Research On Link Prediction Based On Transitivity And Symmetry Of Role In Complex Networks

Recently, with the development of the information age, research on a complex network has drawn more and more attention. The so-called complex network is a large mesh structure which is composed of a variety of individuals and their mutual relationships. In the field of data mining, the complex network research, also known as link mining, it is designed to find the missing links or predict the emergence of new links that do not present currently in a complex network. Link prediction as an important hot spot in the study of link mining, it has practical application value in many research fields. For example, link prediction which is applied to the friend recommendation in the field of social network can be used to predict the likelihood of becoming friends between people in the future, and link prediction which is applied to neural network in the field of biological information can be used to find the potential relationship between neurons in neural network, and so on. Therefore, link prediction provides an effective means in exploring possible relationship but having not been discovered links in the future.A lot of methods of link prediction on complex networks have been put forward, which is mainly based on similarity between two end nodes. These methods are very considerable in the aspect of accuracy and timeliness. We found that these links prediction methods are widespread in a fact by analysis, that is, when calculating the similarity between nodes, they treat the role of each node in the network the same, which the role value of each node is regarded as 1. In practice, the function of each node in the network is different, and thus, the role of each node can’t be the same. Therefore, we focus on improving the situation and promoting the accuracy of link prediction, this paper puts forward two methods of similarity, namely CorpSim and AconSim.The CorpSim method first introduces Google’s PageRank algorithm to calculate the real role value of each node in a network. This can effectively solve the situation which the existing methods treat the role of each node in the network the same. At the same time, The CorpSim method defines and calculates the similarity between nodes in the network by the role of transitivity. The CorpSim method consists of two parts, that’s the intermediate node and two endpoints role’s contribution to the similarity, and eventually the CorpSim method uses weighting factor to connect these two parts. As same as the CorpSim, the AconSim method also firstly uses PageRank algorithm to calculate the real role value of each node in the network. Then, The AconSim method defines and calculates the similarity between nodes in the network by the role of symmetry which is very different from the CorpSim method. Experimental results on real data sets show that the presented measures can greatly improve the accuracy of prediction.