Research On Link Prediction Algorithms Based On The Similarity

Link prediction of complex network is a crossed research direction of the data mining and the complex network, having important applications in different domains like bio-informatics, human social networks, traffic networks, etc. Link prediction aims at estimating the likelihood of the existence of links between nodes, based on the known attribute information of nodes or the network topology. At present, the research on link prediction algorithms based on the similarity is a hot direction, these algorithms have a very important premise that the more high similarity between two nodes, the bigger probability they link. Similarity-based link prediction algorithms can make full use of the related information such as node properties and network topology, and they have low computing time complexity, relatively easy to implement, suitable for large-scale network, and can achieve high prediction accuracy.However, the current classical algorithms for the unweighted and undirected networks can not fully use the information inherent in the topology. Most of them are based on common neighbors of the two predicted nodes, and only consider the number of common neighbor nodes and the individual characteristics of common neighbor nodes, without attention to the influence on the prediction results by the interaction between nodes. In the paper, a new algorithm named Individual Attraction Based on Local Path Index(IALP) is proposed to solve the problem mentioned above. The algorithm not only use the degree of common neighbor nodes and indirect neighbor nodes, but also consider the tightness of each node interaction in the neighbor node set. Experimental results show that the algorithm can improve prediction accuracy.In addition, a growing number of experimental studies of the real networks for recent years show that, using simple unweighted and undirected network to describe the structure of the real network is not good enough. Many important information, such as link intensity, link type, etc., have been ignored due to simply focusing on the topological structure. Weighted networks can be more comprehensive and profound to depict the real network systems. Therefore, this paper also studies the link prediction problem of weighted networks, and put forward a link prediction method for weighted social networks. Using the user information and the interrelated information crawling from the sina micro blog, we built a weighted micro blog network considering the factors such as the network topology and the user’s interest classification. And we extended three existing similarity index for unweighted and undirected networks (i.e., CN, AA and RA) to the link prediction algorithms for weighted networks (i.e.,WCN, WAA and WRA). Link prediction experiments for the weighted sina micro blog network show that the extended algorithms can achieve better prediction effect.