KR20120038575A - Method of automatic patent document categorization adjusting association rules and frequent itemset - Google Patents

Abstract

PURPOSE: A method for automatically categorizing patent documents using a frequency item list and association rules is provided to automatize processes of manually categorizing IPCs of applied patent documents, thereby reducing manpower. CONSTITUTION: Transactions of patent documents are configured by extracting content included in an identification item from mother group patent documents whose class is determined according to an IPC(International Patent Classification). A keyword is extracted from content included in the transactions through a morpheme analyzer(S202). A frequency item list and association rules about the class are generated from the keyword(S206). The frequency item list and the association rules are applied to categorized patent documents to classify the class(S208).

Description

Method of automatic patent document categorization adjusting association rules and frequent itemset}

The present invention relates to a method for classifying patent documents, and more particularly, to a method for automatically classifying patent documents using frequent items and related rules, which can automatically classify patent documents based on international patent classification criteria using frequent items and related rules. It is about.

As the Korea-US Free Trade Agreement (FTA) concludes, the protection of patents is strengthened in accordance with the extension of the right to intellectual property rights consisting of patents, trademarks, and copyrights. have.

Patent information refers to information related to industrial property rights and information about the technical content and rights claimed, personal information of applicants and inventors, and other bibliographic matters. As the industry is advanced, complicated, and diversified, a tremendous amount of information on patent technology is pouring in. In order for a company to survive in a changing industrial society, this information must be appropriately reflected in corporate management strategies.

Currently, the Patent Office of major countries such as the United States, Japan, and Europe, including Korea, operates a search site for searching such patent information on the Internet, and there are also a large number of paid search sites developed for commercial purposes. Among them, the patent classification system based on the International Patent Classification (IPC) classification system is a classification system operated by international agreements such as the United States, Europe, International (WIPO-PCT), and Japan.

However, until now, it is taking a lot of time because IPC classification is performed by person rather than by machine. In addition, the number of classifications is considerable, and the classification is required to understand the entire detailed description including the claims in order to give a significant stress to the classifier.

Accordingly, there is a demand for a method for automatically and accurately classifying patent documents.

An object of the present invention is to provide a method for automatically classifying patent documents using frequent items and related rules that can automatically classify applied patents according to international patent classification.

In addition, an object of the present invention is to provide a method for automatically classifying patent documents using frequent items and association rules that can be effectively classified using the 'Apriori' algorithm in classifying the applied patent.

In order to achieve the above object, the method according to the present invention extracts the contents contained in at least one identification item from the N population patent documents, each class is determined according to the international patent classification to form a transaction for each patent document step; Extracting a keyword from a content included in each transaction through a stemmer; Generating frequent items and association rules for the corresponding classes from the extracted keywords; And classifying the class by applying the classification target patent document to the generated frequent items and association rules for each class.

The extracting of the keyword may further include removing stopwords from the contents included in each transaction.

At this time, the identification item to be extracted is characterized in that any one or more selected from the application number, IPC classification, the name of the invention, the representative claims, key keywords, and identification in the technical field.

The extracting of the keyword may further include removing duplicate keywords from the extracted keywords for each patent document.

The extracting of the keyword may further include replacing a similar keyword with a representative word among the extracted keywords for each patent document.

In this case, the frequent items may be configured by a combination having an occurrence map of at least the minimum support by the 'Apriori' algorithm.

In addition, after generating the frequent item and the association rule, weights may be assigned to duplicate keywords among the classes in consideration of the number of occurrences of the keyword.

According to the present invention, it is possible to reduce the consumption of manpower by automatically classifying what was conventionally done manually by IPC classification of the applied patent document, there is an advantage that can be classified within a short time.

In addition, according to the present invention, an 'Apriori' algorithm is applied when classifying a patent document, and weighting is based on support and reliability, thereby increasing the accuracy of patent classification.

1 is a block diagram illustrating the concept of patent document classification according to an embodiment of the present invention.
2 is a flow chart showing a patent document automatic classification procedure using frequent items and association rules according to an embodiment of the present invention.

In the present invention, the claim 1 and the name of the invention for each of the eight classes of patent documents A to H of the already completed classification system are extracted for each document to remove stopwords, and frequent occurrences generated from keyword data combinations extracted using a morphological analyzer. We propose a method of automatically classifying and recommending patent documents using a set of items and association rules.

Document classification is the task of grouping similar documents together by assigning them to one or more categories related to the contents of the documents by defining several classification categories.

As described above, the classification of patent documents has been performed by humans in most cases, but as the number of patents applied increases and the scope and form of the patents are diversified, the need for automation of document classification is widely recognized. Various studies have been made.

Unlike a statistical classification method that specifies a classification category by simply using the frequency of occurrence of keywords appearing in a document, the present invention uses the Apriori algorithm, which is an association rule analysis technique, based on the contents of a patent document and documents in the same international patent classification. We extract a set of keywords that are related to each other, generate a classification rule with keywords that are semantically representative of each category, and use this to predict international patent classification of new documents. In addition, the present invention proposes a data preprocessing method for efficient keyword generation, and verifies performance through data mining experiments based on support and reliability.

In the present invention, IPC automatic classification is recommended through the generation of frequent items and association rules associated with each class using data mining techniques for texts of documents of eight classes of inventions up to A-H and the texts of the representative claims.

The present invention extracts the name of the invention, representative claim data from the patent information already registered IPC classification. Next, we use the representative algorithm of data mining, 'Apriori', which is an association rule exploration technique, to remove the redundant words that only patent documents have, and then apply the morphological analyzer to remove the redundant words. We propose an automatic classification recommendation method. Detailed implementation method of the present invention in this regard will be described in the following order.

First, the basic concepts and techniques of association rule exploration techniques, Apriori algorithm, morphological analysis, maximum frequent itemsets, keyword search techniques, etc. required for the subject of the present invention will be described. Next, in order to generate data for exploration of association rules, data conversion, refinement, complementary processing and morphological analysis of the name of the invention and the representative claims, word extraction methods that eliminate the redundancy between words in a transaction (document), each Applying 'Apriori' algorithm, the correlation analysis algorithm, to the list of words with redundancy removed for each transaction, to find the frequent items for each minimum map, and the method for obtaining the maximum frequent item sets by removing the redundancy between the item sets from each set of frequent items. Explain.

Finally, in order to verify the efficiency of IPC automatic classification, the results of the present invention are examined through various experiments to grasp the frequent items, the results of the association rules, and the hardware processing efficiency such as CPU and memory through comparison with the documents already classified. Verify efficiency

Patent information refers to information about the technical content and rights claimed, personal information of applicants and inventors, and other bibliographic details. In the present invention, the expression "patent information" is used for convenience, but information about a utility model whose technical content is the subject of rights, like the patent, is also included in the scope of the patent information, and it is obvious that the present invention can be applied equally. In a broader sense, the term patent information may be used as a concept including all information on four industrial property rights such as patents, utility models, designs, and trademarks. On the other hand, such patent information is used as the right analysis technology data for understanding and analyzing the prior art in research or technology development.

The International Patent Classification (IPC) to which the present invention is applied has been using different classification systems in each country such as the US (USPC), Japan (JPC), and Europe (ECLA), but in 1968, as the internationally unified patent classification system is needed. Was introduced.

The international patent classification has been introduced to systematically organize patent documents so as to provide easy access to technology and rights information contained in patent documents, and to selectively distribute information to all users of patent information. It can also be used to investigate publicly known technologies in a given technical field, and can also be used to generate industrial property statistics to evaluate technological developments in various areas.

Such an International Patent Classification (IPC) consists of a hierarchy of sections, classes, subclasses and main groups or subgroups. <Table 1> is a table showing the structure of the international patent classification (IPC) F16K 1/00 (or 1/02) as an example.

Classification code F - 16 K 1/100 1/20 division section Subsection class Subclass Main group Subgroup Classification Mechanical engineering General industry Mechanical element valve Lift valve Screw spindle

Meanwhile, the section items are divided as follows.

-Section breaks (section titles)

Section A: Necessities of Life

Section B: Processing Operations, Transportation

Section C: Chemistry, Metallurgy

Section D: Fiber, Paper

Section E: Fixed Structure

Section F: Mechanical Engineering, Lighting, Heating, Weapons, Blasting

Section G: Physics

Section H: Electric

In the present invention described below, the classification by section in the IPC classification is described as an example, and the same may be applied to the classification of subsections, classes, subclasses, main groups, and subgroups, which are subclasses.

On the other hand, document classification is the task of grouping documents by assigning each document to be classified into the most suitable category under the defined classification system. In the past, document sorting was possible by hand because the amount of information to be handled was small. However, in the modern situation where a large amount of information is generated every day, such as on the web (WWW), manual sorting of all documents is impossible. Therefore, research on automatic classification of documents is actively conducted.

However, since patent documents have some characteristics different from general documents and web pages, it is difficult to classify them with existing methods and various characteristics should be considered. The features of the patent document are summarized as follows.

1. IPC is classified according to the name of the invention and claim 1.

2. Abstracts, claims, technical fields, and names of inventions are semantically related.

3. In order to broaden the scope of the invention, some broad and general terms are used in the claims.

4. Includes many technical terms. Other patents use or define terms that are not used.

5. Patent documents vary in length.

Among the above features, the present invention performs patent classification on patent documents focusing on the first and second features.

The patent document targeted by the present invention consists of six large areas such as bibliographic information, summary, representation, claim, specification, and drawings as shown in Table 2 below. The specification is again made up of details such as the technical field, the detailed description of the invention, examples, and a brief description of the drawings. The technical field contains information that enhances the technical background and understanding of the technology, which may help categorize more than other parts. Therefore, if we consider these subdivided factors as the qualities of classification, we can get good performance from patent classification.

1. Bibliography Application Information, Disclosure Information, Applicant, Inventor 2. Summary Full summary of patent documents 3. Representative Representative Drawing Information 4. Scope of Claims Claims 1 to n 5. Specification Brief Description of the Technical Field, Invention, Embodiment, and Drawings 6. Drawings Drawing 1, Drawing 2, Drawing 3, ...

As such, the attributes of the patent data specification consist of bibliographic information such as application number, publication number, publication date, registration number, and registration date, and the name of the invention, abstract, claims, and detailed description. Basic information of IPC classification.

As described above, the patent data has characteristics different from those of general web data and data used in daily life. Therefore, according to an exemplary embodiment of the present invention, purified data that has undergone appropriate processing is required for the user to intuitively recognize and to facilitate the analysis. To this end, the present invention performs the data conversion, data purification and correction step for the name of the invention of the patent data and claim 1.

In other words, a patent file is extracted to make one document into one transaction. The application number, IPC, the name of the invention, and claim 1 are extracted from the actual document referred to by each transaction, and constitute one transaction. After that, the stopwords of each document are removed and the document is composed of one transaction through the stemmer. In this case, the keywords extracted from the document for each transaction are not duplicated.

DETAILED DESCRIPTION Hereinafter, a detailed description of a preferred embodiment of the present invention will be described with reference to the accompanying 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.

1 is a block diagram illustrating the concept of patent document classification according to an embodiment of the present invention. First, the plurality of patent documents 100 that have been felt are applied to each identification item such as the application number 111, the IPC 112, the invention name 113, the drawing 114, the abstract 115, and the claim 116. Extract to construct transaction 110. Thereafter, the classification algorithm 120 is applied to the keywords extracted from the contents of the respective identification items to generate the frequent items 121 and the association rules 122. Next, the score 131 and the weight 132 are set on the frequent items 121 and the association rule 122 based on the support and the reliability using the association rule set 130 according to the embodiment of the present invention. do. As described above, the target patent document is finally classified by applying the frequent items and the association rules 122 generated for each classification category to the classification target patent.

Hereinafter, a method of classifying a patent document according to an embodiment of the present invention will be described in more detail.

Meanwhile, the patent data used for the test is based on the KIPRIS search system operated by the Korea Institute of Patent Information, and is registered in January and February 2010.

<Data processing>

More than 200,000 patent data are filed per year, and IPC classification, which is a system of patent documents, is made regardless of registration. In addition, the IPC classification criteria depend on the name of the invention and claim 1, but are actually classified according to the subjective judgment of the classifier. However, in the present invention, the classification system is set based on academic theory.

Accordingly, a process of removing unnecessary words (investigation or specific words) from all items of patent data is established, and a data model is constructed for exploration of association rules required for the present invention. Meanwhile, the items applied in the present invention are application number, IPC classification, name of the invention, claim 1, abstract, key keywords and the like.

In the present invention, the modeling is completed based on the name of the invention as the IPC classification criteria and claim 1, and this is defined as one transaction (S201). That is, one transaction consists of application number, IPC classification, name of invention, claim 1, abstract, key keywords, etc., and these items are parsed through 'Java XML Sax Parser' because they are in the form of one publication in XML format. Proper conversion is necessary to make the process easy to understand and analyze.

<Table 3> is extracted XML attribute information of the Korean registered patent, and <Table 4> is extracted XML results.

 1. Bibliography  <KR_Application_Number> </ KR_Application_Number>  2. Summary  <KR_abstract> </ KR_abstract>  3. Representative  <KR_draw> </ KR_draw>  4. Scope of Claims  <KR_cliaims> </ KR_cliaims>  5. Specification  <KR_description> </ KR_description>  6. Drawings  <KR_figure> </ KR_figure>

Application number IPC Name of invention Claim 1 Key keywords Technical Field --- --- --- --- --- ---

<Purification and correction of patent document data>

Patent data has a variety of information such as abstract, claims, detailed description, technical information, as well as bibliographic information, so the purification process to select only the necessary data (S202). In the present invention, a transaction is defined based on the name of the invention, claim 1, key keywords, and technical information. In addition, due to the nature of patent data, "manufacturing method", "thing", "1", "2", "method", "manufacturing", "thing", "bone", "my", "this", "number" ”,“ Above ”,“ invention ”,“ dragon ”,“ etc ”,“ four ”,“ include ”,“ yi ”, etc., are preliminarily removed (S203). In addition, the present invention defines an important document that is meaningful only for data in which the number of extracted word groups is more than 20, and extracts words specified as key keywords and adds them to the word groups. Due to the nature of the data, words having the same meaning may be marked differently, and a correction operation for representing the words with similar meanings as representative words is performed (S205). Table 5 below shows a table processed by one transaction of a document whose number of words is 20 or more and the corrected work is completed as a representative word.

TID Application number IPC Name of invention Claim 1 Key keywords Transaction number

Stemming and Keyword Extraction

We define this as an important document that is meaningful only for data with more than 20 keyword groups, and represent it as a single transaction. In the result data of applying the Hangul morpheme analyzer, the morphemes with 'Score' of 100 or more include the noun, the pronoun, the rhetoric, and the dependent nouns, so the morphemes with scores of 100 or more are extracted as keywords. At this time, since each morpheme is duplicated in one document and duplicated in another document, keywords are extracted so as not to be duplicated in one transaction (S204).

The implementation of the Apriori algorithm used in the experiments of the present invention consists of one transaction item set in one row of an Oracle (10g) database, created with varchar2 data type, and applied to 255 characters or less. It was determined that one keyword could be a criterion for classification, and the number of keywords was not limited. In the present invention, a keyword list in which a keyword in a transaction is not duplicated is defined as a 'duplicate exclusion keyword'. <Table 6> is a result table of duplicate exclusion keyword lists extracted by applying Hangul morpheme analyzer to documents included in each transaction.

T1 Keyword 1 Keyword 2 Keyword 3 Keyword 4 Keyword 5 Keyword 6 Keyword 7 Keyword 8 T2 keyword Keyword 2 Keyword 3 Keyword 4 Keyword 5 Keyword 6 Keyword 7 Keyword 8 ? ? ? ? ? ? ? ? ? - - - - - - - - -

<Frequent item set extraction and association rule generation>

For each keyword of each transaction shown in Table 6, the abstract, claim 1, and the technical field are analyzed by applying the Apriori algorithm, which is an association analysis algorithm. At this time, the minimum support map is changed to obtain frequent item sets for each support degree. In the embodiment of the present invention it was defined as a support of 0.5%. There is redundancy between items because all item sets in the frequent item set are higher than the minimum map. In the present invention, for the purpose of researching the correlation between keywords, a unique keyword of each transaction should exist. A maximum frequency item set (MFI), which is not included in other frequent item sets other than itself, is obtained from a frequent item set having redundancy (S206).

Association Analysis Using Apriori Algorithm

The association rule mining algorithm, Apriori, performs an association analysis in two stages. The first stage constructs a frequent word item by finding a combination that has more than minimum support of transactions. The second step is to find all non-empty subsets of the frequent item set for the frequent item set (L) to generate association rules from the database. For each such subset A, output a rule of the form A-> (LA) if the ratio of Support (L) to Support (A) is at least above minimum confidence. The support of this rule is support (L) and the reliability is support (L) / support (A). The generation of candidate sets in the Apriori algorithm can reduce the number of candidate items by using Apriori-gen to make new candidate item sets. This reduces the time to find the association rule. At this time, the data mining Apriori algorithm applied to the present invention is as follows.

Step 0. Determine the minimum support map s _min .

   k = 1

C₁ = [{i₁}, {i₂}, ..., {i _m }]

L₁ = {c∈C₁ | supp (c) ≥ s _min

Step 1. k = k + 1

C _k formation from L _{k -1} (apriori-gen function)

Step 1-1. (join) Join sets of L _{k -1} to form a set of k- items.

C = L _{k -1} * L _{k -1}

Step 1-2. (prune) When the (k-1) -item subset of C does not belong to L _{k -1} , remove all of it and form C _k . Stop if C _k = Φ.

Step 2. Generate L _k by collecting the support of the set of C _k equal to or greater than the minimum map.

L _k = {c∈C _k | supp (c) ≥ s _min }

In this case, L _k means a candidate k-item set, and C _k Denotes a frequent K-item set.

Apply the result of each transaction of <Table 6> by applying the Apriori algorithm to each of the minimum maps to obtain Frequent Itemsets by each support degree as shown in <Table 7>, and <Table 8 Create an association rule as shown in>.

Keyword 1 (support map)
Keyword 1, Keyword 2 (support map)
Keyword 2, Keyword 3 (support map)
Keyword 1, Keyword 2, Keyword 3, Keyword 4 (support map)
XXX, XXX, XXX, XXX, XXX (support)
XXX, XXX, XXX, XXX, XXX (support)

Keyword1 <-Keyword1, Keyword3 (0.5, 100.0)
Keyword 3 <-Keyword 5, Keyword 6 (0.5, 100.0)
Keyword 4 <-Keyword 5, Keyword 7 (0.5, 100.0)
Keyword 6 <-Keyword 5, Keyword 8 (0.5, 100.0)
Keyword 7 <-Keyword 6, Keyword 8 (0.6, 84.6)
Keyword8 <-Keyword6, Keyword9 (0.6, 84.6)

<Automatic Classification of Patent Documents-Calculation of Accuracy by Creating Frequent Items and Association Rules>

Keywords in the patent data are words representing the contents of the document. Extracting the exact keywords maximizes the efficiency of the automatic patent document classification system. Therefore, in order to overcome the shortcomings of the method of extracting frequent items by simply applying the Apriori algorithm to the keywords existing in the full text of each document, the present invention uses weights based on the support and the reliability based on the association rule rule set. Suggest ways to give. In the present invention, such a method is defined as an accuracy calculation method based on the frequent items and the association rules.

<Calculation of accuracy by deduplication and weighting of keywords between classes>

In patent documents, duplicate keywords must be considered. That is, the increase in the number of occurrences of a keyword increases the probability of being an important keyword of a patent document. Duplicate keywords within one document are important, but keyword overlap with other classes must also be considered. Because it can be a unique keyword or a double keyword.

At this time, the analysis method that can differentiate from the existing words by increasing the accuracy through the elimination of the association rule set of the duplicate keywords and the duplicate keyword item set of duplicate keywords generated between each class is defined as a weight-based keyword extraction method in the present invention. .

The concept of extracting and removing frequent duplicate keywords is shown in Table 9 below.

A class B class C class D class
A
A, B (X)
A1, C1, D1
A, C, D, F

A2
A, B (X)
A1, C1, D1
A1, C2, D2, F2

A3
A, B (X)
A3, C3, D3
A3, C3, D3, F3

A4
A, B (X)
A4, C4, D4
A4, C4, D4, F4

In this case, in Table 9, if the item set is an intersection of class A and class B such as A1, C1, and D1, a weight that can clearly distinguish between A class and B class is necessary. Therefore, in the present invention, the weight is given based on a numerical value twice that of the original value.

Table 10 below shows a concept of weighting.

A class B class A, B (X)
A1, C1, D1
A1, C1, D1, F1 A, B (X)
A1, C1, D1
A1, C2, D2, F2

Referring to <Table 10>, it is possible to clarify the distinction between Class A or Class B by calculating the accuracy by adding the number of A1, the number of C1, and the number of D1 in A1, C1, and D1.

<Patent Document Classification Standards by Accuracy Calculation>

In the experimental example to which the present invention is applied, the data used are divided into two categories, which are applied to the accuracy calculation method by the frequent item set and association rule generation and the accuracy calculation method by keyword deduplication and weighting between classes. The full text of the document is defined as the experimental data, and the full text of the patent document for verifying and comparing the efficiency of the accuracy calculation method is defined as the verification data. In other words, the experimental data are assumed to be 8,000 out of the 10,000 registered IPCs in 2010, which were used to obtain the frequent itemsets, and the validation data are already granted IPCs to evaluate and verify the maximum frequent itemsets. In other words, the remaining 2,000 cases already know the answer.

When an item of a frequent item set of experimental data belongs to a transaction keyword item of validation data, this frequent item set item is defined as a complete match. In addition, in order to find the matching points between the frequent item set and the verification data, the frequent item match degree and the association rule rule match degree are obtained as follows for the abstract, claim 1, and technical background of the patent document.

<Set weights of frequent item sets and item sets>

A set of frequent items that perfectly match the transaction of verification data is extracted for each item set. The weight of the item set is determined by two or three kinds. In the embodiment of the present invention, two weights were defined by the method of 1, 3, 9, 27, 81 and 1, 5, 10, 300, and 500. In this case, FIC is defined as a frequent item set number, Sup is defined as support, Conf is reliability, and FIA is defined as a frequent item set weight.

<Validation Data Matches>

When the number of transactions of the verification data including all the frequent item set keywords is defined as the verification data match number, the verification data match number Score1 may be calculated as in Equation 1 below.

At this time, the item set of the frequent itemsets for each support obtained from the experimental data is composed of keywords and the support of the keywords, and the measure of correlation between these keywords is a single transaction of the keywords and the validation data of the single itemsets of the frequent itemsets. The judgment is made by comparing with the keyword shown in the following. In the embodiment of the present invention, in order to increase the accuracy of the determination of the matching degree between the frequent item set and the verification data, which is obtained from the correlation analysis for each of the minimum support maps, the accuracy obtained by multiplying the frequent item set number and the verification data match size is proposed as a comparison criterion of the present invention. .

<Reliability based on association rule rule, support based on match degree>

It represents the numerical value of the association measure based on the support and reliability of the association rule rollset based on the validation data, and the match degree score2 expressed as a match for each transaction of the validation data is calculated as shown in Equation 2 below. Can be.

<Keyword deduplication and weighting between classes>

In order to ensure the distinction between duplicate keywords generated in all classes, and to ensure distinction of partially generated keywords, a double weight is added to the duplicated classes to increase accuracy (S207). In this case, the addition point may be calculated as follows.

Weight = IF (Each class Score [A1, C1, D1]> Each class Score [A1, C1, D1])

TRUE A Score * 2, FALSE B Score * 2

<Abstract, claim 1, Match degree calculation by sum of each item of technical background>

The classification of patent documents does not depend on any one item, so it is important to increase the accuracy by calculating the sum of each item. This is an important feature of patent documents and can be used as a way to increase accuracy. For each abstract, claim 1, and technical background, a match degree may be calculated as shown in Equation 3 below (S208).

In this way, the IPC classification of the patent document can be automatically performed by calculating the match degree by applying the frequent items and the association rules between classes for each classification target patent document (S208).

On the other hand, in the embodiment of the present invention has been described with respect to specific embodiments, various modifications are possible without departing from the scope of the invention. Therefore, the scope of the present invention should not be limited by the illustrated embodiments, but should be determined by the scope of the appended claims and equivalents thereof.

100: Patent Document 110: Transaction
120: classification algorithm 130: association rule ruleset
140: classification

Claims (7)

Composing a transaction for each patent document by extracting content included in at least one identification item from N population patent documents for which each class has already been determined according to the international patent classification;
Extracting a keyword from a content included in each transaction through a stemmer;
Generating frequent items and association rules for the corresponding classes from the extracted keywords; And
Classifying a class by applying a classification target patent document to each of the generated frequent items and association rules for each class; and automatically classifying patent documents using the frequent items and association rules.
The method of claim 1, wherein the extracting of the keyword comprises:
Removing the stopwords from the contents included in each of the transactions; Patent document automatic classification method using frequent items and association rules further comprising.
The method of claim 1, wherein the identification item to be extracted,
Method for automatically classifying patent documents using frequent items and association rules, characterized in that at least one selected from the application number, IPC classification, name of the invention, representative claims, key keywords, and identification in the technical field.
The method of claim 1, wherein the extracting of the keyword comprises:
Method for automatically classifying patent documents using frequent items and association rules, further comprising the step of: removing duplicate keywords from the extracted keywords for each patent document.
The method of claim 1, wherein the extracting of the keyword comprises:
Among the extracted keywords for each patent document, a similar keyword is replaced with a representative word; Patent document automatic classification method using frequent items and association rules further comprising.
According to claim 1, The frequent items,
A method for automatically classifying patent documents using frequent items and association rules, comprising a combination having an origin map of at least a minimum of support by an 'Apriori' algorithm.
The method of claim 1, wherein after generating the frequent item and the association rule,
The patent document automatic classification method using frequent items and an association rule, wherein weights are assigned to the duplicate keywords of each class in consideration of the number of occurrences of the keywords.

Images