Technology assisted research assessment: algorithmic bias and transparency issues
| Date | 02 October 2023 |
| Pages | 175-190 |
| DOI | https://doi.org/10.1108/AJIM-04-2023-0119 |
| Published date | 02 October 2023 |
| Author | Mike Thelwall,Kayvan Kousha |
Technology assisted research
assessment: algorithmic bias
and transparency issues
Mike Thelwall and Kayvan Kousha
Statistical Cybermetrics Research Group, University of Wolverhampton,
Wolverhampton, UK
Abstract
Purpose –Technology is sometimes used to support assessments of academic research in the form of
automatically generated bibliometrics for reviewers to consult during their evaluations or by replacing
some or all human judgements. With artificial intelligence (AI), there is increasing scope to use technology
to assist research assessment processes in new ways. Since transparency and fairness are widely
considered important for research assessment and AI introduces new issues, this review investigates
their implications.
Design/methodology/approach –This article reviews and briefly summarises transparency and fairness
concerns in general terms and through the issues that they raise for various types of Technology Assisted
Research Assessment (TARA).
Findings –Whilst TARA can have varying levels of problems with both transparency and bias, in most
contexts it is unclear whether it worsens the transparency and bias problems that are inherent in peer review.
Originality/value –This is the first analysis that focuses on algorithmic bias and transparency issues for
technology assisted research assessment.
Keywords Technology assisted research assessment, Bibliometrics, Research evaluation, Machine learning,
Algorithmic bias, Transparency
Paper type Research paper
1. Introduction
Technology Assisted Research Assessment (TARA) refers to the use of routine computer
automation or artificial intelligence to generate information to support or replace human
judgement for research evaluations. TARA may have value if it improves outcomes or saves
the time of administrators or skilled researchers without introducing perverse incentives for
researchers (Wilsdon et al., 2015). TARA has previously taken the form of mostly hidden
computerisation of bibliometric databases and bibliometric indicator calculations (e.g. article
citation counts, journal impact factors) but with the rise of artificial intelligence (AI), it has
become possible to provide a wider range of functionalities, such as identifying or selecting
evaluators (Fiez et al., 2020), detecting plagiarism (Zhang, 2010), checking methods details
(Wren, 2018) and estimating the overall quality of articles from bibliometrics and/or text and/
or other metadata (Thelwall et al., 2023).
Both transparency and bias are important concerns for research assessment (Hicks
et al., 2015). Whilst bias against researchers or institutions is undesirable from a natural
human justice perspective, bias against aspects of research, such as fields, methods or
output types, is undesirable from a systemic perspective if it provides perverse incentives
Technology
assisted
research
assessment
175
Funding: This study was funded by Research England, Scottish Funding Council, Higher Education
Funding Council for Wales, and Department for the Economy, Northern Ireland as part of the Future
Research Assessment Programme (https://www.jisc.ac.uk/future-research-assessment-programme).
The content is solely the responsibility of the authors and does not necessarily represent the official
views of the funders.
The current issue and full text archive of this journal is available on Emerald Insight at:
https://www.emerald.com/insight/2050-3806.htm
Received 13 April 2023
Revised 25 July 2023
Accepted 9 September 2023
Aslib Journal of Information
Management
Vol. 77 No. 1, 2025
pp. 175-190
© Emerald Publishing Limited
2050-3806
DOI 10.1108/AJIM-04-2023-0119
to researchers to alter their behaviours in ways that do not benefit science, such as by
changing to a higher citation field. Transparency is also important from a natural justice
perspective to allow those evaluated to check the key assumptions and calculations in
TARA data and, if necessary, challenge errors or inappropriate calculations.
It is already recognised that bibliometric data often requires extensive computer
processing, potentially reducing transparency and generating biases (Hicks et al., 2015). This
review extends previous discussions of bibliometric transparency and bias to the wider
context of TARA, incorporating insights from analyses of general AI transparency and bias
issues.
2. Transparency in technology assisted research assessment
This section covers transparency in the sense of the ability of the people assessed to fully
understand the procedures used to assess them. Whilst reproducibility is also important for
research assessment, a fully reproducible complex assessment may not be transparent to the
person assessed. As mentioned above, transparency in assessment allows those assessed to
check the results and suggest corrections for mistakes, when necessary. One of the ten
principles of the influential Leiden Manifesto for research evaluation is, “Keep data collection
and analytical processes open, transparent and simple”(Hicks et al., 2015). In practice,
simplicity is a core aspect of transparency because those evaluated may not be able to
understand complex or large-scale computing solutions even if they are fully public.
Unfortunately, since research is complex and carried out on a large scale internationally, all
TARA probably has either obvious or hidden complexity, as described next, limiting its
transparency. In situations where adequate transparency is impossible because complexity is
necessary for sufficient accuracy, reproducibility might sometimes be judged to be an
acceptable alternative.
2.1 Publication databases
2.1.1 Coverage. The publication databases used in research evaluation are mostly controlled
by commercial organisations including Dimensions.ai (Digital Science), Scopus (Elsevier) and
the Web of Science (Clarivate). These organisations broadly publish their methodologies for
finding and including journal articles. The main sources are manually curated lists of
academic journals for Scopus [1] and the Web of Science [2], which are published and public.
The process of choosing these journals is human-based and private, although the outcome is
public.
Elsevier and Clarivate presumably have agreements with the publishers to harvest
relevant information about the journals from the publishers’websites and then use their
own private algorithms to transform the rawdataintobibliometricinformation,by
extracting titles, matching articles with journals and metadata such as DOIs, author
names and affiliations, assuming that they are not already supplied in XML or other forms
by the publisher. This basic processing seems uncontroversial, even if the precise
algorithms are not published. Nevertheless, there can be errors for unusual cases, such as
for conference papers dual published as journal articles and for consortia listed as authors
of academic papers.
Dimensions.ai uses public Crossref data provided freely by publishers as well as
arrangements with other publishers to directly harvest their bibliometric metadata, and
crawlers to harvest various repositories, such as PubMed and arXiv [3].Itdoesnot
publish a list of journals indexed but explains how to check if a journal is indexed [4].
2.1.2 Metadata and citation indexing. In addition to ingesting publications, bibliometric
databases typically have layers of extra information from largely or completely automated
AJIM
77,1
176
Get this document and AI-powered insights with a free trial of vLex and Vincent AI
Get Started for FreeStart Your 3-day Free Trial of vLex and Vincent AI, Your Precision-Engineered Legal Assistant
-
Access comprehensive legal content with no limitations across vLex's unparalleled global legal database
-
Build stronger arguments with verified citations and CERT citator that tracks case history and precedential strength
-
Transform your legal research from hours to minutes with Vincent AI's intelligent search and analysis capabilities
-
Elevate your practice by focusing your expertise where it matters most while Vincent handles the heavy lifting
Start Your 3-day Free Trial of vLex and Vincent AI, Your Precision-Engineered Legal Assistant
-
Access comprehensive legal content with no limitations across vLex's unparalleled global legal database
-
Build stronger arguments with verified citations and CERT citator that tracks case history and precedential strength
-
Transform your legal research from hours to minutes with Vincent AI's intelligent search and analysis capabilities
-
Elevate your practice by focusing your expertise where it matters most while Vincent handles the heavy lifting
Start Your 3-day Free Trial of vLex and Vincent AI, Your Precision-Engineered Legal Assistant
-
Access comprehensive legal content with no limitations across vLex's unparalleled global legal database
-
Build stronger arguments with verified citations and CERT citator that tracks case history and precedential strength
-
Transform your legal research from hours to minutes with Vincent AI's intelligent search and analysis capabilities
-
Elevate your practice by focusing your expertise where it matters most while Vincent handles the heavy lifting
Start Your 3-day Free Trial of vLex and Vincent AI, Your Precision-Engineered Legal Assistant
-
Access comprehensive legal content with no limitations across vLex's unparalleled global legal database
-
Build stronger arguments with verified citations and CERT citator that tracks case history and precedential strength
-
Transform your legal research from hours to minutes with Vincent AI's intelligent search and analysis capabilities
-
Elevate your practice by focusing your expertise where it matters most while Vincent handles the heavy lifting
Start Your 3-day Free Trial of vLex and Vincent AI, Your Precision-Engineered Legal Assistant
-
Access comprehensive legal content with no limitations across vLex's unparalleled global legal database
-
Build stronger arguments with verified citations and CERT citator that tracks case history and precedential strength
-
Transform your legal research from hours to minutes with Vincent AI's intelligent search and analysis capabilities
-
Elevate your practice by focusing your expertise where it matters most while Vincent handles the heavy lifting