Research On Cross-graph Relational Learning Method And Its Application Based On Graph Representation Technology

With the gradual maturity of information technology and the widespread popularity of smart terminals,the manifestations of information data are becoming increasingly diversified.For example,multi-modal heterogeneous data such as numbers,images,text,audio,and voice have gradually become the main carriers of information.The main form of connecting these multi-modal data is heterogeneous graphs.Therefore,how to conduct data mining based on this heterogeneous graph and analyze the connections between various entities in the network has become an important research direction in the current cross-media and data mining fields.Based on heterogeneous graph data,this paper proposes a novel cross-graph relational learning model CGRLGE(Cross-graph Relational Learning based on Graph Embedding).The model is based on graph representation technology.It first uses a meta-path-based random walk method to sample the nodes in the graph,and then builds various types of nodes in the heterogeneous graph and multiple association relationships between nodes.Modular learning,the model finally learns the vector representation of the node in the embedding space by solving the optimal solution of a specific objective function,and realizes the mapping of node features from the heterogeneous graph space with complex structure to the low-dimensional embedding space.In the mapped embedding space,based on the learned node representation vector,node classification and network recovery experiments were performed on multiple real data sets.The experimental results show that the CGRLGE model can fully mine the relationship between various types of nodes in a heterogeneous graph,and the learned node representation vector can fully capture the characteristics of the nodes and the correlation between nodes,in node classification and link recovery,etc.In actual task analysis,the CGRLGE model can achieve better performance than other algorithm models in the same field,and can obtain more prominent experimental results under multiple dimensional evaluation standards.