Research On Overlapping Community Detection Algorithms In Heterogeneous Networks

Recently, the research on the relationship between users in the networks becomes more and more important to our understanding of the functions of social networks as the web 2.0 develops. Community structure is a crucial character which is groups of vertices that have more intra-connections than inter-connections. It is a mesoscope structure of networks with respect to vertices and the whole network. And we can realize how the network works and the roles each vertex plays after we get the community structure of the network. Community detection has become quite popular in the field of social network research.Usually vertices are shared between communities rather than independent. For example,if the research fields are seen as communities, some scholars who have achievements in more than one fields are the overlapping vertices of different communities. Overlapping vertices act as bridges between different communities and play an important role in communication between communities as weak ties in networks.However, most methods can only deal with the problem of community detection in the networks in which there is only one kind of vertex with only one kind of interaction. The results can not explain the reason why the nodes are in the same community as well as the meanings of the overlapping nodes. And only a single kind of interaction fails to express the relationship between nodes accurately so that it may miss some important data. In this paper,we work for the problem of overlapping community detection in heterogeneous networks which traditional methods can not deal with.Heterogeneous networks contains vertex heterogeneous networks and edge heterogeneous networks(also relation heterogeneous). In this paper, we use nonnegative matrix tri-factorization to co-cluster the vertex heterogeneous network that the result contains communities of both types of vertices with which they have interactions. And we can explain the physical meaning of the communities of one type of vertex via the communities of the other type of vertex.Also there may exist various types of interactions between vertices that the relationships between vertices can not be represented by any single interaction. So in this paper, we also did some research on the problem of overlapping community detection in edge heterogeneous networks. There are more than one form of interactions between vertices in edge heterogeneous networks, and each can be represented as a graph. We present an algorithm to integrate relation heterogeneous networks into a single network through ranking the neighbours of each vertex that can express the relationship between nodes more accurately.And then overlapping community detection is done by an improved label propagation algorithm.In this paper, we do research on the characters of vertex and relation heterogeneous networks and propose an overlapping community detection algorithm for the two kinds of heterogeneous networks respectively. Experiments in both synthetic and real world networks show that the methods presented in this paper has an excellent performance in identifying overlapping communities in heterogeneous networks.