🛡️ EU Fact Force

A shared hub for coordinated response to health misinformation

Table of Contents

• Quick Start
• About
• Key Features
• Architecture
• Contributing
• Seeding the database

Quick Start

The fastest way to run the full stack locally is via Docker Compose. The only prerequisite is Docker Desktop.

1. Start all services

docker compose up --build -d

This builds the app image, starts PostgreSQL, MinIO (S3-compatible storage), the Django backend, and the Dash frontend. Database migrations run automatically on first start. Avoid having DATABASE_URL set in your .env for this step, to avoid a connection timeout when the app tries to reach the DB.

2. Seed the database

docker compose exec app uv run --no-sync python manage.py seed_db --csv data/seed/vaccine_autism_evidence_curated.csv -e DATABASE_URL=postgresql://eu_fact_force:eu_fact_force@localhost:5432/eu_fact_force

This ingests a curated set of vaccine/autism research articles (fetches PDFs and metadata from the internet). See Seeding the database for other input options.

3. Create an admin user

docker compose exec app uv run --no-sync python manage.py createsuperuser

4. Open in your browser

Service	URL
Dash frontend	http://localhost:8050
Django admin	http://localhost:8000/admin/
MinIO console	http://localhost:9001 (user: minioadmin / pass: minioadmin)

Stopping

docker compose down

---

About

EU Fact Force is a collaborative platform developed by EUPHA (European Public Health Association) with support from Data For Good volunteers. The platform empowers public health professionals to counter health misinformation by:

• Connecting scientific evidence with disinformation narratives
• Visualizing knowledge graphs of research articles, claims, and concepts
• Tracking disinformation trends through integration with PGP (The Public Good Projects) monitoring data
• Enabling rapid response with validated counter-narratives

Use Case

Marie, a health communicator at a national public health association, sees a viral post claiming "vaccines cause autism." She needs to respond quickly with solid evidence.

She searches "vaccines autism" on EU Fact Force and immediately sees:

• An interactive graph showing 15+ peer-reviewed articles that refute this claim
• The scientific consensus: "Refuted with high confidence"
• Current disinformation trends: 1,200 mentions this week, peak in France/Belgium
• Key evidence to cite in her response

Time to find relevant evidence: <30 s

Key Features

V0 (Minimum Viable Product - Target: April 2026)

• Semantic Search (FR/EN): Find relevant scientific articles even without exact keyword matches
• Interactive Knowledge Graph: Explore connections between articles, claims, and narratives
• Disinformation Trends: Visualize PGP monitoring data (volume, geography, examples)
• Researcher Upload: Members can upload scientific articles with auto-extracted metadata
• Multilingual: Interface and search in French and English
• Access Control: Authentication system for EUPHA members

Scope V0

• 3 priority narratives (e.g., vaccines-autism, moderate alcohol benefits, COVID misinformation)
• 100-150 scientific articles from EJPH, WHO, ECDC, and other trusted sources
• 20-30 claims per narrative (confirmed/refuted/nuanced)
• Integration with PGP data (weekly batch updates)

Contributing

Project structure

14_EUFactForce/
├── .github/
│   └── workflows/
│       ├── d4g-utils.yml
│       └── pre-commit.yaml
├── docs/
├── eu_fact_force/
│   ├── exploration/ # code to keep track of benchmarks
│   ├── ingestion/ # ingestion and indexing of documents
│   └── app/ # web app
├── tests/

Installation

• Installation de Python

Ce projet utilise uv pour la gestion des dépendances Python. Il est préréquis pour l'installation de ce projet.

Une fois installé, il suffit de lancer la commande suivante pour installer la version de Python adéquate, créer un environnement virtuel et installer les dépendances du projet.

uv sync
uv run pre-commit install

Pour exécuter le pipeline d'ingestion avec parsing Docling, installez aussi les dépendances de parsing :

uv sync --group parsing

Pour exécuter l'étape d'embedding dans le pipeline d'ingestion :

• le modèle utilisé est intfloat/multilingual-e5-base
• sentence-transformers est requis (installé via uv sync)
• prévoir plus de RAM/temps au premier chargement du modèle

A l'usage, si vous utilisez VSCode, l'environnement virtuel sera automatiquement activé lorsque vous ouvrirez le projet. Sinon, il suffit de l'activer manuellement avec la commande suivante :

source .venv/bin/activate

Tests et formattage

Installer les precommit

uv run pre-commit run --all-files

Utiliser pytest pour tester votre code

uv run pytest

Déploiement de l’application

L’application se compose d’un serveur Django, d’une base PostgreSQL (avec pgvector), de MinIO pour le stockage S3 (compatible AWS), et d’un frontend Dash.

Pour un démarrage rapide, voir la section Quick Start en haut de ce document.

Développement hors Docker (Django au host)

Si vous souhaitez lancer Django directement sur votre machine (sans le conteneur app) :

1. Prérequis

• Python 3.12+ et uv
• Docker et Docker Compose (pour Postgres et MinIO)

2. Démarrer uniquement les services d’infrastructure

docker compose up -d postgres minio minio-init

3. Variables d’environnement

cp .env.template .env

Pour pointer Django vers MinIO local, définissez dans .env :

DATABASE_URL=postgresql://eu_fact_force:eu_fact_force@localhost:5432/eu_fact_force
AWS_S3_ENDPOINT_URL=http://localhost:9000
AWS_ACCESS_KEY_ID=minioadmin
AWS_SECRET_ACCESS_KEY=minioadmin
AWS_STORAGE_BUCKET_NAME=eu-fact-force-files
AWS_S3_REGION_NAME=eu-west-1

4. Installer les dépendances et appliquer les migrations

uv sync
uv run python manage.py migrate

5. (Optionnel) Créer un superutilisateur

uv run python manage.py createsuperuser

6. Démarrer le serveur Django

uv run python manage.py runserver

7. Lancer le frontend Dash

uv run python eu_fact_force/dash-app/app.py

Puis ouvrir : http://127.0.0.1:8050/

Pour utiliser le JSON par défaut (default_search.json) côté backend, définir dans .env :

FLAG_RETRIEVE_DEFAULT_JSON=1

Seeding the database

The seed_db management command populates the database with scientific articles. It supports two input modes.

From a CSV of DOIs (fetches PDFs from the internet):

uv run python manage.py seed_db --csv data/seed/vaccine_autism_evidence_curated.csv

The CSV must have a doi column. An optional pdf_url column can provide a direct download link. A curated list of vaccine/autism articles is included at data/seed/vaccine_autism_evidence_curated.csv.

From a zip archive of pre-downloaded PDFs (skips the download step):

uv run python manage.py seed_db --zip archive.zip

The DOI is extracted automatically from the text of each PDF (first 3 pages). PDFs where no DOI can be found are skipped and reported. Metadata is still fetched from the API using the extracted DOI.

Dry run (preview without writing to the database):

uv run python manage.py seed_db --csv data/seed/vaccine_autism_evidence_curated.csv --dry-run
uv run python manage.py seed_db --zip archive.zip --dry-run

Duplicate DOIs (already in the database or repeated in the input) are skipped and reported without causing the command to fail.