KR20180137387A - Apparatus and method for detecting overlapping community - Google Patents

Abstract

The present invention relates to an apparatus and method for detecting an overlapping community. The apparatus comprises: a core node extraction unit extracting a predetermined number of nodes as core nodes based on degree of connection in a node relation diagram composed of a plurality of nodes and a plurality of edges connecting the nodes; an edge relationship diagram generation unit generating an edge relation diagram in which the edges in the node relation diagram are transformed into nodes; a community extraction unit extracting a community for each core node set corresponding to each core node in the edge relation diagram; and a transformation unit transforming the communities extracted in the edge relation diagram into communities constituted by the nodes of the node relation diagram.

Description

[0001] APPARATUS AND METHOD FOR DETECTING OVERLAPPING COMMUNITY [0002]

More particularly, the present invention relates to an apparatus and method for detecting overlapping communities, and more particularly, to a method and apparatus for detecting overlapping communities by extracting communities of each core node set in an edge relation diagram where edges representing node- And more particularly, to an apparatus and method for detecting an overlapping community capable of detecting a plurality of communities to which each node belongs.

A network is composed of nodes and edges. In a network, a community is a set of nodes having a high edge density between nodes internally and an edge connected to a node existing outside the community. On the other hand,

When data exists, it is important to find communities that exist in the network. Therefore, there have been many attempts to find the community to which each node belongs to the network. Conventionally, however, each node was not allowed to belong to multiple communities at the same time. However, in a real network, each node can have multiple relationships or have different characteristics and functions. For example, if a node is a person, you can belong to multiple communities at the same time, and if it is an item, you can belong to multiple categories or have characteristics at the same time. That is, most nodes can belong to more than one community at the same time.

Recently, research on detection of overlapping communities in which each node belongs to several communities at the same time while finding the communities to which each node belongs is under study. In the study of overlapping community detection, the research that extracts the community from the nodes representing each individual, and the research that extracts the community from the edge after considering that the node belongs to several communities at the same time and the edge belongs only to one community have. However, in an actual network, edge - to - node relationships represented by edges can belong to more than one community, and the community overlooks the fact that relationships are gathered together.

On the other hand, communities in the real world are made up of people-human relationships, and in the network, communities are organized by gathering nodes and nodes. Also, relationships between people and people can belong to more than one community at the same time. For example, some of the middle school alumni relationships that make up the middle school alumni community are high school alumni associations that make up the high school alumni community. Therefore, it is necessary to detect overlapping communities so that each relationship belongs to several communities at the same time. In other words, since the edges that are the minimum units connecting the node and the node are gathered and become a community, the community should be extracted for the edge.

Conventional overlapping community detection studies, however, did not allow an edge to belong to more than one community at the same time, although extracting communities to edges. This is not appropriate because the edges representing the inter-node relationships do not reflect the actual network that can belong to multiple communities.

Therefore, it is required to develop a technique for detecting a community to which each node belongs to an edge represented by a relationship between nodes.

Related Prior Art Korean Patent Publication No. 10-2016-0083017 entitled " System and Method for Identifying Influencers and Their Communities in a Social Data Network, Public Date: July 11, 2016) .

It is an object of the present invention to provide an apparatus and method for allowing an edge to belong to two or more communities at the same time and detecting overlapping communities existing in the network with respect to edges.

The problems to be solved by the present invention are not limited to the above-mentioned problem (s), and another problem (s) not mentioned can be clearly understood by those skilled in the art from the following description.

The overlapping community detection apparatus provided in the present invention includes a core node extracting a certain number of nodes into a core node based on degree of connection in a node relation diagram composed of a plurality of nodes and a plurality of edges connecting the plurality of nodes, An edge relation diagram generation unit for generating an edge relation diagram in which a plurality of edges in the node relation diagram are respectively converted into node shapes, and in the edge relation diagram, for each core node set corresponding to each of the core nodes, A community extracting unit for extracting a community, and a converting unit for converting the communities extracted in the edge relation diagram into communities constituted by nodes of the node relation degree.

Preferably, the core node extracting unit may calculate a connection degree of each of the plurality of nodes, and may extract a predetermined number of nodes as a core node in order of the calculated connection degree.

Preferably, each time the core node is extracted, the core node extractor extracts a core node extracted from the extracted core node and a neighboring node that is one level adjacent to the extracted core node, It can be excluded from candidates.

Advantageously, said edge relationship generator is operable to convert edges into said nodes, respectively, and to connect said transformed nodes based on adjacent edge information in said node relationship diagram to generate an edge relationship diagram .

Preferably, the edge relation generator may extract the nodes of the node relation diagram corresponding to the edges connected to the respective key nodes of the node relation diagram as key node sets.

Preferably, the community extractor executes PPR (Personalized PageRank) in the edge relation diagram to calculate PPR scores of all the nodes based on the set of core nodes, It can be added to the community of node sets.

Preferably, the community extractor calculates a conductance indicating a ratio of an edge connected to the outside of the community each time a node is added to the community of the core node set, and based on the calculated conductance, The configuration of the community can be terminated.

In addition, the overlapping community detection method provided by the present invention extracts a certain number of nodes into core nodes based on degree of connection in a node relation diagram composed of a plurality of nodes and a plurality of edges connecting the plurality of nodes Generating an edge relation diagram in which a plurality of edges in the node relation diagram are transformed into respective node types, and in the edge relation diagram, determining a community for each key node set corresponding to each of the core nodes, Extracting each of the communities extracted from the edge relation diagrams into communities composed of the nodes of the node relation diagrams.

Preferably, the step of extracting to the core node calculates the degree of connection for each of the plurality of nodes, and extracts a certain number of nodes to the core node in order of the node having the calculated degree of connection.

Preferably, each time the core node is extracted, it is possible to exclude the core node extracted in accordance with the characteristics and the number of the community from the node candidate adjacent to the next node, have.

Advantageously, the step of generating the edge relationship diagram further comprises the steps of converting edges into the respective nodes in the node relationship diagram, and connecting the transformed nodes based on adjacent edge information in the node relation diagram .

Preferably, in the step of generating the edge relation diagram, nodes of the node relation diagram corresponding to edges connected to each key node of the node relation diagram may be extracted as a core node set.

Preferably, the step of extracting each of the communities includes executing a PPR (Personalized PageRank) in the edge relation diagram, calculating PPR scores of all nodes based on each core node set, calculating the PPR score To the community of the core node set in ascending order of node.

Preferably, each time a node is added to the community of the core node set, calculating a conductance indicating a ratio of an edge connected to the outside of the community, if the calculated conductance is larger than the previous value, And terminating the configuration of the community.

Meanwhile, the 'overlapping community detection method' described above can be recorded in a recording medium readable by an electronic device after being implemented in the form of a program, or distributed through a program download management device (a server or the like).

The present invention extracts the communities of each core node set in the edge relation diagram where the edge representing the relationship between the nodes is transformed into the form of the node, transforms the communities extracted in the edge relation diagram into communities composed of nodes of the degree of node relation , It is possible to detect a plurality of communities to which each node belongs.

In addition, it is possible to detect overlapping communities existing on the network with respect to the edge, while allowing one edge to belong to more than two communities at the same time, thereby identifying a plurality of communities to which each node belongs.

The effects of the present invention are not limited to the above-mentioned effects, and various effects can be included within the scope of what is well known to a person skilled in the art from the following description.

FIG. 1 is a block diagram schematically showing a configuration of an overlapping community detection apparatus according to an embodiment of the present invention.
FIG. 2 is a diagram for explaining a core node extracting method according to an embodiment of the present invention. Referring to FIG.
FIG. 3 is a flowchart for explaining an overlapping community detection method according to an embodiment of the present invention.
4 is an exemplary diagram for explaining an overlapping community detection method according to an embodiment of the present invention.
5 is a flowchart illustrating a method of extracting a community from a set of core nodes according to an embodiment of the present invention.

While the invention is susceptible to various modifications and alternative forms, specific embodiments thereof are shown by way of example in the drawings and will herein be described in detail. It should be understood, however, that the invention is not intended to be limited to the particular embodiments, but includes all modifications, equivalents, and alternatives falling within the spirit and scope of the invention. Like reference numerals are used for like elements in describing each drawing.

The terms first, second, A, B, etc. may be used to describe various elements, but the elements should not be limited by the terms. The terms are used only for the purpose of distinguishing one component from another. For example, without departing from the scope of the present invention, the first component may be referred to as a second component, and similarly, the second component may also be referred to as a first component. And / or < / RTI > includes any combination of a plurality of related listed items or any of a plurality of related listed items.

It is to be understood that when an element is referred to as being "connected" or "connected" to another element, it may be directly connected or connected to the other element, . On the other hand, when an element is referred to as being "directly connected" or "directly connected" to another element, it should be understood that there are no other elements in between.

Also, the terms "to "," unit ", "module ", and the like described in the specification mean units for processing at least one function or operation and may be implemented by hardware or software or a combination of hardware and software .

The terminology used in this application is used only to describe a specific embodiment and is not intended to limit the invention. The singular expressions include plural expressions unless the context clearly dictates otherwise. In the present application, the terms "comprises" or "having" and the like are used to specify that there is a feature, a number, a step, an operation, an element, a component or a combination thereof described in the specification, But do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, or combinations thereof.

Unless defined otherwise, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. Terms such as those defined in commonly used dictionaries are to be interpreted as having a meaning consistent with the contextual meaning of the related art and are to be interpreted as either ideal or overly formal in the sense of the present application Do not.

Hereinafter, preferred embodiments according to the present invention will be described in detail with reference to the accompanying drawings.

FIG. 1 is a block diagram schematically showing a configuration of an overlapping community detection apparatus according to an embodiment of the present invention. FIG. 2 is a diagram for explaining a method of extracting a key node according to an embodiment of the present invention.

Referring to FIG. 1, the overlapping community detection apparatus 100 includes a core extracting unit 110, an edge relation generating unit 120, a community extracting unit 130, and a converting unit 140.

The core node extracting unit 110 extracts a certain number of nodes as core nodes based on degree of connection in a node relation diagram composed of a plurality of nodes and a plurality of edges connecting the plurality of nodes. Here, the degree of node relation may be a graph composed of a plurality of nodes and edges connecting the nodes, and the degree of connection may be the number of edges (or nodes) connected to one node. A core node may be a node that is the core of a community, and a community may be a set of nodes and inter-node edges that have weak connections internally with strong connections internally. Accordingly, the core node extracting unit 110 should extract a core node that allows the community to have a strong connection to the nodes internally and a weak connection with the external community.

On the other hand, nodes with a high degree of connection in the node relation diagram are likely to become the core (center) of the community because there are many connections with other nodes.

Accordingly, the core node extracting unit 110 may calculate the degree of connection for each of the plurality of nodes, and may extract a predetermined number of nodes to the core node in order of the node having the calculated degree of connection. At this time, the core node extracting unit 110 may extract a core node using a gravity centers method, a spread hubs method, or the like. The Graclus centers method finds k clusters through the NEO k-means, selects the closest nodes to the centroid of each cluster, and the Spread hubs method extracts k key nodes in order of degree. have. Spread hubs scheme selects nodes that have many connections with other nodes, so it can have high coverage when searching for a community based on a set of key nodes in the graph, and it is likely to be an important core of the community . Therefore, in the present invention, the key nodes are extracted using the spread hubs method, but the present invention is not limited thereto.

Hereinafter, a method of extracting a core node by using the spread node method will be described.

In the case of using the spread hubs scheme, the core node extracting unit 110 can extract a core node in a different manner according to the characteristics and the number of communities. For example, the core node extracting unit 110 may extract a core node by one-step neighboring node removal method and two-step neighboring node removal method according to the characteristics and the number of the community.

One-step neighbor node removal method is a method to exclude nodes extracted directly from the core node candidates to be extracted later whenever one node is extracted to the core node in descending order of degree in node relation diagram. This method may be appropriate when there are a large number of communities to look for and connections within the community and between the community are relatively weak.

A single-step neighboring node removal method will be described in detail with reference to FIG. 2 (a). When the node 'I' having the highest degree is extracted as the first key node in the node relation diagram, the key node extracting unit 110 extracts the node I (the solid lines X, D, E, F, G , H, J, K, L) are extracted from the core node candidate group, and then the second key node is extracted. When the second key node 'B' having the highest degree is extracted from the core node candidate group, 110) can extract the next key node after excluding the node (X, A, C) adjacent to the node 'B' from the key node candidate group. As described above, the one- In this paper, we propose a method for extracting core nodes from the extracted node nodes.

The two-step neighbor node removal method is to exclude the extracted key node and two adjacent nodes from the candidate node to be extracted later, whenever one node is extracted as the key node in descending order of degrees. This method requires a small number of communities to be found, a strong link between the community and the community Can be suitable.

The two-step neighboring node removal method will be described in detail with reference to FIG. 2 (b). If the node I having the highest degree is extracted as the first key node, the core node extracting unit 110 extracts the neighboring nodes (the solid lines X, A, B, C, The node N is excluded from the candidate node group, and the second node is extracted. When the second node 'O' having the highest degree is extracted from the core node candidate group, the core node extracting unit 110 extracts 'O' and adjacent nodes (dotted line X (P, R) And then the next key node can be extracted. In this way, the two-step neighbor node removal method can be a method of excluding a node extracted as a core node and a node adjacent to two adjacent nodes from a candidate node to be extracted later, whenever a node having a higher degree is extracted.

In the above-described method, the core node extracting unit 110 can extract the core nodes as many as the number of communities.

On the other hand, a community is a group of relationships with similar characteristics, and since the minimum unit that constitutes a community is an edge, it is necessary to search the community for the edges and search for the communities to which the nodes belong. However, since the key node of the node relation diagram is not an edge, it is necessary to convert the edge of the node relation diagram to be a node.

Accordingly, the edge relation diagram generation unit 120 generates an edge relation diagram in which a plurality of edges in the node relation diagram are respectively converted into node shapes. That is, the edge relation diagram generation unit 120 converts the edges of the node relation diagrams into the nodes of the edge relation diagram, connects the nodes of the edge relation diagram based on the adjacent edge information in the node relation diagram, Can be generated. At this time, the edge relation diagram generation unit 120 may extract the nodes of the edge relation diagram corresponding to the edges connected to the respective key nodes of the node relation diagram as core node sets. In other words, the edge connected to each core node in the node relation diagram can be one core node set in the edge relation diagram.

The edge relation diagram generation unit 120 transforms the node relation diagram to an edge relation so that the community can be detected with respect to the edge (relation), and the edge relation (relationships) belonging to the core node extracted from the node relation diagram is referred to as an edge relation You can do this with your core set of nodes.

When the node relation is transformed into the edge relation, the edge in the node relation diagram is represented as a node in the edge relation diagram. Therefore, the community can be extracted from each core node set in the edge relation diagram.

The community extracting unit 130 extracts a community for each key node set in the edge relation diagram.

The community extractor 130 extends the community from the core node set composed of the central nodes of each community to the community having the characteristics of the community, Can be configured. At this time, the community extracting unit 130 may execute the PPR (Personalized PageRank) to extend the community based on each core node set.

That is, the community extracting unit 130 executes the PPR (Personalized PageRank) in the edge relation diagram to calculate the PPR scores of all the nodes based on the set of the core nodes, Add to the community of the set. At this time, each time the community extractor 130 adds a node to a community of the core node set, the community extractor 130 may terminate the community structure based on conductance indicating the ratio of edges connected to the outside of the community.

This approach reflects the fact that the closer the node from the core node is to the node, the more frequently it communicates with the nearer one.

Hereinafter, a method of extracting a community by executing the PPR will be described in detail.

가 수렴할 때까지 수학식 1을 반복 계산하여, 수렴된 확률 분포

를 구하고, 수렴된 확률 분포

의 각 요소를 각 노드의 PPR 점수로 산출할 수 있다. First, the community extractor 130 executes the PPR in the edge relation diagram, and calculates the PPR score of all the nodes based on the set of the core nodes. At this time, the community extracting unit 130 can calculate the PPR score of each node using Equation (1) described below. That is, the community extracting unit 130 extracts

(1) is repeated until the convergence probability distribution

And the converged probability distribution

Can be calculated by the PPR score of each node.

은 n*1 랭킹 결과 벡터, c는 리스타트 확률, W은 n*n 정규화된 인접 행렬,

은 n*1 리스타트 벡터를 의미할 수 있다.

의 값은 각 노드에서 리스타트 할 확률을 의미하며, 핵심노드가 집합(다수 존재)이기 때문에 PPR 수행 시에

의 값은 핵심노드 집합내(에지 관계도)의 전체 노드 수가 n개일 때,

의 각 인자는 '1/n'의 균일한 값일 수 있고, 나머지는 '0' 값을 갖게 될 수 있다. here,

N is a ranking result vector, c is a restart probability, W is n * n normalized adjacent matrix,

May refer to an n * 1 restart vector.

Means the probability of restarting at each node, and since the core node is a set (many exist), the value of

Is the number of nodes in the core node set (edge relation diagram) n,

May be a uniform value of '1 / n', and the rest may have a value of '0'.

When the PPR score of the nodes is calculated through Equation 1, the community extractor 130 considers that the association with the community of the corresponding core node set is high in order of nodes having a high PPR score, and expands the community by adding nodes to the corresponding community . At this time, the community extracting unit 130 may add the node until the nodes in the community are strong and the connection with the outside nodes is weak, and the characteristics of the community are conductance, . ≪ / RTI > Here, conductance may refer to the ratio of the edge directed to the outside of the community versus the edge connected to all the nodes in the community.

That is, each time the community extractor 130 adds a node to the community of the core node set, the community extractor 130 calculates the conductance and ends the community structure based on the calculated conductance. At this time, the community extracting unit 130 may calculate the conductance? (S) using Equation 2 described below.

는 커뮤니티 S안에 한 노드만 속해 있는 에지들의 수를 의미하며, Vol(V|S)는 커뮤니티 S안의 노드들을 제외한 모든 노드들의 집합을 의미한다. Vol(S)는 커뮤니티 S안에 속해 있는 노드들의 degree를 의미하는 것으로 수학식 3과 같다.Where u and v are nodes,

(V | S) denotes a set of all nodes except the nodes in the community S, and V (S) denotes the number of edges belonging to only one node in the community S. Vol (S) denotes the degree of nodes belonging to the community S, and is expressed by Equation (3).

Here, d (u) denotes the degree of the node u.

Conductance is a value calculated for a community. It can mean the ratio of the edges connecting the inside and outside of the community S among all the edges of all the nodes in a certain community S.

The community extracting unit 130 calculates the conductance by adding one to the community of the corresponding core node set in order of the node having the highest PPR score, and when the calculated conductance starts to increase, the community of the corresponding core node set is terminated do.

Specifically, the community extractor 130 calculates the conductance for the community S including only the core node set, and calculates the conductance whenever the node S is added to the community S. Then, the community extracting unit 140 compares the calculated conductance with the previously calculated conductance, and when the value starts to increase, the community extracting unit 140 stops adding the node to the community S and finishes the configuration of the community S. [ Conductance can be reduced while adding nodes with high PPR scores due to the edges in the community S of the newly added node. In other words, the conductance may decrease while adding well connected nodes to the community, and may increase if you begin to add more connected nodes to the outside of the community because there are no more well connected nodes inside. Accordingly, when the conductance starts to increase, the community extractor 130 stops adding the node to the community of the corresponding core node, and terminates the configuration of the community.

In another embodiment, when the conductance becomes greater than '0', the community extractor 130 may stop adding the node to the community of the corresponding core node and terminate the configuration of the community.

As described above, the community extractor 130 extends the community from the core node set composed of the central nodes in each community until the characteristics of the community are maximized, and when the characteristics of the community start to grow, Can be stopped.

As a result, the community extractor 130 executes the PPR in the edge relation diagram to extend the community based on each core node set, and as a result, one core node set can constitute one community. That is, one community per core node set can be constructed.

The conversion unit 140 converts the communities extracted in the edge relation diagram into the communities constituted by the nodes having the degree of node relation. Since the communities extracted from the edge relation are the community for the edges, they must be transformed into the community form of the node relation. Then, finally, the community to which each node belongs is derived, and the overlapping community can be detected through the derived community.

Each of the core node extracting unit 110, the edge relation generating unit 120, the community extracting unit 130, and the converting unit 140 may be implemented by a processor or the like necessary for executing a program on a computing device . As described above, the core node extracting unit 110, the edge relation generating unit 120, the community extracting unit 130, and the converting unit 140 may be implemented by physically independent configurations, As shown in FIG.

FIG. 3 is a flowchart illustrating an overlapping community detection method according to an embodiment of the present invention. FIG. 4 is a diagram illustrating an overlapping community detection method according to an embodiment of the present invention.

Referring to FIG. 3, the overlapping community detection apparatus extracts at least one key node among a plurality of nodes constituting the node relation diagram (S310). At this time, the overlapping community detection apparatus calculates the degree of connection to each of the nodes constituting the degree of node relationship, and extracts a certain number of nodes as key nodes in order of the calculated degree of connection. For example, the overlapping community detection apparatus can extract a core node using a spread hubs method, a gravity centers method, or the like.

When step S310 is performed, the overlapping community detection apparatus generates an edge relation diagram in which a plurality of edges in the node relation diagram are converted into the form of each node (S320). That is, since the overlapping community detection apparatus searches the community for edges and finds the communities to which the nodes belong, the node relation degree should be converted to an edge relation. Therefore, the overlapping community detection apparatus can convert the edges of the node relation diagrams into the nodes of the edge relation diagram, and connect the nodes of the edge relation diagram based on the adjacent edge information in the node relation diagram to generate the edge relation diagram . At this time, the overlapping community detection apparatus can extract the nodes of the edge relation map corresponding to the edges connected to the respective key nodes of the node relation diagram as key node sets.

When the step S320 is performed, the overlapping community detection apparatus extracts a community for each core node set in the edge relation diagram (S330). A detailed description of how the overlapping community detection apparatus extracts a community by a set of core nodes will be described with reference to FIG.

When the step S330 is performed, the overlapping community detection apparatus converts the communities extracted in the edge relation diagram into the communities constituted by the nodes of the node relation degree (S340). In order to detect the community to which each node belongs, it is necessary to convert it into a community form composed of nodes of degree of node relation. Then, finally, the community to which the node belongs is derived, and the overlapping community can be detected through the derived community.

A method by which the overlapping community detection apparatus detects overlapping communities will be described with reference to FIG.

Referring to FIG. 4, (a) and (b) are obtained by transforming the node relation diagram of FIG. That is, in the node relation diagram, edges A, B, C, D, E, F, G, H, I, J, K, L and M are transformed into nodes in the edge relation diagram, The nodes of the edge relation are connected. Then, the two key nodes 'a' and 'b' in the node relation diagram are adjacent to each other, and the edge nodes become two key node sets in the edge relation diagram. E, F, G, H, I), which are the edges (D, E, F, G, (D, E, F, G, H, I) to derive the community of Also, the edge relationship of the edge (A, C) connected to the core node 'b' may be the core node set (A, C) for the inner node (A, C) to derive another community. In other words, the key node a extracted from the node relation diagram becomes the key node set {D, E, F, G, H, I} in the edge relation diagram and the key node b becomes the key node set {A, C} .

Since the key nodes extracted from the node relation diagram are represented as edges in the edge relation diagram, the edge relation diagram is obtained by using the edge nodes (each node in the edge relation diagram) connected with the extracted core node as the core node set Perform a community search. (c), two communities can be extracted by performing a community search from each core node set in the edge relation diagram as shown in (b). E, E, F, G, H, I, J, K, and K can be obtained by performing a community search from the set of key nodes {D, E, F, The community of nodes A, B, and C can be extracted by performing a community search from the core node set {A, C} .

(c), the edge relation including two communities can be transformed into a node relation, as shown in (d). (d), the communities to which each node belongs are deduced, and the overlapping communities can be identified through this.

5 is a flowchart illustrating a method of extracting a community from a set of core nodes according to an embodiment of the present invention.

Referring to FIG. 5, the overlapping community detection apparatus executes a PPR in an edge relation diagram (S510), and calculates a PPR score of all nodes based on the set of core nodes (S520). Since the method of calculating the PPR score uses Equation (1), a detailed description will be omitted.

The overlapping community detection apparatus calculates the conductance by adding the nodes one by one to the community of the corresponding core node set in descending order of the calculated PPR score (S530). Since the method of calculating the conductance uses Equation (2), a detailed description will be omitted.

The overlapping community detection apparatus determines whether the calculated conductance at step S530 increases (S540). That is, the overlapping community detection apparatus determines whether the calculated conductance is greater than the previous conductance.

As a result of the determination in step S540, when the conductance starts to increase, the overlapping community detection apparatus stops the configuration of the community (S550) and determines whether there is a next core node set (S560). At this time, the overlapping community detection apparatus may stop the configuration of the community when the conductance becomes larger than '0'.

If it is determined in step S560 that the next key node set exists, the overlapping community detection apparatus performs step S520.

The above-described embodiments of the present invention can be embodied in a general-purpose digital computer that can be embodied as a program that can be executed by a computer and operates the program using a computer-readable recording medium.

The computer readable recording medium includes a magnetic storage medium (e.g., ROM, floppy disk, hard disk, etc.), optical reading medium (e.g., CD ROM, DVD, etc.).

The present invention has been described with reference to the preferred embodiments. It will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention as defined by the appended claims. Therefore, the disclosed embodiments should be considered in an illustrative rather than a restrictive sense. The scope of the present invention is defined by the appended claims rather than by the foregoing description, and all differences within the scope of equivalents thereof should be construed as being included in the present invention.

100: overlapping community detection device
110: Core node extracting unit
120: Edge relation diagram generation unit
130: community extraction unit
140:

Claims (14)

A core node extracting unit for extracting a certain number of nodes to a core node based on degree of connection in a node relation diagram composed of a plurality of nodes and a plurality of edges connecting the plurality of nodes;
An edge relation diagram generation unit which generates an edge relation diagram in which a plurality of edges in the node relation diagram are transformed into respective node shapes;
A community extractor for extracting a community for each core node set corresponding to each of the core nodes in the edge relation diagram; And
A transformation unit for transforming the communities extracted in the edge relation diagram into communities constituted by nodes of the node degree diagram,
And an overlapping community detecting unit for detecting the overlapping community. The method according to claim 1,
The core node extracting unit
Calculates a connection degree for each of the plurality of nodes, and extracts a predetermined number of nodes as a core node in order of the node having the calculated connection degree. 3. The method of claim 2,
The core node extracting unit
Wherein each time the core node is extracted, the core node extracted in accordance with the characteristics and the number of the community, and the nodes adjacent to the one-step adjacent nodes or the two adjacent steps are excluded from the core node candidate group for extracting the next core node. Community detection device. The method according to claim 1,
The edge relation generating unit
Wherein each edge in the node relationship diagram is transformed into a respective node and the transformed nodes are connected based on adjacent edge information in the node relation diagram to generate an edge relation diagram. 5. The method of claim 4,
The edge relation generating unit
And extracts the nodes of the node relation diagram corresponding to edges connected to each of the key nodes of the node relation diagram as a core node aggregate. The method according to claim 1,
The community extractor
In the edge relation diagram, PPR (Personalized PageRank) is executed to calculate the PPR scores of all the nodes based on the set of core nodes, and to add the calculated PPR scores to the community of the corresponding core node set in descending order of nodes Wherein the overlapping community detection unit detects the overlapping community. The method according to claim 6,
The community extractor
Each time a node is added to the community of the core node set, a conductance indicating a ratio of an edge connected to the outside of the community is calculated, and the configuration of the community is terminated based on the calculated conductance The overlapping community detection device. A method for an overlapping community detection device detecting overlapping communities,
Extracting a certain number of nodes as a core node based on degree of connection in a node relation diagram composed of a plurality of nodes and a plurality of edges connecting the plurality of nodes;
Generating an edge relationship diagram in which a plurality of edges in the node relationship diagram are transformed into respective node shapes;
Extracting a community for each core node set corresponding to each of the core nodes in the edge relation diagram; And
Transforming the communities extracted in the edge relation diagram into communities constituted by nodes of the node degree diagram
And detecting the overlapping community. 9. The method of claim 8,
The step of extracting to the core node comprises:
Calculating a connection degree for each of the plurality of nodes, and extracting a predetermined number of nodes as a key node in order of the node having the calculated connection degree. 10. The method of claim 9,
Wherein each time the core node is extracted, the core node extracted in accordance with the characteristics and the number of the community, and the nodes adjacent to the one-step adjacent nodes or the two adjacent steps are excluded from the core node candidate group for extracting the next core node. Community detection method. 9. The method of claim 8,
Wherein the step of generating the edge relationship comprises:
Converting each of the edges in the node relationship diagram to a respective node; And
And connecting the transformed nodes based on neighboring edge information in the node relationship diagram. 12. The method of claim 11,
In the step of creating the edge relation,
And extracting, as a core node set, nodes of the node relation diagram corresponding to edges connected to each of the key nodes of the node relation diagram. 9. The method of claim 8,
The step of extracting each of the communities includes:
Executing PPR (Personalized PageRank) in the edge relation diagram;
Calculating a PPR score of each node based on each core node set; And
And adding the computed PPR score to the community of the core node set in descending order of nodes. 14. The method of claim 13,
Calculating a conductance indicating a ratio of an edge connected to the outside of the community each time a node is added to the community of the core node set; And
And terminating the configuration of the community when the calculated conductance is greater than a previous value.

Images