging to the technology group is 0.36 for the technology group 3 (cluster3). As a result, It can be confirmed that a plurality of technical documents or keywords can belong to a plurality of technical groups at the same time if the technical group membership probability for each of a plurality of technical documents or keywords is determined based on the model.

By using the technical document analysis method and apparatus according to the embodiment of the present invention, it is possible to classify the technical document, to confirm the relationship between the keyword and the technology, and to cluster the technical document, Etc., can contribute to the establishment of research and development plans, and can be applied to various technical management (MOT) issues.

While the present invention has been described in connection with what is presently considered to be practical exemplary embodiments, it is to be understood that the invention is not limited to the disclosed embodiments, but many variations and modifications may be made without departing from the scope of the present invention. It will be understood that the invention may be practiced.

1: Technical Document Database
100: Technical Document Analysis Device
110: Technical Document Collecting Department
120: Keyword extracting unit
130: Technical Document Analysis Department
131: Technical Document Classification
133: Appearance Frequency Classification
135: belonging technology group decision unit

Claims (8)

Collecting a plurality of technical documents to be analyzed from a technical document database;
Extracting a plurality of keywords from the plurality of descriptive documents based on the number of occurrences of each of at least one word included in each of the plurality of descriptive documents; And
And analyzing the plurality of technical documents by inputting the plurality of keywords into at least one technical document analysis model generated in advance on the basis of the machine learning. The method according to claim 1,
Wherein the extracting of the plurality of keywords comprises:
Converting the entire text information into corpus information by removing a stopword from the entire text information included in each of the plurality of technical documents;
Generating a word matrix for each descriptive document which is a matrix representing the number of occurrences of each of the at least one words included in the corpus information for each of the plurality of technical documents; And
And extracting the plurality of keywords by selecting a predetermined number of words in descending order of the number of occurrences of each of the at least one words contained in the word matrix of each descriptive document. The method according to claim 1,
Wherein analyzing the plurality of technical documents comprises:
Wherein the technical document classification section includes inputting the plurality of keywords into a technical document classification model generated in advance based on machine learning and classifying the plurality of technical documents by bibliographic information. The method of claim 3,
Wherein the step of classifying the plurality of technical documents by bibliographic information comprises:
Based on a degree of belonging that indicates the degree to which each of the plurality of keywords belongs to each of a predetermined number of appearance frequency classes for classifying the plurality of keywords included in each of the plurality of technical documents by an appearance frequency interval, Determining said appearance frequency rating of said frequency;
Inputting an appearance frequency rating corresponding to each of the plurality of keywords and the plurality of keywords into the technical document classification model; And
Classifying the plurality of technical documents according to bibliographic information based on predetermined conditions for a relationship between the plurality of keywords and the appearance frequency grades corresponding to each of the plurality of keywords. The method according to claim 1,
Wherein analyzing the plurality of technical documents comprises:
Wherein the appearance frequency class estimating section includes a step of estimating an appearance frequency class of a specific estimation target keyword by inputting the plurality of keywords into a keyword analysis model previously generated based on machine learning. 6. The method of claim 5,
Estimating an appearance frequency class of the specific estimation target keyword,
Based on a degree of belonging that indicates the degree to which each of the plurality of keywords belongs to each of a predetermined number of appearance frequency classes for classifying the plurality of keywords included in each of the plurality of technical documents by an appearance frequency interval, Determining said appearance frequency rating of said frequency;
Inputting an appearance frequency rating corresponding to each of the plurality of keywords and the plurality of keywords into the keyword analysis model; And
Estimating an appearance frequency class of the specific estimation target keyword based on a predetermined condition for a relationship between the plurality of keywords and the appearance frequency grades corresponding to each of the plurality of keywords. The method according to claim 1,
Wherein analyzing the plurality of technical documents comprises:
Belonging technology group determining section is configured to input the plurality of keywords into a belonging skill group decision model generated in advance based on the machine learning and to classify a plurality of technical groups belonging to a technology group belonging to a technology group indicating the degree to which each of the plurality of technical documents belongs to each of the predetermined number of technology groups And determining a coefficient. A technical document collection unit for collecting a plurality of technical documents to be analyzed from a technical document database;
A keyword extracting unit that extracts a plurality of keywords from the plurality of technical documents based on the number of occurrences of each of at least one word included in each of the plurality of technical documents; And
And a technical document analysis unit for analyzing the plurality of technical documents by inputting the plurality of keywords into at least one technical document analysis model generated in advance on the basis of the machine learning.

Images