Skinwise: AI-Powered Skincare Recommendation System
Overview
AI-powered skincare recommendation system using product metadata, reviews, and medical knowledge to match users with best products tending to their dermatological needs. Direct Link to SkinWise chatbot
As part of this collaborative project, I performed sentiment analysis on 581K+ reviews from a Sephora product reviews dataset, to identify relevant feedback for product recommendation.
Sentiment Analysis Key Results
- 86% classification accuracy using TF-IDF + Logistic Regression
- 46% improvement in negative feedback detection (61% → 88% precision) from baseline model (VADER)
- Multi-platform validated on 4K additional Ulta reviews
- Automated quality scoring to surface top-5 representative reviews per product
Click to view full presentation
Technical Stack
Python • scikit-learn • pandas • NumPy • VADER • TF-IDF • Logistic Regression
What I Learned
This project taught me how to extract meaningful signals from noisy user feedback, and practice with Natural Language Processing through TF-IDF implementation. The 46% improvement in detecting negative reviews came from careful handling of class imbalance and optimizing for precision on the minority class.