A Decentralized Approach for Negative Link Prediction in Large Graphs

Abstract

Social network analytics is an important research area and attracts a lot of attention from researchers. Extraction of meaningful information from linked structures such as graph is known as link analysis. The emergence of signed social networks gives interesting insights into the social networks as the signed networks have the ability to represent various real-world relationships with positive (friend) and negative (foe) links. One interesting issue in signed networks is edge sign prediction among the members of the network. Negative link prediction is challenging due to the limited availability of the training data and also due to extracting a graph embedding that represents the negative links in a sparse graph. This study is focused on the prediction of the negative links across the signed network using a decentralized approach. For learning latent factors across the network, we use probabilistic matrix factorization. A detailed experimental study is performed to evaluate the accuracy of the proposed model. The results show that negative link prediction using matrix factorization is a promising approach and negative links can be predicted with high accuracy.