Search arXiv⌕ Search
Home/Archive/arXiv

arXiv · 2504.04464

In which fields do ChatGPT scores align better than citations with research quality?

Mike Thelwall
View original source ↗Read source document ↗Export metadata ↓

Abstract

Although citation-based indicators are widely used for research evaluation, they are not useful for recently published research, reflect only one of the three common dimensions of research quality, and have little value in some social sciences, arts and humanities. Large Language Models (LLMs) have been shown to address some of these weaknesses, with ChatGPT-4o mini showing the most promising results, although on incomplete data. This article reports by far the largest scale evaluation of ChatGPT-4o mini yet and also evaluates its larger sibling ChatGPT-4o and ChatGPT-5 mini. Based on comparisons between LLM scores, averaged over 5 repetitions, and departmental average quality scores for 107,212 UK-based refereed journal articles, ChatGPT-4o is marginally better than ChatGPT-4o mini in most of the 34 field-based Units of Assessment (UoAs) tested, although combining both gives better results than either one. ChatGPT-4o scores have a positive correlation with research quality in 33 of the 34 UoAs, with the results being statistically significant in 31. The most substantial exception is Physics, for which citations are more useful. ChatGPT-4o scores had a higher correlation with research quality than long term citation rates in 21 out of 34 UoAs and a higher correlation than short term citation rates in 26 out of 34 UoAs. ChatGPT-5 mini has even stronger correlations overall. In summary, the results give the first large scale evidence that ChatGPT-4o and ChatGPT-5 mini are competitive with citations as new research quality indicator sources.

Explore related subjects

cs.DL

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Mike Thelwall. 2026-02-08. In which fields do ChatGPT scores align better than citations with research quality?. https://arxiv.org/abs/2504.04464

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

The illusion of neutrality in metric-based research evaluation

Venue prestige and citation counts are two widely used, albeit imperfect, signals of research quality. When the two signals conflict, evaluators must decide how much weight to assign each. Yet, it remains unknown how researchers across disciplines trade off these two signals when evaluating research outcome. To fill this gap, we surveyed 869 researchers using paired choices between hypothetical departmental hiring rules that assigned different weights to venue prestige and citation counts, asking which would produce better science. Among 795 respondents whose choices were largely internally consistent, choices placed nearly equal aggregate weight on the two signals. This apparent balance concealed substantial individual heterogeneity: nearly two in five respondents occupied the most venue-heavy or citation-heavy intervals. Moreover, respondents on average chose more citation-heavy rules than they believed their departments used in hiring. By revealing the subjective judgments that arise when research indicators conflict, our findings reinforce calls for greater caution when quantitative indicators are used to evaluate research.

cs.DL↗
arXiv2026

Reading the Data Back: Enriching Variable-Level Metadata for Model-Data Consistency Checks

Purpose: Research data repositories often lack variable-level metadata. Model-choice screening, checking whether a reported model suits the values of its outcome variable, also requires summaries of those values and links to the analyses that use them. We ask how repositories can represent and acquire them as metadata with evidence and review histories. Methods: We propose a metadata application profile compatible with the Data Documentation Initiative Cross-Domain Integration (DDI-CDI) model, linking versioned variables and empirical profiles to reported analyses, estimators, and analysis roles. Extraction provenance and review decisions are recorded separately. We evaluate streaming data profiling, deterministic extraction from Stata and R code, and language-model extraction of analysis records from papers. Results: A worked export of one replication package illustrates the profile: it passes shape and interchange checks and answers four queries, one of which returns the analyses that raise no alert. Screening 4,868 replication datasets from six political-science journals links outcomes to profiled columns in 1,440 deposits and flags 965, including 14 of the 21 deposits with a hand-verified linear model on a count, proportion, binary, or ordinal outcome. AI-assisted adjudication yields 55% precision on a near-balanced sample of 119 count and proportion candidates; screening rules were developed on the same corpus. Conclusion: The profile makes analysis-variable relationships queryable while preserving evidence and review history; screening yields review candidates with context. Code, metadata, and measurements are released.

cs.DL↗
arXiv2026

From Code Archival to Knowledge Graph: Bridging Software Heritage, COAR Notify and Wikidata

Software is a first-class scientific object, yet validated links between source code and the scholarly record remain largely absent from the Linked Open Data (LOD) cloud, isolating archived artefacts from semantic discovery. This paper presents an end-to-end reconciliation pipeline that harvests, validates, and models publication-to-repository pairs from sources where the link between a paper and its source code is explicit and editorially verified: the software-centric journals JOSS, SoftwareX, and IPOL, together with the reproducibility reports of the SIGMOD Availability and Reproducibility Initiative (ARI). This yields a curated corpus of 4,397 $\langle$DOI, repository-URL$\rangle$ pairs. We design two distinct application profiles grounded in Wikidata classes (one for scholarly articles, one for software instances) aligned with the schema.org and CodeMeta vocabularies. This architectural separation enables rule-based reconciliation at two granularities: lightweight, inline publication references or standalone, first-class Wikidata software nodes equipped with SWHIDs, Software Heritage's content-addressed identifiers. A read-only lookup against Wikidata shows that only 82 of the harvested repositories were already modelled there; human-reviewed batches have since created 4{,}182 new software items cross-linked to their articles. We further show that payloads of the emerging COAR Notify protocol, an external effort we do not develop, map natively onto our input format, so the same backend could later serve a live enrichment stream. Our core contribution is a pair of application profiles that turn Wikidata into a connector between the scholarly record and archived source code; we openly release all code, application profiles, and harvested datasets.

cs.DL↗