On the temporal analysis for improved hybrid recommendations

Recommender systems address the issue of information overload by providing personalized recommendations towards a target user based upon a history of his/her likes and dislikes. Collaborative filtering and content-based methods are two most commonly used approaches in most recommender systems. Although each of them has both advantages and disadvantages in providing high quality recommendations, a hybrid recommendation mechanism incorporating components from both of the methods would yield satisfactory results in many situations. Unfortunately, most hybrid approaches have focused on the contents of items but the temporal feature of them, which is the theme of our study here. In particular, we argue, here in the context of movie recommendation, that movie's production year, which reflects the situational environment where the movies were filmed, might affect the values of the movies being recommended, and in turn significantly affect target user's future preferences. We called it the temporal effects of the items on the performance of the recommender systems. We perform some experiments on the famous MovieLens data sets, and significant results were obtained from our experiments. We believe that the temporal features of items can be exploited to not only scale down the huge amount of data set, especially for Web-based recommender system, but also allow us to quickly select high quality candidate sets to make more accurate recommendations.