Back
Training
Supply Chain · Trustpilot
Customer satisfaction analysis from Trustpilot reviews (BERT, F1 85%).
The project
Final project of the DataScientest bootcamp. Turn Trustpilot reviews (2005) into actionable Supply Chain indicators (delivery, quality, support). Full NLP pipeline (normalisation, lemmatisation, TF-IDF), model comparison (LogReg, Random Forest, XGBoost, BERT) with F1 as primary metric. BERT reaches 85% accuracy.
Highlights
- F1 as primary metric (imbalanced classes 64/36)
- BERT vs XGBoost vs Random Forest vs LogReg
- Streamlit demo app for real-time prediction
Stack
Pythonscikit-learnXGBoostBERTNLTKStreamlit
Team project — modelling and approach comparison.
My role on this project.
Contact
Let's talk?
Easiest route: a 30-minute call to understand your context. Email works just as well.