Finding Domain-based Expert for Improving Collaborative Filtering Algorithm

Traditional neighborhood-based collaborative filtering algorithms are widely used in recommender system field for its accuracy, interpretability and operability. In this paper, we introduce expert user model into collaborative filtering and determine authoritative expert users via expert finding methods in large corpus. We propose a method to produce predications for target user. Instead of the similarity between normal users and target user, we determine target user’s neighborhood based on the similarity be-tween expert users and target user. Experiments on Amazon datasets show that our method has better performance than neighbor-hood-based collaborative filtering on recommendation accuracy, novelty and calculation efficiency.