Improved Memory-based Collaborative Filtering Using Entropy-based Similarity Measures

Accuracy of predicting the user preference score is the most important element of collaborative filtering. This paper proposes novel similarity measures using difference score entropy of common rating items between two users. The proposed similarity measures can apply various weights according to the score difference, to evaluate the similarity. We implemented a recommender system using the proposed similarity measures and, experimented on performance with memory-based collaborative filtering. Based on the experimental results, the proposed similarity measures significantly improve the prediction accuracy with respect to existing similarity measures, and we confirmed that the proposed measure is robust to sparse data sets.