Research On Community Detection In Complex Networks Based On Topological Similarity And Node Attributes

Recently,with the emergence of information age,more and more scholars apply community detection to social network analysis,network visualization,social relationship recommendation system and other specific problems.Community is a kind of clustering structure,which is composed of nodes and their connected edges.It can help us better understand the complex network system and detect the hidden information in the network.In the face of explosive data,how to make full use of and extract these implied data information is the current concern of scholars.Game theory is a theory to study the strategic interaction between game players.It can be used to explain the top-down formation process of communities in complex networks.In recent years,many scholars have combined community detection with game theory to model the detection process of network community structure as a game for the community.This kind of community detection research based on game theory has achieved good results,which proves the rationality and effectiveness of the game theory used in community detection algorithm.Now,most of the community detection algorithms only consider the topological structure of the network and ignore the attribute information of the nodes.In real life,users on social networks leave a lot of information,which not only enriches the structural characteristics of the network,but also reflects the behavior of users.Therefore,the research of community detection combined with node attributes will be a challenging work.For the research of this kind of network with attributes,the pure topology algorithm can not be applied directly.How to effectively integrate network topology information and node attribute information is the key to solve this kind of network community detection problem.In this paper,the existing community detection algorithm and the community detection algorithm based on game theory are analyzed,and a community detection model based on non-cooperative game is proposed.On the basis of the model,a community detection method combining node topological similarity and node attribute information is proposed.The paper mainly focuses on the following three parts:(1)This paper uses an efficient weighting-and-rewiring scheme to enhance the global community structure of a target network by explicitly incorporating higher order node proximities.This could be also used to alleviate the “resolution limit” problem in classic community detection domain.(2)At present,most community detection algorithms are based on optimizing a pre-defined objective function to detect community.This paper formulates the overlapping community detection problem as a non-cooperative game played by N players and proposes a utility function based on the topological similarity of nodes,and divides the weighted and reconnected network into three parts,and prove the game is a potential game that possesses pure Nash equilibrium.(3)Based on the community detection model of non-cooperative game,the likelihood function of network is established by combining the attribute information of nodes.The weight adjacency matrix and the maximum likelihood function of the attribute information of nodes are used to obtain the community membership vector of nodes and finally determine the community ownership of nodes.