Polarity Estimation in a Signed Social Graph Using Graph Features

Abstract

Signed graphs play an important role in social setup. They represent a majority of the networks that model sentiment, trust, opinions and a lot more between entities (people, organizations etc.) represented by the nodes. We propose an approach to predict the polarity of an edge of the graph solely based on the graph features. The key aspect of this experiment lies in the fact that we are trying to model the polarity estimation exclusively based on the structural aspects of the graph while earlier works have focused on using the non-structural, external data like textual attributes from, for example, reviews. Our experiment is to find the latent information that graph features possibly possess. We have also come up with a compact feature vector representation for the task of prediction of the polarity of the edge. We see that it outperforms the selected baseline which is seen to use a bigger feature vector i.e. the feature vector that we are using is almost one-fourth the size of our baseline’s feature vector.