Deep Representation Learning - Information Matters

This study explores how artificial intelligence (AI), specifically deep representation learning, can predict the scholarly impact of new scientific papers without relying on citation data. Using the SciBERT model, the research introduces two key indicators—Topicality (τ) and Originality (σ)—to estimate the potential impact of newly published papers. The approach is validated using the COVID-19 Open Research Dataset, demonstrating that papers with high topicality or originality are more likely to gain scholarly attention. The findings suggest that AI can complement traditional citation-based metrics, particularly for early-stage research, offering insights into knowledge creation dynamics and interdisciplinary research potential.