Improving Overall Recommendation Quality with Convolutional Autoencoder
Zeynep Batmaz · Düzce Üniversitesi Bilim ve Teknoloji Dergisi · 2026
Recommender systems play a crucial role in addressing the information overload problem by providing personalized recommendations to users. Although the accuracy of the generated recommendations is essential for user satisfaction, beyond-accuracy quality of the recommendations significantly impacts user satisfaction. Therefore, researchers strive to enhance user satisfaction by improving not only the accuracy but also the beyond-accuracy quality of recommendations. Multi-criteria recommender systems extend traditional recommendation techniques by allowing the evaluation of an item based on various criteria, enabling more personalized recommendations. Although numerous studies in the literature focus on improving the accuracy of multi-criteria recommendations, there is a limited number of works that aim to enhance beyond-accuracy quality of recommendations as well as accuracy, to increase user-centric personalization and satisfaction. In this study, two new methods, based on a convolutional autoencoder, are proposed to improve accuracy and beyond-accuracy quality of multi-criteria recommender systems. Additionally, to measure the overall quality of the recommendations considering both accuracy and beyond-accuracy metrics, a new metric is presented. Experimental studies conducted on two real datasets show that CAE_MCCF generates strong diversity performance compared to existing state-of-the-art approaches in the field. The other proposed approach CAE_SMCCF shows competitive results on beyond accuracy metrics. In terms of overall quality of the recommendations, the proposed approach CAE_MCCF achieves the best overall performance across all TopN levels. CAE_MCCF provides an increase in overall quality scores. These findings support the effectiveness of convolutional autoencoders in enhancing recommendation quality.