An Efficient Similarity Measure for User-Based Collaborative Filtering Recommender Systems Inspired by the Physical Resonance Principle

User-based collaborative filtering is an important technique used in collaborative filtering recommender systems to recommend items based on the opinions of like-minded nearby users, where similarity computation is the critical component. Traditional similarity measures, such as Pearson's correlation coefficient and cosine Similarity, mainly focus on the directions of co-related rating vectors and have inherent limitations for recommendations. In addition, CF-based recommendation systems always suffer from the cold-start problem, where users do not have enough co-related ratings for prediction. To address these problems, we propose a novel similarity measure inspired by a physical resonance phenomenon, named resonance similarity (RES). We fully consider different personalized situations in RES by mathematically modeling the consistency of users' rating behaviors, the distances between the users' opinions, and the Jaccard factor with both the co-related and non-related ratings. RES is a cumulative sum of the arithmetic product of these three parts and is optimized using learning parameters from data sets. Results evaluated on six real data sets show that RES is robust against the observed problems and has superior predictive accuracy compared with the state-of-the-art similarity measures on full users', grouped users', and cold-start users' evaluations.