Meta-path Based Classification And Computation Approach In Heterogen-ous Information Network

With the rapid development of social networks in recent years,it has become very difficult to find the interested information with the large scale data.Traditional social network analysis is often orientedhomogeneous networks and focuses on the structural characteristics of the network.However,the network drawn in the real world often contains multiple types of nodes and multiple types of relationships,forming heterogeneous information network with rich structure and semantics.The model of heterogeneous information network is undoubtedly more in line with the original reality of the real world.Through the analysis of heterogeneous information network,we can mine more accurate and rich network connotations.For the discovery and analysis of heterogeneous information network,meta-path based method is an effective and important approach.In heterogeneous information network,complex link relationship is formed by linking the nodes through multiple paths,so that complex relationship can be clearly represented and measured by metapath.The study of meta-path semantic information plays an important role in the discovery and analysis of heterogeneous information network.This paper focuses on network construction with meta-path and calculation of different semantic meta-paths to implement classification and community discovery in heterogeneous information network.The main work of the paper is as follows:1)Measure the similarity between nodes and self-learning of meta-path weight.It mainly studies the method of similarity measurement based on meta-path and construct the similarity matrix between nodes.It analyzes the influencing factors of the weight of the meta-path,and proposes the method of actively learning the weight of the metapath.2)Classification of the node based on meta-path in Heterogeneous information network.We construct feature matrix used to classify the nodes based on the meta-path information.In order to solve the problem of matrix sparseness,jump path is added to enrich the heterogeneous information network.On the basis of manually setting of path weight,the method of active learning is used to improve the classification accuracy.3)Meta-path based community discovery and calculation of nodes influence in heterogeneous information network.This paper proposes community discovery algorithm based on meta-path semantics in heterogeneous information network.And it analyzes the impact of different types of meta-path on community discovery.It studies the important nodes in the community and proposes method of weighted cooperative relationships and method of Author-ranking respectively.Experimental results show that the meta-path based classification algorithm proposed in the paper can effectively classify nodes in a heterogeneous information network.After applying the active learning algorithm of the meta-path weight,the classification accuracy is improved and the performance is superior to the traditional classification algorithm.For the community discovery research,the algorithm proposed in the paper can effectively divide the community.Author-ranking algorithm can evaluate the influence of nodes and identify important nodes objectively.