FLP-ID: Fuzzy-based link prediction in multiplex social networks using information diffusion perspective

Abstract

The growing popularity of online social networks is evident nowadays and allows researchers to find solutions for various practical applications. Link prediction is the technique of understanding network structure and identifying the missing links in the social network. The two significant challenges of the link prediction problem are accuracy and efficiency on growing and multiplex networks. Well-known methods for link prediction are the similarity-based methods, which use local, global, and topological features of the network to predict missing links. These approaches ignore critical factors such as different channels of interaction, information diffusion, group norms to form new connections. Therefore, a fuzzy-based link prediction algorithm (FLP-ID) in multiple social networks is proposed using information diffusion. First, FLP-ID generates a multiplex network by combining different types of relationships among users and identifying the community structure. Thereafter, the algorithm computes node and relative relevance for distinct fuzzy criteria under group norms. Finally, the likelihood score of each non-existing link is computed to predict missing links. The experimental results show that the proposed fuzzy algorithm accuracy is better than crisp algorithms over the multiplex network. The prediction rate of FLP-ID with F1-score, AUC, and balanced accuracy is excellent, which are improved compared to related methods up to 30%, 35%, and 30%, respectively, on high density and clustering coefficient datasets under multiplex settings.