Research On The Algorithm Of Overlapping Community Detection Based On Social Networks

With the emergence and development of various social platforms,social network data becomes more abundant.More and more researchers begin to analyze and study complex social networks.Community detection is an important research direction for mining and analyzing hidden information of social networks.Network data is often high-dimensional and very large and complex,which makes it very difficult to process.Therefore,the representation of network structure as a low-dimensional vector is of great significance for community detection.Due to the existence of overlapping communities in many complex social networks,many researchers focus on the research of overlapping community detection algorithms.Overlapping community division can more accurately divide the community structure of social networks.At present,there are still some problems to be paid attention to in the overlapping community detection methods for social networks,such as the fact that social networks are often sparse,nodes in the network are different,and overlapping community detection algorithms still have room for improvement in accuracy and partition efficiency.The research analyzes the characteristics of social networks and the problems of the current overlapping community detection algorithms,this thesis makes innovations in the expression of node similarity and overlapping community detection in social networks.The main work is as follows:1.Aiming at the problems of sparseness in social networks and differences among nodes in the network,this thesis proposes a node intimacy representation method based on network representation learning.Firstly,according to the structural information in the network,the social network is represented by low-dimensional vector,which summarizes the topology structure information of nodes more comprehensively and accurately.Then,considering the differences between nodes,the method of measuring the similarity between nodes is extended,and a new method of expressing the node tightness is proposed,which can reflect the similarity between nodes in the network more accurately.Finally,this method is applied to the community detection algorithm based on node similarity,and the results of community division based on other similarity are compared and analyzed.The experimental results show that this method can better reflect the similarity of nodes.2.Aiming at the problems of insufficient accuracy and efficiency of many current overlapping community division methods and poor performance on weighted networks,this thesis proposes an overlapping community detection algorithm based on node intimacy and density peaks.Firstly,a method of node intimacy proposed in this thesis is used to measure the distance between nodes,and the local density calculation method in the original clustering algorithm is improved.Secondly,this thesis choose a community center through three steps.Then,the remaining nodes are allocated according to the node ownership degree.Finally,the experimental results show that this algorithm can achieve a good effect of community division by comparing with the algorithms in recent years.