Research On Attributed Community Search Method For Heterogeneous Network

The purpose of community search is to find the community that contains the query vertex set.Existing research mainly focuses on community search for simple graphs and attributed graphs.There are many community models,such as k-core,k-truss,k-ECC and k-clique.Besides,existing research mainly focuses on homogeneous networks of the same vertices and link types,and cannot apply to heterogeneous information networks of multiple types of objects and links.In most heterogeneous information networks,such as bibliographic networks and knowledge graphs,there are various network topology information and nodes' attribute information.The attributed community search can often get personalized communities.But,the existing attributed community search researches rarely consider the comprehensive attribute information or only consider the attribute information without considering the topological structure of the node,which will lead to a more serious free-rider effect and cannot guarantee the attribute aggregation of the community.In response to the above problems,this paper studies an attributed community search method for heterogeneous networks,considering the characteristics of the heterogeneous information networks and the nodes' attribute information on the network.The specific work content includes the following aspects:This paper proposes an attributed community search method for heterogeneous information networks.Give a formal description of the problem and elaborate the basic idea and specific steps of the method,and analyze the complexity of the algorithm.The attributed community search mainly includes preprocessing,getting candidate community,and community search.The preprocessing process uses the concept of meta-path,which defined a sequence of relationships between different vertices types;the community search algorithm is based on the branch-and-bound idea and uses k-truss to constrain the structural cohesiveness of the community.The candidate communities were scored by using the scoring function which integrated the community structure cohesiveness and attribute cohesiveness and we finally got the high-quality communities.Finally,we conducted experiments on three datasets with real community structures,Foursquare,DBLP and IMDB datasets,compared and analyzed the experimental results to verify the effectiveness and efficiency of the aforementioned models and methods.