Improving Collaborative Filtering Recommendation via Graph Learning
Yongyu Wang
TL;DR
The paper tackles inefficiency in kNN-based collaborative filtering caused by dense graphs and fixed neighborhood sizes by learning a sparse yet informative user graph with graph signal processing. It formulates CF as graph signal propagation and introduces an edge-pruning strategy guided by a Gaussian Markov random-field prior to obtain a sparse Laplacian that preserves signal smoothness. The proposed two-phase approach first learns the graph from a base kNN graph and then applies neighborhood-based CF on the learned graph, with experiments on MovieLens-100K showing that pruning up to 30% of edges maintains MAE while reducing computation. Overall, the method enhances CF scalability without compromising recommendation accuracy.
Abstract
Recommendation systems aim to provide personalized predictions by identifying items that are most appealing to individual users. Among various recommendation approaches, k-nearest-neighbor (kNN)-based collaborative filtering (CF) remains one of the most widely used in practice. However, the kNN scheme often results in running the algorithm on a highly dense graph, which degrades computational efficiency. In addition, enforcing a uniform neighborhood size is not well suited to capturing the true underlying structure of the data. In this paper, we leverage recent advances in graph signal processing (GSP) to learn a sparse yet high-quality graph, improving the efficiency of collaborative filtering without sacrificing recommendation accuracy. Experiments on benchmark datasets demonstrate that our method can successfully perform CF-based recommendation using an extremely sparse graph while maintaining competitive performance.