Similarity Measure based on Punishing Popular Items for Collaborative Filtering
Xiaotong Gao, Qing Ji, Zhenqiang Mi, Yang Yang, Yu Guo · 2018
Collaborative filtering (CF) helps users to find the items they may like in large quantity of items. It is widely used because of its interpretability and simplicity. However, there's still a challenging problem in calculating similarity between users or items. The existence of popular items will cause the similarity too big, resulting in misjudgment of the user's preferences. In order to alleviate this problem, This paper proposed that adding the punishment to popular items when calculating the similarity between users, so as to minimize the negative influence caused by popular items. The extensibility problem is caused by the growing number of users and items. This paper uses clustering collaborative filtering model to alleviate this problem, which uses the similarity with punishment as the basis of clustering. The combination of clustering collaborative filtering model and popular items punishment achieves the goal of finding more accurate nearest neighbors to predict user's preference with reducing the calculation complexity. The movie dataset in real world was used to test the proposed algorithm. The results of experiment show that the model can not only reduce the calculation complexity, but also greatly improve the accuracy of recommendation and surpass traditional item based collaborative filtering.