KR20180072167A - System for extracting similar patents and method thereof - Google Patents

Abstract

When a company commercializes a technical item, barrier patents which induce obstacles such as patent infringement to commercialization can exist. When there is a barrier patent, the company may establish a countermeasure strategy, such as invalidation strategy, avoidance design, and non-infringement logic development. An object of the present invention is to propose a system for evacuating prior patents which can invalidate the barrier patents. On the basis of a patent database (10) comprehensively including bibliographic information and expertise information on patents of US trademark and patent office, the system finally recommends prior patents having a potential capable of invalidating the barrier patents, by way of a keyword group setting module for the barrier patents, an information entropy calculation module for grasping a ratio of keyword group content contained in an arbitrary patent, and a similarity calculation module for calculating a semantic patent similarity between the barrier patents and an arbitrary patent through topic modeling.

Description

[0001] SYSTEM FOR EXTRACTING SIMILAR PATENTS AND METHOD THEREOF [0002]

The present invention relates to a similar patent extraction system and method for automatically extracting patents similar to a reference patent.

Generally, patents that obstruct commercialization, such as patent infringement on technology items to be commercialized, are called barrier patents. Therefore, if a company makes commercialization of a specific item, if it does not consider the patent registered by another applicant, it may cause a problem of patent infringement when selling a product developed on the market in the market. In other words, even if a company is a patentee, it can not be infringed by another patentee's patent. Therefore, it is necessary to identify what kind of patents may be a problem for the products currently under development, establish a countermeasure strategy for them, and develop inventions accordingly. For example, in the presence of barrier patents for development technology items, countermeasures such as invalidation strategies for barrier patents, avoidance designs, and development of non-infringement logic can be established, requiring high level of patent and technical expertise.

Among the countermeasures against the aforementioned barrier patent, the present invention aims at supporting the invalidation strategy. In the case of a patent already obtained a patent right, there may be invalid reasons in all or part of the invention described in the claim, which means that the patent to be denied the patent right has received the patent right due to insufficient examination. If the invalidation of the patent is confirmed, the validity of the patent is deemed to be from the beginning. Therefore, the problem of infringement does not arise. The invalidation strategy is a countermeasure strategy to choose when it is difficult to deny the infringement by judging the patent infringement. These factors are mainly caused by novelty and inventive step. In order to investigate the cause of nullification of key patents recognized as barrier patents, the precedent patent documents related to the barrier patents are collected and analyzed.

In practice, most of the process of analyzing prior patent literature for barrier patents is done by qualitative work by experts. In other words, after a technical expert reviews the technology described in the barrier patent, it creates a patent search formula, collects related patents from the patent search service, and then selects patents having technological relevance. However, this task involves a lot of time and manpower because it targets thousands of related patents, which number to tens of thousands. According to industry experience, one person can review roughly 1,000 patent data per day (assuming that one case is reviewed every 30 seconds), and review details including the claims for prior patents It is not easy to get up to 300 cases a day. In addition, a problem of frequent misjudgment may occur due to the limit of repetitive review work.

Next, from the inventive point of view, the existing inventive methods are a type of analyzing the keyword similarity or structural similarity to the collected patent set to grasp the similarity relationship between the patents, and they are not included in the collection range or exist in potentially different fields It is difficult to analyze the prior patent technologies that can invalidate the barrier patents. Therefore, it has limitations to apply to the methodology of finding the prior patent technology for barrier patents.

Therefore, in order to extract similar patents, a lot of time and manpower are injected because the similar patents are filtered through a simple repetitive operation in which a person directly examines a patent patent document directly in a population and extracts a similar patent.

The following patent documents relate to a patent document search and classification method and system thereof, but fail to provide a solution to the above-mentioned problems.

Korean Patent Publication No. 10-2006-0047306

SUMMARY OF THE INVENTION The present invention has been made in view of the above problems, and it is an object of the present invention to provide a similar patent extracting system for extracting similar patents on a reference patent by calculating information entropy such as existence / And a method thereof.

The present invention also provides a similar patent extraction system and a method for extracting similar patents from a text viewpoint by calculating text similarity through subject modeling analysis between a candidate patent extracted from a keyword base and a reference patent.

According to another aspect of the present invention, there is provided a method of extracting similar patents, comprising: (a) selecting at least one reference patent as a target of interest among a plurality of patent documents, Setting information; (b) calculating information entropy for the patent on all patents recorded in the patent population or the patent database using the keyword set in the step (b); And (c) selecting and providing at least one candidate patent similar to the reference patent based on the information entropy calculated in the step (a).

The extraction method includes analyzing a topic of each candidate patent based on the word frequency information of the candidate patents selected in the above process to calculate the subject probability distribution, Calculating a text similarity with the patent; And extracting and providing a similar patent having a text structure similar to the reference patent according to the similarity calculated in the above.

The patent information may include at least one of a patent number, an inspection period, and an IPC classification code of a reference patent.

According to another aspect of the present invention, there is provided a system for extracting similar patents, comprising: a processor for selecting at least one reference patent among a plurality of patent documents, A reference patent setting section for setting a reference patent setting section; A frequency measuring unit for extracting key keywords for each document from a plurality of patent documents and measuring frequency of occurrence in the document; An information amount calculating unit for calculating information entropy of each document through a document occurrence frequency probability of a main keyword with respect to the number of occurrences of all words in the document measured by the frequency measuring unit; And a candidate patent selection section for selecting at least one candidate patent similar to the reference patent based on the information entropy calculated in the above.

Then, the extraction system analyzes the topic of each candidate patent based on the word frequency information of the candidate patents selected in the above-described manner, calculates the subject-specific probability distribution, and uses the subject-specific probability distribution of each candidate patent A similarity calculating unit for calculating a text similarity with a reference patent; And a similar patent extracting unit for extracting and providing a similar patent having a text structure similar to the reference patent according to the similarity calculated in the above.

According to an embodiment of the present invention, the prior patent documents can be analyzed in accordance with the degree of inclusion of the technical viewpoints to be analyzed by the user. While conventional patent search services are based on Boolean search, they do not consider how much the search results contain the technical viewpoints that the user wants to analyze. On the other hand, And the weights for them, so that it is possible to understand how arbitrary patent documents are related to the technical viewpoints that the user sees.

In addition, this system can reduce the reliance on research experts in the work such as the creation of retrieval formulas and selection of prior patent documents, and dramatically shortens the time required for the collection and analysis of prior arts through the automation process of this system can do.

1 is a conceptual diagram for explaining a similar patent extraction system according to the present invention.
2 is a block diagram illustrating a similar patent extraction system according to an embodiment of the present invention.
3 is a diagram showing a detailed configuration of the similarity degree calculating unit shown in FIG.
4 is a flowchart for explaining a similar patent extracting method according to an embodiment of the present invention.
5 is a UI screen showing a similar patent extraction system according to an embodiment of the present invention.
FIG. 6 is a diagram showing the reference patent information input and keyword recommendation screen of FIG.
FIG. 7 is a screen showing the result of the calculation of the information amount in FIG.
FIGS. 8A to 8C are diagrams for explaining the LDA applied to the computation of the text similarity degree in FIG.
9 is a view showing the result of calculation of the text similarity degree in FIG.
10 is a final screen for extracting and displaying a patent document similar to the reference patent according to the embodiment of the present invention.

Hereinafter, various embodiments of the present invention will be described with reference to the accompanying drawings. It is to be understood that the embodiments and terminology used herein are not intended to limit the invention to the particular embodiments, but are to be construed to cover various modifications, equivalents, and / or alternatives of the embodiments. In connection with the description of the drawings, like reference numerals may be used for similar components. The singular expressions may include plural expressions unless the context clearly dictates otherwise.

1 is a conceptual diagram for explaining a similar patent extraction system according to the present invention.

In general, in order to obtain the widest range of populations in a patent or invention search, a search formula is roughly prepared, and then various patent documents having various application technologies or heterogeneous features are collected. As a result, it is inevitable that noise is included in the population and the process of removing noise or extracting the patent concerned.

The present invention proposes a method for selecting a reference patent object of interest from a patent database or a population and efficiently extracting a similar patent similar thereto. Here, the reference patent may be a barrier patent obstructing the commercialization of a certain company or a patent or noise patent of interest. The important thing is to automatically search for and extract patents similar to pre-established reference patents. In the embodiment of the present invention, the case where the reference patent is a barrier patent will be described as an example.

The method of extracting similar patents is to collectively group the patent documents having high similarity by using standard patent, keyword frequency and semantic similarity analysis using Shannon information entropy.

The present invention aims at proposing a system for automatically discovering prior patents with potential to disable them, from a patent database, on the assumption that a given barrier patent exists. The proposed system consists of 1) an interest group setting module that extracts a keyword (hereinafter, also referred to as 'a word of interest') for a barrier patent and sets a technical viewpoint to be analyzed by the analyst, 2) An information entropy calculation module for analyzing how much the information of the interest group is described about an arbitrary patent, and 3) a module for calculating the semantic patent similarity degree between the barrier patent and the arbitrary patent using the topic modeling .

2 is a block diagram illustrating a similar patent extraction system according to an embodiment of the present invention.

2, a similar patent extraction system 1 according to an embodiment of the present invention includes a patent database 10, a reference patent setting unit 110, an information processing unit 130, and a similar patent extracting unit 150 The information processing unit 130 may include a keyword recommendation unit 131, a frequency measurement unit 133, an information amount calculation unit 135, a candidate patent specification unit 137, and a similarity calculation unit 139.

The patent database 10 may be a database storing patent information for each country or a patent list (Excel) collected through a keyword search formula by a patent document search system. The patent document search system may be a search system such as Google Patent, Delphi-in, Kipris, WIPSON, WINTELIPS, NDSL, and the like.

The reference patent setting unit 110 is configured to receive and set at least one of the registration number of the reference patent, the irradiation interval considering the filing date of the reference patent, the core keyword, and the IPC classification code to which the reference patent belongs. The keyword is composed of at least one or more keywords, and the keyword is used for information entropy calculation of a patent document to be described later.

The keyword recommendation unit 131 automatically extracts the keyword from the patent and provides it to the user when the registration number of the reference patent is input. There are two methods of automatically extracting keywords from a document: WordNet (https://wordnet.princeton.edu/), which is a word dictionary, and using a learned model based on deep learning. In this system, We use AlchemyAPI (http://www.alchemyapi.com/), an extractable deep-run-based solution. AlchemyAPI is a tool that can analyze images, voice, text, etc., contained in IBM's artificial intelligence-based data analysis platform, Watson. Especially, it supports various functions such as object recognition, emotional analysis, and abstract in text. The system extracts keywords having importance in reference patent by using Alchemy API 's keyword extraction function. In order to improve the accuracy of the keyword selection, the user directly grasps the content of the reference patent and selects an appropriate keyword.

The frequency measurement unit 133 is configured to extract a set keyword of the reference patent for each patent document from a plurality of patent documents stored in the patent database 10 and to measure the appearance frequency of the extracted keyword in the patent document.

The information amount calculation unit 135 is configured to calculate the information entropy of each patent document by summing the appearance frequency probabilities of the predetermined keywords in the patent document with respect to the number of occurrences of all words in the measured patent document. The keywords set in the above can be set to different weights for each keyword and reflected in the information entropy calculation.

Information entropy is an important concept of information theory, in which uncertainty is measured. That is, it has a high information entropy value in a high uncertainty state and a low information entropy value in a low uncertainty state. For example, a coin throwing event has lower uncertainty and information entropy values than six cases where there are two cases of uncertainty, that is, events that can occur, than the case of throwing a die. Also, the amount of information of one system changes according to the probability of occurrence of each event in the same situation. In other words, assuming that the probability of occurrence of each event is different in the same situation with the number of events, if the occurrence probability of each event is the same, prediction of the event becomes more difficult, and in this case, the highest information entropy value is obtained.

Such information entropy can be measured using the Shannon entropy algorithm of the following equation (1) for the discrete probability distribution. Using the mathematical expression, the amount of information about one system can be expressed as a numerical value, which means the degree of information diversity of the system. That is, since the probability that a patent document includes information desired by the user, that is, a word of interest evenly, is high, the information entropy of the patent document is calculated by measuring the amount of information on the word of interest.

Where i is the word of interest and pi is the probability value of the word i of interest.

The candidate patent selection unit 137 is configured to select and provide at least one candidate patent similar to the reference patent based on the calculated information entropy. Candidate patent selection criteria can be determined by preset reference value or number of orders.

The similarity calculation unit 139 calculates a subject-specific probability distribution based on the subject analysis of each patent document based on the word frequency information on the plurality of patent documents generated by the frequency measurement unit 133, And is configured to calculate the degree of similarity with the selected reference patent using the probability distribution.

Here, as shown in FIG. 3, the similarity calculating unit 139 may include a subject probability distribution calculating unit 140, a similarity calculating unit 142, and a similar patent clustering unit 144. The subject probability distribution calculating unit 140 is configured to calculate a probability distribution in which each patent document belongs to each topic through the subject modeling algorithm for the word frequency information generated by the frequency measuring unit 133, 142 compares the subject probability distribution of the patent document selected as the standard patent with the subject probability distribution of the other comparison target patent documents in the candidate patent application management section 137 and determines whether the patent document selected as the reference patent and the other comparison target patent document And the similar patent clustering unit 144 is configured to cluster and provide patent documents similar to the reference patent based on the similarity calculated by the similarity calculating unit 139. [ Here, the comparison target patent document may be a candidate patent selected through the candidate patent selection unit 137. [

As the subject modeling algorithm, any one of LDA (Latent Dirichlet Allocation), LSA (Latent Semantic Analysis), and PLSA (Probabilistic Latent Semantic Analysis) may be used, and the similarity measurement algorithm may be Hellinger distance, cosine similarity, and Jaccard similarity coefficient may be used. The above-described algorithms are well known in the art and will not be described in detail.

The similar patent extraction unit 150 extracts similar patents having a text structure similar to that of the reference patent according to the similarity information for each patent document calculated by the similar patent clustering unit 144. [ Here, the similar patent extracting unit 150 may extract similar similar patents similar to the reference patent by comparing the calculated similarity degree information per patent document with a preset threshold value. At this time, the similar patent can extract the bibliography such as the patent number (registration number, application number, etc.), the name of the invention, the applicant information, and provide it on the screen 170 of the client.

As described above, in the present invention, the information entropy for each patent document is obtained using the keyword of the reference patent, and the candidate patent is selected based on the information entropy for each patent document. The text similarity degree between the selected candidate patent and the reference patent is recalculated, and the similar patent is finally extracted according to the calculated similarity value and the information entropy. Of course, similar patents can be selected and extracted using only information entropy.

Here, the keyword is used to calculate information entropy for each patent document, the similar patent is selected based on the information entropy of each patent document, the similarity is measured between the subject probability distribution of the reference patent and the subject probability distribution of another patent document .

In addition, the Shannon entropy algorithm can be used to calculate the information entropy of each patent document, and Latent Dirichlet Allocation, Latent Semantic Analysis or Probabilistic Latent Semantic Analysis algorithms can be used to obtain the subject probability distribution of each patent document , Hellinger distance, Cosine similarity or Jaccard similarity coefficient algorithm can be used to measure the similarity between patent documents.

A detailed operation process of the interest document extraction system configured as above will be described with reference to FIG. 4 to FIG.

4 is a flowchart illustrating a similar patent extracting method as an interest document according to an embodiment of the present invention.

First, the user connects the patent extraction system 1 by driving the program with the similar patent extraction program installed on the client. As shown in FIG. 5, the user interface of the extraction program is provided with an input window for inputting reference patent information such as reference patent number (registration number or application number), irradiation interval, keyword, IPC, etc. (S10, S20).

When the registration number of the reference patent is input and the patent selection button located at the upper right is clicked, the extraction system 1 extracts the basic information of the reference patent from the patent database 10, (S30). Then, the information processing unit 130 transmits document information (abstract, claim range, etc.) of the reference patent to an external keyword extraction system (not shown) such as Alchemy API to recommend the keyword and display the recommended keyword in the keyword recommendation window You may. Since the keyword extraction system is a known technology, a detailed description thereof will be omitted.

The user searches for a similar patent stored in the patent database 10 on the basis of the keyword of the reference patent by referring to a keyword recommended by the keyword extraction system or by arbitrarily selecting a keyword (S40). In the reference patent information, the investigation section refers to the examination of patents disclosed before the filing date of the reference patent, and the IPC code means that the patent having the IPC is examined. By setting the IPC code and the irradiation interval in this manner, the patent to be researched can be arbitrarily adjusted, and the calculation speed of the extraction system 1 can be reduced as needed.

Then, the frequency measurement unit 133 of the extraction system 1 extracts a set keyword included in a plurality of patent documents stored in the patent database 10, calculates the occurrence frequency of the keyword, Frequency information can be generated. Here, the appearance frequency (tf) of the keyword is a value indicating how often a specific word appears in the patent document, and can be calculated by adjusting the frequency value of the word according to the length of the patent document. For example, the frequency measurement unit 133 may use the frequency of appearance as the frequency of occurrence of the word tf, but may also be calculated by the following equation (2). At this time, the word having the highest occurrence frequency in the patent document will have a value of '1', and other words will have a value smaller than 1.

Where t is any word, d is any patent document, w is any word in patent document d, and f (t, d) is the frequency of the word t in patent document d.

The information amount calculating unit 135 can calculate the information entropy of each patent based on the set keyword. The keyword is a keyword related to the reference patent and can be set by the user, and the weight may be set differently for each keyword. Here, the weight means that weights can be applied to the occurrence frequency of the keyword.

Specifically, the information amount calculating unit 135 can calculate the information entropy of the keyword by using the frequency of the word included in the keyword and the frequency of the word not included in the keyword included in the patent document. The frequency information of each word can be used as the frequency information measured by the frequency measuring unit 133.

Here, the information entropy can be calculated by the following equation (3).

Where, n: mean value of the required amount of information of each of the patent documents, k: number of classification of the keyword, h _i: a probability of occurrence of each keyword, the word that a corresponds to the keyword i of the total number of word appearance in the patent document Occurrence frequency probability.

For example, in the case where the keyword of the keyword is three in Stereo, Lithography and 3D, the frequency of words included in the keyword of the specific patent document and not included in the keyword is as shown in Table 1 below.

Here, the hi of Stereo in a specific patent document may be a value obtained by dividing the frequency of appearance of Stereo by 50, which is the appearance frequency of the entire word. Based on this, if the keywords (k1 to k3) of Table 2 are applied to Equation (3), the information entropy of the patent document can be '1.08'.

The matrix of Table 1 occurs for each patent document. In the present invention, words other than the keyword list are treated as one word to form a word group (non-keyword), thereby maximizing the composition of the information amount for the core words (keyword).

Next, the candidate patent selecting unit 137 selects a candidate patent similar to the reference patent based on the information entropy of each calculated patent document and a predetermined reference value (information entropy) (S50). Here, only when the information entropy of each patent document is larger than a preset reference value, the patent document will be selected as a candidate patent similar to the reference patent. The selected candidate patents can be provided in the order of information entropy as shown in FIG. At this time, the candidate patent information may be displayed with the patent number, the name of the invention, and the information entropy (component similarity value). Here, the reference value may be set to rank rather than information entropy. For example, when the reference value is set to 500, the first to the 500th information entropy ranking is selected as the candidate patent.

Similar patents can be extracted using only information entropy as needed.

For example, the information amount calculating unit 135 calculates information entropy for the keywords using the keywords selected by the reference patent setting unit 110 for all patents. As the stored keywords in the reference patent setting unit 110 are uniformly present in the patent, they have a high entropy, and when they are biased to one or two of some keywords, they have a relatively low entropy. From the viewpoint of the present system, patents in which all the keywords that technically have an interest in the analysts exhibit a high entropy, and patents in which only a specific keyword has a relatively low entropy. Therefore, a patent having a high information entropy can be regarded as a patent related to the contents of a barrier patent which the analyst desires to invalidate.

On the other hand, a patent with a low information entropy is considered to be a patent that can not invalidate a barrier patent by including only a part of what the analyst wants to invalidate. First, this module combines the text information related to the content of the patent, such as title, abstract, full claim, and detailed description among the information of each patent. The number of appearing words for the total text combined for each patent is obtained, and the frequency of occurrence of the word of interest is calculated therefrom. Therefore, each patent can be represented by a vector of the interest, and the information amount of each patent can be obtained by using the expressed patent-interest vector and Equation (3).

Equation (3) calculates the amount of information that is interested in the modified information amount calculation formula. Therefore, by using Equation (3), it is possible to calculate an index indicating how colorful the information that the analyst is interested in, excluding the unnecessary information. After calculating the amount of information, the present candidate patent selection unit 137 sorts the information amounts in descending order to select a predetermined number of candidate patents, and displays them to the user. The degree of similarity is calculated by the degree-of-similarity calculating section 139, which is the next module, without further processing on the selected candidate patents.

The similarity calculation unit 139 calculates a topic probability distribution by analyzing a topic of each patent document based on the word frequency information about the selected candidate patents, The similarity degree of the text with the patent is calculated (S60).

The similarity calculating unit 139 may include a subject probability distribution calculating unit 140, a similarity calculating unit 142, and a similar patent clustering unit 144.

The subject probability distribution calculating unit 140 may consider the potential relationship between patent documents by analyzing the vector similarity between patent documents using a subject modeling algorithm to extract similar patents similar to the reference patent. If a plurality of candidate patents are selected by the candidate patent selecting unit 137, the similar patent clustering unit 144, for example, determines whether or not each candidate generated through the frequency measuring unit 133 A subject modeling algorithm can be applied to the weighting of each word in a patent, so that a probability distribution of each patent document belonging to each subject can be calculated. For example, the subject modeling algorithm may be a Latent Dirichlet Allocation (LDA) algorithm.

The LDA algorithm is a tool generally used in classification according to a topic of a patent document as a known technology, and will be briefly described with reference to a MATLAB code as shown in FIG. Basically, the LDA algorithm begins on the premise that a patent document is a bundle of words, a patent document has a specific subject, and a subject is shared by patent documents. For example, if there are eight patent documents as shown in FIG. 8A, and each patent document is made up of a total of 16 words, it can be displayed in color according to the occurrence frequency of words. For example, the higher the green color, the higher the occurrence frequency of words, and the higher the blue color, the lower the appearance frequency. In the case of the seventh patent document of FIG. 8A, only the word of the matrix (3, 4) appears at a significantly high frequency.

FIG. 8B shows a distribution on a topic, which has eight topics (Topic 1 to Topic 8) and indicates which words have a topic. That is, the subject means a distribution of words. For example, in the case of topic 1, the frequency of appearance of words from the first to the fourth ((1,1) to (1,4)) is high. Therefore, when LDA is applied to the weight for each word in each patent document, a similar pattern as shown in FIG. 8B is obtained, thereby finding each topic.

FIG. 8C shows the distribution of the subject matter for each patent document. Red is used to make the data, and blue is found through the LDA. In other words, when the order of the subject corresponding to the x-axis is ignored, it can be seen that the LDA can eventually find the subject of each patent document similarly.

In addition to the Latitude Dirichlet Allocation (LDA), Latent Semantic Analysis (LSA) or Probabilistic Latent Semantic Analysis (PLSA) may be used as a subject modeling algorithm.

The number of topics (corresponding to the 'technical field') can be preset in the extraction system 1, and it has been found most appropriate to classify the number of subjects into 8 to 10 over several tests. Therefore, as shown in Table 3 below, the subjects are first classified into 9 categories, and LDA is applied to a large number of patent documents to be classified into respective subjects.

As shown in Table 2 below, the number of patent patent documents belonging to each topic derived as a result of the LDA and major keyword information constituting each topic can be displayed. Using the main keyword information corresponding to each topic, It is possible to judge the characteristics. For example, in Topic 1, it can be inferred that this is a technical clustering of adhesive particulate bonding.

In this manner, the subject probability distribution calculating unit 140 extracts the subject keywords and calculates the probability distribution belonging to each subject in each patent patent document as shown in Table 3 below.

Then, the similarity calculator 142 can calculate the similarity between the patent documents by performing the similarity analysis between the patent documents using the probability distribution belonging to each subject. The similarity degree is calculated by using Hellinger distance, Cosine Similarity ) And a Jaccard similarity coefficient.

For example, the similarity calculation unit 142 calculates the similarity between the reference patent selected by the candidate patent selecting unit 137 and the other comparison target patent document by the Hellinger distance H (P, Q) in the following equation (4) The degree of similarity can be calculated.

Here, i is the subject, k is the number of subjects, pi is the subject probability distribution of the reference patent document, and qi is the subject probability distribution of the compared patent document.

The result value (H (P, Q)) calculated by Hellinger's distance has a value between 0 and 1. The smaller the result is, the greater the degree of similarity between the two patent documents. The larger the result, Therefore, the final similarity value S (P, Q) is obtained by subtracting the result value H (P, Q) of the hellenistance to 1 as shown in the following equation (5) to facilitate intuitive understanding, Can be used as the similarity value.

Table 4 below shows an example of a similarity matrix for each patent document calculated by Equations (4) and (5).

The similar patent clustering unit 144 groups and provides similar patents similar to the reference patent based on the similarity information calculated by the similarity calculating unit 142. [ At this time, if there are a plurality of reference patents, it is also possible to group similar patents for each reference patent.

The similarity calculation module, which is the last module of the system proposed in the present invention, computes the text structural similarity with the barrier patent based on the topic modeling for the candidate patents judged as having potential barrier invalidation based on the information amount in the previous step do. Topic modeling is an algorithm that can identify potential topics based on the appearance of words in a set of documents and infer which topics each document contains by using the association between topics and words. The final computation of topic modeling can be used to obtain a probability distribution between a document and a topic. In this module, the similarity of the text structure between patents is calculated using the probability distribution of the topics of each patent. Therefore, patents composed of texts similar to standard patents will show a high degree of similarity, and as the words constituting the content of the patent are different from the barrier patents, the subject covered by the patent will change. Various methods such as cosine similarity, Jaccard coefficient, and Euclidean distance can be used to calculate the similarity. However, in this system, considering that the patent topic distribution is a probability distribution, The Hellinger distance used to calculate the similarity between the probability distributions is used. Therefore, the final calculation of this similarity calculation module is the hellinger distance between each candidate patent and the barrier patent, and the formula of the hellinger distance is shown in Equation (4). In addition, similarity is closer to 1, which means similar, closer to 0, meaning different. On the contrary, Hellinger's distance is closer to 0 and closer to 1, 5 to convert the hellinger distance to have the meaning of similarity.

Then, the similar patent extracting unit 150 may filter or remove the clustered reference patent and the similar patent through the similar patent clustering unit 144 based on a preset threshold value. That is, the similar patent extracting unit 150 may compare the similarity information between the reference patent and the comparative patent document with a predetermined threshold value, and recommend interest documents similar to the reference patent to recommend.

In other words, as shown in FIG. 9, in the extraction system 1, a patent document similar to the reference patent is clustered and provided, and information on the similarity degree between similar patents similar to the reference patent is provided.

9, the information of the similar patents extracted by the similar patent extracting unit 150 and the similarity values are stored by rank, as shown in FIG. 10, Can be displayed through the similar patent output section (170). The similarity value may be an information entropy-based component similarity value and a text similarity value calculated by the similarity degree calculating unit 139.

Prior to the similarity analysis process, the user can select the analysis range (abstract, claims, detailed description, etc.) of the patent specification in FIG. 5 when analyzing the similarity with the reference patent. In the case of investigating similar patents, the entire patent specification including the detailed description should be used. In the case of investigating similar patents in the claims unit, only the whole claims should be used so that the results that meet the purpose of the analyst can be derived . However, as the scope of analysis increases, the time required to analyze the similarity increases.

After the similarity analysis, the ranking of the candidate patents can be determined based on various criteria. However, similarity computed by this similarity module is structural similarity according to occurrence frequency of words in text structure rather than technical similarity. Therefore, rather than judging the possibility of similar technology by itself, similarity degree is centered on information amount, Is preferably used. Therefore, the ranking is based on the amount of information in the final ranking, and the degree of similarity can be used as a reference for future analysts to check individual patents.

It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory and are not intended to limit the invention to the particular forms disclosed. It will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention as defined by the appended claims.

1: Patent extraction system 10: Patent database
110: Reference patent setting unit 130: Information processing unit
131: Keyword recommendation unit 133: Frequency measurement unit
135: information amount calculating section 137: candidate patent specification section
139: similarity calculating unit 140: subject probability distribution calculating unit
142: similarity calculation unit 144: similar patent clustering unit
150: similar patent extracting unit 170: similar patent outputting unit

Claims (14)

(a) selecting at least one reference patent as a target of interest among a plurality of patent documents, and then setting patent information including a main keyword for the reference patent;
(b) calculating information entropy for the patent on all patents recorded in the patent population or the patent database using the keyword set in the step (b); And
(c) selecting and providing at least one candidate patent similar to the reference patent based on the information entropy calculated in the step (c). The method according to claim 1,
(d) analyzing the topic of each candidate patent based on the word frequency information of the selected candidate patents, calculating the probability distribution of themes, and using the topic-specific probability distribution of each candidate patent, Calculating a text similarity degree of the text; And
(e) extracting and providing a similar patent having a text structure similar to the reference patent according to the similarity calculated in the above. The method according to claim 1,
Wherein the patent information includes at least one of a patent number, an irradiation interval, and an IPC classification code of the reference patent in the step (a). The method of claim 1, wherein the information entropy (n)
A similar patent extraction method calculated by Shannon entropy, which is the following equation.
[Equation]

Here, n represents the value of the necessary information amount of each patent document, k represents the number of classified keywords, and h _{i represents the} probability of occurrence of each keyword, a word corresponding to the keyword i for the total number of word occurrences in one patent document The probability of occurrence is. The method according to claim 1,
Wherein the reference patent is a barrier patent or a patent of interest that is an obstacle to commercialization. The method of claim 2,
The step (d)
Calculating a probability distribution in which the candidate patent selected by applying a topic modeling algorithm to the word frequency information belongs to each topic;
Calculating a degree of similarity between the reference patent and the comparative candidate patent using the subject-specific probability distribution of each candidate patent and the subject probability distribution of the reference patent; And
And clustering and providing patent documents similar to the reference patent based on the calculated similarity. The method of claim 6,
The subject modeling algorithm is any one of Latent Dirichlet Allocation (LDA), Latent Semantic Analysis (LSA) and Probabilistic Latent Semantic Analysis (PLSA), and the similarity is determined by Hellinger distance, Cosine Similarity, And a Jacquard similarity coefficient. The method according to claim 1, The method of claim 2,
The degree of similarity in the step (e)
Wherein the probability distribution of each patent document belonging to the subject is calculated by applying the Hellinger distance (H (P, Q)).
[Equation]

Here, k being a subject, t is the number of topics, p _k are probability distributions of the topics based on patent, q _k is _the probability of the subject to be compared patent document distribution. The method of claim 8,
Since the value of the hellinger distance H (P, Q) is equal to 0 when the probability distributions of the patent documents are the same, the similarity can be expressed by the following formula (S (P, Q), similarity) Determine a similar patent extraction method.
[Equation]

A reference patent setting unit for setting patent information including a main keyword for a reference patent after selecting at least one reference patent as a target of interest among a plurality of patent documents;
A frequency measuring unit for extracting key keywords for each document from a plurality of patent documents and measuring frequency of occurrence in the document;
An information amount calculating unit for calculating information entropy of each document through a document occurrence frequency probability of a main keyword with respect to the number of occurrences of all words in the document measured by the frequency measuring unit; And
And a candidate patent selection unit that selects at least one candidate patent similar to the reference patent based on the information entropy calculated in the above. The method of claim 10,
Based on the word frequency information of the candidate patents selected above, a topic analysis of each candidate patent is performed to calculate the subject probability distribution, and the text similarity with the reference patent is calculated using the subject-specific probability distribution of each candidate patent A similarity degree calculating unit for calculating a similarity degree; And
And a similar patent extracting unit for extracting and providing similar patents having a text structure similar to the reference patent according to the similarity calculated in the above. The method of claim 10,
The information entropy (n)
A similar patent extraction system calculated by Shannon entropy, which is the following equation.
[Equation]

Here, n represents the value of the necessary information amount of each patent document, k represents the number of classified keywords, and h _{i represents the} probability of occurrence of each keyword, a word corresponding to the keyword i for the total number of word occurrences in one patent document The probability of occurrence is. The method of claim 11,
Wherein the similarity-
A subject probability distribution calculating unit for calculating a probability distribution of each patent document belonging to each subject through a subject modeling algorithm with respect to the generated word frequency information;
A similarity calculation unit for comparing a subject probability distribution of a patent document selected as a reference patent in the reference patent setting unit with a subject probability distribution of another comparison target document to calculate a similarity degree between the reference patent and another patent document; And
And a similar patent clustering unit for clustering and providing patent documents similar to the reference patent based on the similarity calculated by the similarity calculating unit. The method of claim 11,
Wherein the similarity-
Wherein the probability distribution of each patent document belonging to the subject is calculated by applying the Hellinger distance H (P, Q) of the following equation.
[Equation]

Here, k being a subject, t is the number of topics, p _k are probability distributions of the topics based on patent, q _k is _the probability of the subject to be compared patent document distribution.

Images