A Novel Link Prediction Method Integrated Link Attributes For Directed Graph

Link prediction is an important research direction in complex network analysis.Its purpose is to predict the possible links between two nodes according to the known network information.Link prediction is very important for understanding the structure of complex networks.As an effective means of complex network analysis,link prediction method has important research value in the fields of social network research,recommendation system,drug action research and so on.In real society,links in most networks not only have direction,but also often have some specific attributes.For example,in the voting network,some comments will be generated during the voting process,which can be understood as the voter's support for the voter.In directed networks with semantic information,the semantic information on links will have a decisive impact on the results of link prediction.Therefore,based on the node structure prediction,this paper mainly integrates the link attributes to predict the directed network,and carries out the research work from the following aspects:1.Link prediction methods based on network structure similarity often ignore the impact of link attributes on link prediction results.Therefore,BILSTM emotion analysis model is introduced into link prediction problem,and a new emotion analysis method is proposed.This method uses BILSTM model to quantify the weight of each link according to the semantics of each link.This weight will participate in the link prediction method as an important parameter.We use common neighbor similarity and preferential attraction similarity to evaluate the relationship between nodes.Based on the similarity of the two nodes and the weight of the above links,two new link prediction methods are proposed.2.The method proposed in this paper is tested on public data sets.After experimental verification and analysis,the method proposed in this paper combines the advantages of node structure similarity method and link attribute,and achieves good results in the prediction of directed and weighted networks.In this paper,a new link prediction method framework is proposed.In this method framework,the BILSTM emotion analysis model is introduced to quantify the text attribute characteristics of links.Combined with the node structural similarity method:common neighbor link prediction method and priority attraction link prediction method,two link prediction methods are proposed respectively: common neighbor link prediction method integrating link attributes and priority attraction link prediction method integrating link attributes,Experimental results in real data sets verify the effectiveness of the proposed method.