Research And Implementation Of Co-authorship Prediction Algorithm Based On Temporal Knowledge Graph

In recent decades,the development of modern Internet and related technologies has led to the rapid expansion of online social networks.The social network analysis has received significant interests and concerns of researchers,such as public opinion analysis and control in public opinion social networks,real-time user recommendation and influence analysis of important people in online social systems.As an important sub network of social networks,co-authorship network has also attracted great attention in recent years.With the gradual popularization of academic big data,the vast majority of academic information can be obtained through the internet.Cooperation has always been a key factor for sociologists to analyze social models.At the same time,cooperation is also a necessary choice for researchers to carry out scientific research activities.As an important sub task of the co-authorship network,the co-authorship prediction can provide strong suggestions for the project and paper cooperation among researchers,so as to recommend suitable co-authors for them,and enhance the academic exchange of ideas among researchers.This paper proposes a co-authorship prediction algorithm based on temporal coauthorship knowledge graph.And the algorithm consists of three modules.Firstly,The previous embedding representation methods of knowledge graph are based on graph structure when initializing embeddings,without considering the natural semantic information of entities and relations.This paper designs a semantic encoder based on the pre training language model BERT to capture the semantic information of entities and relations as their initial embedding representation.Secondly,considering that there are a large number of temporal interactions among the neighbors of entities in the temporal co-authorship knowledge graph,this paper designs a temporal graph attention network based on Bi-LSTM neural network,which can effectively capture the time context information between entity neighbors.Considering how to effectively aggregate the multi-hop neighborhood information of entities in time and space,this paper constructs an embedding encoder to aggregate the neighborhood information and obtains the spatial context information based on the proposed temporal graph attention mechanism and the neighborhood strategy.Through the spatiotemporal context information,the embeddings of entities and relations in the knowledge graph can be well updated.Finally,considering how to score the triples to predict co-authorship,this paper designs a decoder,which uses the convolution neural network based on multi-scale convolution kernel strategy to score each triple in the knowledge graph.Experiments on the temporal co-authorship knowledge graphs extracted from academic relations in the real world prove the effectiveness of the co-authorship prediction algorithm.And based on the algorithm,this paper provides a visual interface for coauthorship prediction.Users can query and predict co-authorship on the interface,and the visual interface provides more perfect and intelligent services.