Link Prediction:According To The Contribution Of Method Research

As an interdisciplinary research problem,which is in many areas,such as complex network?data mining and information science,link prediction plays an increasingly prominent role in theory and application.In order to tap the network structures effectively and do more in-depth understanding of the complex network evolution mechanisms in recent years,researchers have put forward to a lot of link prediction methods with network structures.However,these algorithms are still much room for improvement in terms of performance,the fundamental reason lies in the existence of vast research space of network structure.Tracking domestic and international research frontiers,this work devotes to deep exploration and experiment in view of link prediction problem based on the network structure as follows:(1)Making more in-depth exploration and studying the current domestic and international popular link prediction methods,we propose the motivation and objection of the novel algorithm based on analyzing and listing the advantages and disadvantages of the various methods comparatively.Most of the link prediction approaches are measuring the similarity between two nodes through using structural features of their own.However,they often overlooked the tightness of the common nodes,and this is an important network structure feature.Thus,this paper defines the node contribution,integrates the information of edge into the definition of point,which is not only able to describe the tightness of the nodes,but also to portray the differences between nodes and measure the similarity scores between two nodes,and finally proposes a new index,which is called a similarity index on the basis of node contribution.Experiments on six real data sets show that the proposed methods have much better performance than each of three classical ones.(2)In considering links between common neighbor nodes,we do future study in the connection between neighbor nodes and summarize this structure,called the links across two neighbor sets.And by defining the link contribution, depicting the links across two neighbor sets,and describing the likelihood of candidate links existing,this paper proposes another link prediction method,called a similarity index on the basis of link contribution.Because different networks are presenting different structure features,in order to in-depth study the importance of different network structural features on the node connection,we adopt a parameter ? to distinguish the contribution of different features to the final connection likelihood and call a similarity index based on the node and link contribution.Comparisons with other methods,we illustrated the feasibility of the new methods on ten real network data sets.In conclusion,the paper describes the link prediction methods from different point of views and proposes three new similarity indexes,and validates the effectiveness and feasibility of new methods on real network data sets.The research results provide new methods and ideas for link prediction analysis,and have theory and application values in some domains such as complex networks and network science.