Latent context-aware recommender systems

The emergence of smart mobile devices has given rise to the development of context-aware systems that utilize sensors to collect available data about users. This data, in turn, is used in order to improve various services for the user. The development of such applications is inherently complex, since these applications adapt to changing context information, such as: physical context, computational context, and user tasks. Context information is gathered from a variety of sources that differ in the quality of information they produce and that are often failure-prone. Our study is part of a growing research effort that examines how data collected from mobile devices can be utilized to infer users' behavior and environment. We propose novel approaches that use a rich set of mobile sensors in order to infer unexplored users' contexts in personal models. We also suggest utilizing these high dimensional sensors, which represent users' context for a CARS (context-aware recommender system). For this purpose, we suggest several methods for reducing the dimensionality space by extracting latent contexts from data collected by mobile device sensors. Latent contexts are hidden context patterns, modeled as numeric vectors that are learned for each user automatically, by utilizing unsupervised deep learning techniques on the collected data. We also describe a novel latent context recommendation technique that uses latent contexts and improves the accuracy of state-of-the-art CARS. A preliminary analysis reveals encouraging insights regarding the feasibility of latent contexts and their utilization for context-aware recommendation systems.