Kayvan Kousha

Mike Thelwall, Kayvan Kousha, Mahshid Abdoli et al.

National research evaluation initiatives and incentive schemes have previously chosen between simplistic quantitative indicators and time-consuming peer review, sometimes supported by bibliometrics. Here we assess whether artificial intelligence (AI) could provide a third alternative, estimating article quality using more multiple bibliometric and metadata inputs. We investigated this using provisional three-level REF2021 peer review scores for 84,966 articles submitted to the UK Research Excellence Framework 2021, matching a Scopus record 2014-18 and with a substantial abstract. We found that accuracy is highest in the medical and physical sciences Units of Assessment (UoAs) and economics, reaching 42% above the baseline (72% overall) in the best case. This is based on 1000 bibliometric inputs and half of the articles used for training in each UoA. Prediction accuracies above the baseline for the social science, mathematics, engineering, arts, and humanities UoAs were much lower or close to zero. The Random Forest Classifier (standard or ordinal) and Extreme Gradient Boosting Classifier algorithms performed best from the 32 tested. Accuracy was lower if UoAs were merged or replaced by Scopus broad categories. We increased accuracy with an active learning strategy and by selecting articles with higher prediction probabilities, as estimated by the algorithms, but this substantially reduced the number of scores predicted.

Mike Thelwall, Kayvan Kousha, Mahshid Abdoli et al.

This document describes strategies for using Artificial Intelligence (AI) to predict some journal article scores in future research assessment exercises. Five strategies have been assessed.

Kayvan Kousha, Mike Thelwall

This literature review identifies indicators that associate with higher impact or higher quality research from article text (e.g., titles, abstracts, lengths, cited references and readability) or metadata (e.g., the number of authors, international or domestic collaborations, journal impact factors and authors' h-index). This includes studies that used machine learning techniques to predict citation counts or quality scores for journal articles or conference papers. The literature review also includes evidence about the strength of association between bibliometric indicators and quality score rankings from previous UK Research Assessment Exercises (RAEs) and REFs in different subjects and years and similar evidence from other countries (e.g., Australia and Italy). In support of this, the document also surveys studies that used public datasets of citations, social media indictors or open review texts (e.g., Dimensions, OpenCitations, Altmetric.com and Publons) to help predict the scholarly impact of articles. The results of this part of the literature review were used to inform the experiments using machine learning to predict REF journal article quality scores, as reported in the AI experiments report for this project. The literature review also covers technology to automate editorial processes, to provide quality control for papers and reviewers' suggestions, to match reviewers with articles, and to automatically categorise journal articles into fields. Bias and transparency in technology assisted assessment are also discussed.

Kayvan Kousha, Mike Thelwall

This study investigates how Large Language Models (LLMs) are influencing the language of academic papers by tracking 12 LLM-associated terms across six major scholarly databases (Scopus, Web of Science, PubMed, PubMed Central (PMC), Dimensions, and OpenAlex) from 2015 to 2024. Using over 2.4 million PMC open-access publications (2021-July 2025), we also analysed full texts to assess changes in the frequency and co-occurrence of these terms before and after ChatGPT's initial public release. Across databases, delve (+1,500%), underscore (+1,000%), and intricate (+700%) had the largest increases between 2022 and 2024. Growth in LLM-term usage was much higher in STEM fields than in social sciences and arts and humanities. In PMC full texts, the proportion of papers using underscore six or more times increased by over 10,000% from 2022 to 2025, followed by intricate (+5,400%) and meticulous (+2,800%). Nearly half of all 2024 PMC papers using any LLM term also included underscore, compared with only 3%-14% of papers before ChatGPT in 2022. Papers using one LLM term are now much more likely to include other terms. For example, in 2024, underscore strongly correlated with pivotal (0.449) and delve (0.311), compared with very weak associations in 2022 (0.032 and 0.018, respectively). These findings provide the first large-scale evidence based on full-text publications and multiple databases that some LLM-related terms are now being used much more frequently and together. The rapid uptake of LLMs to support scholarly publishing is a welcome development reducing the language barrier to academic publishing for non-English speakers.

Mike Thelwall, Kayvan Kousha

The use of Large Language Models (LLMs) like ChatGPT and DeepSeek for translation and language polishing is a welcome development, reducing the longstanding publishing barrier to non-English speakers. Assessing the uptake of this facility is useful to give insights into changing nature of scientific writing. Although the prevalence of LLM-associated terms has been tracked across science in abstracts and for full text biomedical research, their science-wide prevalence in full texts is unknown. In response, this article investigates an expanded set of 80 potentially LLM-associated terms during 2021-2025 in a science-wide full text collection from the publisher MDPI (1.25 million articles), partly focusing on the 73 journals that published at least 500 articles in 2021. The results demonstrate the increasing prevalence of LLM-associated terms science-wide in full texts to 2024, with some terms declining from 2024 to 2025 and others continuing to increase. LLMs seem to avoid some terms (e.g., thus, moreover) and a few terms have stronger associations with abstracts than full texts (e.g., enhanced) or the opposite (e.g., leveraged). The term family "underscore" had the biggest increase: up to 29-fold. There are substantial differences between journals in the apparent use of LLMs for writing, from lower uptake in the life sciences to higher uptake in social sciences, electronic engineering and environmental science. Fields in which there is currently low uptake may need improved or specialist support, such as for reliably translating complex formulae, before the full benefits of automatic translation can be realised.

Kayvan Kousha, Mike Thelwall

Large Language Models (LLMs) have the potential to be used to support research evaluation and have a moderate capability to estimate the research quality of a journal article from its title and abstract. This paper assesses whether there are language-related factors unrelated to the quality of the research that influence ChatGPT's scores. Using a dataset of 99,277 journal articles submitted to the UK-wide Research Excellence Framework (REF) 2021 assessments, we calculated several readability indicators from abstracts and correlated them with ChatGPT scores and departmental REF scores. From the results, linguistic complexity and length were more strongly associated with ChatGPT research quality scores than with REF expert scores in many subject areas. Although cause-and-effect was not tested, these results suggest that ChatGPT may be more likely than human experts to reward linguistic complexity, with a potential bias towards longer and less readable abstracts in many fields. The apparent preference of LLMs for complex language is an undesirable feature for practical applications of LLMs for research quality evaluation, unless solutions can be found.

Kayvan Kousha, Mike Thelwall

Academics and departments are sometimes judged by how their research has benefitted society. For example, the UK Research Excellence Framework (REF) assesses Impact Case Studies (ICS), which are five-page evidence-based claims of societal impacts. This study investigates whether ChatGPT can evaluate societal impact claims and therefore potentially support expert human assessors. For this, various parts of 6,220 public ICS from REF2021 were fed to ChatGPT 4o-mini along with the REF2021 evaluation guidelines, comparing the results with published departmental average ICS scores. The results suggest that the optimal strategy for high correlations with expert scores is to input the title and summary of an ICS but not the remaining text, and to modify the original REF guidelines to encourage a stricter evaluation. The scores generated by this approach correlated positively with departmental average scores in all 34 Units of Assessment (UoAs), with values between 0.18 (Economics and Econometrics) and 0.56 (Psychology, Psychiatry and Neuroscience). At the departmental level, the corresponding correlations were higher, reaching 0.71 for Sport and Exercise Sciences, Leisure and Tourism. Thus, ChatGPT-based ICS evaluations are simple and viable to support or cross-check expert judgments, although their value varies substantially between fields.