We present a novel latent recommendation model, which is trained by a contextual modeling recommendation algorithm. Specifically, we suggest a new method for extracting latent contexts from a rich set of mobile sensors and utilizing them to enrich the recommendation algorithm.