Determine whether recency bias explains arXiv-enriched retrieval degradation

Determine whether the recency of arXiv papers causes retrieval degradation by conducting a temporal ablation that separates recent papers from older papers in the academic advisor discovery system.

Background

The paper finds that concatenating arXiv abstracts with faculty profile text substantially reduces retrieval quality on a controlled 296-professor subpopulation. It proposes two possible explanations: jargon dilution, whereby technical vocabulary weakens the general research-interest signal, and recency bias, whereby papers reflect current trends rather than a professor’s established research identity. The authors describe the recency-bias explanation as plausible but unresolved because the available analysis does not compare papers across publication periods. A temporal ablation is therefore needed to determine whether paper recency contributes to the observed degradation.

References

H2 is plausible but not directly testable without temporal ablation (future work).

— Comparing Retrieval Methods for Academic Advisor Discovery: A Six-Method Study of 768 CS Faculty Profiles Across 9 US Universities  (2609.03901 - Subedi, 3 Sep 2026) in Section 6.2, subsection “arXiv Concatenation Experiment (Controlled)”; discussion in Section 10.2, subsection “Why arXiv Concatenation Hurts Performance”