Flow-Based Representations Outperform Counts for Modeling Research Attention Over Time
June 5, 2026
Replacing aggregated citation counts with contextually structured attention flows improves structural comparison in analogy-style benchmarks, particularly where attention evolves temporally or across context distributions. Learned flow representations also show better robustness under partial observation.
HOW THIS AFFECTS YOU
●
researcherRelevant if you work on bibliometrics, research impact modeling, or scientometrics — the flow representation approach offers a more structurally faithful alternative to count-based citation signals.