A location-aware matrix factorisation approach for collaborative web service QoS prediction

Predicting the unknown QoS is often required due to the fact that most users would have invoked only a small fraction of web services. Previous prediction methods benefit from mining neighbourhood interest from explicit user QoS ratings. However, the implicitly existing but significant location information that would potentially tackle the data sparsity problem is overlooked. In this paper, we propose a unified matrix factorisation model that fully capitalises on the advantages of both location-aware neighbourhood and latent factor approach. We first develop a multiview-based neighbourhood selection method that clusters neighbours from the views of both geographical distance and rating similarity relationships. Then a personalised prediction model is built up by transforming the wisdom of neighbourhoods. Experimental results have demonstrated that our method can achieve higher prediction accuracy than other competitive approaches and as well as better alleviating the data sparsity issue.