Exploring Set-Inspired Similarity Measures for Collaborative Filtering Recommendation

The similarity measure is an important component used in collaborative filtering recommender systems (CFRSs) to determine the set of users having the same behavior with regard to the selected items. The measure is typically defined on sets of real-valued or discrete-valued vectors. For discrete-valued vectors, similarity measures are inspired by the comparison of sets and the cardinality of sets. In this paper, we aim to explore set-inspired similarity measures for CFRSs, including Fuzzy sets index, Jaccard index, Sorensen coefficient, and Symmetric difference, with four collaborative filtering methods: (i) user-based, (ii) item-based, (iii) user clustering-based, and (iv) item clustering-based methods. We conduct extensive experiments to evaluate the effect of different measures on the benchmark datasets. An important result is that all four of these measures outperform the Pearson coefficient and Cosine measures in both recommendation effectiveness and computation time. Empirical evidence also shows that the Symmetric difference measure provides better results than all remaining measures.