Interpretable Hybrid Recommendation System For Personalized Skincare Products

Abstract

This thesis develops an interpretable hybrid recommendation system for personalized skincare product selection. The system combines ingredient-based analysis with user review sentiment extraction to generate balanced, data-driven recommendations. Using Natural Language Processing techniques, both lexicon-based (VADER) and AI-driven (GPT) methods are applied to assess user sentiment, while dermatological ingredient data are scored using a logarithmic weighting scheme. A tunable parameter (α) controls the influence between objective formulation data and subjective user feedback. Experiments show stable and interpretable ranking behavior, with optimal performance around α = 0.6. The results demonstrate the feasibility of transparent, AI-driven skincare recommendation systems that integrate scientific formulation knowledge with real-world user experience.