A Hybrid Reinforcement Learning and Multi-Criteria Decision-Making Approach for Enhanced Personalized Product Recommendations in E-Commerce

This study presents a hybrid recommendation model that integrates Reinforcement Learning (RL) with the Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) to address the limitations of traditional recommender systems in handling dynamic user behavior and multi-criteria decision-making. The proposed system employs a Deep Q-Network (DQN)-based RL agent to learn and adapt to user preferences in real-time, while TOPSIS is used to evaluate candidate items based on multiple objective criteria such as price, quality, popularity, and user rating. Experimental evaluations conducted on a benchmark e-commerce dataset demonstrate that the RL+TOPSIS approach outperforms content-based, collaborative filtering, and standalone RL models across key performance metrics, achieving an F1-score of 0.833, RMSE of 0.5639, and notable improvements in Hit Rate, NDCG, and diversity. The integration of adaptive learning with structured decision analysis enables not only more accurate and explainable recommendations but also improved user satisfaction and platform efficiency. By bridging RL with MCDM in a unified architecture, this work contributes a scalable, interpretable, and effective solution for next-generation recommender systems.