
Paper Download Counts Predict Citation Impact Five Years Later: Evidence from 2.3 Million Papers
A Cornell University team has shown that download counts within the first 30 days after publication can predict citation counts five years later with an AUC of 0.80, based on an analysis of 2.3 million papers on arXiv. A new problem formulation called "lead-lag forecasting" and a large-scale benchmark dataset open the door to compressing the timeline of scientific evaluation.








