Community-guided link prediction in multiplex networks

Multiplex link prediction is the problem of finding missing links between nodes based on information from other layers. Although the link prediction problem in the online social networks is studied comprehensively, most approaches only employ internal features of the under prediction layer and do not consider additional link information from other networks. Also, many existing link prediction techniques are only based on the extracted information from links or nodes. However, the information flow in many real-world systems like social networks is considered as collaborative relations on correlated groups as an alternative for individual relations. In this research, we have proposed a Community-guided Link Prediction based on External Similarity (CLPES) method for multiplex networks in which, beside nodes and links information, community information is also employed. In our proposed method, we used an evolutionary algorithm (MOEA/D-TS) for specifying the community structure of the desired network. Next, the incorporation of internal features of each layer with a new external similarity metric (ExSim) obtains the final values for the likelihood of link formation in the network. Experiments on various real-world multiplex networks prove the capability of the proposed CLPES method for producing improved results and its superiority against other link prediction algorithms.

Automated summary

The research proposes a link prediction method CLPES for multiplex networks that utilizes community information along with internal features and a new external similarity metric to predict link formation probabilities, demonstrating its superiority through experiments.