Research On Link Prediction Methods Based On Complex Networks

With the rapid development of complex network in recent years, it become more and more popular for the link prediction by using complex network. Based on the research of the complex nature of the network topology, it revealed the relationship between the nodes in the network and made it possible for further prediction. However, most existing link prediction algorithms ignored the differences between the nodes, which lead to inaccurate result when calculate the similarity between nodes. For example, users in social network had different influence between each others.There were a lot of methods of link prediction, such as the methods based on path, methods based on random walk and so on. However, those ways did not consider the different influence between two directed nodes in the networks. In fact, different nodes represented different meanings,which had the asymmetric effects between each other. The main work of this paper includes:(1) The asymmetric maximum transfer similarity index is put forward. Starting with the degree and path of the nodes in the network and considering the asymmetric effects, this article put forward a new local topological similarity method. It regarded the degree as the kinds of things' interesting and computed the transitive similarity according to the shortest path between two nodes.(2) Testing AMTS on the data sets including of traditional networks(Facebook, Celegans) and binary networks(baidu films). This part is mainly to test the validation of AMTS. Compared with the existing link prediction methods, this paper took 10-cross validation to avoid the accident. As a result,it inspected the accuracy,recall and F1 to get that ATMS to get good ascension.(3) Applying AMTS on predicting the candidate gene. It compute the distance that from the core gene RHO to other candidate genes as the similarity. The bigger the similarity was, the more possible to be a candidate gene. It checked the exact gene to get the precision. In this chapter, we got good results by using AMTS.