An Enhanced Collaborative Filtering with Flexible Item Popularity Control for Recommender Systems

With the emerging and rapid development of Internet applications like social networks, E-commerce and so on, massive information has been created and stored. Recommender systems have been developed to deal with the information overload problem. Various recommendation algorithms have been proposed and Collaborative Filtering (CF) is one of the most remarkable. However, many similarity-based CFs suffer from a popularity bias problem: Popular items are frequently recommended not necessarily promoting the accuracy but making the recommendation lacking diversity. In this paper, we firstly explain how the item popularity impacts on the recommendation. Secondly, we propose an enhanced collaborative filtering approach (ECF) by adding item popularity control into a user-taste based method. Different from some other existing item popularity based CF methods, the popularity control in our approach is flexible and tunable. After that, extensive experiments are performed on two real benchmark datasets where the relationship between item popularity control and recommendation accuracy and diversity is investigated and turns out to be nonlinear. Experimental results demonstrate that temperate item popularity control can further improve the recommendation accuracy and diversity significantly, compared with the pure user-taste based method. But intemperate control makes the performance even worse Thus the flexibility is indeed essential and valuable.