Research On The Method Of Co-Author Relationship Analysis Based On Graph Neural Network

Scientific research plays a key role in Chinese development to become a scientific and technological strength in the world.The vigorous arise of a new round of scientific and technological revolution makes the interdisciplinary integration closer and the demand for scientific research cooperation stronger.As an important illustration of scientific research cooperation,the co-author relationship brings up core or key technologies which can not only contribute to national development and the improvement of social living standards,but also contribute to the improvement of China’s scientific and technological competitiveness on a global scale.Based on the graph neural network,this paper mainly analyzes the co-author relationship from the two perspectives of node and relationship,and explores the potential co-author relationship in the academic heterogeneous information network.The research contents of this paper are as following.Firstly,according to the heterogeneity of academic information network and the richness of semantic information,this paper proposes a node-classification model of graph attention neural network based on hierarchical attention mechanism,including node-level attention mechanism and semantic-level attention mechanism.The node-level attention is designed to understand the importance between nodes and their meta-path based neighbors,while the semantic-level attention mechanism is designed to understand the importance between different meta-paths.By learning the importance of node features and semantic features,the importance of nodes and meta-paths is fully considered,and the neighbor features based on meta-paths are hierarchically aggregated to generate node embedding.Thus,the purpose of node classification is realized.Then,based on heuristic similarity measure function of predefined limits and the consumption of high order characteristic learning problems,this paper combines decay theory and the heuristic algorithm facing the network relations and puts forward a new method based on the theory of the attenuation of heuristic algorithm.The algorithm can unify the link prediction higher-order heuristic similarity measurement method.By extracting local subgraph around each target link,the method learn subgraph patterns mapped to link the existence of the function,thereby automatically heuristic learning is suitable for the current network.At the same time,a heuristic model of learning from local subgraph by graph neural network is constructed.Finally,this paper constructs a real academic heterogeneous information network on the open data set DBLP,and carries out experimental verification on the two co-author relationship prediction models and related algorithms proposed above.The validity of the method is proved by comparing the results of the experiment.