Promising Methods for Discovery-oriented Collaborative Filtering
Takuya Shimizu, Yoshinori Hijikata, Shogo Nishida · 2007
A number of recommender systems employed in commercial websites use collaborative filtering. The main goal of traditional techniques of collaborative filtering is improving the accuracy of the recommendation. Though, they have a problem that they include many items the user has already known. When we consider only the accuracy, these recommendations appear good. On the other hand, when we consider users' satisfactions, they are not necessarily good becouse of the lack of discovery. In our work, we infer items which a user does not know by computing the similarity of users or items based on what items the user has already known. We try to recommend items which the user likes and does not know combining this method and the most popular method of collaborative filtering. We hope that users' satisfactions will improve.