関係性マイニングと協調フィルタリングを用いた情報推薦手法

抄録

User-based collaborative filtering is one of the most popular recommendation methods, however, it has been pointed out that it is difficult to provide recommendation results with good recommendation accuracy and high recommendation variety simultaneously. To recommend good items for a target user, we consider that the tendency of the target user's Kansei evaluation to items should be explicitly reflected to recommendation methods. In this paper, we focus on pairs of items that there are big differences between target user's evaluation scores of the two items. We regard such pairs as the target user's preference patters, and in this paper, we propose a revised user-based collaborative filtering approach that reflects the tendency of target user's Kansei evaluation to recommendation. The proposed method is based on explicit extraction of target user's preference patterns as interrelated attributes in rough-set-based interrelationship mining and comparison of preference patterns between the target user and other users instead of direct comparison of evaluation scores of items. Experimental results indicate that, in comparison with collaborative filtering, our proposed method can recommend appropriate items for users with at least equal or better accuracy and high variety.