Recommender systems are intelligent tools that help on-line users to tame information overload. Most of the current recommender systems assume that users' preferences do not change quickly and previously recorded observations about what users like or dislike can help in predicting their future choices. This assumption is valid only to some extent. In fact, users' general interests can be relatively stable, but they can also be influenced by many additional and varying factors, here referred as context. For example, weather could play an important role while choosing between going to a beach or visiting an indoor museum.
In the thesis I have studied how contextual information could be exploited to improve the prediction accuracy of Recommender Systems based on the Collaborative Filtering technique. Within this thesis we analyzed several new methods and techniques to perform context-aware Collaborative Filtering. We have also introduced and investigated a comprehensive development methodology for context-aware system. It consists of a two steps data gathering process and a specialized model-based rating prediction algorithm. We validated this methodology on a mobile recommender that provides real-time context-aware recommendations and can explain such recommendations on the base of the acquired contextual information. We have shown in an experimental study that our context-aware methods can increase the prediction accuracy compared to some context-free approaches and other state of the art context-aware methods. In addition, we carried out a user study that showed that the proposed techniques can also increase the user satisfaction.