ELP: Link prediction in social networks based on ego network perspective

Abstract

Social network analysis has recently been of much interest to researchers in diverse fields. This increased attention is due to its broad applicability in modeling complex real-world scenarios (problems). Link prediction is a crucial issue in social network analysis, one that finds the likelihood of having a link between two nodes in the network. Of the existing methods, many use topological network properties, while others use algebraic methods, statistical models, node embeddings and, community information. Although some path-based approaches can be said to deal with some nodes’ commutative effect at some point, they are not designed to infer the total community effect of all local nodes on a specific link. Hence we present ELP, a link prediction method based on the Ego perspective. First, this approach computes each existing edge's ego strength using ego networks, which can be construed as regions of influence of specific nodes. These ego strengths can be abstracted as the total effect of all local nodes on a particular edge. Then we utilize a topological feature set to estimate the prediction scores for target links. This feature set is selected after observing the performance of five different possible topological feature sets. Finally, we perform experiments on real-world networks to validate our algorithm's performance and compare it with state-of-the-art algorithms. The statistical tests justify the significant difference of our proposed method from the state-of-the-art algorithms. © 2022 Elsevier B.V.