KR101078978B1 - System for grouping documents - Google Patents

Abstract

A document classification system according to the present embodiment includes a database in which patent documents are stored; A document classification module that reads the indirect citation relationship between the patent documents and performs clustering on the first group of patent documents using the read indirect citation relationship; And UI output means for providing a user with information of the results clustered by the document classification module.

Document classification

Description

System for grouping documents

This embodiment relates to a system capable of providing a user with a relationship between a plurality of patent documents or an indirect citation relationship between patent documents.

When a patent applicant intends to obtain a patent, it is necessary to prepare a document that satisfies predetermined requirements and submit it to the JPO. The patent application documents submitted to the KIPO are published after a predetermined time or requirement is met. Such documents can be referred to as patent documents.

In general, a person who intends to apply for a patent goes through a process of searching / searching these patent documents in order to confirm the existence of the prior art, and most patent document searches are made by inputting keywords.

That is, with the recent development of technology, the number of patent applications is enormous, and with this the amount of patent documents is also enormous. Accordingly, it is not easy to investigate patent documents to prevent duplicate studies, to check for infringement of rights, to investigate prior art prior to patent application, to identify trends of technology development of other companies, or to improve research and development.

In a conventional retrieval system for searching or searching such a patent document, if a keyword is incorrectly selected, unnecessary information is often included. In this case, there is a problem that the time of irradiation itself is enormous.

An embodiment of the present invention is to propose a document classification system capable of performing more efficient classification between patent documents by deriving a reference or citation relationship between a plurality of patent documents or by reading an indirect citation relationship even if not a direct citation relationship.

A document classification system according to the present embodiment includes a database in which patent documents are stored; A document classification module that reads the indirect citation relationship between the patent documents and performs clustering on the first group of patent documents using the read indirect citation relationship; And UI output means for providing a user with information of the results clustered by the document classification module.

The document classification module classifies the first to third patent documents into the same cluster when the first patent document is cited in the second patent document and the second patent document is cited in the third patent document. do.

The document classification module may further include a document clustering means for performing clustering using an indirect citation relationship read to the patent documents of the first group, and a second document using the information of the clustering result by the document clustering means. And document classification means for performing classification on the group's patent documents.

The document clustering means generates a specific patent document as a representative document among the patent documents clustered as a result of the clustering, and the document classification means is a predetermined range with the representative document for the second group of patent documents. Patent documents having similarity within are classified into a cluster containing the representative document.

The document clustering means generates a classification name as a result of the clustering, and the UI output means provides a UI through which the user can input the classification name.

According to the embodiment as proposed, by deriving a reference or citation relationship between a plurality of patent documents or by reading an indirect citation relationship even if not a direct citation relationship, there is an advantage that more efficient classification between patent documents can be performed.

In addition, as the efficient document classification is performed, there is an advantage that the patent development through the patent document can be effectively performed.

Hereinafter, with reference to the accompanying drawings for the present embodiment will be described in detail. However, the scope of the idea of the present invention may be determined from the matters disclosed by the present embodiment, and the idea of the invention of the present embodiment may be performed by adding, deleting, or modifying components to the proposed embodiment. It will be said to include variations.

In the following description, the word 'comprising' does not exclude the presence of other elements or steps than those listed.

1 is an example showing the configuration of a document classification system according to the present embodiment.

The system as shown in FIG. 1 has a predetermined evaluation factor for a database 130 in which patent documents are stored, and patent documents stored in the database 130 or other patent documents accessible through a network. Document evaluation module 140 which assigns an evaluation value by using a user, and derives direct and indirect citation relations of patent documents designated by a user or stored in a database, thereby classifying and clustering patent documents. Document classification module 150 to be made.

In addition, this document classification system performs classification on patent documents using indirect citations between patent documents, and unclassified patent documents by using the result of classification (representation document in classified patent documents). As a clustering can be performed with respect to the fields, it can be realized by a server device or a computer. The input / output module 110, the document retrieval module 120, the document feature creation module 160 and the document feature DB 170 can be implemented. It may further include.

The query receiving means 111 of the input / output module 110 receives a query input by a user using a keyboard or a mouse to perform an action such as searching or analyzing a document. The query input by the user includes the database ( 130 may be a keyword recorded in the patent document stored in (or network-connected). In addition to the characters, the keyword also includes numbers such as an application number and a publication number constituting the patent document.

In addition, the UI (User Interface) output unit 112 of the input / output module 110, the user information is calculated or extracted by the document search module 120, document classification module 150 or document evaluation module 140 to the user side Although it is described as a device for providing a variety of UI to be described later, which serves to provide, depending on the embodiment may be provided in other components of the evaluation system, of course.

Further, the patent document data is stored in the database 130 of the embodiment, and the patent document data group is a database configured to store document data of a specification related to a patent application or a patent to be digitized. This patent document data is data containing text data which describes the content of the specification by the character code. Document data containing descriptions by other general tag languages such as Standard Generalized Markup Language (SGML), HyperText Markup Language (HTML), and eXtensible Markup Language (XML) are also possible. If text data can be extracted, other formats such as PDF (Portable Document Format), a general word processor document format, and RTF (Rich TextFormat) format can be used.

The patent document database 130 may be provided outside the patent document evaluation system. In that case, the patent document evaluation system connects to the database and acquires the document data of the patent document using the network.

The document search module 120 searches for patent documents to be called from among patent documents stored in the database 130 based on a query input by a user. In the search of the patent document by the document search module 120, the document feature creation module 160 and the document feature DB 170 may be used.

The document feature creation module 160 may obtain text from the documents stored in the database 130 and provide the document feature DB 170 with index information on the frequency of each keyword. The document search module 120 searches for documents including a query by using an index file of each document stored in the document feature DB 170 when a query is received by the query receiving unit 111. can do.

In the document of the result searched by the document search module 120, a UI as shown in FIG. 3 may be provided to the user through the UI output unit 112.

The document feature creation module 160 is an index file for the documents when a predetermined query is received through the query receiving means 111 or when a new document is stored in the database 130 by a web robot. We can then write and use it to determine the feature vector for each document. See FIG. 8 for a description thereof.

FIG. 8 is a diagram illustrating attribute information of each document, and the attribute information of the documents illustrated in FIG. 8 may be created in the form of an index file by the document feature creation module 160, and the created index file is the document feature DB. Stored at 170.

The document feature creation module 160 may determine a feature vector of each document using the index file stored in the document feature DB 170, and the feature vector may also be stored in the document feature DB 170.

In FIG. 8, information on the frequency of occurrence of each keyword (A, B, C, D, M, I, K, O, P, Q, Z) is shown for each document. A (where A means nouns, proper nouns, compound nouns, etc., does not mean alphabet A) is 35 times, keyword B is 19, keyword C is 15, keyword D is 13 It is included.

In addition, the frequency table for each keyword included in each document may be prepared to be sequentially arranged from the keyword having the highest frequency to the keyword having the low frequency as shown in FIG. 8.

For example, in Document 1, in order to indicate that keyword A includes 4.5%, keyword B 2.4%, keyword C 1.9%, and keyword D 1.7%, the index file for Document 1 is (A, B, C, D) → (4.5%, 2.4%, 1.9%, 1.7%).

The index files of the respective documents are created by various methods as described above, and the feature vector of each document can be extracted using the created index file.

In detail, the document feature creation module 160 creates a table based on the frequency of occurrence of each keyword in each document, and creates a feature vector of each document using the document.

Here, the feature vector determined by the document feature creation module 160 is an evaluation value of a keyword for each document, for example, when the total number of keywords included in each document is n, the feature vector of each document Is a vector of n-dimensional space and can be expressed as Equation (1) below.

Feature vector = (evaluation value w1 of keyword A, evaluation value w2 of keyword B, ... wn evaluation value of word n) wn) --- (1)

For the calculation of the evaluation value, for example, the tf-idf method disclosed in Document (Salton, G: Automatic Text Processing: The transformation, Analysis, and Retrieval of Information by Computer, Addison-Wesley) can be used. According to the tf-idf method, among the n-dimensional feature vectors corresponding to the first document, for the element corresponding to the keyword included in the first document, a value other than 0 is calculated as the evaluation value and is included in the first document. 0 is calculated as an evaluation value for the element corresponding to the keyword (word with a frequency of 0) which is not.

In this respect, the evaluation value of the keyword as an element of the feature vector may be the frequency rate of each keyword appearing in each document. For example, by the document retrieval module 120, keywords A, keywords B, and keywords C may be grouped into similar words from the first document, and the grouped similar words may be stored in a separate similar word DB.

That is, the predetermined keyword A and the keyword B are clustered by the document search module 120, and the grouped keyword A and the keyword B are stored in the similar word DB.

When the extracted keyword includes any one of the keyword A and the keyword B, the document search module 120 searches for similar documents including the remaining keywords.

The search that is limited to the extracted keywords is not performed, but similar documents may be searched based on the attributes of the patent documents.

When keyword A is included in the query received through the query receiving means 111, a document including keyword B and keyword C may be searched together with keyword A when searching for similar documents.

Since retrieval of such similar documents is related to clustering the documents in this embodiment, the description of the operation of grouping the documents after classification of patent documents through citation relationship analysis will be described in more detail. An operation of evaluating patent documents, classifying the patent documents selected by a user using indirect citations, and grouping other documents after the classification of the documents will be described.

First, a document evaluation module 140 for assigning an evaluation value according to a predetermined evaluation factor for patent documents stored in the database 130 or accessible through a network will be described.

The document evaluation module 140 of this embodiment evaluates a patent document with respect to the patent document stored in the database 130 or the network-connectable patent documents using the attribute information of the patent document, and also results of the evaluation. Provides the result to the UI output means 112 so that is shown to the user. In addition, the UI output unit 112 may provide the user with information on the evaluation value of the patent documents of the searched object together with the search result list of the patent document, and as a pop-up or OSD separate from the inspection result list. Evaluation information for patent documents can also be provided.

The document evaluation module 140 creates an evaluation item table using evaluation items set for patent documents stored in the database 130 or patent documents capable of network connection. It may be performed whenever a new patent document is stored in the database 130.

However, the evaluation of the patent document by the document evaluation module 140 may be performed when there is a user's document search request and the document to be searched exists. It should be noted that it will be described regardless of the constraint of.

The document evaluation module 140 includes an evaluation factor management means 141 for managing the characteristics of the patent document as an evaluation factor and a document for performing evaluation on the patent document stored in the database 130 using the evaluation factor. The evaluation means 142 and the DB document management means 143 may be included so that the evaluation value which is the result of document evaluation by the document evaluation means 142 corresponds to each patent document.

The evaluation factor management means 141 manages items for internal and external characteristics of the patent document stored in the database 130, and these characteristics can be edited by a user.

That is, the structure of the evaluation factors for the internal and external characteristics of the patent document by the evaluation factor management means 141 is shown in FIG. 2 is a diagram showing the structure of an evaluation factor of a patent document.

As shown in FIG. 2, a plurality of tables of attributes related to patents described by the evaluation factor management unit 141 may be linked for each country, and each table includes internal characteristics derived from matters recorded in a patent document. It also includes external characteristics that can be derived by considering the characteristics of the cited document cited by the patent document.

Internal characteristics that can be derived from the matter recorded in the patent document refer to keywords or information of the patent document that can be extracted through text mining of the description of the patent document.

For example, the maintenance period in which the period from the registration date recorded in the patent document to the current date is calculated can be derived from the matters described in the patent document, and thus can be an internal characteristic of the patent document.

In addition, the claim length which may be determined according to the historical information, the number of independent terms of the patent document, the number of keywords to be read as a result of text mining for a specific independent term, and the period from the filing date to the current date described in the patent document, Because a particular phrase is entered, such as in a term 'or' according to claim 1 ', the number of subordinate terms that can be identified as subordinate terms may also be an internal characteristic of a patent document.

In addition, the number of inventors described in the patent document may also be an internal characteristic of the patent document.

However, with respect to the number of patents filed by A as an inventor in the first patent document, it is an external characteristic of the patent document because the inventor A must search other patent documents recorded as the inventor.

When there is another patent document cited in the patent document, the number of patent documents cited and the period of citation / citation become external characteristics of the patent document.

In order to calculate an evaluation value for scoring a patent document, an evaluation factor for the patent document must be defined, and the evaluation value for the corresponding patent is ultimately calculated by calculating respective weighting values for the defined evaluation factors. Can be.

In view of this, using the example table as shown in FIG. 2, the evaluation factor managing means 141 prepares evaluation factor items for each of the patent documents stored in the database 130. In Fig. 2, internal and external characteristics are randomly arranged, but evaluation values for internal characteristics that can be obtained from information extracted from a patent document, and corresponding patent documents and other patent documents (with other patent documents in search results) Evaluation values calculated from the relationship between the other patent documents of the same technical field stored in the database may be distinguished as separate items.

After the values of the characteristics read out from the respective patent documents are recorded in a table as shown in Fig. 2, the evaluation value of the patent document is calculated by the document evaluating means 142.

For example, a predetermined weight may be given to each of the evaluation factors, and in this case, the weights are calculated on the values of the internal and external characteristics extracted from the patent document, so that the sum of the scores of each of the evaluation factors is corresponding. It can be an evaluation of a patent document.

The evaluation values for the patent documents calculated in this way can be managed separately by the DB document management means 143, and the evaluation values calculated for each patent document included in the searched results together when the information of the patent document search results is shown to the user. Make it visible.

Accordingly, the UI output unit 112 of the input / output module 110 provides the user with an item or table of evaluation factors managed by the evaluation factor managing unit 141, and the user adds, edits, and deletes the evaluation factor. The content of is stored and managed by the evaluation factor management means (141).

Each of the patent documents may be given an evaluation value by the document evaluation module 140 as described above. When the patent document is displayed to the user as a result of the search, the patent document is evaluated. As can be seen in FIG. 10.

For reference, FIG. 3 illustrates a list of document search results provided to a user's computer or server. For example, when seven search results of a patent document stored in the database 130 are read by a query input by a user, bibliographic information on patent documents to be searched is read. (E.g., patent number, status, filing date, patent date, name of invention, IPC) together with an evaluation value for each patent document.

In addition, the document evaluating means 142 provides an evaluation value for the patent document to the UI output means 112 so that a user can quickly distinguish between a patent having the highest value and a patent which is not of the searched patent documents. . The average evaluation value of the retrieved patent documents together with the evaluation value for each patent document may be calculated to provide the average evaluation value to the UI output unit 112 as well.

When the average evaluation values of the patent documents of the searched results are shown together, the user can easily determine the superiority of each patent document of the search results, and according to this embodiment, the user can easily determine the patent documents having high evaluation value. By checking them first, the search efficiency can be improved.

In this regard, the document evaluating means 142 may calculate an average evaluation value in the technical field in which the patent documents of the search results are included, and the UI output means 112 is associated with the evaluation value of each patent document of the search results. Together, the average estimates in the technical field to which the patent documents belong can be provided together.

In this case, whether or not the common technical field to which the searched documents belong is determined by the IPC classification, which is an international classification, or determined by an F-term classified by the Japan Patent Office. When patent documents classified in different technical fields are to be output as search results, an average value of evaluation values for the technical fields to which the patent documents which occupy a large proportion in the search results belong may be provided.

In this case, the user can easily grasp how important the patent documents of the searched results are by patent documents by comparing the evaluation value assigned to each of the patent documents of the searched result with the average evaluation value of patent documents of the technical field. Will be.

Meanwhile, a function of selectively downloading a list of search results may be provided by a user, and when downloading a list of search results, information on evaluation values performed by the document evaluation module 140 may also be provided. To be provided to the server.

In addition, in the UI of the search result as shown in FIG. 3, when the user clicks on a specific weighting value in order to confirm the detailed item of the evaluation value assigned to each patent document, the evaluation value is configured. The evaluation factors to provide a separate UI that can be confirmed in detail for each evaluation factor and the score given to the corresponding patent document.

In addition, in the UI as shown in FIG. 3 including a list of search results, when a user selects a specific patent document, a separate window (UI) showing a summary of the patent document may be generated. have. That is, as shown in FIG. 4, the patent document analysis UI may be provided to the user side, and evaluation value information on the patent document is also provided to the UI.

For example, the item of the evaluation factor applied to the patent document and the score of each item may be provided along with the name, representative drawing and summary of the invention for the selected patent document. As described above, the average evaluation factor values of the patent documents of the searched result or patent documents of the same technical field as the corresponding patent may be provided together.

Then, the user can operate his or her own server or computer to modify and edit the displayed evaluation factor item, and also edit the assigned score separately. To this end, the evaluation factor management means 141 and the DB document management means 143 of the document evaluation module 140 change the information of the patent document so as to correspond to the items and scores of the evaluation factor changed by the user.

On the other hand, when the graphic of the results classified by the document classification module 150 according to the present embodiment is shown on the user side, the list of patent documents of the clustered results may be shown to the user side in the form as shown in FIG. 11 or 12, the patent document (representative document) having the highest evaluation value given in the form of a graph or a matrix map can be representatively shown.

Here, the document retrieval module 120, the document evaluation module 140, and the document classification module 150 of the present embodiment are preset together with each other so that the retrieval, classification, and clustering of documents can be performed more effectively than the separate operations. It can be seen that the algorithm operates in a complex manner.

Hereinafter, when predetermined patent documents are searched by the document search module 120 and the document feature creation module 160 with respect to a query input by a user, and the result of the search is displayed as a list as shown in FIG. 3, With regard to the patent documents of the document search results, the operation of classifying documents having similar technical problems (problems of the prior art) or solutions (task solutions) will be described.

That is, according to the present exemplary embodiment, documents may be classified by using indirect citations between patent documents, and patent documents having such citations tend to have common technical problems or solutions, and thus, the database 130 It is more advantageous to classify the patent documents of the results of document search (including similar search) for the query input by the user, rather than classifying all of the patent documents stored in c).

In this regard, the operation of the document classification module 150 will be described with reference to patent documents belonging to a predetermined similar range as an example of document search. However, the document evaluation module 140 operates in clustering of documents after classification of patent documents, but the evaluation value information given as shown in FIGS. 3 and 10 may be provided even in the document searching step before document classification and document clustering. will be.

On the other hand, the UI output means 112 is a tag (34, FIG. 3) for guiding a user to perform classification and clustering for some patent documents or all patent documents in the search results from the patent document list of the search results. Can be provided.

When a key for requesting such document classification and clustering is input, the document classification module 150 performs derivation of an indirect citation relationship for selected patents and document classification using the same. For example, when the first patent document is cited in the second patent document and the second patent document is cited in the third patent document, the first patent document and the third patent document are in an indirect citation relationship. The document classification module 150 classifies the first and third patent documents together with the second patent document in the same category.

Here, the citation relationship according to the present embodiment, that is, the indirect citation relationship will be described. A citation relationship refers to a patent document cited if a reference document number (patent application number, patent publication number, registration number, etc.) of another patent document is described in the patent document to describe a problem of the prior art. The relationship between and the cited patent document can be formed.

In addition, only the patent documents mentioned or described in the patent documents need not be limited to the documents cited, and documents referred to in the prior art / cited inventions in the examination, objection or invalidation trial of the patent documents are also referred to. It can be said to be in a citation relationship. Therefore, it can be said that not only the bibliographic information of other patent documents is described in the patent document, but also other patent documents that can be used indirectly by the examiner or another third party during the examination. .

In order to expand such a citation relationship, the database 130 may be provided with a citation and reference document storage unit in which information is stored in each citation of the patent documents, and in this case, the citation relationship is derived from the matters described in the patent document. In addition to reading means for reading, reading means for deriving citation relations from documents used in the examination or post-registration procedure may be provided from materials provided by the KIPO.

For example, if a publication number of another B patent document is described in the A patent document, a direct citation relationship between the A patent document and the B patent document can be read. If the C patent document is presented by the examiner as the citation invention during the examination of the A patent document, it can be said that the C patent document also has a citation relationship with the A patent document.

In addition, among the matters set forth in the claims, there is a patent document of the first group and a patent document of the second group. It is a document group formed by performing. The second group refers to patent documents designated by the user or other patent documents stored in the database 130, and refers to a group of patent documents for which document classification is not performed by the document classification module 150 of the embodiment. It can be seen that.

Therefore, when the user requests to perform document classification on the patent documents of the search result, after document classification is performed by the document classification module 150, at least one or more like the first group described above. A group can be created. If the user wants to perform document classification or clustering on other patent documents (second group) after document classification, the documents belonging to the unclassified or US grouped second group are the characteristics of the first group (representative document). Or a representative vector) to be classified and clustered into classifications belonging to the first group.

However, for the sake of understanding, the documents belonging to the first group were defined as document classification using indirect citation, and the documents belonging to the second group were not described as being classified or clustered yet. Documents belonging to are also classified or clustered according to the classification criteria of the first group even if classification or clustering has already been performed, and thus, it is not necessarily limited to this definition.

In addition, in the detailed description of the present invention, the terms classification and cluster may be used interchangeably, which is sufficient to be interpreted in connection with the operation of the document classification module 150 or the document retrieval module 120. It should be noted.

Meanwhile, in addition to the reading of the citation relationship, the present exemplary embodiment may classify the patent documents using the indirect citation relationship, which will be described with reference to FIGS. 5 to 7.

5 is a view showing an example of a document grouping means of a document classification module according to the present embodiment, and FIG. 6 is a view for explaining a configuration for deriving an indirect citation relationship by the document classification module according to the present embodiment. 7 is a view for explaining a configuration of grouping similar documents into classified groups by the document classification module according to the present embodiment.

First, referring to FIG. 6, a configuration of deriving an indirect citation relationship by the document classification module 150 of the present embodiment will be described.

The user can obtain the information on the indirect citation relationship by the document classification module 150 for the document of the searched result or directly designated documents. As shown in FIG. 6, a user can set a period (period A to period B) for documents to be classified, in which case classification is performed for documents within the set period among patent documents to be classified. do.

That is, even if a direct citation relationship (a citation relationship formed by recording bibliographic information or a citation relationship formed by referencing by an examiner, etc.) does not hold between patent documents belonging to a set period, the cited patent document or citation If there is an association between patent documents, the patent documents may be classified in the same category as indirect citations.

For example, the period set by the user for document analysis and classification is period A to period B, and the patent documents belonging to the period (Base Patent, Patent 5, Patent 6, Patent 7, Patent 8, Patent 9) If the first patent document (Patent 1) outside the set period is cited in the fifth patent document (Patent 5) and the base patent, and is not in a direct citation relationship, the fifth patent document (Patent 5) and the base patent document (Base Patent) Is an indirect citation relationship with each other.

As another example, if the third patent document (Patent 3) directly refers to the seventh patent document (Patent 7) and the base patent document (Base Patent) within the period, the third patent document (Patent 3) and the third patent Patent Document (Patent 7) is classified into the same category according to the present embodiment, because indirect citation relationship between each other is established.

Through this method, in the case of FIG. 6, since the base patent document forms an indirect citation relationship in all of the fifth to ninth patent documents Patents 5 to 9, the representative document or the base patent document is It can be.

Then, the user can directly create a classification name for the category units of the patent documents classified by this method in order to easily grasp the contents. For example, as shown in FIG. 11, when the patent documents of the categorized category have a common problem (or a corresponding task) of 'noise reduction', the category name is referred to as' noise reduction (ex. Can be written as'.

The categories categorized by these methods may be provided to the user in the form of FIG. 11 or a graph or a diagram, and of course, it is also possible to be shown as a bubble graphic or the like.

Referring to FIG. 12, when the user names the categories classified as the tasks 1, 2, 3, and the solutions 1, 2, 3, and the like, the user corresponds to each task and the solution. Images 410 and 420 may be displayed as indications of the categories. In this case, the image displayed in the graph may be expressed in a different color or size depending on the number of patent documents included in each category, and also the sum of the evaluation values (or average evaluation values) of the patent documents included in each category is displayed. It may be expressed in different colors or in different sizes.

As a result of document classification or document clustering provided to the user, when providing data in the form as shown in FIG. 11 or 12, the user may select a specific category (solution 1, solution 2, solution 3, corresponding In the case of selecting the task 1, the subject 2, and the subject 3), the user is provided with information on the patent document with the highest evaluation value given by the representative patent document (base patent document) described above or the document evaluation module.

By this process, the user can perform document classification on the documents of the searched result. Furthermore, after performing document classification using indirect citations, patent documents that are not classified or classified as other indirect citations, which may belong to the second group, may be classified and clustered.

In the document clustering process, the similarity between documents by the document feature creation module 160 may be used, and the document classification module 150 may include a second group based on patent documents of a first group that are already classified. Classify and cluster patent documents. In addition, the document grouping means 152 of the document classification module 150 may determine the degree of similarity between the patent document belonging to the first category of the first group (which may be a representative document of the first category) and the patent document of the second group. By judging, it is determined to which category of the first group the patent document belonging to the second group is to be classified.

The document grouping means 152 may include a representative vector calculator 1521 that calculates a representative vector required for clustering by using a representative document in a classified category or by using a plurality of documents belonging to the category.

The document grouping means 152 may include a field-by-field clustering unit 1522 for clustering similar documents by field (or by identification item) constituting the patent document.

The representative vector calculating unit 1521 creates the document feature based on the occurrence frequency for each keyword from a representative document (a patent document selected by using a base patent document or an evaluation value) in a previously formed category or documents belonging to the same category. Use the index file created by module 160. For example, the representative vector calculator 1521 may extract representative keywords having a high frequency among keywords appearing in each document, and select the top several keywords in the order of high frequency from the index file of each document. Can be selected.

By this selection operation on the keyword distribution chart as shown in FIG. 8, feature vectors of each document as shown in FIG. 9 can be formed.

In addition, the representative vector calculator 1521 may calculate a percentage of each document with respect to the keywords selected in the order of high frequency. For example, in Documents 1, keyword A is 4.5%, keyword B is 2.4%, and keyword E is The percentage of occurrences for each keyword can be calculated with 1.9% and 1.7% for keyword D.

In this process, the percentage of occurrence frequency for each keyword is calculated for documents in the corresponding category or representative documents (hereinafter referred to as 'category documents').

8 and 9, after this process is performed on the category documents, a percentage of each keyword is added to the entire category document, and a predetermined number of specific keywords are added in order of increasing percentage of the combined keywords. Can be elected as a representative keyword.

For example, among the keywords shown in FIG. 8, the sum of the percentages occupied by each keyword in all ten category documents is keyword B, keyword A, keyword E, keyword D, keyword O, keyword C, keyword K. In the case of having a high value in order, keywords B, keywords A, keywords E and keywords D can be selected as representative keywords for clustering the selected documents. Then, a feature vector for each document is calculated using the selected representative keyword as a component of the representative vector. That is, the selected representative keywords are arranged in a high probability distribution order, and these are selected as components of the representative vector. A process of creating a feature vector of each document is performed based on the selected keyword B, keyword A, keyword E, and keyword D. This process is performed for the top four keywords in the index file of each document. However, four keywords are selected as representative keywords constituting the components of the representative vector, and four document keywords with high frequency in each document are compared to create a feature vector of each document. However, this is merely an example. Can be modified by any number.

When the selected representative keyword is included in each document, the vector component can be set to '1', and when not included, it can be set to '0'.

However, this may also be a vector component, and instead of 1 and 0, a vector component may be created with a weighted value for each keyword.

As shown in FIG. 9, the feature vector of each document thus prepared is '1' when the representative keyword is included, and '0' when it is not included, thereby completing the feature vector of each document.

By this process, the feature vector of Document 1 becomes (1,1,1,1) and the feature vector of Document 2 becomes (1,1,0,1). Although the component of each feature vector is set to 1 or 0, each vector component may be given a different value according to the frequency of occurrence of each representative keyword.

In the case of using a plurality of category documents, a process of selecting a representative vector (or a center vector) using the feature vectors of each document is performed. Here, the representative vector for clustering the largest vector among the feature vectors is performed. Can be selected as.

In this case, among the respective feature vectors shown in FIG. 9, the feature vector (1,1,1,1) of Document 1 may be a representative vector, and the second group of patents unclassified by using the selected representative vector Documents can be clustered.

By using the representative vector derived from the category document, it is possible to check whether the patent document having a predetermined degree of similarity with the specific category is included in the second group, and the similarity is determined by using the feature vector or the representative vector as described above in the second group. It can also be determined by performing the group patent documents.

That is, the similarity between the category document belonging to the predetermined category of the first group and the document unclassified in the second group may be calculated using the dot product by the respective feature vector or the representative vector. When the value of the operation that is internalized through the dot product with the feature vector for the two groups of patent documents falls within the preset range, the group may be clustered together with the representative vector. That is, they may be classified and clustered within the category to which the representative vector belongs.

The document grouping means 152 sets the dot product of the vector A and the vector B when the representative vector is A and the feature vector of the document to be compared with similarity is B. ^The degree of similarity between the document corresponding to the vector A and the document corresponding to the vector B is determined according to how far the value divided by ² is from '1'.

However, when the value of the internalized operation between the feature vectors of the second group document with the representative vector deviates from the reference value, it is not clustered together with the representative vector, and is used as a document for another cluster.

According to this similarity calculation and determination between the representative vector representing the category and the feature vector of the second group document, as shown in FIG. 7, the 20th document P20 belonging to the second group can be clustered into the A group of the first group. The twenty-first document P21 of the second group may be grouped into the B category of the first group.

However, when document classification is performed by the document classification module 150 in addition to the above-described embodiment, as a result, the document classification module 150 may select a technical classification code (IPC or F-term) representing a category. have. In this case, the classification and grouping of the second group documents by the document grouping means 152 use the technical classification code in addition to the above similarity determination.

For example, the document grouping means 152 uses F-terms that record high frequency for each category that is a result of classifying documents using indirect citation, so that the second group documents have F-terms. The similarity with the term can be determined.

In the case of F-terms, they are classified according to a task or a solution, and if used together with similarity determination using vectorization of documents, more efficient document clustering can be performed.

As described above, the result of document classification or document clustering provided by the document classification system of the present embodiment can be stored and shared with other users according to the setting of the system. Or it will be very useful to the team.

1 is a diagram for explaining a document classification system according to the present embodiment.

2 is an example of an evaluation factor table of the document evaluation module according to the present embodiment.

3 and 10 are examples showing the search and evaluation results of the document according to the present embodiment.

4 is an example of a UI in which information of a document is shown according to the present embodiment.

5 is a diagram showing the configuration of a document grouping means according to the present embodiment;

6 is a view for explaining the indirect citation relationship according to the present embodiment.

7 is a view for explaining that a second group document is classified and grouped into categories of a first group according to the present embodiment;

8 is an example showing attribute information of a category document or a second group document according to the present embodiment.

9 is an example showing a feature vector calculated from a category document or a second group document according to the present embodiment.

11 and 12 are examples of a UI provided to a user as a result of document classification or clustering according to the present embodiment.

Claims (12)

A database in which patent documents are stored; A document classification module that reads the indirect citation relationship between the patent documents and performs clustering on the first group of patent documents using the read indirect citation relationship; And A UI output means for providing a user with information of the results clustered by the document classification module, The document classification module includes a document grouping means for performing clustering by using the indirect citation relationship read out for the first group of patent documents, and information of the grouping result by the document grouping means. A document classification system comprising document classification means for performing classification on patent documents. The method of claim 1, The document classification module classifies the first to third patent documents into the same cluster when the first patent document is cited in the second patent document and the second patent document is cited in the third patent document. Document classification system characterized by. delete The method of claim 1, The document clustering means generates a specific patent document as a representative document among the patent documents clustered as a result of the clustering, And said document classification means classifies a patent document having similarity within a predetermined range with respect to said representative document with respect to said second group of patent documents into a cluster containing said representative document. The method of claim 1, The document clustering means generates a classification name as a result of the clustering, And the UI output means provides a UI through which the user can input the classification name. The method of claim 1, The document clustering means generates a classification name corresponding to an F-term code as a result of the clustering, And the document classification means classifies a patent document belonging to the second group into a cluster having the F-term code in common. The method of claim 1, The information of the clustered result by the document clustering means becomes the content of the problem or the solution constituting the patent document, The UI output means is a document classification system, characterized in that for providing the user with the information of the patent document classified according to the above-mentioned problem or solution. The method of claim 1, And the UI output means displays the number of patent documents classified based on the above-mentioned problem or solution means as information of the patent document in a matrix map. The method of claim 1, Further comprising a document evaluation module for calculating the evaluation value for the patent document by performing an evaluation of the patent document using the characteristics of the patent document, And the UI output means provides the user with a map in which the classification criteria of the patent document by the document classification module and the information of the patent documents grouped or classified according to the classification criteria are recorded. The method of claim 9, And the UI output means provides information on the number of patent documents classified in the same cluster or evaluation values of the grouped patent documents as information on the classified patent documents. The method of claim 9, When a specific classification area of the map is selected by the user, The UI output means may provide the user with the information of the clustered patent document included in the selected classification area, the evaluation value of the clustered patent documents, or the content of the patent document having the highest evaluation value among the clustered patent documents. Classification system. The method of claim 9, The UI output means provides the user with information on the evaluation item of the document evaluation module, and when the information on the evaluation item is changed by the user, the document classification module re-executes the grouping or classification of the patent document. Document classification system characterized by.

Images