Exploring Matrix Decomposition Methods for Recommender Systems

Abstract

Matrix factorization, a pivotal technique in recommender systems, is used to uncover latent patterns within user-item interaction matrices. This technique, which reduces the user-product rating matrix into separate lower dimensional matrices, is instrumental in generating personalized recommendations. Our study explores three matrix factorization techniques: U.V. decomposition, singular value decomposition, and the CUR algorithm. We perform a rigorous experimental evaluation on two real-world datasets, ensuring the thoroughness and reliability of our research. We perform a rigorous experimental evaluation to compare the three techniques on two datasets. This provides the thoroughness and reliability of the experimentation.