NLP Sentiment Modeling on OpenGov.gr (Greek e-Government)

This repository contains the code and supporting files for the MSc project: “Natural Language Processing tools in e-Government: Comparative Sentiment Modeling on OpenGov.gr”.

Repository purpose

• Scrape and structure public consultation comments from OpenGov.gr.
• Train and benchmark supervised sentiment models (binary P/N).
• Compare model performance and export scored outputs.

Project snapshot

• Comments in labeled corpus: 14,953
• Unique consultations represented: 57
• Class distribution: {'N': 12281, 'P': 2672}
• Validation split: 90/10 stratified

Best validation result (from included metrics)

Model	Threshold	Accuracy	Macro-F1	F1 (N)	F1 (P)
Logistic Regression	0.60	0.888	0.799	0.933	0.664
Linear SVM (calibrated)	0.44	0.892	0.788	0.936	0.640
Random Forest	0.27	0.859	0.772	0.913	0.632

Folder structure

.
├─ data/
│  ├─ raw/
│  │  ├─ consultations_index.xlsx
│  │  └─ comments_sample_anonymized.xlsx
│  └─ processed/
├─ docs/
│  └─ Report.docx
├─ notebooks/
├─ outputs/
│  ├─ metrics/
│  │  ├─ validation_metrics.xlsx
│  │  ├─ model_validation_metrics.xlsx
│  │  └─ model_comparison.csv
│  └─ predictions/
│     └─ comments_with_predictions_sample.xlsx
├─ scripts/
│  └─ create_public_sample.py
└─ src/
   ├─ data_collection/
   │  └─ extract_opengov_comments.py
   └─ modeling/
      └─ train_and_evaluate.py

Quick start

1) Create environment

python -m venv .venv
source .venv/bin/activate  # Windows: .venv\Scripts\activate
pip install -r requirements.txt

2) Train + evaluate models

Place the full labeled file as Comments(1st code).xlsx in the repository root (or edit paths in src/modeling/train_and_evaluate.py), then run:

python src/modeling/train_and_evaluate.py

This script:

• cleans text,
• builds dual TF-IDF features (word 1–2 grams + char 3–5 grams),
• trains Logistic Regression, calibrated Linear SVM, and Random Forest,
• tunes threshold on validation macro-F1,
• writes metrics tables and confusion matrices.

3) Scrape comments (collector)

python src/data_collection/extract_opengov_comments.py