Context Aware Recommendation

Recommender system is to automatically generate relevant information to users,including products, news, movies and so on. Broadly speaking, recommender sys-tems are based on two strategies, the content fltering and the collaborative fltering.Traditional collaborative fltering approaches only focus on the rating matrix, payingno attention to the context of users and items. However, user’s interest always hasclose relation with his age, occupation and environment around him/her. At the sametime, the popularity of items changes with time and environment. Incorporating thesecontextual information into recommendation will signifcantly improve recommenda-tion accuracy and user satisfaction. With the growing usage of e-commerce as well asentertainment products such as movies, music, and TV shows, personalization has be-come more and more important. Besides, the recent popularization of mobile serviceattaches even more importance to personalized recommendation. On the other hand,there are some datasets available now which contain contextual information, Eg., timeand social network information in movie recommendation; user mood and music genrein music recommendation; location and weather information in tour recommendation.These contextual information refects the real environment of users and items, whichcan help to better capture user preference and item characteristics.Inthiswork,weexploretheimportanceofcontextualinformationinrecommendersystem. We incorporate context into the traditional recommendation algorithms to geta better personalized results. The experimental results show that context aware rec-ommendation can signifcantly improve the accuracy of the recommender system. Be-sides, we propose a novel approach alleviating the cold start problem in recommendersystem, by transferring contextual information patterns from a dense auxiliary ratingmatrix in other domains to a sparse or totally new rating matrix in a target domain. Infact, our model provides a single unifed framework to address totally new, cold and warm start scenarios in a smooth way.