Cold-start link prediction in multi-relational networks

During the last decade, interaction data have accumulated exponentially in many fields and provide a new opportunity for cold start link prediction. It seems necessarily to take full advantages of diversified information. However, correlation between different interactions has to be pre-tested. Therefore, this paper abstracts complex systems as multi-relational networks, and employs latent space network model to extract low-dimensional factors of sub-networks and adopts likelihood ratio test to examine correlation between factors. Then, regression between target sub-networks and correlated auxiliary sub-networks could be established for cold start link prediction. Experiments on 8 bioinformatic data sets validate the effectiveness and potential of our strategy for network correlation analysis and cold-start link prediction.