Similarity community clustering method and system capable of defining targets

Abstract

The invention relates to a similarity community clustering method and system capable of defining targets, and belongs to the field of computer networks. The clustering method comprises the steps that a user customizes a similarity measurement method and a threshold value, constructs a fully connected graph, calculates the number of expected communities, aggregates attributes through a neural network, iteratively updates a loss function, and outputs a division result. The system framework comprises a data input module, a user definition module, a community trial calculation module, a community clustering module and a community output module. According to the method, the target is customized by the user, the number of communities does not need to be specified in advance, the privacy disclosure risk can be reduced, and the community clustering accuracy can be improved.

Description

Technical Field

[0001] The present invention relates to the field of computer networks, and in particular to a similarity community clustering method and system capable of defining targets. Background Art

[0002] Community clustering is the process of dividing a network into several subgroups (called communities). In social networks, community clustering helps to reveal the structure and patterns in the network, and can also be used to analyze the behavior and characteristics of individuals in the network. The number of communities in a network is usually unknown and difficult to obtain through simple observation or analysis. The traditional K-means clustering method requires the pre-specification of the number of communities K, but this step is not only costly, but also may bring the risk of privacy leakage. Therefore, automatically determining the appropriate number of communities is a challenging problem. To this end, the present invention proposes a similarity community clustering method and system with definable targets, which can not only perform community clustering according to user goals, but also does not need to pre-specify the number of communities K. This method can not only reduce the risk of privacy leakage, but also improve the accuracy of community clustering. Summary of the invention

[0003] In view of the problem that the K-means clustering method requires the pre-specification of the number of communities K and may cause privacy leakage risks, the present invention proposes a similarity community clustering method and system with definable targets, which reduces the risk of privacy leakage and improves the accuracy of community clustering by allowing users to define targets. The technical solutions adopted by the present invention are as follows:

[0004] A similarity community clustering method capable of defining a target, the method comprising the following steps:

[0005] (1) User-defined similarity measurement method and threshold;

[0006] (2) Construct a fully connected graph and calculate the expected number of communities K based on the user-defined similarity measurement method and threshold;

[0007] (3) Using neural networks to aggregate the attribute information of each node;

[0008] (4) Iteratively adjust the number of communities K according to the loss function;

[0009] (5) After iterating for a specified number of times, the community clustering results are output.

[0010] The user-defined similarity measurement method and threshold in the above step (1) means that the user can customize the similarity measurement method and threshold according to his / her own segmentation target.

[0011] The expected number of communities K mentioned in step (2) above refers to the similarity between any two nodes calculated according to the user-defined similarity measurement method. If the similarity between the two nodes is lower than the specified threshold, the edge between the two nodes is deleted, otherwise the edge is retained until all nodes are traversed to obtain the expected number of communities K.

[0012] The iterative adjustment of the number of communities K according to the loss function in the above step (4) refers to updating the parameters of the neural network according to the feedback result of the loss function, and repeating steps (3) and (4) until the loss function no longer decreases significantly.

[0013] The present invention is also used for a similarity community clustering system capable of defining a target, comprising the following modules:

[0014] (1) Data input module: the starting point of this system, used to obtain the number of nodes and their attribute information;

[0015] (2) User-defined module: obtains user-defined similarity measurement methods and thresholds;

[0016] (3) Community calculation module: construct a fully connected graph, and then calculate the similarity between any two nodes according to the user-defined similarity measurement method. If the similarity between the two nodes is lower than the specified threshold, the edge between the two nodes is deleted, otherwise the edge is retained until all nodes are traversed and the expected number of communities K is obtained;

[0017] (4) Community clustering module: A neural network is used to aggregate the attribute information of each node, and then the parameters of the neural network are updated according to the feedback results of the loss function. The process of neural network training and parameter updating is iteratively executed until the loss function no longer decreases significantly.

[0018] (5) Community output module: The terminal of this system is used to output the community clustering results that reach the specified number of iterations.

[0019] Beneficial Effects

[0020] According to the technical solution provided by the present invention, the present invention discloses the following technical advantages:

[0021] (1) User definable: The present invention allows users to customize similarity measurement methods and thresholds to calculate the expected number of communities K, thereby achieving community clustering consistent with the user's goals.

[0022] (2) High degree of automation: There is no need to pre-specify the number of communities K. Through the user-defined similarity measurement method and threshold, the present invention can automatically calculate the expected number of communities K without pre-specifying. This feature not only reduces the cost of determining the number of communities K, but also effectively reduces the risk of privacy leakage.

[0023] (3) High accuracy of division: Based on the number of communities expected by the user, the present invention performs community clustering by iteratively adjusting the neural network through the loss function, thereby ensuring high accuracy of community clustering. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 Flowchart of similarity community clustering method with definable targets;

[0025] Figure 2 This is an example diagram of similarity measurement method;

[0026] Figure 3 The architecture diagram of the similarity community clustering system with definable targets. DETAILED DESCRIPTION

[0027] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0028] Embodiment 1:

[0029] This embodiment is used to provide a user-defined similarity measurement method and threshold, such as Figure 2 As shown, the method comprises the following steps:

[0030] (1) Assuming that the user wants to find friends with similar attributes in the community, the user first customizes the similarity measurement method as a weighted combination of the Jaccard similarity coefficient and the cosine similarity, and sets the weights to 0.4 and 0.6 respectively. The Jaccard similarity coefficient measures the similarity of two sets by calculating the ratio of the intersection size to the union size, and the cosine similarity measures the similarity between two vectors by calculating the ratio of the dot product of two vectors to the product of their norms. Finally, the user sets the similarity threshold to 0.6. If the value calculated by the customized similarity measurement method is lower than 0.6, it is considered that the two people do not have similar attributes. Each lowercase letter represents a different attribute.

[0031] (2) Construct a fully connected graph.

[0032] (3) The similarity between any two nodes is calculated based on the user-defined similarity measurement method. If the similarity between the two nodes is lower than the specified threshold, the edge between the two nodes is deleted. Otherwise, the edge is retained until all nodes are traversed and the expected number of communities K is obtained.

[0033] Although the embodiments of the present invention are described in conjunction with the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present invention, and such modifications and variations are all within the scope defined by the claims of the present invention.

Claims

1. A similarity community clustering method with a definable target, characterized in that: The following steps are involved: (1) User-defined similarity measurement method and threshold; (2) Construct a fully connected graph and calculate the expected number of communities K based on the user-defined similarity measurement method and threshold; (3) Using neural networks to aggregate the attribute information of each node; (4) Iteratively adjust the number of communities K according to the loss function; (5) Output the community clustering results that reach the specified number of iterations.

2. A similarity community clustering method with a definable target based on claim 1, characterized in that: The user-defined similarity measurement method and threshold in step (1) means that the user can customize the similarity measurement method and threshold according to his / her own segmentation target.

3. A similarity community clustering method with a definable target based on claim 1, characterized in that: Calculating the expected number of communities K in step (2) refers to calculating the similarity between any two nodes according to a user-defined similarity measurement method. If the similarity between the two nodes is lower than a specified threshold, the edge between the two nodes is deleted; otherwise, the edge is retained until all nodes are traversed to obtain the expected number of communities K.

4. A similarity community clustering method with a definable target based on claim 1, characterized in that: The iterative adjustment of the number of communities K according to the loss function in step (4) refers to updating the parameters of the neural network according to the feedback result of the loss function, and repeating steps (3) and (4) until the loss function no longer decreases significantly.

5. A similarity community clustering system with definable targets, characterized in that: Includes the following modules: (1) Data input module: the starting point of this system, used to obtain the number of nodes and their attribute information; (2) User-defined module: obtains user-defined similarity measurement methods and thresholds; (3) Community calculation module: construct a fully connected graph, and then calculate the similarity between any two nodes according to the user-defined similarity measurement method. If the similarity between the two nodes is lower than the specified threshold, the edge between the two nodes is deleted, otherwise the edge is retained until all nodes are traversed and the expected number of communities K is obtained; (4) Community clustering module: A neural network is used to aggregate the attribute information of each node, and then the parameters of the neural network are updated according to the feedback results of the loss function. The process of neural network training and parameter updating is iteratively executed until the loss function no longer decreases significantly. (5) Community output module: The terminal of this system is used to output the community clustering results that reach the specified number of iterations.

Images