Collaborative filtering algorithm based on rating distance
Yi Liu, Jun Feng, Jiamin Lu · 2017
Collaborative filtering, a successful and wildly used technique in personalized recommender systems, generates recommendations by similar users. Cosine similarity and Pearson correlation coefficient are widely used in collaborative filtering to calculate the similarity; however, the similarity is not accurate in some cases because of the defects of the algorithm. To solve these issues, this paper proposes a novel similarity calculation method which combined information entropy with compressive distance weight based on the probability distribution of rating distance. Experiment results show that the proposed method get better performance than conventional Pearson correlation coefficient method.