Improving Deep Reinforcement Learning-Based Recommender Systems: Overcoming Issues with Usability, Profitability, and User Preferences

Recommender systems play a important role in personalizing user experiences across digital platforms. Usual recommendation methods often struggle with balancing both user satisfaction and business profitability, leading to inefficiencies in recommendation accuracy and engagement. This study proposes a Deep Reinforcement Learning (DRL)- based recommender system that integrates utility-based optimization, profitability-driven strategies, and user preferences. The methodology uses real-world Amazon product review data to find key insights into review helpfulness, brand engagement, and category-based preferences. Data analysis revealed strong correlations between review helpfulness and product ratings. Experiment was conducted by simulating environment on python using DQN. The recommendations were predicted based on data and user interactions. The experiment highlighted the importance of leveraging high-quality reviews in recommendations. Additionally, brand popularity was identified as a significant factor influencing user engagement, emphasizing the need for brand-aware recommendation strategies. The study introduces a framework that balances utility, business profitability, and consumer effort. By incorporating reinforcement learning techniques, the proposed model adapts to evolving user preferences while improving recommendation efficiency. Experimental results find that the DRL-based system enhances recommendation accuracy, improves long-term engagement and increasing overall business profitability. This research contributes to the improvement of AI-driven recommendation models by offering a scalable, adaptive, and viable solution for recommender systems. Future work will explore real-time adaptability and further refinements in reward modeling to enhance computational efficiency and user experience.