KR100816934B1 - Clustering System and Method Using Document Search Results - Google Patents

Abstract

Clustering system according to an embodiment of the present invention includes a user input means for inputting a search word; Retrieving means for retrieving a document including a search term inputted by accessing a database in which the document is stored; Output means for displaying a search result to a user; Analysis means for storing the output document of the search result in a cluster storage space of the database, deriving a feature from the document, vectorizing it, and deriving a similarity between documents; Calculating means for calculating a virtual center between the documents based on the similarity between the documents derived by the analyzing means; And clustering means for grouping documents in the cluster storage space based on a virtual center between the documents.

Clusters, documents, databases

Description

Clustering system and method using document search results {Clustering system and method using search result document}

1 is a block diagram illustrating a clustering system according to an embodiment of the present invention.

2 is a view for explaining how a property of a document is vectorized according to an embodiment of the present invention.

3 is a view for explaining an input interface and an output interface provided to a user according to an embodiment of the present invention.

4 illustrates a method for clustering according to an embodiment of the present invention.

5 shows a clustered result screen.

The present invention relates to a system for clustering search results for a user-entered search word, and to search for documents to be clustered based on similarity between documents based on a vector space model and to update similar documents according to a set cluster. The clustering system and the method using the results.

As the exchange of information through the Internet has become common, it has led to a rapid increase in information. However, it has become difficult to search for the most appropriate information desired by the user, and much effort has been required to store and manage necessary documents.

In addition, various methods for storing and grouping documents retrieved using a predetermined web server have been proposed. However, in the grouping of these groups, the documents classified by predetermined classification means are stored separately. It is common to complete.

Therefore, it is difficult for a user to edit structured clusters, and when a document corresponding to a cluster condition is newly generated, it is impossible to update the clusters.

The present invention has been proposed to solve the above-described problem, and an object of the present invention is to propose a clustering system and a method using a document search result for clustering documents based on the similarity between documents retrieved according to a user's request. do.

In addition, the clustering system and method using a document search results for updating the clusters by using a new document provided from the outside by classifying the clustered documents, and by the machine learning based on the classification system. It is for the purpose of suggestion.

Clustering system according to an embodiment of the present invention for achieving the above object is input means for the user inputs a search word; Retrieving means for retrieving a document including a search term inputted by accessing a database in which the document is stored; Output means for displaying a search result to a user; Analysis means for storing the output document of the search result in a cluster storage space of the database, deriving a feature from the document, vectorizing it, and deriving a similarity between documents; Calculating means for calculating a virtual center between the documents based on the similarity between the documents derived by the analyzing means; And clustering means for grouping documents in the cluster storage space based on a virtual center between the documents.

According to another aspect of the present invention, a clustering method includes: reading and outputting a document including a search word from a database according to a search word input by a user; Storing the documents read from the database in a cluster storage space of the database; Deriving a characteristic from documents stored in the cluster storage space and vectorizing the same to derive similarity between the documents; Calculating a virtual center between the documents based on the similarity between the documents; And clustering the documents based on a virtual center between the documents.

Hereinafter, with reference to the accompanying drawings, preferred embodiments of the present invention will be described in detail. However, the spirit of the present invention is not limited to the embodiments in which the present invention is presented, and those skilled in the art who understand the spirit of the present invention may easily add other embodiments by adding, changing, deleting, and adding components within the scope of the same idea. It may be suggested, but this will also fall within the scope of the spirit of the present invention.

The document used below is used to mean a document corresponding to a result of a document search requested from a client for a document stored in a database or a document stored in the database.

FIG. 1 is a block diagram illustrating a clustering system according to an embodiment of the present invention, and FIG. 2 is a view for explaining how a property of a document is vectorized according to an embodiment of the present invention.

1 and 2, a clustering system according to the present invention includes a clustering server 2 for performing a document search and a clustering of documents using the results, and a plurality of networked networks connected to the clustering server 2. It consists of the client 1.

In detail, the client 1 has input means 110 for transmitting a predetermined search word to the clustering server 2 by the user, and output means for outputting a result of a document search transmitted from the clustering server 2. 120 is provided.

In addition, the input unit 110 may be a user interface for a user to input a predetermined search word, and the search word may be a word that is a keyword or a sentence unit.

When a search word is input in units of sentences, a keyword is extracted from a sentence input by the search means 220, and a document including the extracted keyword or a document including a similar word is searched for.

In addition, the client 1 connected to the clustering server 2 may be connected in plural through an IP network. In the drawing, the clustering server 2 and the client 1 may be separately configured, but the input may be performed. It is also possible for the means 110 and the output means 120 to be formed in the clustering server 2.

In addition, the clustering server 2 includes a database 210 in which a plurality of documents are stored, and search means for searching for a document corresponding to a search word requested from the client 1 from the database 210 or another web server. And analysis means for storing the document of the search result output through the output means 120 in a cluster storage space of the database 210 and deriving a characteristic from the document to vectorize it to determine the similarity between the documents ( 230 and an arithmetic means 240 for calculating a virtual center between the documents based on the similarity between the documents derived by the analyzing means 230, and the cluster storage space based on the virtual center between the documents. Clustering means 250 for clustering the documents stored in the, and learning means 260 for machine learning based on the classification system classified in the cluster storage space.

In more detail, a plurality of documents are stored in the database 210, and the documents may be data such as patent documents or papers. In addition, the database 210 may be connected to a web server (not shown) capable of providing a plurality of documents through a predetermined network interface (not shown), and may be provided from a connected web server. May be stored in the database 210.

For example, the search means 220 may access a database of the Korea Intellectual Property Office, the US Patent Office, or the World Intellectual Property Organization (WIPO) to download patent documents in the form of a hypertext transfer protocol, and the documents may be downloaded to the database 210. Can be stored.

The search means 220 performs a search for a document from the database 210 or a predetermined web server connected to a network by using a search word input through the input means 110. The document including the searched word may be searched or the document including the keyword related to the input search word may be searched.

In addition, the document searched by the search means 220 is provided to the output means 120 with predetermined bibliographic information, through which the user can check the information of the searched document.

The analyzing means 230 derives the characteristics of the document from the documents searched by the search means 220, that is, the result of the document search provided to the output means 120 and vectorizes it. The similarity between documents is determined based on the derived characteristics.

The analyzing means 230 may select a representative document from among documents of the search result by using similarity between documents formed based on a feature extracted from a document, for example, a keyword that may be a technical feature. In this case, the representative document selected by the analyzing means 230 may be associated with a virtual center to be described later.

The vector consisting of the features of the document derived by the analysis means 230 is a word representing the feature of the document and the weight of the words as a component of the group, the number of elements constituting the vector is formed differently depending on the document Can be.

For example, referring to FIG. 2 for description, with respect to characteristics derived from a search word input through the input means 110, in document 1, the first characteristic is 19 times, the second characteristic is 35 times, and the last characteristic is 15. It is included in frequency of times.

In the same way, a vector consisting of the characteristics of documents to be analyzed can be formed.

In addition, the analyzing means 230 may determine the similarity between documents based on the derived vector.

The vectorization of document features by the analyzing means 230 may be performed in a vector space model to express text and categories as weight vectors of index words, and the similarity therebetween may be calculated by cosines of both vectors. A characteristic for deriving from the document, for example, a weighted keyword, may be assigned a number for identifying the keyword, in which case the calculation means 240 may use the number assigned to the keyword to determine a relationship between the documents. Calculate the virtual center

In addition, since the analysis means 230 is generally not structured text document stored in the database 210, it also plays a role for converting the structured data by a predetermined text mining engine.

In addition, the calculation means 240 serves to calculate a virtual center from the vector of each document formed by the analysis means 230, the virtual center formed by the calculation means 240 is The user can set through the input means 110. The description thereof will be described later with reference to the accompanying drawings.

The virtual center formed by the calculating means 240 may be made by referring to a characteristic derived from each document and a number assigned to the characteristic.

In addition, the clustering means 250 serves to cluster documents stored in a predetermined cluster storage space (not shown) in the database 210 based on the virtual center between the documents. To this end, a cluster storage space partitioned into a predetermined area may be formed in the database 210, and the result of the document search when there is a clustering request for the result of the document search displayed on the output means 120. Corresponding documents are stored in the cluster storage space.

In addition, the clustering means 250 may serve to classify documents stored in the cluster storage space, and in this case, the clustering means 250 may simultaneously perform a role as a predetermined sorting means. .

The clustering means 250 may classify documents stored in the cluster storage space according to a predetermined classification code or a user's designation through the input means 110.

In this case, the classification code may be a predetermined map or table in which each word (or keyword) is formed according to their similarity, and a plurality of words are connected by nodes so that the similarity between words from a lower level of similarity between words is high. Is structured at a relatively low upper level.

In addition, the learning means 260 machine-learns a classification system that is a standard for classifying documents in the cluster storage space, and a new document provided from a predetermined web server is classified into a classification system based on the mechanized learning result. To be stored in the database 210.

In addition, the classification of the new document by the learning means 260 may be performed periodically or in real time according to the user's setting.

3 is a diagram illustrating an input interface and an output interface provided to a user according to an embodiment of the present invention, FIG. 4 is a diagram illustrating a method for clustering according to an embodiment of the present invention, and FIG. Shows the result screen.

3 to 5, an input interface 111 provided by the input means 110 for inputting a search word by a user and an output interface provided by the output means 120 for displaying a search result ( 121) is shown to the user.

A user may input a condition of a document search through the input interface 111, and may input a document number or a keyword.

In addition, when a search for a search word input through the input interface 111 is performed, the result of the document search by the search means 220 is output to the output interface 121.

In addition, the user may check the documents of the search result provided to the output interface 121 and select and group predetermined documents. In this case, the user selects a specific document from among the items of the displayed document through the output interface 121 and inputs a clustering execution request key (for example, shown as 'my project transmission') by the analyzing means 230. The clustering of documents by the computing means 240 and the clustering means 250 may be performed.

Then, as a clustering procedure using the document search result, the result screen shown in FIG. 4 is output to the user. In this case, the user can designate the number of clusters and the number of documents per cluster, and the virtual center between the documents is calculated according to the number of clusters specified by the user.

And, as shown in FIG. 5, the clustered results may separately output the characteristic values used for clustering, and the clustered list, which is a clustered result, may be divided into large and small categories. In this case, when the user selects a specific list from the cluster list, for example, 'Korea 1 (9)', bibliographic information for nine clustered documents is output.

According to the embodiment of the present invention as described above, the user can set the documents used in the cluster by performing a grouping of documents using the document search result, and there is an effect of updating the cluster.

According to an embodiment of the present invention as proposed, the documents are clustered based on the similarity between documents retrieved at the request of the user, and the user can easily check the clustered results and edit the structured cluster. This has the advantage of being easy.

In addition, by grouping the clustered documents and by the machine learning based on the classification system, there is an advantage that the clusters are updated by using a new document provided from the outside.

Claims (15)

Input means for a user to input a search word; Retrieving means for retrieving a document including a search term inputted by accessing a database in which the document is stored; Analysis means for expressing a keyword extracted from the searched document as a weight vector, and reading the similarity between the searched documents using the vector; Calculating means for calculating a virtual center of the retrieved documents using the weight vectors of the documents represented by the analyzing means; And And clustering means for grouping documents of a search result based on a virtual center calculated by the calculating means. And the number of virtual centers calculated by the calculating means is determined according to the number of clusters designated by the user. The method of claim 1, And said calculating means extracts a particular representative document from the documents of said search result based on the similarity between the documents read by said analyzing means. delete The method of claim 1, And said clustering means classifies documents stored in said cluster storage space according to a pre-stored classification code or a user's designation. The method of claim 4, wherein And a learning means connected to the grouping means and configured to perform machine learning based on the classification code. The method of claim 5, wherein Clustering system, characterized in that the documents provided from a predetermined web server is automatically classified based on the learning result by the learning means. The method of claim 1, Clustering of documents by the clustering means is updated by documents provided from a web server connected to the clustering means through a predetermined network interface. The method of claim 7, wherein And when the clustering of the documents stored in the cluster storage space is updated by the documents provided from the web server, the clustering means notifying the user of the updated information through a mail. Retrieving a document including the search word from a database according to a search word input by a user; Expressing a keyword extracted from the searched documents as a weight vector, and reading the similarity between the searched documents using the vector; Calculating a virtual center of the retrieved documents using the weight vector of the document; And And grouping documents of a search result based on the calculated virtual center. And the number of the computed virtual centers is determined according to the number of clusters designated by the user. delete The method of claim 9, The grouping of the documents may include classifying the virtual centers between the documents according to a predetermined classification scheme code or a user's setting, thereby clustering the documents connected to the virtual centers. The method of claim 9, And after the clustering of the documents is performed, grouping the clustered results into the cluster storage space. The method of claim 9, Clustering characterized in that the automatic classification of the documents provided by the learning means for the machine learning based on a predetermined classification system is performed when a new document is provided to the database after the grouping of the documents is performed. Way. The method of claim 13, And the information on which the document is automatically sorted by the learning means is transmitted to the user using a predetermined mail service. The method of claim 9, And when new documents are downloaded to the database, the new document information is periodically provided to the user.

Images