DeepNewsDef

Resources and tools for the DeepNewsDef project — a dataset and toolchain for deep-fake Romanian news detection.

---

Modules

Module	Description
synthetic_news_text_generator/	Generates synthetic Romanian news articles using local Ollama models
metadata_validator/	Validates dataset JSON files against the TextMetadata / image schemas

---

Setup

python -m venv .venv
.venv\Scripts\activate        # Windows
pip install -r requirements.txt

---

Synthetic News Text Generator

→ Full documentation

# Single model, 10 articles
python -m synthetic_news_text_generator.main --count 10

# All models, 20 articles each, free disk after each batch
python synthetic_news_text_generator/run_multi_model.py --per-model 20 --delete-after

---

Metadata Validator

# Single file
python -m metadata_validator.validate --type text --file path/to/metadata.json

# Entire folder
python -m metadata_validator.validate --type image --folder path/to/dataset/

A report is written to validation_report.json (override with --report).