Link prediction algorithm based on time-step weights and node similarities

Abstract

Social network node link prediction is an important topic in the field of data mining, and the link prediction algorithm based on node similarity is one of the popular algorithms. In this paper, we propose a link prediction algorithm based on time-step weights and node similarities (TSW). The proposed TSW algorithm improves the accuracy of link prediction from three aspects. Firstly, different neighbor nodes are treated differently considering common neighbor nodes to highlight the degree of contribution of different neighbor nodes to calculate the similarity between nodes, which helps to improve the accuracy of link prediction. Secondly, the temporal properties of social networks are combined, i.e., multiple consecutive network snapshots are used to help predict links. What’s more, the weights are assigned to network snapshots based on the distance in time to highlight the influence of network snapshots from different periods on the prediction. Finally, a moving average of the degrees of the nodes is applied to attenuate the influence of noisy data, making more accurate prediction results. The experimental results show that our proposed TSW algorithm obtains higher accuracy compared with most existing link prediction algorithms.