Figures and CSVs for figures in paper

Full description

Research on automated assessment spans nearly five decades and is currently experiencing a surge of publications using Large Language Models (LLMs). There has not been a previous quantitative analysis of how technology trends have shaped the automated assessment field over time, or whether the current concentration on a single technology is unprecedented. I present a bibliometric study of technology trends in automated assessment to address this gap. A broad search of six literature databases returned 169,521 records, which I reduced reduced to a corpus of 9,963 on-topic publications. Technology mentions were extracted with LLM assistance, consolidated into 458 keywords, and mapped to a taxonomy of 13 categories. The taxonomy was validated computationally: held-out papers' categories were predicted from their text with high accuracy, and textually similar papers shared categories far more often than chance. I applied this taxonomy to the corpus and analysed the results using growth modelling, growth decomposition, diversity indices, burst detection, and rolling rank correlations. The results show that the field has grown near-exponentially, with doubling time falling from 2.8 years (2000–2010) to 1.4 years (2015–2025). Several growth phases were identified along with driving technologies, however the LLM trend is without precedent. LLM-related publications grew by an average of 73.6\\% per year from 2023 to 2025 and accounted for 69\\% of publications in 2025, while publications not mentioning LLMs declined. Several other categories show concurrent bursts of activity, but topic diversity is falling and LLM growth is negatively correlated with other AI and machine learning approaches, suggesting these are being displaced. The analysis is limited by the impacts of external shocks to publication rates, such as the COVID-19 pandemic, and by the inability of a keyword classifier to fully represent the context of technology use in each publication. The results are significant despite these limitations, warranting further investigation.