The Study And Application Of Entity Association Representation Learning Method On Heterogeneous Network

The network usually refers to complex graph with a larger scale,which includes various entities with rich attribute characteristics and the associated relations between them.A heterogeneous network is one kind of network.The core of heterogeneous network entity association representation learning is to effectively vectorize the nodes in the network to facilitate various follow-up tasks.The study of entity association representation learning in heterogeneous networks is not only conducive to the understanding and mining of heterogeneous network information,but also conducive to the related application of heterogeneous networks and the study of dynamic heterogeneous networks.Based on the graph neural network,the existing model is improved and innovated in two aspects.Methods not dependent on the generating Homogeneous network,are only considering the node heterogeneity alone,it is difficult to further mining network heterogeneity.Through modeling in a heterogeneous network edge pattern,the network heterogeneity can be considered under the actions of nodes and edges at the same time.Based on key-value-based attention network KAHN,heterogeneous networks can be effectively represented and processed.By summarizing and analyzing the existing ideas to solve the problem,a combined attention-based network CAN is proposed.By combining different models,two kinds of network information can be learned simultaneously,increasing the amount of network information that the model can use.KAHN can be used as an implementation model of CAN and CAN could improve the network learning ability of KAHN and similar models through further utilization of heterogeneity.Through node classification experiments and analysis in multiple datasets,KAHN’s better classification effect is demonstrated,and KAHN model’s better representation learning ability is verified,indicating that heterogeneity in edge mode can improve network information mining.Compared with the basic model,CAN constructed by the basic model has higher node classification accuracy,which verifies the performance improvement effect of CAN framework on the model and the two types of information can complement each other.Meanwhile,it indicates that the information enhancement under the meta-path can indeed bring about the improvement of network learning ability.