Towards a semi-local random walk technique through multilayer social networks to improve link prediction

Abstract

Abstract The rapid expansion of social networks has generated a growing need for scalable algorithms capable of effectively predicting links. Link prediction is a crucial area of study within complex networks research. Link prediction aims to predict future connections between nodes from the current snapshot of the network and plays a vital role in estimating the growth of social networks. This article introduces an improved approach to link prediction in social networks by exploiting an extended version of local random walk as semi-local random walk (SLRW) for multilayer social networks. Here, taking into account the connectivity and structural similarity of the involved nodes, we propose the SLRW method to acquire nodes sequence with the highest similarity. Also, SLRW metric includes a distributed technique to identify the nearest neighbours by considering the extended neighbourhood concept. To ensure optimal performance, we conduct extensive studies on various hyperparameters of the proposed metric. The experimental results conducted on different datasets demonstrate that the proposed metric achieves improvements in the field of link prediction compared to the state-of-the-art baselines.