Research On Semantic Prediction Algorithm For Searching And Clustering In Heterogenerous Network

With the increasing scale of heterogeneous networks,how to effectively carry out data mining on heterogeneous networks has become a hot research topic.As a basic component of heterogeneous network,meta-path contains rich semantics.When it is applied to data mining operations such as similarity search and clustering,can effectively use these semantic information to analyze heterogeneous network,which has been paid more and more attention by the researchers and many algorithm based on meta-path have been proposed.However,these algorithms still need to be improved in terms of performance.For example,the similarity search algorithm based on meta-path involves a large number of matrix calculations;the clustering algorithm based on meta-path does not filter paths,and irrelated paths will have interference on cluster effect,etc.In view of these problems,this paper proposes a semantic prediction algorithm for heterogeneous network search and clustering based on user guide to alleviate the above problems.Namely,the user to provide a sample of results as a guide while searching or clustering,and then predict the meta-path associated with the user search or clustering based on user guide.It will eliminates irrelevant meta-paths at the beginning of the algorithm,thereby reducing the impact of irrelevant paths on the results.The semantic prediction algorithm for heterogeneous network search carry out path backtracking in the graph based on user guide,to obtains the meta-path related to the search and the number of the path instance on the corresponding path,and the path length is constrained according to the number of regular paths,the probability of the corresponding path is obtained by the machine learning method,and the similarity calculation is carried out;the semantic prediction algorithm for heterogeneous network clustering takes into account the relationship between user guide?clustering nodes and clustering categories,and use the meta-path ratio to calculate the relational weights between user guide and clustered nodes,and then calculate the probability of generation between the user guide and the clustering nodes to clustering.Experimental results show that the semantic prediction algorithm proposed in this paper can not only predict the semantics of nodes in heterogeneous networks,but also has better performance when apply to practical problems combining with heterogeneous network search and clustering algorithm.