A Systematic Review of AI Personalization for Data Privacy and Recommendation Quality
TL;DR
The findings indicate that hybrid deep learning architectures improve recommendation performance under data sparsity and cold-start conditions, while context-aware approaches enhance personalization through dynamic adaptation to user behavior and contextual information.
Abstract
Artificial Intelligence (AI)-based recommendation systems have become essential tools for delivering personalized services across digital platforms. However, persistent challenges related to data sparsity, cold-start conditions, dynamic user preferences, and data privacy continue to limit recommendation effectiveness. Existing studies have investigated these issues from different perspectives, yet the evidence remains fragmented across research streams focusing separately on recommendation accuracy, personalization, and privacy preservation. This study aims to provide a comprehensive synthesis of current developments in AI-based recommendation systems by examining the integration of hybrid deep learning approaches, context-aware recommendation mechanisms, and privacy-preserving techniques. A Systematic Literature Review (SLR) was conducted following the PRISMA 2020 guidelines. An initial search of the Scopus database identified 176 records published between 2000 and 2025. After a multi-stage screening, eligibility assessment, and quality evaluation process, 68 studies were selected for detailed analysis. Descriptive, bibliometric, and thematic synthesis methods were employed to identify technological trends, implementation approaches, and emerging research directions. The findings indicate that hybrid deep learning architectures improve recommendation performance under data sparsity and cold-start conditions, while context-aware approaches enhance personalization through dynamic adaptation to user behavior and contextual information. Privacy-preserving techniques, including differential privacy, cryptographic methods, and secure recommendation architectures, strengthen data protection without substantially reducing recommendation effectiveness. This review contributes an integrated analytical framework that links recommendation accuracy, personalization, and privacy preservation, providing guidance for the development of adaptive, trustworthy, and privacy-aware recommendation systems.