Social Network Link Prediction Algorithm Based On Node Similarity

With the rapid development of social, social network analysis has become an important research focus. As one of social network analysis of research tasks, link prediction based on the existing network structure and node information, forecasting two no link exists between nodes. It is the object of study related to a variety of social networks, such as the development of social networks in recent years, including research cooperation network, business cooperation network. Link prediction can be applied to friend recommendation, community discovery, business decisions and other areas, and play an important role. Therefore, it is highly theoretical study and practical value.At present, link prediction algorithm based on node similarity is calculated by the similarity of nodes, judging whether there is a link between nodes. However, these algorithms only consider the number of neighbor in the network, and ingore relationship between neighbor nodes, leading to inaccurate prediction of nodes in the network. In this paper, the node similarity will be studied based on the existing node similarity algorithm. First, research has been based on node similarity algorithm defects. We analyze effect of relationship of the predicted node linking to nodes based on the structural characteristics of the social network and relationship between different nodes. Basing on the node information and characteristics of the network, a new link prediction algorithm named neighbor relationships weights algorithm will be proposed. First, the algorithm considers neighbors of predicted node, and neighbor nodes are classified into common neighbor nodes and other neighbors; assign different weights to the edges between the values of common neighbor nodes and other nodes,while the definition of a common neighbor node weight, given the neighbor node similarity weight calculation; then the similarity of the right common to all neighbor nodes in a weight added value of the similarity obtained two nodes of the node; finally, judge the existence possibility of links between nodes by calculating the similarity score of predicted node and order the score, therefore given the link prediction results.Finally, common social network dataset are collected in this paper, and algorithm is compared with three existing link prediction algorithms in the experiment, and we used two indices to evaluate the prediction accuracy of the algorithm. It is turned out that new algorithm is higher than the existing three link prediction algorithm in the prediction accuracy,and it maintaining lower time complexity.