NLP Sentiment & Topic Classifier
A multilingual NLP model (EL/EN) for classifying citizen feedback by sentiment, topic, urgency, and department — trained on anonymised municipal data.
Summary
A multilingual NLP model (EL/EN) for classifying citizen feedback by sentiment, topic, urgency, and department — trained on anonymised municipal data.
Description
This research project trains and evaluates a transformer-based NLP model capable of processing citizen feedback in both Greek and English. The model outputs sentiment polarity, topic category, urgency score, and suggested department routing.
Model Architecture
- Base: XLM-RoBERTa fine-tuned on 23 000 annotated municipal texts
- Multi-task head: sentiment (3-class), topic (18-class), urgency (1–5), department (12-class)
- Human-in-the-loop annotation pipeline with civil-servant reviewers
Ethics & Bias Auditing
An independent ethics review board evaluates model outputs at each training cycle. Bias reports are published internally, with mitigation strategies documented before any model version is promoted to production.
Bootcamp Curriculum Platform
Completed