LLM-Style Explainable E-Commerce Recommendation Cards: A UI/UX Design Framework for Trust-Calibrated Product Recommendation

This paper presents and empirically evaluates a UI/UX design framework for explainable e-commerce recommendation cards. The framework addresses a practical visual-communication problem: product lists can be useful but opaque, while explanation-heavy cards can create unwarranted confidence when the system has weak evidence. The revised study therefore uses the term LLM-style for the language condition and treats it as a grounded card-generation and confidence-display policy rather than as evidence of a black-box large-language-model recommender. Experiments were conducted on Amazon Reviews'23 All_Beauty raw reviews and item metadata, together with the Beauty_and_Personal_Care 5-core benchmark split as a larger same-domain warm-user check. The All_Beauty review file contains 701,528 review records from 631,986 users and 112,565 parent items, and the metadata file contains 112,590 parent items with near-complete title and image coverage. On the sparse All_Beauty all-user test, Recall@10 remained low for all methods, with the LLM-style reciprocal-rank reranker reaching 0.007945. On the All_Beauty warm-user slice, the same reranker reached Recall@10 of 0.008079. On the larger Beauty_and_Personal_Care 5-core test, it reached Recall@10 of 0.021463, improving over popularity and last-item co-history baselines but still indicating modest recommendation effectiveness. Card-level evaluation on All_Beauty shows that the LLM-style explanation plus confidence card achieved the highest confidence-discrimination AUC (0.700), while the review-evidence card offered a simpler evidence-forward alternative. The results support an interface-oriented conclusion: recommendation cards should separate ranking quality, grounded evidence, and confidence display, and UI/UX claims should be framed as proxy-based evidence until validated with a controlled user study.