Prioritising explainable AI-driven recommendations with knowledge graphs and reinforcement learning

Explainable Recommender Systems (XRSs) assist users in decision-making by offering precise and interpretable recommendations. However, most existing XRSs tend to prioritise either accuracy or explainability in isolation, with limited attention to aligning recommendations with individual user preferences while providing well-substantiated justifications. To address this gap, we propose the Explainable Artificial Intelligence-driven Recommender System (XAIRec), which incorporates explainability mechanisms into knowledge graphs (KGs) and leverages knowledge graph embeddings (KGEs) alongside reinforcement learning (RL) to enhance both transparency and interpretability. By applying optimisation techniques that consider rewards, probabilities, information values, and affinity scores while minimising traversal hops within KG paths, XAIRec learns user preferences from historical data, determines optimal reasoning paths, and enhances recommendation quality by prioritising products with higher relevance. A core innovation within XAIRec is the Product Prioritisation Score (PPS) algorithm, which ranks recommendations based on their user-specific relevance. Extensive evaluations across Amazon domains, including CDs & Vinyl, Beauty, Clothing, and Cellphones, demonstrate XAIRec’s strong performance, with significant improvements in key metrics such as NDCG (27.5% for CDs & Vinyl, 8.4% for Clothing, 5.8% for Cellphones, and 6.1% for Beauty), along with comparable gains in Recall, Hit Ratio, and Precision. These results underscore XAIRec’s advancement in developing explainable, user-centric recommender systems. Code Repository: https://github.com/Neeraj-Tiwary/PhD-ExplainableRecommendation