Interlayer co-similarity matrices for link prediction in multiplex networks

Given a target layer in a multiplex network, the objective of link prediction is to predict future relationships by utilizing information from all layers. Generally, the proposed techniques consider two significant steps: (i) evaluating the similarity within each layer, and (ii) assessing the similarity between layers. While measuring the intra-layer similarity can be achieved using basic and monoplex link predictors, determining the inter-layer similarity is a more complex task. In this paper, a new similarity measure, denoted as co-occurrence matrix based Inter-layer similarity (CMIS), is proposed for link prediction in multiplex networks. The main idea behind the CMIS is to capture the inter-layer similarity by utilizing co-occurrence matrices that consider the common neighbors of each node. Subsequently, the amount of information present in the neighborhoods of nodes is quantified and regarded as the measure of similarity. This approach differs from previous proposed similarity measures which usually rely solely on the number of common neighbors between pairs of nodes. Experimental results on several real networks in terms of the Precision and AUC scores affirm the effectiveness of the proposed measure.