SAM: a Similarity Measure for Link Prediction in Social Network

Abstract

Research in the field of social network analysis attracting majority of the researchers nowadays. Out of many social network analysis problems, link prediction gaining high attention due to a growing number of social network users. Link prediction is a task to predict which new interaction is going to be occurring in the future. Traditional link prediction techniques considered pair of node as one unit and make decisions based on the commonality between them. We argued that both nodes in a pair have their own similarity to each other. It may be that one person is 100% similar to another, but the other person is not the same as the first. Moreover, we have proposed a similarity measure SAM for link prediction in the social network. We have compared SAM similarity with four other state-of-the-art link prediction techniques (i.e., Jaccard, Salton Index, Salton Cosine and Resource Allocation). The experiments in this paper are performed on five different datasets (i.e., Astro, CondMat, GrQc, HepPh and HepTh). Our results show that SAM performs better than rest of the link prediction techniques on all datasets.