Sign Prediction in Signed Social Networks Using Inverse Squared Metric

This paper investigates the edge sign prediction problem in signed social networks, in which edges have either positive and negative sign. The main goal of this research is to effectively predict the sign of the edges using a newly proposed metric and with an emphasis on reducing the computational cost. In this study a new metric is introduced based on both neighbourhood and distance based link prediction measures. The sign of a connection between two users is subjected to the context of their relations. In the absence of these information, we can utilize topological and structural information of the network. The proposed metric has two components: (i) the importance of a node which is measured by node degree and (ii) the distance between two nodes which is penalizing the first component can be estimated by any shortest path algorithms. The new metric outperforms other sign prediction methods and also is computationally more affordable.