An ensemble framework for link prediction in signed graph

Sociological study is an important research area which has been center of interest from past few years. A large number of applications have proved that predicting missing links from a signed social network is very essential. A variety of approaches to link prediction have been adopted that focused on positive link prediction while task of missing negative link prediction in signed network is neglected. However, intrinsic characteristics of negative relations pose number of challenges in link prediction task such as fewer negative relations and sparsity of negative links. In this work, we introduced an ensemble-based learning framework in order to scale up negative link prediction task. In order to predict negative links, a low-dimensional network representation is learned using alternating least square matrix factorisation approach. The low-dimensional representation is provided to ensemble framework that is able to predict negative links. We evaluate our approach using three real-world datasets to demonstrate the scalability, robustness and correctness of approach.