KR101620841B1 - Patent Analysis Method using A Hierarchical Diagram of Technology based on Statistical Patent Analysis - Google Patents

Abstract

The present invention relates to a method of analyzing a technology using a patent document, comprising the steps of extracting words from documents to be analyzed, generating a document-word matrix using the extracted words, Generating a document-keyword matrix by selecting a keyword from a word matrix, generating a regression model using the document-keyword matrix, calculating a significance probability (p (n)) of the parameters of the generated regression model, selecting a keyword corresponding to a parameter having a value of -value, and deriving a relation between the keywords by using the significance probability between the selected keywords, thereby deriving a technology layer using statistical analysis, By using the hierarchy, objective and accurate patent analysis is possible.

Description

TECHNICAL FIELD The present invention relates to a patent analysis method using a technology hierarchical diagram model based on statistical analysis,

The present invention relates to a method for analyzing a patent, and more particularly, to a method for deriving a technology layer using statistical analysis and analyzing a patent using the derived technology layer.

Recent Technology Forecasting (TF) is an important issue in technology management. Most MOT R & D policies depend on the outcome of various technology forecasts. Traditional technology forecasts use qualitative approaches such as the Delphi method. Although the Delphi method is a representative method in technology prediction, it is not a suitable method due to the disadvantage that the result is subjective.

 Due to these problems, most of the technology prediction techniques have recently been focusing on quantitative approaches such as patent analysis. Patents have a variety of information on developed technologies and are one of the objective data formats used in technology prediction.

Korea Open Patent "Document Characteristic Analysis Apparatus for Investigated Documents (10-2006-0095565)"

A problem to be solved by the present invention is to provide a method of deriving a technology layer using statistical analysis and analyzing a patent using the derived technology layer.

According to an aspect of the present invention, there is provided a method of analyzing a technology using a patent document, the method comprising: extracting words from documents to be analyzed; Generating a document-word matrix using the extracted words; Generating a document-keyword matrix by selecting a keyword from the generated document-word matrix; Generating a regression model using the document-keyword matrix; Selecting a keyword corresponding to a parameter having a significance probability (p-value) below a first threshold among the parameters of the generated regression model; And deriving a relationship between the keywords by using the significance probability between the selected keywords.

According to another embodiment of the present invention, the step of deriving the relationship between the keywords may include: selecting one or more keywords corresponding to the parameters having the significance probabilities below the second threshold among the selected keywords as the first keyword; And selecting, as a keyword having a relation with the first keyword, a keyword corresponding to a parameter having a significance probability below a third threshold through a regression model analysis between the first keyword and the keywords other than the first keyword . ≪ / RTI >

According to another embodiment of the present invention, the step of deriving the relation between the keywords may be a method of dividing the hierarchy by using the significance probability between the keywords.

According to another embodiment of the present invention, the step of generating the document-keyword matrix includes: selecting words having an occurrence frequency value of the fourth threshold or more from the document-word matrix; And removing words of the selected words that are not related to the description.

According to another embodiment of the present invention, the document to be analyzed is a patent document, and the step of extracting the word may include extracting at least one of a name, an abstract, a patent claim, And extracting the words.

According to another embodiment of the present invention, the step of extracting the words may be performed using sentence parsing and corpus analysis.

According to the present invention, it is possible to derive a technology layer using a statistical analysis, and to analyze a patent using the derived technology layer, in order to derive a core keyword, out of a method of qualitatively extracting expert opinions and Delphi techniques have. Thus, objective patent analysis is possible.

1 is a flowchart of a patent analysis method according to an embodiment of the present invention.
2 to 3 are flowcharts of a patent analysis method according to another embodiment of the present invention.
4 is a flowchart of a method for analyzing a patent according to another embodiment of the present invention.
5 to 6 illustrate a technology layer derived according to a patent analysis method according to an embodiment of the present invention.

Prior to the description of the concrete contents of the present invention, for the sake of understanding, the outline of the solution of the problem to be solved by the present invention or the core of the technical idea is first given.

A method for analyzing a technology using a patent document according to an embodiment of the present invention includes extracting words from documents to be analyzed, generating a document-word matrix using the extracted words, Generating a document-keyword matrix by selecting a keyword from the generated document-word matrix, generating a regression model using the document-keyword matrix, generating a regression model with a first threshold of the parameters of the generated regression model Selecting a keyword corresponding to a parameter having a significance value (p-value), and deriving a relationship between the keywords using the significance probability between the selected keywords.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. It will be apparent to those skilled in the art, however, that these examples are provided to further illustrate the present invention, and the scope of the present invention is not limited thereto.

BRIEF DESCRIPTION OF THE DRAWINGS The above and other features and advantages of the present invention will become more apparent by describing in detail preferred embodiments thereof with reference to the attached drawings in which: It is possible to quote the above. In the following detailed description of the principles of operation of the preferred embodiments of the present invention, it is to be understood that the present invention is not limited to the details of the known functions and configurations, and other matters may be unnecessarily obscured, A detailed description thereof will be omitted.

1 is a flowchart of a patent analysis method according to an embodiment of the present invention.

A method of analyzing a technology using a patent document according to an embodiment of the present invention is implemented through the following series of processes.

Step 110 is a step of extracting words from the documents to be analyzed.

More specifically, words are extracted from documents to be analyzed using a text mining method or the like. Text mining is a mining technique that extracts patterns or relationships from unstructured text data composed of natural language and finds value and meaningful information. It is a technology based on natural language processing technology that can understand people's language. The words can be extracted using sentence parsing and corpus analysis.

Or may be documents that have been investigated for the technology or technical field to analyze the documents.

The documents to be analyzed may be textual documents, in particular patent documents. In extracting words for a patent document, for the sake of accuracy and efficiency in extracting significant words, it is possible to extract words from at least one part of the title, summary, patent claim, or inventive description of the patent document have. The title, abstract, and claims of the patent document include the essential parts of the invention to be covered by the patent document, and the details of the invention are included in the detailed description of the invention, Extracts words from one or more portions of the abstract, the claims, or the detailed description of the invention. Among the above patent documents, it is preferable to use a summary of the patent (supplement).

Step 120 is a step of generating a document-word matrix using the extracted words.

More specifically, a document-term matrix (DTM) is generated using the analyzed document and the extracted words. Corpus and text mining technique text repository are used to construct document-word matrix. Document - A word matrix is a matrix that describes the relationship between a document and words, and represents the frequency with which each word appears in the document. The rows and columns of a matrix consist of words and documents. Each element is a word frequency value in each document.

Step 130 is a step of generating a document-keyword matrix by selecting a keyword from the generated document-word matrix.

More specifically, a keyword of the words is selected to generate a document-keyword matrix. By selecting keywords among the words, the number of lines is reduced to enable fast and accurate analysis.

In generating the document-keyword matrix, steps 310 to 320 of FIG. 3 may be performed. In step 310, words having an occurrence frequency value equal to or higher than the fourth threshold value can be selected from the document-word matrix. The fourth threshold may be predetermined or may be determined according to the number of rows of the document-word matrix. The higher the frequency of occurrence is, the more likely the keyword is, and the word can be selected according to the occurrence frequency value. The document-keyword matrix may be generated by removing words not related to the description among the selected words in step 320. [ The frequency of occurrence is high, but accuracy can be improved by removing words that are not related to technology. For example, you can remove meaningless words such as "is" and "the". The words to be removed may be preset or specific words may be removed according to the setting input of the user.

Step 140 is a step of generating a regression model using the document-keyword matrix.

More specifically, a regression model is generated using a document-keyword matrix. A regression model is a statistical technique that can estimate the impact of one or more independent variables on dependent variables. That is, a regression model can be generated to find statistically significant keywords among the selected keywords. The keyword selection is not statistically significant but statistically significant by using the regression model. A regression model is generated from the document-keyword matrix. The regression model can be expressed as:

Z is a dependent variable, and X _s is an independent variable. Dependent variables are the affected technologies and independent variables are the developed technologies. That is, the dependent variable is a technique to be analyzed, and the independent variable corresponds to a keyword derived from a document analysis by techniques developed in relation to the technology. ε is the error term. The regression parameter β _k represents the strength of the causal relationship between the two techniques. In the full regression model, all independent variables are used and hypothesized as follows.

Hypothesis testing can be conducted to find X technologies that affect Z technology. The null hypothesis H ₀ indicates that the ith regression parameter is zero. This means that the i-th technology does not affect the technology Z. Also, the alternative hypothesis H ₁ indicates that the i-th regression variable is not zero. That is, the technology Z depends on the technology Xi. To determine if the technique X _i has a significant effect on Z, H ₀ can be rejected through a statistical test. For the hypothesis test, we use the t distribution with n - (k + 1) degrees of freedom. n and k are the data size and the number of variables. If H0 is true, the test statistic can be calculated as follows.

는

의 추정치이고,

는

의 표준오차이다. H₀를 거절하기 위해서는 다음조건을 만족시켜야 한다.

The

≪ / RTI >

The

. To reject H ₀ , the following conditions must be satisfied.

의 값을 얻을 수 있다.α is a significant level, and using a t-distribution table

Can be obtained.

Step 150 is a step of selecting a keyword corresponding to a parameter having a significant probability (p-value) below the first threshold among the parameters of the generated regression model.

More specifically, for fast and accurate analysis, it is possible to reduce the number of keywords by selecting significant keywords. For this, a keyword corresponding to a parameter having a significant probability (p-value) below the first threshold value among the parameters of the regression model generated in step 140 may be selected. The significance probability is the lowest significance level required to see if the parameter is significant. If the significance probability is less than 0.05 in the regression result of the parameter, this parameter can be judged to be significant. When the significance probability of the test statistic is less than 0.05, the description of X _i is judged to be significant. That is, in order to generate a reduced regression model, keywords corresponding to parameters having a probability of less than 0.05 are selected. Selecting key keywords is very efficient in terms of accuracy and speed.

Step 160 is a step of deriving a relation between the keywords using the probabilities of the selected keywords.

More specifically, it is possible to derive the relationship between the selected significant keywords and the technique for analyzing the selected significant keywords using the significance probability between the keywords. The relationship between the keywords can be expressed by dividing the hierarchy by using the significance probability between the keywords. The hierarchy can be divided into a hierarchical diagram of technology (HDT). The relation between the upper layer technology and the lower layer technology can be expressed as follows.

X ₁ is the affected technology, and W ₁ is the developed technology. u ₁₁ is the causal relationship strength between X ₁ and W ₁ . The model is to find all causal intensities to predict technology Z. W _m technology affects the development of technology Z depending on causality intensity in (ump * vpn * rn). Therefore, all causal strengths of all techniques except for Z of TF can be calculated.

In order to derive the relationship between the keywords, steps 210 to 220 of FIG. 2 may be performed.

Step 210 is a step of selecting at least one keyword corresponding to a parameter having a significance probability below a second threshold among the selected keywords as a first keyword.

One or more keywords corresponding to parameters having significance probabilities below the second threshold among the keywords corresponding to the parameters having the significance probabilities below the first threshold value are selected as the first keywords. The first keyword is a technique most related to the technology to be analyzed, and selects a keyword corresponding to a parameter having a probability of less than a second threshold as a first keyword corresponding to an upper layer. The second threshold value may be preset or may be set according to the user's input. Alternatively, the first keyword may be selected in a predetermined number in the order of lower probability of significance.

In operation 220, a keyword corresponding to a parameter having a significance probability less than a third threshold value is selected as a keyword having a relation with the first keyword through a regression model analysis between keywords other than the first keyword and the first keyword .

More specifically, after selecting the first keyword, a keyword of the next layer is selected. The depth of the hierarchy may be predetermined, or may vary depending on the number of keywords, or the degree to be analyzed. In order to select a keyword in the next layer related to the first keyword, a regression model analysis is performed between keywords other than the first keyword and the first keyword. A keyword corresponding to a parameter having a significance probability less than or equal to a third threshold value is selected as a keyword having a relation with the first keyword through a regression model analysis between keywords other than the first keyword and the first keyword. This step corresponds to steps 140 to 150.

4 is a flowchart of a method for analyzing a patent according to another embodiment of the present invention.

A document-word matrix (DTM) is constructed by retrieving a patent document in a predetermined TF field and extracting a summary (abstract) of the patent data. Reconstruct from document-word matrix (DTM) to document-keyword matrix. To do this, the ranked terms are searched for in the document-word matrix (DTM), meaningless words are removed, and keywords are selected using the remaining words to form a document-keyword matrix. Hierarchical regression analysis is then performed. That is, a full regression model is constructed using all the keywords, a reduced regression model is constructed using the selected keywords, and a technique to analyze and analyze the constructed HDT model by analyzing the constructed HDT model is analyzed.

5 to 6 illustrate a technology layer derived according to a patent analysis method according to an embodiment of the present invention.

The results of the patent analysis according to the embodiment of the present invention are shown in FIG. The results were derived using patent documents retrieved from KIPRIS and USPTO (KIPRIS 2013; USPTO 2013). I chose telematics as a technology field that uses TF. The data set consisted of telematics patents filed in the US, Europe, and China, and a total of 474 patent documents were used. The number of applications has increased significantly since the 2000s. Using the abstract of the patent document retrieved for the DTM configuration, the rows and columns of the matrix are represented by documents and words, respectively. The elements of the matrix represent the frequency of word occurrences in each document. Hierarchical regression was performed for TF of telematics using DTM. The top 20 keywords including telematics were selected for variable selection for path analysis. We selected the keywords center, communication, control, device, diagnostic, information, message, mobile, network, receiving, request, service, signal, telematics, terminal, transmit, unit, user, vehicle. We obtained the following full regression.

In the regression model, the dependent variable is telematics and all keywords except telematics are used as independent variables. Table 1 shows the results of full regression analysis.

Independent variable Beta p-value center 0.126 0.001 communication -0.004 0.857 connection -0.009 0.642 control 0.037 0.072 device 0.286 0.001 diagnostic -0.044 0.024 information 0.022 0.346 message 0.024 0.236 mobile 0.003 0.893 network 0.060 0.003 receiving 0.086 0.001 request 0.063 0.004 service 0.169 0.001 signal 0.019 0.440 terminal 0.219 0.001 transmit -0.154 0.001 unit 0.321 0.001 user 0.067 0.001 vehicle 0.152 0.001

Beta represents regression parameters of standardized variables and can compare the effect of all independent variables. That is, beta is a change in "telematics" associated with unit change in each independent variable. For example, the unit change of "center" increases by 0.126 in the occurrence of "telematics". Furthermore, we have found that "unit" has the greatest effect on "telematics". The significance of independent variables for the dependent variable is determined by the p-value. If the p-value is less than 0.05 in the 95% confidence interval, it can be determined to be a significant variable. Therefore, you can use the "center", "device", "diagnostic", "network", "receiving", "request", "service", "terminal" Were used as variables in the HDT model.

The following is a reduced regression model for HDT generation for TF of "telematics" as shown below.

Independent variables Beta p-value center 0.124 0.001 device 0.285 0.001 diagnostic -0.045 0.024 network 0.061 0.019 receiving 0.101 0.002 request 0.063 0.001 service 0.171 0.003 terminal 0.220 0.001 transmit -0.134 0.001 unit 0.325 0.001 user 0.069 0.001 vehicle 0.165 0.001

The independent variables are all significant because the p-value of all independent variables is less than 0.05. Since the beta of the device, the unit, and the terminal has a larger value than the others, the first HDT model is constructed by selecting the first keyword as shown in FIG.

As a result, we found that the variables "device", "unit" and "terminal" directly affect "telematics". Other independent variables also indirectly affected "telematics" via "device", "unit", and "terminal". Therefore, three regression models of three dependent variables (ie device, unit, terminal) were constructed to complete the HDT model for "telematics" TF.

Independent variable Dependent variable device unit terminal Beta p-value Beta p-value Beta p-value center -0.093 0.001 0.093 0.001 0.001 0.962 diagnostic 0.040 0.140 -0.093 0.001 0.015 0.579 network 0.123 0.001 0.019 0.464 0.068 0.011 receiving 0.060 0.048 0.070 0.020 0.131 0.001 request -0.020 0.253 0.025 0.391 0.055 0.062 service 0.033 0.253 0.044 0.125 0.055 0.001 transmit -0.096 0.001 0.096 0.001 0.100 0.001 user 0.048 0.001 0.182 0.001 0.124 0.388 vehicle 0.016 0.001 0.079 0.005 -0.053 0.066

Here, statistically significant independent variables were selected using statistical significance based on three independent variables, p-value. We selected center, network, receiving, transmit, user, and vehicle with p-value of 0.05 or less as an independent variable for the device, considering those with p-value less than 0.05 (confidence level 95%). In the same way, the independent variables for unit and terminal were determined as follows.

Dependent variable Selected independent variable device center, network, receiving, transmit, user, vehicle unit center, diagnostic, receiving, transmit, user, vehicle terminal network, receiving, service, transmit

From the results in Table 4, it can be seen that the technology of center, network, receiving, transmit, user, and vehicle influenced the development of device technology, and it can be seen that this development affected the technology development of telematics. Unit and terminal technologies are similar to devices. Using the results of Table 4, a complete hierarchical model of the telematics TF of FIG. 6 was constructed.

This shows the technical relevance of the "telematics" TF. Each connection strength means the strength from bottom to top. For example, the "receiving" technology affects the "terminal" technology as much as the connection strength of 0.133. In addition, this weight is significant because the p-value is less than 0.05. The "terminal" technology continuously affects the "telematics" technology as much as the connection strength of 0.422. If we want to know the direct effect of the "receiving" technique on the "telematics" technique, we must multiply two weights: the weights of "receiving" (0.133) in "terminal" and "terminal" (0.422) in "telematics" . Therefore, the "receiving" technology indirectly affects the "telematics" technology with a connection strength of 0.056 (0.133 × 0.422). The rest of the hierarchical diagram can be explained in the same way.

In selecting the keyword of step 130, a keyword may be selected through the following series of steps (step 410).

In operation 410, a document-principal component analysis matrix is generated from the generated document-word matrix through principal component analysis.

More specifically, it is very difficult to select a core keyword among the extracted words, and the number of words can not be reduced because the selected keyword has a great influence on subsequent results. In order to extract objective and key keywords, principal component analysis (PCA) can be used to solve the scarcity of the matrix while preserving information that can be lost at the same time. Principal component analysis is a so-called multivariate analysis that analyzes various variables. It extracts several principal components from many variables. Principal component analysis is for dimension reduction, where the principal component is the principal component that describes many variables. The principal component analysis may be performed using a correlation matrix. Or a covariance matrix. The number of principal components (PC) in the document - principal component analysis matrix generated by principal component analysis is generated as many as the number of words (Term) as the first variable.

Here, for efficient analysis, the number of PCs can be reduced. As a result of the principal component analysis, a reduced document-principal component analysis matrix can be generated using only principal components having eigenvalues equal to or greater than the threshold value. In order to perform the analysis using only the significant principal components, the number of columns of the matrix can be reduced by using only the principal component having the eigenvalue of the principal component equal to or greater than the threshold value. The eigen value for each principal component and the principal component score for each principal component in the principal component can be calculated through principal component analysis. And the document-principal component analysis matrix is reduced using only the principal component when the eigenvalues of the calculated principal components are equal to or greater than the threshold value. The threshold value may be one. That is, it is possible to reduce the size by reducing the number of principal components using only the principal component having an eigenvalue of 1 or more.

Step 420 is a step of generating a regression model using the document-principal component analysis matrix. More specifically, a regression model is generated using a document-principal component analysis matrix. A regression model is a statistical technique that can estimate the effect of one or more independent variables on dependent variables. Statistical analysis of the principal components is possible. In other words, a statistically significant principal component of the selected principal components can be found by creating a regression model. Principal component analysis is not statistically significant but statistically significant by using regression model. A regression model is generated from the document-principal analysis matrix. The regression model can be expressed as:

Step 430 is a step of selecting a principal component corresponding to a parameter having a significant probability (p-value) below the threshold value among the parameters of the generated regression model. More specifically, in order to select a statistically significant principal component in the regression model generated in step 420, the parameters of the regression model are used. In Equation (7), β ₁ and β ₂ have a significant probability (p-value). A principal component corresponding to a parameter having a significance probability below the threshold value can be selected as a significant principal component. The threshold may be 0.05. The principal component corresponding to the parameter having the probability of 0.05 or less can be selected as a significant principal component. If there are several words, it is difficult to randomize the importance of all the words and it is difficult to confirm whether each word actually affects it. Therefore, selecting key keywords through principal component analysis and regression model is very efficient in terms of accuracy and speed.

In step 440, one or more words belonging to the selected principal component are selected as key keywords. More specifically, one or more words included in the principal component are selected as key keywords. Select key words among the words included in the main component. If there are a plurality of words included in the principal component, a word having a principal component score of a threshold value or more among the words belonging to the principal component or a predetermined word may be selected as a core keyword according to the order in which the principal component score is high. The principal component may include a plurality of words, and the core keyword may be finally selected using the principal component load (linear coefficient) of the words, that is, the principal component score. A word having the principal component score of the principal component score equal to or greater than the threshold value can be selected as a key keyword. The principal component score may be from -1 to 1, and the threshold may be zero. Or key words in order of the words having a high principal component score. A predetermined number of words may be selected, or the number of words selected may be varied according to the amount of words and documents. It may be chosen according to both the threshold and the number of words.

By using principal component analysis (PCA) and regression model, statistically significant parameter β can be found and quantitative technology trend analysis can be done by selecting important keywords through selected parameters.

An embodiment of the present invention includes at least one processing unit (processor) for extracting words of a document, generating a document-word matrix, generating a document-keyword matrix, generating a regression model, One or more storage units (database) for storing the results to be calculated and words to be removed.

Embodiments of the present invention may be implemented in the form of program instructions that can be executed on various computer means and recorded on a computer readable medium. The computer-readable medium may include program instructions, data files, data structures, and the like, alone or in combination. The program instructions recorded on the medium may be those specially designed and constructed for the present invention or may be available to those skilled in the art of computer software. Examples of computer-readable media include magnetic media such as hard disks, floppy disks and magnetic tape; optical media such as CD-ROMs and DVDs; magnetic media such as floppy disks; Magneto-optical media, and hardware devices specifically configured to store and execute program instructions such as ROM, RAM, flash memory, and the like. Examples of program instructions include machine language code such as those produced by a compiler, as well as high-level language code that can be executed by a computer using an interpreter or the like. The hardware devices described above may be configured to operate as one or more software modules to perform the operations of the present invention, and vice versa.

As described above, the present invention has been described with reference to particular embodiments, such as specific elements, and specific embodiments and drawings. However, it should be understood that the present invention is not limited to the above- And various modifications and changes may be made thereto by those skilled in the art to which the present invention pertains.

Accordingly, the spirit of the present invention should not be construed as being limited to the embodiments described, and all of the equivalents or equivalents of the claims, as well as the following claims, belong to the scope of the present invention .

Claims (6)

A method for analyzing a technology using a patent document,
Extracting words from documents to be analyzed;
Generating a document-word matrix using the extracted words;
Generating a document-keyword matrix by selecting a keyword from the generated document-word matrix;
Generating a regression model using the document-keyword matrix;
Selecting a keyword corresponding to a parameter having a significance probability (p-value) below a first threshold among the parameters of the generated regression model; And
And deriving a relationship between the keywords using the probability of the selected keyword,
Deriving a relationship between the keywords,
Selecting at least one keyword corresponding to a parameter having a significance probability less than a second threshold among the selected keywords as a first keyword; And
Selecting a keyword corresponding to a parameter having a significance probability less than a third threshold value as a keyword having a relation with the first keyword through a regression model analysis between the first keyword and the keywords other than the first keyword, ≪ / RTI > delete The method according to claim 1,
Deriving a relationship between the keywords,
And dividing the hierarchy by using the significance probability between the keywords. The method according to claim 1,
Wherein the generating the document-keyword matrix comprises:
Selecting words having an occurrence frequency value equal to or higher than a fourth threshold value from the document-word matrix; And
And removing words of the selected words that are not relevant to the description. The method according to claim 1,
The documents to be analyzed are patent documents,
The step of extracting the word comprises:
Extracting words from at least one part of the title, abstract, patent claim, or inventive description of the patent document. The method according to claim 1,
The step of extracting the words comprises:
Parsing and corpus. ≪ RTI ID = 0.0 > 8. < / RTI >

Images