Supervised link prediction using forecasting models on weighted online social network

Abstract

With the increase in size of online social network, the need to predict the future links among the nodes is enlarged. In this paper, an efficient prediction of links in the online social network is performed by considering link weights along with the temporal information. Although the existing technique is based on either weighted networks or time series based, the link prediction is based on the combination of weighted network and temporal data; and then applying supervised and unsupervised learning algorithm to predict the future link among the nodes (users) in the online social networking sites. Our task is to investigate that a weighted temporal network can be used with supervised learning to achieve a high-performance link prediction. Here research focus is to take weighted as well as unweighted network and to apply a similarity function for generating a set of connected nodes, then a time series is built for every pair of nonconnected nodes, and forecasting model is deployed on the time series. The final results obtained using supervised and unsupervised learning shown acceptable results when a weighted temporal network is used.