Prediction of Binary Labels for Edges in Signed Networks: A Random-Walk Based Approach

TLDR: A semi-supervised approach is proposed that uses the concept of random-walk for prediction of binary labels for edges in undirected and unweighted networks.

Abstract: Mining of signed networks where the links/edges between nodes have a positive or negative sign/label, is getting the attention of researchers and practitioners due to its wide realworld applicability in various domains. Label prediction for nodes in a network is a well-known and explored problem. However, the prediction of labels for edges in a network is relatively less explored, and very challenging and interesting problem. In this paper, we consider the problem of binary label prediction for edges in undirected and unweighted networks. The prediction of binary labels has a number of applications in realworld like friend/foe prediction, recommendation, trust/distrust prediction in social networks, and categorization. In this work, a semi-supervised approach is proposed that uses the concept of random-walk for prediction of binary labels for edges. In this paper, we demonstrate the viability and the effectiveness of the proposed approach using a real-world network.