Global research output in Transformers and Denoising Generative Models in Construction Engineering: A bibliometric analysis

Transformers and denoising diffusion models have revolutionised computer vision and language processing, yet their penetration into construction engineering remains unclear. To chart this emerging landscape, we carried out the first global bibliometric review dedicated to these two paradigms. A PRISMA‑guided search of Scopus retrieved 1,765 english‑language journal articles; title–abstract screening reduced the articles to 200 records that explicitly implemented either transformers or diffusion techniques. Analysis and science mapping were conducted in VOSviewer to uncover publication growth, geographic distribution, key thematic areas, use cases and future directions. Annual output has risen twelve‑fold since 2019, with Automation in Construction, Applied Sciences and Sensors accounting for half of all papers. China generates 49 % of the literature, followed by Japan (10 %) and Canada (9 %). Keyword clustering reveals six thematic communities centred on (i) vision‑based safety and quality monitoring, (ii) core algorithmic development, (iii) generative material and layout design, (iv) carbon‑oriented risk analytics, (v) natural‑language knowledge extraction and (vi) electrical‑equipment diagnostics. Vision transformers deliver up to 44 % fewer false alarms in 3‑D concrete‑printing defect detection, while diffusion models shorten shear‑wall optimisation runtimes by an order of magnitude. Gaps persist in longitudinal impact assessment, multimodal model fusion and data‑set availability. We conclude with a research roadmap that prioritises uncertainty‑aware hybrid architectures, shared multimodal benchmarks and industry-academic test beds to accelerate trustworthy, carbon‑aware automation across the construction life‑cycle.