Extracting User's Preference and Selecting Users in Collaborative Filtering

摘要

This paper proposes a novel method in collaborative filtering in terms of the accuracy of prediction and calculation cost. The effectiveness of collaborative filtering can be assessed by its accuracy of prediction, depending on the way to seek similar users in preference. Previous works seek these users by the similarity between the users' rating. But our method selects users and measures the similarity by the users' preference. To extract these users' preference, we propose a weighting method for item, introduction of users' hidden preference model and a formula to predict users' rating. Besides these, we propose the formula to select users from all users quantitatively for extracting user's preference. Through the experiments, we can confirm the accuracy of prediction based on extracted user's preference and reduction of calculation cost by selecting users.