Insights into the South African research landscape through mining theses and dissertations using transformer-based language models

Postgraduate research plays a critical role in the development of national research capacity and advanced knowledge production. In South Africa, electronic theses and dissertations (ETDs) constitute a substantial yet underutilised body of scholarly output for analysing postgraduate training, knowledge production, and scholarly influence. Despite their significance, ETDs are rarely incorporated into large-scale scientometric analyses due to fragmented institutional repositories, limited standardisation, and weak integration with global bibliographic infrastructures. This study develops a methodology for harvesting, enriching, and analysing ETDs using a combination of metadata integration, full-text mining, and citation analysis. Institutional ETD metadata are integrated with OpenAlex, an open-access scholarly knowledge graph, using a stemming-assisted title matching approach and persistent identifier mapping. In addition, the study applies automated PDF mining techniques to extract reference lists and citation contexts directly from ETD full texts, enabling fine-grained analysis of citation behaviour beyond aggregate citation counts. The enriched dataset supports citation network construction, concept mapping, supervisor–student linkage, and longitudinal analysis of postgraduate research output. Empirical analyses focus on South Africa’s research-intensive universities and examine institutional productivity, temporal growth patterns, language use, retention dynamics, and citation characteristics of postgraduate research between 2000 and 2024. The results reveal a concentration of postgraduate research output among a small number of institutions, sustained growth prior to 2020, and a marked decline thereafter. This post-2020 downturn is likely influenced by economic pressures, funding constraints, and the disruptive effects of the COVID-19 pandemic, with implications for future doctoral production and national policy targets. By exploring ETD metadata integration, full-text citation mining, and open bibliographic enrichment, this study extends traditional publication-based scientometrics and demonstrates the value of ETDs as instruments for monitoring postgraduate research training capacity and informing evidence-based higher education policy in South Africa.

Grantor dc:publisher

Stellenbosch : Stellenbosch University

Year dc:date.issued

2026

Author dc:creator

• Khanyi, Masana Hlengiwe Michelle

Advisors dc:contributor.advisor

• Dunaiski, Marcel
• Van Lill, Milandre

Language dc:language.iso

en

Repository record dc:identifier.uri

https://scholar.sun.ac.za/handle/10019.1/136187

OAI identifier oai:identifier

oai:scholar.sun.ac.za:10019.1/136187