Research On Citation Network Link Prediction And Community Division Based On SBERT And Common Neighbors

As the carrier of the publication and dissemination of scientific research results,academic articles play an important role in the road of scientific research.The references of the article represent the research background of the current article and the previous work,so that the researchers can stand on the "shoulders of giants" and deepen the research and exploration in the field.The citation network synthesizes the citation and citation relationships among multiple articles,takes a single article as a node,and the citation relationships among articles as a directional edge to connect,forming the mutual citation relationship of academic articles.As a node,articles also contain their own content information.By mining citation network information and applying it to its community division,it is of great practical significance to the research of citation network information transmission and interdisciplinary problems.The main work of this paper is as follows:(1)BERT pre-training model is used to extract semantic information from the literature text,SBERT is used to vectorize,and the pre-training model is trained and fineadjusted with its own data.Each document in the citation network is transformed into a vector with the same growth degree,and cosine similarity is used to calculate the distance between the vectorized nodes.Transform the powerless citation network into the empowered network.(2)Combined with the book classification and time information in the network,the nodes in the citation network processed by the SBERT pre-training model are transformed into vectors,and the links in the network are partially connected or disconnected,which is convenient for further link prediction.(3)The WBCNI algorithm is proposed for entitled network local network information.The algorithm takes common neighbors between nodes as the research object.On the one hand,the closeness of common neighbors between unconnected nodes is considered,and on the other hand,the unequal allocation of resources between common neighbors as a bridge between unconnected nodes is considered.The WBCNI algorithm is proved to be effective in improving the accuracy of link prediction through experiments on 5 real network data in Collaboration Networks database.(4)The experiments are conducted using real citation network data.This paper selects2410 journal literatures from January 2010 to October 2022 of China National Knowledge Network,and uses Louvain algorithm in community division method to conduct community division experiment on the improved weighted citation network.Through quantitative analysis of module degree Q index and qualitative analysis,it is proved that the proposed method can improve the accuracy of community division in citation network.