Semi-supervised context-aware matrix factorization: using contexts in a way of "latent" factors

Context-aware recommender systems (CARS) additionally take contexts into consideration and try to adapt users' preferences according to their contextual situations. In the traditional recommender systems (RS), latent factor models, such as matrix factorization and latent dirichlet allocation, have demonstrated their efficiencies. Apparently, contexts could be those possible latent factors -- they are "latent" ones if we have no pre-knowledge of them, but currently we have explicit contextual information at hand, why not treat and use them in a similar way as the latent factors? Most research in CARS seeks ways to incorporate contexts in the recommendation process, but none of them continue to use the contexts in a way of "latent" factors. In this work, the research ideas, relevant challenges and expected outcomes about using contexts in a way of "latent" factors are introduced and discussed as one novel research direction in the CARS domain, and a semi-supervised context-aware matrix factorization approach is proposed as a result.