<p>Recommendation systems play a pivotal role in enhancing user experiences by personalizing content suggestions to individual preferences. However, conventional systems predominantly focus on accuracy, often at the expense of variety and novelty, which are crucial for maintaining user engagement and satisfaction. While widely used, traditional methods like collaborative filtering struggle with challenges such as high dimensionality and data sparsity, limiting their effectiveness. Advances in deep learning have addressed some of these issues, making it a valuable tool for enhancing recommendation techniques. In this paper, we introduce Resc-Auto-MOEA/D, a novel hybrid recommendation model that integrates an autoencoder with a multi-objective evolutionary algorithm (MOEA/D) to optimize recommendations across accuracy, novelty, serendipity, and retention. By decomposing the optimization problem into subproblems and evolving solutions in parallel, our approach achieves a fine-grained balance between personalization and exploration. Experimental evaluations on benchmark datasets demonstrate that Resc-Auto-MOEA/D significantly outperforms existing baselines in both relevance and engagement metrics. The results highlight the model’s ability to adapt to varying user behaviors and data distributions, confirming its effectiveness in generating diverse, surprising, and retention-promoting recommendations. This work contributes a scalable and principled framework for advancing multi-objective personalization in real-world recommendation environments.</p>