NSGA-II optimized deep autoencoders for enhanced multi-criteria recommendation system

Abstract

Recommendation systems are decision-support systems used by e-commerce enterprises to evaluate customers’ preferences and suggest items based on their interests. Moreover, they also tackle the problem of information overload. Multi-criteria recommendation systems differ from standard approaches by using multiple-criterion ratings instead of single-criterion ratings while rating products. Multi-criteria recommendation systems has attracted significant attention in the field of recommendation systems due to the lower accuracy of single-criteria recommendation systems. Furthermore, deep learning models have demonstrated encouraging results in several domains including image processing, computer vision, pattern recognition, and natural language processing. This paper introduces a novel methodology leveraging deep autoencoders optimized using the Non-dominated sorting genetic algorithm (NSGA-II) a meta-heuristic optimization technique to enhance the accuracy of multi-criteria recommendation systems. In the first stage of the proposed method, NSGA-II is employed to optimize the weights of the deep autoencoder for enhancing the fine-tuning of the hyperparameters. Secondly, missing ratings and overall ratings are predicted using autoencoders for more precise top-N recommendations. The model is validated using two real-world multi-criteria datasets Yahoo! Movies and TripAdvisor. Experimental results demonstrate that the model achieves significant improvements in prediction accuracy, with a Mean Absolute Error (MAE) of 0.6012 and 0.7215, and Root Mean Squared Error (RMSE) of 0.6137 and 0.7362 on Yahoo! Movies and TripAdvisor datasets, respectively. These findings indicate that the model outperforms both single-criteria recommendation algorithms and other state-of-the-art multi-criteria recommendation models in accuracy.