aiqbench analysis
This study demonstrates that evaluation methodology significantly influences perceived performance gaps between expert-indexed and automatic Medical Subject Headings (MeSH) in biomedical text classification. By varying cross-validation folds and corpus sizes, the authors reveal that apparent advantages for transformer models like BiomedBERT over bag-of-words baselines may largely stem from experimental design rather than inherent model superiority. For researchers relying on benchmark results to guide screening pipeline decisions, these findings underscore the critical need to account for evaluation artifacts when interpreting comparative metrics in NLP applications.
Evaluation design conditions the expert-vs-auto MeSH gap: a controlled comparison of bag-of-words and BiomedBERT on the Cohen benchmark
arXiv:2607.21685v1 Announce Type: new Abstract: A systematic review begins with someone reading thousands of abstracts to identify the few that are relevant, and classifiers are used to prioritise that reading. Their inputs are often augmented with Medical Subject Headings (MeSH), assigned either by expert indexers weeks or months after publication or by automatic tools at once. To our knowledge the two have not been compared directly as classifier features, and no previous work has asked whether that comparison's outcome depends on how the classifier is evaluated. Using the Cohen et al. (2006) drug-class benchmark on three topics, we characterise a bag-of-words logistic regression classifier (seven reruns) and BiomedBERT (five seeds), then examine how the Statins result changes under alternative designs. Under the canonical 5-fold full-corpus design, the bag-of-words expert-vs-auto gap on Statins is +0.096 WSS@95%. Matching the corpus size to the smaller topics (n = 803) reduces it to +0.033 (95% bootstrap CI includes zero), and 10-fold cross-validation at full size to +0.021 (CI narrowly excludes zero). Under canonical evaluation BiomedBERT gives +0.020, within sampling noise of the bag-of-words 10-fold result. A power analysis indicates a Statins-sized effect would not have been detectable at the Opioids or ADHD variance, so those nulls are design-limited rather than informative. A representation asymmetry remains: 15.1% of Statins inputs exceed BiomedBERT's 512-token limit when expert MeSH terms are appended, so truncation may contribute to the smaller transformer gap, although this cannot be separated from training volume here. In screening pipelines using transformers or 10-fold bag-of-words, the gap on the topics tested is about 0.02 WSS@95%, with CIs spanning zero on at least one bound. More broadly, benchmark conclusions about feature sources can change substantially under reasonable changes to the evaluation design.
Abstract:A systematic review begins with someone reading thousands of abstracts to identify the few that are relevant, and classifiers are used to prioritise that reading. Their inputs are often augmented with Medical Subject Headings (MeSH), assigned either by expert indexers weeks or months after publication or by automatic tools at once. To our knowledge the two have not been compared directly as classifier features, and no previous work has asked whether that comparison's outcome depends on how the classifier is evaluated. Using the Cohen et al. (2006) drug-class benchmark on three topics, we characterise a bag-of-words logistic regression classifier (seven reruns) and BiomedBERT (five seeds), then examine how the Statins result changes under alternative designs. Under the canonical 5-fold full-corpus design, the bag-of-words expert-vs-auto gap on Statins is +0.096 WSS@95%. Matching the corpus size to the smaller topics (n = 803) reduces it to +0.033 (95% bootstrap CI includes zero), and 10-fold cross-validation at full size to +0.021 (CI narrowly excludes zero). Under canonical evaluation BiomedBERT gives +0.020, within sampling noise of the bag-of-words 10-fold result. A power analysis indicates a Statins-sized effect would not have been detectable at the Opioids or ADHD variance, so those nulls are design-limited rather than informative. A representation asymmetry remains: 15.1% of Statins inputs exceed BiomedBERT's 512-token limit when expert MeSH terms are appended, so truncation may contribute to the smaller transformer gap, although this cannot be separated from training volume here. In screening pipelines using transformers or 10-fold bag-of-words, the gap on the topics tested is about 0.02 WSS@95%, with CIs spanning zero on at least one bound. More broadly, benchmark conclusions about feature sources can change substantially under reasonable changes to the evaluation design.
Computation and Language (cs.CL)
arXiv:2607.21685 [cs.CL]
(or arXiv:2607.21685v1 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2607.21685 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 375 words
Related coverage
[AINews] not much happened today
Latent Space · Oct 3, 2026
[AINews] Here are 6 Clones of Jev in 2 days
Latent Space · Sep 19, 2026
Dropless MoE Training in JAX with Primus-Turbo
AMD · Jun 10, 2026
[AINews] Reality Checks on AI News (Yegge shuts down Gas Town, Databricks’ +60% Astra cost)
Latent Space · Sep 17, 2026