Multilevel learning based modeling for link prediction and users’ consumption preference in Online Social Networks

The problem with predicting links in Online Social Networks (OSNs) is having to estimate the value of a link that can represent the relationship between social media users. The evolution of the OSN is influenced by the structure of the social network and the interaction between the preferential behaviors of users that have long converged by sociologists. However, conventional methods treat these behaviors in isolation. Therefore, the roles of users’ historical preferences and the dynamic structure of the social network are still not clear as to how these things affect the evolution of the OSN. Link prediction for new users who have not created a link or a small network is a fundamental problem in OSNs. To start creating social networks for such users, these behaviors can be used to recommend friends and user consumption preferences. In this paper, we propose novel direct and latent models to represent link prediction and a user's consumption preferences in an OSN platform. We also introduce a multilevel deep belief network learning-based model for link prediction and a user's consumption preferences to achieve high accuracy. To evaluate the performance of our model, we elaborated several performance measures and used datasets from Facebook, Amazon and Google+ to validate the accuracy. The result of our evaluation shows that our proposed model provides significantly improved performance for link prediction and user preferences over other methods.