Fuzzy Side Information Clustering-Based Framework for Effective Recommendations

keywords: Recommender systems, collaborative filtering, Mahalanobis distance, k-means clustering, multi-criteria, demographic recommender
Collaborative filtering (CF) is the most successful and widely implemented algorithm in the area of recommender systems (RSs). It generates recommendations using a set of user-product ratings by matching similarity between the profiles of different users. Computing similarity among user profiles efficiently in case of sparse data is the most crucial component of the CF technique. Data sparsity and accuracy are the two major issues associated with the classical CF approach. In this paper, we try to solve these issues using a novel approach based on the side information (user-product background content) and the Mahalanobis distance measure. The side information has been incorporated into RSs to further improve their performance, especially in the case of data sparsity. However, incorporation of side information into traditional two-dimensional recommender systems would increase the dimensionality and complexity of the system. Therefore, to alleviate the problem of dimensionality, we cluster users based on their side information using k-means clustering algorithm and each user's similarity is computed using the Mahalanobis distance method. Additionally, we use fuzzy sets to represent the side information more efficiently. Results of the experimentation with two benchmark datasets show that our framework improves the recommendations quality and predictive accuracy of both traditional and clustering-based collaborative recommendations.
reference: Vol. 38, 2019, No. 3, pp. 597–620