Link prediction in complex networks has found applications in a wide range of real-world domains involving relational data. The goal is to predict some hidden relations between individuals based on the observed relations. Existing models are unsatisfactory when more general multiple membership in latent groups can be found in the network data. Taking the nonparametric Bayesian approach, we propose a multiple membership latent group model for link prediction. Besides, we argue that existing performance evaluation methods for link prediction, which regard it as a binary classification problem, do not satisfy the nature of the problem. As another contribution of this work, we propose a new evaluation method by regarding link prediction as ranking. Based on this new evaluation method, we com...[ Read more ]

Link prediction in complex networks has found applications in a wide range of real-world domains involving relational data. The goal is to predict some hidden relations between individuals based on the observed relations. Existing models are unsatisfactory when more general multiple membership in latent groups can be found in the network data. Taking the nonparametric Bayesian approach, we propose a multiple membership latent group model for link prediction. Besides, we argue that existing performance evaluation methods for link prediction, which regard it as a binary classification problem, do not satisfy the nature of the problem. As another contribution of this work, we propose a new evaluation method by regarding link prediction as ranking. Based on this new evaluation method, we compare the proposed model with two related state-of-the-art models and find that the proposed model can learn more compact structure from the network data.