A novel rating style mining method to improve collaborative filtering algorithm
Yang Wei, Sheng Hui Guo, Chun Jin Zhang · Journal of Physics Conference Series · 2019
Collaborative filtering (CF) algorithm is widely used in recommendation systems, which makes recommendation based on the neighbors' interests. Therefore, how to discover the neighbors with similar interests to target user is the core of the CF algorithm. Most existing algorithms discover neighbors by using rating similarity measure, which ignore the differences of users' rating styles. In this paper, we propose a user rating style mining method and use it to eliminate the rating style differences before calculating a similarity measure. Comparing with the raw similarity measure and another rating style mining method with the Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) over amazon movie dataset, we conclude that (i) use our method to eliminate the rating styles differences can improve the prediction accuracy and (ii) our method outperforms other rating style mining method.