Determining Optimal Number of Neighbors in Item-based kNN Collaborative Filtering Algorithm for Learning Preferences of New Users

Abstract

Although the collaborative filtering (CF) is one of the efficient techniques to develop recommender systems, it suffers from a well-known problem called cold start which is a challenge to know the new user preferences. Ask To Rate technique is a simple way to solve this problem. In this technique, some items are shown to the new user, and ask her/him to rate them. Usually, Ask To Rate technique selects the items using kNN algorithm. However, determining k or number of the new user's neighbors in this algorithm is critical, because it affects the accuracy of recommender system. In this paper, a CF based recommender system is improved by Ask To Rate technique to solve cold start problem. Consequently, k or number of the new user's neighbors is determined by an experimental evaluation. The experimental results on MovieLens dataset show that the highest accuracy of recommendations can be seen when the number of neighbors is set by a low value e.g. 10-15 neighbors.