Automated Political Stance Identifier (APSI)

A Django-based web application for automated political text analysis. APSI scores political statements across three ideological dimensions using Natural Language Inference (NLI).

Dimensions

• Left – Right (economic positioning)
• Liberal – Illiberal (democratic commitment)(WIP)
• Populism – Pluralism (elite vs. people rhetoric)

Key Features

• Continuous 0–10 scores per dimension (not just labels)
• Hypothesis-based NLI scoring with per-hypothesis explainability
• Confidence scores with contradiction detection
• Automatic filtering of non-political text
• Docker-based, so easy to run yourself (docker compose build && docker compose up -d)

Tech Stack

• Backend: Django, PyTorch, HuggingFace Transformers
• Base model: DeBERTaV3
• NLI model: mlburnham/Political_DEBATE_large_v1.0

Hosted Instance

Available at: apsi.sc.hpi.de

Citation

If you use APSI in your research, please cite:

@software{apsi,
  title   = {Automated Political Stance Identifier},
  author  = {Juan S. Gómez-Cruces, Yorick Scheffler, and Ewan Thomas-Colquhoun},
  year    = {2025},
  url     = {https://apsi.sc.hpi.de}
}

License

See LICENSE for details.