A Similarity Measure Using Fuzzy User Rating Patterns in Collaborative Filtering Systems

Collaborative filtering has been a representative technique for recommender systems. Performance of this technique largely relies on its underlying similarity measure which computes similarity using the user rating records. This paper deals with one of the fundamental problems with the existing similarity measures, i.e., their dependence on the exact ratings of common items. A new model of user rating pattern is suggested by exploiting the fuzzy logic and the information entropy. Using this model, we propose similarity measures incorporated with the previous measures. Performance of the proposed measures is investigated through various experiments to demonstrate that they yield significant improvement over the existing measures in terms of both prediction and recommendation accuracy.