Peer review is the quiet architecture of science: it catches missteps, gauges novelty, and helps journals steer toward credible conclusions. Yet the system is famously fractious—slow, opaque, and sometimes biased toward the most visible names in a field. A new study from researchers at Green University of Bangladesh, in collaboration with the University of Dhaka and Dhaka International University, proposes a bold tweak: could a software assistant actually pick better reviewers by tracing the web of allied references behind a paper? The authors, led by Tamim Al Mahmud with colleagues B M Mainul Hossain and Dilshad Ara, sketch a pipeline that reads a manuscript’s footprints and returns a ranked pool of potential referees. It is not a replacement for human editors, they argue, but a turbocharged starting point for a process that has become unwieldy as science multiplies.
What makes this idea feel almost provocative is its simplicity: use what a paper cites to map who is actively shaping the same conversation, then measure those researchers by the very metrics skeptics rely on—how often they publish, how influential their work is, and even how to contact them. The project—presented as a concrete software toolkit with names like AFFA, ACFA, ARA, and AEFA—presents a reproducible recipe for auto-suggesting reviewers. It’s a glimpse of the future where editors tap a computer-assisted roadmap to assemble a committee of experts within minutes, not days. The study is a proof-of-concept, but its implications ripple through how we think about publishing, expertise, and trust in a data-driven era of science.
Two things stand out from the outset. First, the study locates its power in the ordinary: every scholarly article comes with references, and those references form a map of who already thinks about the topic. Second, the project is a collaboration across institutions in Bangladesh—Green University of Bangladesh, University of Dhaka, and Dhaka International University—reflecting a rising regional contribution to AI-enabled research workflows. The authors behind this exploratory work say the goal is practical: to help editors cope with swelling submission streams while keeping reviewer quality in view. In their own words, they want to automate a piece of editorial labor that historically demanded long hours, careful judgment, and sometimes luck.
What the paper does and why it works
At the core, the authors outline a sentence-by-sentence pipeline that turns a manuscript into a candidate reviewer roster. It starts with the obvious input: a paper, preferably in text form. The software then homes in on the References section, parsing author names, publication titles, and venues to assemble a list of allied scholars who’ve already weighed in on related topics. The central idea is simple: the more frequently an author appears in the references of a paper, the more their work touches the same material, and thus the more relevant they might be as a reviewer. This is the essence of what they call the Automatic Frequency Finding Algorithm, AFFA.
AFFA is not just counting names; it creates a structured representation of which researchers are anchored to a given topic through the paper’s own cited literature. The team stores authors in a hash map—an efficient data structure that keeps track of how often each name appears. As the reference list grows, the map tallies frequencies, and those tallies become the first-pass signal of topical centrality. It’s a bit like building a social map from a single apartment listing: who shows up most often in the neighbor’s book club? Those people are likely deeply embedded in the same conversation.
But frequency alone isn’t enough. The next stage, ACFA, steps in to weigh a short list of top candidates by their scholarly impact. The authors describe trying to pull publication metrics from public profiles—Google Scholar being the most accessible—and then extracting h-index, i10-index, and citation counts for the researchers’ first authors. In practice, the system searches for each candidate’s profile, discovers a unique identifier, and pulls the metrics that help differentiate a prolific contributor from a prolific influencer who rarely engages with new literature. The aim is to quantify “fit” not just by topical proximity but by demonstrated engagement with the kinds of papers the target manuscript builds upon.
The trickier part is emails. AEFA, the Automatic Email Finding Algorithm, tries to locate valid contact information to ensure editors can reach the potential reviewer. This step is messy in the real world—profile pages move, emails change, and some scholars keep their digital traces sparse. The authors acknowledge that not every top candidate will yield a contact email, and they describe strategies to cope with incomplete data, including fallback heuristics and cautious rejection of dubious signals. The combination of AFFA, ACFA, and AEFA then feeds into ARA, the ranking algorithm that threads everything into a single, ranked list of candidates with practical email links when possible. Put together, the quartet is designed to take a manuscript from submission to a suggested reviewer pool in what the authors frame as “within a minute” in ideal circumstances.
Why it matters for science
Speed is not the only virtue here. The authors argue that this approach reframes how we think about expertise in peer review. Traditional editor assignments often lean on a circle of well-known names or manual searches that reflect long-standing hierarchies within a field. By contrast, the allied-reference approach treats the literature itself as a guide to who might understand a given paper’s nuances. If a reference list nods repeatedly to a cluster of researchers, those researchers are likely to be steeped in the same subtopic, standard methods, and debates. The system thus approximates a topic-aware search for reviewers that scales with the literature’s growth—an antidote to the brittleness of hand-picked panels when a field fractures into new subfields.
In their experimental lineup, the authors tested the method on three published articles. They report strong performance in the early stages of the pipeline: frequency finding (AFFA) reaches near-perfect accuracy in identifying the dominant allied authors, and the later stages—capturing the right publication metrics (ACFA) and locating contact points (AEFA)—achieve impressive, though more nuanced, success. The results aren’t a guarantees of flawless reviewer selection, but they do illustrate a path toward materially aiding editors facing an avalanche of submissions. When the tool points editors to a pool of candidates who have both topical legitimacy and demonstrable impact, it can compress an often lengthy triage process into something more predictable and repeatable.
The study also helps illuminate the ethical and practical design space editors must navigate. The approach leans on publicly available profiles and citations—data that many researchers have willingly exposed to the world. Yet there’s a broader question about consent, privacy, and bias: will such a system systematically privilege senior, highly cited researchers over rising voices? The authors acknowledge these tensions and frame their work as a supplement to human judgment, not a replacement for it. In other words, the software can offer a data-informed starting point, but a thoughtful editor’s eye remains essential to interpret, contextualize, and balance conflicts of interest, diversity of perspectives, and disciplinary norms.
Surprising findings and the road ahead
One of the study’s most striking implications is the idea that a paper’s own reference network can illuminate the most appropriate future reviewers. It’s a bit like using a family tree to predict whom might best appreciate a new family story; the closer you are to the shared lineage of ideas, the more likely you are to provide an informed, constructive critique. This insight reframes reviewer selection as a problem of relational intelligence—where the topology of citation networks helps us infer expertise, collaboration patterns, and intellectual alignment with the manuscript’s themes.
The results from the three datasets show promising trends: high accuracy in frequency-based identification of allied authors, strong signals in author metrics, and a substantial portion of candidate emails retrievable from public sources. Yet the authors are careful about overinterpreting the numbers. They frame their work as a proof of concept—an encouraging demonstration that a computational assistant can do meaningful editorial work, but not a final adjudicator of who deserves a review. Real-world deployment will demand refinements: more robust data sourcing beyond a single platform, better handling of disambiguation for common names, and more nuanced treatment of authors who work across multiple subfields or in emerging areas without a long track record.
From a broader perspective, the project hints at a future where editorial workflows become more standardized and auditable. If journals adopt tools like AFFA, ACFA, ARA, and AEFA as part of a submission workflow, editors could gain a transparent, data-driven baseline for reviewer invitations, followed by human judgment to handle nuance, preferences, and ethics. That envisioned future would resemble modern software development, where automated tests and human reviews complement each other to ensure quality and accountability. The institutions behind the study—Green University of Bangladesh in Dhaka, with collaboration from the University of Dhaka and Dhaka International University—are signaling a growing regional interest in AI-assisted scholarly infrastructure, an encouraging sign of global diversification in the science of science itself.
Lead authors Tamim Al Mahmud, B M Mainul Hossain, and Dilshad Ara from the participating institutions lay out a pragmatic blueprint: a set of modular algorithms that can be integrated into existing manuscript-management systems, offering editors a smarter place to start. The ambition isn’t to reinvent peer review with a single miracle tool, but to reduce the drudgery, accelerate initial triage, and thereby free editors to focus on the human elements that matter most—context, fairness, and the cultivation of trustworthy scholarship.
In the end, the work is a reminder that science, at its core, is a cooperative enterprise across people, papers, and ideas. If the references a paper tethers itself to can guide us toward the right reviewers, then perhaps the future of peer review lies not in a crowd of nameless algorithms but in a thoughtful blend of data-driven assists and human discernment. That balance—between speed and care, between automation and accountability—holds the potential to strengthen the trust readers place in scientific publications, one carefully chosen reviewer at a time.
