E-commerce Personalization Technologies: Algorithms, Techniques, and Consumer Perceptions

Background: Personalization is central to contemporary e-commerce, yet evidence on how algorithmic choices intersect with consumer perceptions (trust, privacy, satisfaction) remains fragmented. Objective: To synthesize research (2015–2025) on e-commerce personalization technologies—covering algorithms, implementation techniques, and consumer perceptions—and to identify performance patterns, ethical concerns, and gaps for future work. Data Sources: Google Scholar, Scopus, and Web of Science. Study Eligibility Criteria: English-language studies (2015–2025) examining e-commerce personalization algorithms/techniques and/or consumer perceptions. Non-e-commerce or non-personalization studies were excluded. Methods: A PRISMA-guided search and screening identified relevant records. Data were extracted on (i) personalization category (product, content, search, layout), (ii) algorithms (e.g., collaborative filtering, hybrid, deep learning, context-aware models), (iii) data types (behavioral, transactional, demographic), (iv) evaluation metrics (Precision@K, Recall@K, MRR, NDCG), and (v) consumer outcomes (trust, perceived intrusiveness, relevance, satisfaction). Narrative synthesis was used due to heterogeneity; where comparable, quantitative patterns were summarized. Results: Forty-nine studies met inclusion criteria. Collaborative filtering remains foundational, with strong movement toward hybrid and deep-learning approaches that fuse contextual signals (e.g., device, location) to model user behavior. Performance is predominantly reported via ranking metrics (Precision@K, Recall@K, MRR, NDCG). Included studies commonly link personalization to improved engagement and satisfaction; several report notable uplifts (e.g., ~66% of users indicating higher satisfaction and ~58% greater loyalty with AI-driven recommendations), while simultaneously highlighting persistent concerns over privacy, transparency, and fairness (privacy paradox). Practical challenges include cold-start, data sparsity, explainability, and the data/compute demands of advanced models. Evidence on long-term outcomes (loyalty, equity) and culturally diverse contexts is comparatively limited. Limitations:  Heterogeneous designs and metrics constrained meta-analysis; some domains and regions are under-represented, and few studies jointly measure algorithmic performance and human-centered outcomes in live settings. Conclusions: Personalization technologies are advancing toward hybrid, context-aware, and deep-learning models with measurable ranking gains. However, user trust and privacy remain pivotal determinants of acceptance. Future research should prioritize privacy-preserving and explainable designs, standardized evaluation that pairs system metrics with user outcomes, and rigorous A/B testing in production to balance benefits against ethical risks.