Research On Tensor-based Community Detection In Complex Networks

Complex information networks generally have a large scale of entities,and the data represented for them are simultaneously diverse and heterogeneous.With the development of science and technology,the interaction information between many entities can be recorded.The community discovery technology of complex networks has attracted more attention,due to its wide applications in group research,smart recommendation,and information dissemination control.People are no longer satisfied with the simple single-graph analysis,and become more interested in synthesizing various information in a multi-relational network,to obtain the group that is more closely linked to the real situation through a series of complex features.Thus,this dissertation focuses on the study of multi-relational networks with heterogeneous,sparse characteristics and complex topological structures.The tensor is used to represent the network structure.With the results of the tensor Markov chain probability estimation model,an improved seed community discovery algorithm based on selecting probabilities is proposed.The main improvements and innovations of this algorithm are:For the selection of initial seed nodes,a node tightness evaluation algorithm based on link correlation is proposed.By calculating the degree of local network topological convergence among the neighbor nodes of potential core nodes,the core nodes satisfying the multi-level strong structure conditions are finally selected as seeds to ensure the centrality of the seed nodes.For the seed expansion strategy,a collaborative probability estimation model is proposed for optimization.Taking the core nodes as the initial seeds,several closely related seed nodes are iteratively generated,and several sub-communities are generated based on the maximum value of the steady-state access probability of the nodes.The normal distribution functions defined by the cooperative probability estimation are used to estimate the probability of merging into the target community for all the nodes in the sub-community.It aims to remove redundantly incorporated nodes and ensure the community expansion accuracy.At last,comparative experiments on multi-relational real-world network data are designed to examine the performance of algorithms under different kinds of assessment standards.Results show our solutions effectively improve the robustness and accuracy of the seed community discovery algorithm.