Research On Graph Query And Visualization For Heterogeneous Information Network

Graph has very strong expressive power.Graph structure data usually contains a lot of useful information.Therefore,graph query technology,which is used to obtain information from graph structure data,has attracted extensive attention in various fields.Traditional graph query is mainly aimed at homogeneous networks,but graph data in real life are mostly heterogeneous,such as bibliographical network,social network,protein interaction network,etc.There are many types of nodes and links in these networks,which are defined as heterogeneous information networks.The multi-type and heterogeneous characteristics of heterogeneous information networks have brought challenges to the task of graph query.Based on the concept of metapath and short text semantics,this paper designs and implements a graph query visual analysis system based on heterogeneous film data.The work of this paper mainly includes:(1)Combining short text for generating meta-path and importance.By analyzing the factors that affect the importance of meta-path and the role of short text semantics in the process of generating meta-path,we have defined the indexes related to the length,rarity and strength of meta-path,and obtained the calculation function of meta-path importance by combining the three indexes.In this paper,greedy tree data structure is used to generate metapaths and calculate the importance of metapaths.(2)Instance generation and vectorization with multi-metapaths.In this paper,we have calculated the similarity of node pairs based on different meta-paths.And the similarity of node pairs is weighted by the importance of meta-paths and summed,thus instance node pairs that conform to the meta-path relation pattern and semantics can be generated.Furthermore,we have combined the patterns of multi-metapaths to construct subgraphs,extracted topological structure features in subgraphs,and generated the feature matrix.In order to further mine the information in the query results,we have reduced and clustered the feature matrix.Finally,a verification experiment based on author tag semantics has been carried out with DBLP dataset.The experimental result shows that the meta-path generated by combining short text semantics can generate instance node pairs containing short text semantics.(3)Query-driven visualization of heterogeneous information networks.On the basis of the above research,we design and implement a visual analysis system of graph query based on heterogeneous Douban movie network.The visual design of the system is driven by graph query tasks.In order to meet the task requirements of the three stages of graph query input,subgraph query and result analysis,we have already designed input view,pattern view,summary view,sub-graph view and attribute view.Furthermore,case analysis and user experiments are carried out to verify that the system can effectively complete the tasks of graph query.