How an algorithm can lift dissertation supervision allocation above mere logistics
You may also like
How do you pair dissertation students with the right supervisors in a way that is fair and consistent and doesn’t require weeks of manual sorting? Here is how a data-driven matching method can turn this administrative chore into an educational opportunity.
Shortly after being appointed dissertation convener for a postgraduate programme, I faced exactly this problem; I had to allocate more than 600 students across about 100 supervisors. The usual tools were available to me: sign-up sheets, preference surveys and manual sorting.
Each tool has its flaws. Sign-up sheets reward whoever clicks first. Preference surveys turn into popularity contests that overload some supervisors and overlook others. Manual sorting works while the matches are obvious, but at some point, it becomes arbitrary. A systematic approach solves much of this, and algorithms that produce optimal allocations do exist, but their complexity and black-box logic have kept them out of institutional practice.
- How to supervise a student systematic review without losing a semester
- Spotlight guide: Research supervision that supports and inspires
- Campus Talks podcast: Why it takes more than a PhD to become a good doctoral supervisor
The deeper issue, however, is that we often treat allocation as logistics. It is not. Decades of research show that the student-supervisor relationship shapes engagement, satisfaction, performance and well-being. The same literature finds that a good topic fit between student and supervisor underpins a productive supervisory process.
Who supervises whom is an educational design decision, and it deserves to be made with the same care as curriculum development or assessment design.
How the supervisor-student matching method works
My aim was to base the decision on research interests in an intuitive and transparent way. That meant measuring how well each student’s dissertation topic aligned with each supervisor’s expertise. In economics, JEL codes classify research in three layers, moving from broad areas to more than 800 specific topics. Take the code E41 as an example:
- E stands for the broad area: macroeconomics and monetary economics
- E4 narrows the subfield to money and interest rates
- E41 reaches the specific topic of the demand for money.
This hierarchy makes it possible to prioritise exact matches while allowing for broader ones when necessary. It is what stops a student who works on, say, health economics from ending up with a macroeconomist. Each pairing gets a matching score, between 0 and 1, that quantifies this closeness, with higher values meaning better alignment. A rule-based algorithm then turns the scores into a full allocation in seconds.
The process also needs a clear picture of each supervisor’s agreed workload.
For the convenor, this is how it works:
- Use the lists of JEL codes that students and supervisors provide to give a matching score to every possible pairing.
- Rank the students from highest to lowest average score across all supervisors. When two students are tied, the one who could be matched with fewer supervisors is ranked first. A supervisor counts as a possible match when they share at least one broad area with the student’s topic.
- Working down that list, assign each student to the supervisor with whom they score highest among those who still have places within their agreed workload.
Why the algorithm method improved matches
The question, of course, is whether any of this made a difference. Two results suggest it did. First, the algorithm produced far better matches. In 82 per cent of pairs, student and supervisor shared at least one specific research topic, and the average matching score was more than three times higher than under a random allocation. Second, in a survey of 102 postgraduate students at the University of Glasgow, those whose topics aligned more closely with their supervisor’s expertise tended to report a more positive supervision experience and to rate the quality and relevance of feedback more highly.
How to apply it in your programme
The approach can be applied in any discipline. All it needs is a hierarchical classification of topics, whether an established one in your field or one you assemble yourself from staff research profiles and recent dissertation titles.
Practical lessons from implementation of the matching method include:
- Most students have never seen a hierarchical classification of topics, and the quality of the allocation depends directly on how well their selections describe their topic. You will need to guide them through the classification, with a short explanation and worked examples.
- To ensure that matching scores stay comparable across the cohort, set a range for how many topics to list. For example, three to five for students and four to six for supervisors.
- Treat the algorithm’s output as a starting point, refine the process, then automate.
Treating supervisor allocation as the educational decision it is, rather than an administrative routine, is both possible and worthwhile; students gain better-matched supervision, and programmes save staff time on the process.
Ana Santos will start as a lecturer in economics the School of Economics at the University of Sheffield in January 2027. She is a fellow of the Global Labor Organization and the Rimini Centre for Economic Analysis. This article is based on her research paper “Improving student-supervisor matching: educational gains from a data-driven approach”, published in August 2026 in Research in Learning Technology. The replication code is available online.
If you would like advice and insight from academics and university staff delivered direct to your inbox each week, sign up for the Campus newsletter.
