| As one of the most important characteristics of complex networks,community structure plays a very important role in network analysis.Identification and analysis of communities of networks can help comprehend how the network is organized and how individual modules function.The existing community detection methods generally thought the connection between nodes is generated by a single relationship,which did not consider the more common phenomenon in the real world,i.e.,the connection between nodes is usually generated by multiple underlying relationships(for example,the underlying relationships of two connected people in social networks may be colleagues as well as relatives).In addition,the existing methods usually assume that the observed networks are complete,i.e.,there are no missing edges in networks.However,in reality some edges in the network maybe cannot be fully observed due to their own limitations or acquisition techniques.To solve the above problems,this paper carries out research work in the following two aspects:1)Considering the situation that there may be multiple underlying relationships between nodes in the network,this paper proposes a Poisson-Gamma topic model that characterizes multiple underlying relationships.This model describes the generation process of node contents and link structure(edges)under different underlying relationships,and further the content information and topological information of all underlying relationships are aggregated using the law of total expectation,so as to explore multiple underlying relationships in the link structure between nodes and interpret their semantics.As for the model inference,this paper further proposes a closed Gibbs sampling algorithm.Compared with eight representative community detection methods on eight real data sets,experimental results show the effectiveness of the method proposed.In addition,visualization and case study of link structures in all potential relations are carried out to show that the proposed method in this paper can effectively explore the link structure between nodes in multiple underlying relationships,and interpret semantic information in link relationships by using node contents.2)The method can more accurately discover the community structure by characterizing multiple underlying relationships between nodes,and meanwhile can also interpret the link relationships between nodes.However,there are usually missing edges in many real-world networks.In order to make the model be suitable for the case of missing edges,this paper further proposes a multi-relational Poisson-Gamma model on edges-missing networks.This model first adopts a community self-guided generative model to effectively complement the edges which plays an important role in the community structure by using the membership relationship between the community and nodes,and then the completed network is used as the input of the Poisson-Gamma topic model describing multiple underlying relationships to realize community detection on edges-missing networks.Compared with eight representative methods,experimental results showed that the proposed method can effectively reveal the community structure on edge-missing networks.In addition,the visualization and case study of the detected link relationships showed that the proposed method in this paper can not only complete the missing edges in the network,but also can give explanation on which underlying relationships jointly generate these existing edges. |
