E-commerce web system with personalized recommendations based on purchase history for American Eagle in Maná, Ecuador

Introduction: in Ecuador, fashion retailers face volatile catalogs and low-density transaction records, which hinders the personalization of the customer experience. This study aimed to evaluate whether a recommendation engine based solely on purchase history improves performance and user experience for the American Eagle store in La Maná.Methods: A web-based e-commerce system was implemented using an MVC architecture and a two-stage framework: candidate generation using co-occurrences and collaborative item-to-item filtering (cosine/Jaccard), followed by ordering with BPR-MF based on implicit feedback. The offline evaluation included ranking metrics (Precision@k, Recall@k, MAP, NDCG@k), coverage, and diversity. Uncertainty was calculated with 95% confidence intervals using bootstrap, and significance was calculated using the paired Wilcoxon test. Popularity, Item-CF, and User-CF were used as references. In addition, an A/B test was applied measuring CTR, search time and usability (SUS/UEQ).Results: the proposed system showed consistent improvements in NDCG@10 and Recall@10 compared to the baseline models, maintained comparable Precision@10, and increased both catalog coverage and diversity. In user testing, it increased CTR, reduced search time, and achieved favorable usability.Conclusions: a purely historical recommendation approach is viable for SMEs with moderate resources, improves product discovery, and supports evidence-based merchandising decisions. The study provides a replicable protocol applicable to other contexts and seasons in Latin America.or unstructured, with a length of no more than 250 words; written in the past tense and in the third person singular.