Abstract
Non-negative latent factor analysis (NLFA) can high-efficiently extract useful information from high-dimensional and sparse (HiDS) matrices often encountered in recommender systems (RSs). However, an NLFA-based model requires careful tuning of regularization coefficients, which is highly expensive in both time and computation. To address this issue, this study proposes an adaptive NLFA-based model whose regularization coefficients become self-adaptive via particle swarm optimization. Experimental results on two HiDS matrices indicate that owing to such self-adaptation, it outperforms an NLFA model in terms of both convergence rate and prediction accuracy for missing data estimation.