Natural Noise Correction in Collaborative Recommender Systems by Considering Change of Userʹs Preferences

Recommender systems help users to find their desired items from an enormous volume of available information. Recommender systems often act on the basis of user rating. However, users may act unstable in rating items and give different ratings to similar items. Inconsistency may occur for two reasons: user inherent inaccuracy and change of user preferences. The first reason causes the inconsistency of rating in short-term which is called natural noise. Other reason causes the inconsistency of rating in long-term and is one of the challenges that affects the efficiency of recommender systems. So, changing the user preferences should not be confused with the natural noise. This study discriminates between the discrepancies of these two types of rating. In the proposed method, first, changes of user preferences are detected and categorized. Then, natural noise can be detected and corrected by use of these categories. Finally, recommendations are suggested to the users according to the changes of their preferences. The experimental results have shown that the proposed method has improved MAE, precision, recall, and F1 measures compared to former methods.