Network Embedding Based Multi-granularity Community Detection Method

As an important topological structure in the network,community structure has an important theoretical research significance and application value in network analysis tasks.In recent years,research on community detection methods has never stopped,and many new community detection methods are proposed.The current researches focused primarily on the most significant community structure in networks,and the research on the hidden community structures is insufficient.However,the hidden community structures in networks are often of high value in practical applications.Therefore,constructing a multi-granularity community structure from significant to hidden is of great significance as it can provide a more comprehensive perspective for us to analyse and understand the network structure.In view of the above problems,multi-granularity community detection was discussed in this thesis.The main contents include:1.In order to find the hidden community structures in networks,a community structure weakening algorithm based on network embedding is proposed.At first,the algorithm uses network embedding methods to represent each node as a low-dimensional vector,which overcomes the shortcomings of the traditional adjacency matrix based network representation method such as high dimensional sparsity,high computational complexity,low parallelizability,inapplicability of machine learning methods and so on.Then,assuming that the embeddings of all nodes are generated by a same Gaussian mixture model,the embeddings are fitted to a Gaussian mixture model.Finally,by reducing the probability of the node belongs to its original community,the probability that the node belongs to other unknown communities is increased,and the purpose of weakening the community structure is achieved.The experimental results show that the algorithm can effectively weaken the detected community structures,and it is helpful for finding the hidden community structures in networks.2.In order to construct the multi-granularity community structure model,a community structure weakening algorithm based multi-granularity community detection method is proposed.Based on the community structure weakening algorithm proposed in this thesis,the algorithm construct the multi-granularity community structure model by using the community structure weakening algorithm and community detection algorithm iteratively.The multi-granularity community structure not only includes the most significant community structures,but also the hidden community structures that cannot be found by traditional community detection methods.The multi-granularity community detection algorithm provides an effective way for more comprehensive analysis and understanding of networks,and has important practical application value.The experimental results show that the algorithm can construct a multi-granularity community structure of the network,meanwhile,the algorithm has better performance compared with the existing methods on most experimental data sets.