A Comparative Study of Evolutionary Algorithms for Multi-Objective Recommender Systems

Recommender systems face the challenge of optimizing diverse objectives in personalized item recommendations. This study presents a comparative study of evolutionary algorithms for multi-objective optimization in recommender systems, evaluated using RMSE, precision, novelty, diversity, and coverage metrics on Movielens-10OK and Book-Crossing datasets. The support vector decision (SVD) algorithm emerges as superior with RMSE 0.89 and precision 0.42, balancing accuracy and diversity. We identify optimal combinations for each dataset: SVD with NSGA-II for Movielens-10OK and SVD with e-MOEA for Book-Crossing, achieving low RMSE, high precision, and diverse recommendations. The comparative results will offer valuable insights and suggestions to researchers seeking to choose appropriate algorithms that align with specific problem requirements.