AI4citations
Fighting inaccurate citations with AI
The challenge: Researchers spend countless hours manually verifying citations, while misinformation spreads through unchecked claims.
The solution: Shuffled training across multiple datasets lays the groundwork for breakthrough improvements in citation verification.
Impact & innovation:
- pyvers package (based on PyTorch Lightning) automates preprocessing and training on claim verification datasets
- 7% improvement in F1 score over state-of-the-art models through shuffled training methodology
- Real-time web app with feedback collection for model improvement
- Production-ready deployment with CI/CD pipeline ensuring reliability with every update
📝 Related blog posts
Deploying AI4citations: From research to production
Modern understanding of overfitting and generalization in machine learning
Experimenting with transformer models for citation verification