Empirical evaluation of contextual combinations in recommendation system

Context-aware recommender systems try to figure out users' preferences under different context situations and have been proven to be useful in many domains. Different individual context makes different contribution to recommendation performance, but few of researchers have explored how contextual combinations influence recommendation performance. In this paper, we examine empirically the role of contextual combinations in context-aware recommendation algorithms. More specifically, the context-aware recommendation algorithm we used is popular context-aware splitting approaches (CASA). Since CASA needs to be followed by traditional 2D recommendation algorithm, we use item-based collaborative filtering method. We evaluate all the contextual combinations in a context rich movie dataset and experiments show that the best contextual combination chosen by 5-folds cross validation can achieve better predicting performance than other existing evaluated contextual combinations.