Evaluating Explanation Styles in E-Commerce Recommendations: A Comparative Study of User Trust and Understanding

Explainable recommendation systems have gained prominence as transparency becomes critical for user trust in e-commerce platforms. While various explanation styles have been proposed, including feature-based, social proof, and counterfactual explanations, empirical evidence comparing their effectiveness in routine purchase contexts remains limited. This study evaluates three explanation paradigms through a within-subjects experimental design with 70 participants. We developed a Neural Collaborative Filtering model using the Amazon Electronics dataset (192,403 users, 63,001 items) and conducted controlled evaluations measuring trust, understanding, confidence, and decision support. Statistical analysis using Friedman tests revealed no significant differences between explanation styles (χ²=0.54 – 3.70, all p>0.15), with all approaches receiving moderately positive ratings. These null results suggest that in generic, low-involvement e-commerce contexts, explanation presence may matter more than style, contrasting with findings from high-stakes domains. Our work contributes methodological insights on adequate sampling and identifies domain specificity as a critical moderator of explanation effectiveness.Index Terms - Explainable AI, Evaluation Framework, Recommender Systems, User Trust, Neural Collaborative Filtering, Counterfactual Explanations, E-Commerce, Human-Computer Interaction.