Sarcouncil Journal of Engineering and Computer Sciences

Abstract: The digital transformation of beauty retail has created unprecedented challenges for personalized product discovery, necessitating sophisticated artificial intelligence solutions to address the complex nature of cosmetic and skincare recommendations. This article presents a comprehensive dual-tower neural network architecture specifically engineered for beauty product recommendation systems, tackling the inherent difficulties of subjective preferences, intricate ingredient interactions, and rapidly evolving market trends. The implementation showcases how the separate encoding of user preferences and product characteristics through hierarchical explainable networks enables superior performance in modeling temporal dependencies and behavioral patterns. Advanced training methods incorporate sequence-aware negative sampling techniques and demographic-stratified testing to mitigate algorithmic bias while ensuring equitable recommendation quality across diverse user demographics. Performance optimization strategies utilizing approximate nearest neighbor search algorithms and intelligent caching policies achieve rapid response times while processing extensive product catalogs. The system deployment generated substantial improvements in user engagement metrics, conversion rates, and revenue growth while enhancing product discovery experiences. The architecture successfully balances personalization accuracy with demographic fairness requirements, demonstrating the practical viability of sophisticated recommendation technologies in specialized retail domains requiring high levels of personalization and cultural sensitivity.