KR20110001448A - System and method for generating cluster using seed based link - Google Patents

Abstract

PURPOSE: A cluster generation system using a seed according to the link for reducing the necessary operation time for clustering is provided to generate the cluster through the seed by determining an object of high similarity as the seed. CONSTITUTION: A cluster generating system(100) includes a seed determination unit(201) and a cluster generating unit(202). The seed determination unit determines a seed based on link information of objects. A plurality of objects is connected by link. The seed is composed with an object more than the preset similarity. The cluster generating unit generates the cluster in the use of the determined seed.

Description

System and method for creating clusters using seed along links {SYSTEM AND METHOD FOR GENERATING CLUSTER USING SEED BASED LINK}

The present invention relates to a cluster generation system and method using a seed along a link, and more particularly, to a cluster generation system and method for generating a cluster using a hash-based search method.

Clustering means grouping similar objects into clusters. Clustering research has been studied for a long time in various fields such as statistics, database, and data mining. The cluster derived through the clustering process can be applied to services such as advertisement and search to increase the accuracy of the service and easily identify the similarity tendency of objects.

Recently, a study of generating clusters for objects having high similarity to each other by using a large amount of data existing on the Internet during the clustering process has been attracting attention. In particular, link-based clustering, which clusters objects in consideration of links between objects, has emerged.

However, as the use of the Internet increases, creating a cluster of many objects on the Internet requires a lot of execution time. As the execution time increases, more resources for clustering are consumed, so a method of effectively processing the same is needed. In other words, there is a demand for a method of rapidly processing by reducing execution time while ensuring similarity in preparation for a large data environment.

As a result, the need for accurate and efficient clustering of large amounts of data is improved by improving bottlenecks that affect overall performance for clustering large amounts of data.

The present invention provides a cluster generation system and method for reducing the execution time required for clustering by determining a seed having a very high similarity among a plurality of linked objects as a seed and creating a cluster using the seed.

The present invention generates a hash structure from transaction data for a plurality of linked objects, extracts a pattern exceeding the lowest frequency while searching for the hash structure, and determines the seed by determining the seed, thereby ensuring a similar level of cluster generation time. It provides a cluster generation system and method that can reduce the.

The present invention provides a cluster generation system and method capable of improving the similarity of a cluster to be generated by removing objects from transaction data in advance by determining an object having a small number of links as noise and preprocessing the cluster.

The present invention provides a cluster generation system and method that can easily determine the similarity between a plurality of objects by generating a tree hierarchically expressing the similarity between the objects through a plurality of objects and a cluster created from the plurality of objects. do.

According to an embodiment of the present invention, a cluster generation system includes a seed determination unit configured to determine a seed including a predetermined degree of similarity or more based on link information from a plurality of objects connected by a link, and a cluster using the determined seed. It may include a cluster generation unit for generating a.

According to one aspect of the invention, the seed determination unit is a transaction data generation unit for generating transaction data using a plurality of objects connected by the link, the hash data consisting of a candidate pattern based on the pattern length by searching the transaction data to determine A hash structure determination unit and a seed extraction unit for extracting an object corresponding to the candidate pattern exceeding the lowest frequency among the candidate patterns constituting the hash structure as a seed.

According to an aspect of the present invention, the hash structure determination unit generates a candidate pattern corresponding to the pattern length from the transaction data based on the pattern length according to the number of items and the generation from the transaction data. It may include a frequency determination unit for counting the candidate pattern determined to determine the frequency.

According to an aspect of the present invention, the seed determination unit may further include a noise removal unit for determining from the transaction data by determining that the number of objects less than a predetermined number of the link as noise.

The cluster generation system according to an embodiment of the present invention generates a structural tree by setting the plurality of objects to nodes of a lower level for each type, and setting the clusters generated from the plurality of objects to nodes of a higher level. It may further include a tree generator.

The cluster creation method according to an embodiment of the present invention comprises the steps of determining a seed composed of objects having a predetermined similarity or more based on link information from a plurality of objects connected by a link, and generating a cluster using the determined seed. It may include the step.

According to one aspect of the invention, the step of determining the seed consisting of the object of the predetermined similarity or more comprises the steps of generating transaction data using a plurality of objects connected by the link, the transaction data to search the pattern length The method may include determining a hash structure including a candidate pattern based on and extracting, as a seed, an object corresponding to a candidate pattern exceeding a lowest frequency among candidate patterns constituting the hash structure.

According to an aspect of the present disclosure, the determining of the seed may further include determining that the number of objects equal to or less than a predetermined number as the link is noise and removing from the transaction data.

Cluster generation method according to an embodiment of the present invention to generate a structural tree by setting the plurality of objects to each node of a lower level for each type, and the cluster created from the plurality of objects to a node of a higher level It may further comprise a step.

According to an embodiment of the present invention, the execution time required for clustering may be shortened by determining a seed having a very high similarity among a plurality of objects connected by a link as a seed and creating a cluster using the seed.

According to an embodiment of the present invention, by generating a hash structure from transaction data for a plurality of linked objects, searching for a hash structure, extracting a pattern exceeding the lowest frequency and determining the seed as a seed level. It can reduce the cluster creation time while ensuring the similarity of.

According to an embodiment of the present invention, the similarity of a cluster to be generated may be improved by determining that an object having a small number of links is noise and removing it from transaction data through a preprocessing process for cluster generation.

According to an embodiment of the present invention, the similarity between the plurality of objects can be easily identified by generating a tree hierarchically expressing the similarity between the objects through a plurality of objects and a cluster generated from the plurality of objects. have.

Hereinafter, with reference to the contents described in the accompanying drawings will be described in detail an embodiment according to the present invention. However, the present invention is not limited to or limited by the embodiments. Like reference numerals in the drawings denote like elements. Another cluster generation method according to an embodiment of the present invention may be performed by a cluster generation system.

1 is a view for explaining the overall process performed by the cluster creation system according to an embodiment of the present invention.

The cluster generation system 100 may generate the structured tree 103 by clustering the plurality of objects 101 and 102 connected by the link according to the similarity based on the link. Clustering may refer to grouping similar objects into a specific cluster. In this case, the object 101 and the object 102 connected by a link may be data representing different types. For example, object 101 may be a blog, and object 102 may be a post linked by scrap. Although FIG. 1 illustrates two types of objects representing different types, the types of types are not limited.

As the amount of objects (data) targeted for clusters continues to increase, it becomes important to efficiently cluster large amounts of data. Since each object is linked through each other, the amount of data that must be processed for clustering increases as the amount of data increases.

In this regard, the cluster generation system 100 according to an embodiment of the present invention may generate a hash structure using transaction data composed of objects connected by links, and generate a cluster through a hash-based seed exploration method. . In detail, the cluster generation system 100 determines a seed having a very high similarity as a seed in consideration of the fact that the cluster is generated by a few objects 101 and 102 having a high similarity, and another object related to the seed. By including the fields 101 and 102 in the seed, initial clustering can be efficiently and quickly processed.

In addition, the cluster generation system 100 may shorten the execution time of the clustering by repeatedly removing the objects 101 and 102 corresponding to the noise from the transaction data before generating the cluster. The cluster generation system 100 generates a cluster by performing clustering on the objects 101 and 102 through the above process, and reduces the accuracy by generating the trees 103 and 104 for each type using the generated cluster. Link-based clustering can be performed to reduce execution time without the need to do so.

2 is a block diagram showing the overall configuration of a cluster generation system according to an embodiment of the present invention.

Referring to FIG. 2, the cluster generation system 100 may include a seed determiner 201, a cluster generator 202, and a tree generator 203.

The seed determiner 201 may determine a seed composed of objects having a predetermined similarity or more based on link information from the plurality of objects 101 and 102 connected by a link. According to an embodiment of the present invention, the seed may be a reference object for expanding and creating a cluster. In this case, the plurality of objects 101 and the plurality of objects 102 may have different types, and the seed determination unit 201 may determine the seed for each of the objects 101 and 102 having different types.

Cluster generation system 100 according to an embodiment of the present invention can reduce the time to create a cluster from the objects 101 by roughly constructing an initial cluster corresponding to the seed. At this time, the seed may include a small number of objects having a very high similarity among the plurality of objects 101.

For example, the seed determiner 201 may include a transaction data generator 204, a noise remover 205, a hash structure determiner 206, and a seed extractor 207. In detail, the seed determiner 201 may determine seeds from a plurality of objects 101 and 102 having different types through a hash-based seed exploration method.

The transaction data generator 204 may generate transaction data using the plurality of objects 101 connected by a link. In one example, the transaction data generator 204 may set different types of objects 102 each of which the plurality of objects 101 points to via a link.

The transaction data generator 204 may generate transaction data by classifying the plurality of objects 101 into items constituting a set transaction according to a link. That is, the transaction data generator 204 may convert the objects 101 and 102 connected by the link into transaction data. A specific example of generating transaction data is described in FIG. 3.

The noise removing unit 205 may determine that an object having a number equal to or less than a predetermined number as the noise and remove the object from the transaction data. Certain objects 101 may be connected to other types of objects 102 through a few links. The cluster is composed of objects 101 having high similarity, and the similarity is related to the number of links. If the number of links is small, the reliability of the similarity between objects cannot be guaranteed. Thus, when an object having a small number of links is included in a cluster, the accuracy of the cluster may be reduced. In addition, since the number of objects having a small number of links increases the number of data to be processed for clustering, the clustering execution time also increases.

Therefore, the cluster generation system 100 according to an embodiment of the present invention determines the objects 101 having a small number of links as noise and removes them in advance by performing a preprocessing process before generating the clusters, thereby being included in the cluster. It can improve the similarity between objects and shorten the clustering time.

For example, the noise removing unit 205 may repeatedly remove objects in different types for each type. For example, it is assumed that the object 101 (A) corresponding to the type 1 is linked with the objects 102 (a, b) corresponding to the type 2. The noise removing unit 205 may determine and remove objects having one link as noise.

In this case, when the number of links of the object a 102 is one, the noise removing unit 203 may remove the object a 102 from the cluster. As object a 102 is removed, the number of links of object A 102 may also decrease. Then, since the number of links of the object A 101 corresponding to the type 1 is two to one, the noise removing unit 203 may determine that the object A 101 is also removed from the cluster as noise. As a result, the cluster generation system 100 according to an exemplary embodiment of the present invention may repeatedly remove the object corresponding to the noise until the entity does not exist in the cluster.

The hash structure determiner 206 may search the generated transaction data to determine a hash structure composed of candidate patterns based on the pattern length. In this case, the transaction data may search for transaction data from which objects determined as noise are removed through the noise removal unit 205. The pattern length may be the number of objects included in the object pair included in the transaction data. For example, the hash structure determiner 206 may include a candidate pattern generator and a frequency determiner.

The candidate pattern generator may generate a candidate pattern corresponding to the pattern length from the transaction data based on the pattern length according to the number of items. That is, the candidate pattern generator may search for transaction data to generate a candidate pattern corresponding to the pattern length.

In addition, the frequency determining unit may count the candidate pattern from the transaction data to determine a frequency indicating the frequency of occurrence of the candidate pattern in the transaction data. As a result, the hash structure determiner 206 may determine the hash structure by using the candidate pattern and the frequentness corresponding to each of the candidate patterns.

The seed extractor 207 may extract, as a seed, an object corresponding to the candidate pattern exceeding the lowest frequency among candidate patterns constituting the hash structure. That is, the seed extracting unit 207 sets a minimum frequency that can guarantee a degree of similarity or more than a predetermined level, and extracts an object corresponding to a candidate pattern exceeding the minimum frequency among the plurality of objects 101 as a seed. Can be.

Accordingly, the seed extractor 207 may quickly determine a few objects having a very high similarity from the plurality of objects 101 by searching a hash structure composed of one candidate pattern.

The cluster generator 202 may generate a cluster using the seed determined by the seed determiner 201. Clusters can be created for each type of object.

For example, the cluster generator 202 may initially generate a cluster composed of objects corresponding to the seed. In addition, the cluster generation unit 202 may extract and add objects, which are frequently displayed in the same transaction as the seed, to the objects that are not included in the cluster among the plurality of objects and add them to the cluster. In detail, the cluster generator 202 may perform a process of adding an object to the cluster until the number of objects constituting the cluster becomes a preset number.

The tree generator 203 sets the plurality of objects 101 and 102 as nodes of each lower level for each type, and sets the clusters generated from the plurality of objects as nodes of a higher level to structure the trees 103 and 104. Can be generated. That is, the tree X 103 may be generated for the plurality of objects 101, and the tree Y 103 may be generated for the plurality of objects 102.

In this case, each of the plurality of objects 101 and 102 may be determined as an end node in the tree 103 and 104, and the cluster may be determined as a non-terminal node. In addition, the cluster generation system 100 may set a cluster initially determined through the cluster generation process as an object and generate a new cluster from the cluster. That is, the cluster generation system 100 may generate the plurality of objects 101 and 102 and the trees 103 and 104 structured according to levels with respect to the cluster by repeating the cluster generation process.

3 is a diagram for describing transaction data based on an object connected by a link according to an embodiment of the present invention.

Referring to FIG. 3, it can be seen that a plurality of objects 301 and 302 having different types are connected by a link. For example, in FIG. 3, it is assumed that a plurality of objects 301 means a blog and a plurality of objects 302 means a post.

The cluster generation system may determine a seed composed of objects of similarity or higher based on link information from the plurality of objects 301 and 302 connected by a link. This similarity can be determined according to the link of the object. The seed may be extracted from each of the plurality of objects 301 and 302 having different types.

The cluster generation system may generate transaction data using a plurality of objects 301 and 302 connected by a link. In one example, the cluster creation system may set up different types of objects 302 each of which the plurality of objects 301 point to via a link in a transaction. In addition, the cluster generation system may generate the transaction data by classifying the plurality of objects 301 into items constituting the set transaction according to the link.

In FIG. 3, when a plurality of post data is set as a transaction, the cluster generation system may classify blog data connected to each post data as an item 303 for a transaction. For example, since blog data B1, B3, and B4 are linked to post data P3, the item for transaction P3 may be B1, B3, and B4.

Through this process, blog data linked to each of the post data P1 to P8 set as a transaction may be classified as an item 303 to generate transaction data. Conversely, when a plurality of blog data is set up as a transaction, the cluster generation system may classify post data linked to each blog data as an item for the transaction.

The cluster generation system generates a transaction data using a plurality of objects connected by a link, so that the cluster generation system can grasp a pattern of a frequently occurring object. For example, the blog data B1 and B3 are similar to each other because they commonly point to the post data P1 and P3 through a link. In addition, since B1 and B3 are frequently included in the same transactions P1 and P3, the cluster generation system may set B1 and B3 in a frequently occurring pattern.

In addition, the cluster generation system may determine that the object whose number of links is less than or equal to a predetermined number in the generated transaction data may be determined as noise and removed. For example, the cluster generation system may repeatedly remove an object determined as noise for each type. The noise removal process may be repeated until there is no object corresponding to the noise in the transaction data.

As a result, the cluster generation system does not need to generate a cluster for the object determined to be noise by removing an object having fewer links before generating the cluster, so that the cluster generation system can generate the cluster more quickly. Similarity is related to the number of links. That is, the cluster generation system may determine that the number of links is less than the preset number as noise and remove the noise, thereby improving the similarity between the objects included in the cluster and shortening the clustering time.

4 is a diagram for describing a process of determining a hash structure from transaction data according to an embodiment of the present invention.

Referring to FIG. 4, the cluster generation system may generate a hash structure by searching for transaction data 401. In one example, the cluster generation system may search the transaction data 401 to determine a hash structure 402 composed of candidate patterns based on the pattern length. In this case, the transaction data 401 may be an object determined to be noise. That is, the cluster generation system may search for transaction data 401 from which an object determined as noise is removed to generate a hash structure.

In detail, the cluster generation system may generate a candidate pattern corresponding to the pattern length from the transaction data based on the pattern length according to the number of items constituting the transaction data 401. In addition, the cluster generation system may determine the frequency of occurrence by counting candidate patterns from the transaction data.

In FIG. 4, it is assumed that the pattern length is set to two. Then, the cluster generation system may search for the transaction data 401 and generate a candidate pattern having a pattern length of 2 for each transaction. At this time, since the transactions P1, P3, P4, P6, and P8 include two or more items, a candidate pattern having a pattern length of 2 may be generated. A candidate pattern with a pattern length of 2 means that the objects are paired in two.

The cluster generation system searches for the transaction data 401 to generate candidate patterns {B1, B2}, {B1, B3}, {B1, B4}, {B3, B4}, {B4, B5} having a pattern length of 2. can do. In addition, the cluster generation system may determine the frequency of occurrence by counting candidate patterns in the transaction data 401. The frequency of each candidate pattern in the transaction data 401 is 1, 2, 1, 1, 2.

If the lowest frequency is set to 1, the cluster generation system may determine as a seed a candidate pattern exceeding the lowest frequency. That is, the cluster generation system may determine candidate patterns {B1, B3} and {B4, B5} as seeds. The pattern length and the lowest frequency set in FIG. 4 may be changed according to the configuration of the system.

5 is a diagram illustrating a process of creating a cluster and adding an object according to an embodiment of the present invention.

The cluster generation system may extract, as a seed, an object corresponding to the candidate pattern exceeding the lowest frequency among candidate patterns constituting the hash structure. Then, the cluster generation system may generate a cluster using seeds which are objects determined to be similar to each other through link information. In addition, the cluster generation system may add to the cluster objects that appear frequently in the same transaction as the seed among the objects not included in the cluster.

Referring to the example of FIG. 4, candidate patterns {B1, B3} and {B4, B5} that exceed the lowest frequency 1 may be determined as seeds, respectively. Then, as shown by reference numeral 501 of FIG. 5, B1 and B3 may be determined as the cluster X which is one group, and B4 and B5 may be determined as the cluster Y which is the other group. As shown in reference numeral 502 of FIG. 5, the cluster generation system may add B2 included in the same transaction as B1 and B3 to the cluster X.

As a result, the cluster generation system may search a hash structure based on transaction data to determine a small number of objects having a high similarity among a plurality of objects as seeds. In addition, the cluster generation system may generate a cluster by using a seed, and then perform clustering by using a plurality of objects by adding an object, which frequently occurs in the same transaction as an object included in the cluster, to the cluster.

6 illustrates a tree composed of objects and clusters according to an embodiment of the present invention.

6 illustrates a tree in which clusters are generated from a plurality of objects of different types, and are configured of the plurality of objects and clusters. For level 0, the end node of the tree, it consists of a plurality of objects. The non-terminal node of level 1, which is a higher level, is composed of a cluster created from a plurality of objects of level 0. The non-terminal node of the upper level 2 is composed of a cluster created by considering the cluster of the level 1 again as an object. Through this process, the tree can be expanded.

In addition, a plurality of objects corresponding to level 0 in different types of trees may be connected to each other through a link. Through such a link, similarity between nodes may be determined. For example, the tree may store similarity between child nodes belonging to the same parent node. Similarity between child nodes not belonging to the same parent node may be calculated through similarity between ancestor nodes of the child nodes.

That is, the similarity between all nodes in the tree is not calculated, but is calculated through the hierarchical structure of the tree, so that the similarity of nodes included in the tree can be quickly calculated. Based on the similarity finally calculated, the tree can be refined to nodes with more similarity.

7 is a flowchart illustrating the overall process of the cluster creation method according to an embodiment of the present invention.

The cluster generation system may determine a seed composed of objects having a predetermined similarity or more based on link information from a plurality of objects connected by a link (S701).

In one example, the cluster generation system may determine the seed through a hash structure-based search method. In detail, the cluster generation system may generate transaction data using a plurality of objects connected by a link (S704). At this time, the cluster generation system sets different types of objects each of which the plurality of objects point to via a link as a transaction, and classifies the plurality of objects as items constituting the established transaction according to the link to generate transaction data. can do.

The cluster generation system may determine that an object having a number equal to or less than a predetermined number as a noise is removed from the transaction data through a preprocessing operation (S705). The similarity between objects increases as the number of common links increases. However, when an object having a small number of links is included in a cluster, the similarity of the cluster may be reduced. Therefore, the cluster generation system may determine an object having a small number of links as noise through a preprocessing process before generating the cluster. In addition, the cluster generation system may improve the similarity of the cluster generated later by removing the object determined to be noise from the transaction data. At this time, the number of links may be changed according to the configuration of the system.

For example, the cluster generation system may repeatedly remove objects in different types for each type. That is, since objects are linked with objects of different types, when one object is removed, the number of links of other types of objects connected with the corresponding object may be reduced.

The cluster generation system may search the transaction data to determine a hash structure composed of candidate patterns based on the pattern length (S706). Here, the transaction data may be an object determined to be noise. In one example, the cluster generation system may generate a candidate pattern corresponding to the pattern length from the transaction data based on the pattern length according to the number of items. Thereafter, the cluster generation system may determine the frequency of occurrence by counting candidate patterns from the transaction data. Finally, a hash structure consisting of the candidate pattern and the frequency of the candidate pattern may be determined.

The cluster generation system may extract, as a seed, an object corresponding to the candidate pattern exceeding the lowest frequency among candidate patterns constituting the hash structure (S707).

The cluster generation system may generate a cluster using the seed (S702). In one example, the cluster generation system may create a cluster using a seed, and add to the cluster objects frequently appearing in the same transaction as the seed among objects not included in the cluster. That is, the cluster generation system according to an embodiment of the present invention may generate a cluster by determining a seed, which is a small number of objects having high similarity, and shorten an initial execution time for generating a cluster by extending the cluster through the seed. have.

The cluster generation system may generate a structural tree by setting a plurality of objects as nodes of a lower level for each type, and by setting clusters generated from the plurality of objects as nodes of a higher level (S703).

In this case, in the case of level 0, which is an end node of the tree, a plurality of objects are configured. The non-terminal node of level 1, which is a higher level, is composed of a cluster created from a plurality of objects of level 0. The non-terminal node of the upper level 2 is composed of a cluster created by considering the cluster of the level 1 again as an object. Through this process, the tree can be expanded.

Parts not described in FIG. 7 may refer to descriptions of FIGS. 1 to 6.

In addition, the cluster creation method according to an embodiment of the present invention includes a computer readable medium including program instructions for performing operations implemented by various computers. The computer readable medium may include program instructions, data files, data structures, etc. alone or in combination. The media may be program instructions that are specially designed and constructed for the present invention or may be available to those skilled in the art of computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tape, optical media such as CD-ROMs, DVDs, and magnetic disks, such as floppy disks. Magneto-optical media, and hardware devices specifically configured to store and execute program instructions, such as ROM, RAM, flash memory, and the like. Examples of program instructions include not only machine code generated by the compiler, but also high-level language code that can be executed by a computer using an interpreter.

As described above, the present invention has been described by way of limited embodiments and drawings, but the present invention is not limited to the above-described embodiments, which can be variously modified and modified by those skilled in the art to which the present invention pertains. Modifications are possible. Accordingly, the spirit of the present invention should be understood only by the claims set forth below, and all equivalent or equivalent modifications thereof will belong to the scope of the present invention.

1 is a view for explaining the overall process performed by the cluster creation system according to an embodiment of the present invention.

2 is a block diagram showing the overall configuration of a cluster generation system according to an embodiment of the present invention.

3 is a diagram for describing transaction data based on an object connected by a link according to an embodiment of the present invention.

4 is a diagram for describing a process of determining a hash structure from transaction data according to an embodiment of the present invention.

5 is a diagram illustrating a process of creating a cluster and adding an object according to an embodiment of the present invention.

6 illustrates a tree composed of objects and clusters according to an embodiment of the present invention.

7 is a flowchart illustrating the overall process of the cluster creation method according to an embodiment of the present invention.

<Explanation of symbols for the main parts of the drawings>

100: cluster creation system

101: object <type 1> 102: object <type 2>

103: tree <type 1> 104: tree <type 2>

Claims (19)

A seed determination unit configured to determine a seed composed of objects having a predetermined degree of similarity or more based on link information from a plurality of objects connected by a link; And Cluster generation unit for generating a cluster by using the determined seed Cluster generation system comprising a. The method of claim 1, The seed determination unit, A transaction data generation unit generating transaction data using a plurality of objects connected by the link; A hash structure determiner configured to search the transaction data to determine a hash structure composed of candidate patterns based on a pattern length; And A seed extracting unit which extracts, as a seed, an object corresponding to a candidate pattern exceeding a lowest frequency among candidate patterns constituting the hash structure. Cluster generation system comprising a. The method of claim 2, The transaction data generation unit, Set different types of objects pointed to by each of the plurality of objects through a link as a transaction, and classify the plurality of objects into items constituting the established transaction according to the link to generate the transaction data. Cluster creation system. The method of claim 3, The hash structure determination unit, A candidate pattern generation unit generating a candidate pattern corresponding to the pattern length from the transaction data based on the pattern length according to the number of items; And A frequency determination unit for counting the generated candidate pattern from the transaction data to determine the frequency of occurrence Cluster generation system comprising a. The method of claim 4, wherein The hash structure determination unit, And a hash structure is determined using the candidate pattern and the frequentness of the candidate pattern. The method of claim 2, The cluster generation unit, Create a cluster by using the extracted seed, the cluster generation system, characterized in that to add to the cluster objects frequently appearing in the same transaction as the seed among the objects not included in the cluster. The method of claim 2, The seed determination unit, A noise removing unit for removing noise from the transaction data by determining that the number of links is equal to or less than a preset number. More, The hash structure determination unit, And searching for transaction data from which the object determined as noise is removed. The method of claim 7, wherein The noise removing unit, Cluster generation system characterized in that by repeatedly removing the objects of different types for each type. The method of claim 1, A tree generation unit for generating a structural tree by setting the plurality of objects as nodes of a lower level for each type and setting a cluster generated from the plurality of objects as nodes of a higher level. Cluster generation system further comprising. Determining a seed comprising a plurality of objects having a predetermined similarity or more based on link information from a plurality of objects connected by a link; And Creating a cluster using the determined seed Cluster creation method comprising a. The method of claim 10, Determining a seed consisting of the objects of the predetermined similarity or more, Generating transaction data using a plurality of objects connected by the link; Searching the transaction data to determine a hash structure composed of candidate patterns based on pattern lengths; And Extracting, as a seed, an object corresponding to a candidate pattern exceeding a lowest frequency among candidate patterns constituting the hash structure; Cluster creation method comprising a. The method of claim 11, Generating the transaction data, And generating a transaction data by setting different types of objects each of the plurality of objects pointed through a link into a transaction and classifying the plurality of objects into items constituting the set transaction according to the link. How to create a cluster. The method of claim 12, Determining a hash structure composed of candidate patterns based on the pattern length, Generating a candidate pattern corresponding to the pattern length from the transaction data based on the pattern length according to the number of items; And Counting the generated candidate pattern from the transaction data to determine a frequency of occurrence Cluster creation method comprising a. The method of claim 13, Determining a hash structure composed of candidate patterns based on the pattern length, And a hash structure is determined by using the candidate pattern and the frequentness of the candidate pattern. The method of claim 11, Creating the cluster, Creating a cluster using the extracted seeds; And Adding to the cluster objects that appear frequently in the same transaction as the seed among objects not included in the cluster; Cluster creation method comprising a. The method of claim 12, Determining the seed, Determining that the number of objects equal to or less than a predetermined number is the noise and removing from the cluster More, Determining the hash structure, And searching for transaction data from which the object determined to be the noise has been removed. The method of claim 16, Determining the number of objects less than or equal to a predetermined number as the noise to remove from the cluster,  Cluster generation method characterized in that by repeatedly removing the objects of different types for each type. The method of claim 10, Generating a structural tree by setting the plurality of objects as nodes of a lower level for each type, and setting clusters generated from the plurality of objects as nodes of a higher level. Cluster generation method further comprising. A computer-readable recording medium having recorded thereon a program for executing the method of claim 10.

Images