Scaling down candidate sets based on the temporal feature of items for improved hybrid recommendations

The intensive information overload incurred by the growing interest in the Internet as a medium to conduct business has stimulated the adoption of recommender systems. However, scalability still remains an obstacle to applying recommender mechanism for large-scale web-based systems where thousands of items and transactions are readily available. To deal with this issue, data mining techniques have been applied to reduce the dimensions of candidate sets. In this chapter in the context of movie recommendations, we study a different kind of technique to scale down candidate sets by considering the temporal feature of items. In particular, we argue that movies' production year can be regarded as a "temporal context" to which the value (thus the rating) of the movie can be attached; and thus might significantly affect target users' future preferences. We call it the temporal effects of the items on the performance of the recommender systems. We perform some experiments on the MovieLens data sets. The results show that the temporal feature of items can not only be exploited to scale down the candidate sets, but also increase the accuracy of the recommender systems.