Self-citations can inflate research impact metrics by 10-20%, potentially skewing billions in research funding. This session explores the PyTorch-based architecture for analyzing self-citation patterns in massive bibliometric databases with millions of publications across diverse fields.

We’ll tackle computational challenges in developing the Self-Citation Adjusted Index (SCAI), a metric recalibrating citation counts based on discipline-specific patterns using PyTorch’s distributed training. We’ll explore PyTorch-based deep learning, including transformer architectures and graph neural networks via PyTorch Geometric, to distinguish legitimate from problematic self-citations. Hands-on examples will demonstrate training citation classification models with PyTorch’s autograd, emphasizing transparency through interpretable AI.

Attendees will learn to architect large-scale scientific data systems using PyTorch’s ecosystem, integrating Ray for distributed hyperparameter tuning and MLflow for experiment tracking, fostering equitable research evaluation impacting over $100 billion in annual funding.