A Novel Recommendation Approach Using Pre-use Preferences

Abstract

Collaborative filtering, a widely-used approach to recommendation, often suffers from the problem of data sparsity that causes low accuracy in recommendations. To address this problem, this paper proposes an approach that exploits a new notion of pre-use preference for effective collaborative filtering. A pre-use preference on an item by a user, which can be extracted easily from the original rating matrix, indicates the preference that she has in mind on the item before using it. By exploiting the pre-use preferences extracted, our approach predicts a number of items to be disliked (before their usages) by every user and then gives the minimum ratings to those items in the rating matrix. As a result, our approach makes the rating matrix much denser, thereby enabling to solve the problem of data sparsity. Through extensive experiments, we show our approach identifies the items disliked by users accurately and also helps improve the accuracy of collaborative filtering significantly.