A data collection of UML diagrams in PlantUML notation mined from open-source software

Abstract

Unified Modeling Language (UML) diagrams are widely used in software engineering education and practice. However, research on real-world UML usage has been constrained by the limited availability of large-scale, openly accessible, and structurally annotated diagram collections, with existing resources differing in size, format, and the absence of per-diagram type labels. To address this gap, this data article presents a collection of 143,427 UML diagrams in PlantUML notation, mined from open-source software repositories indexed by the World of Code (WoC) infrastructure. The data pipeline identifies PlantUML files by file extension across World of Code, validates the presence of PlantUML content markers, and compiles the validated files with PlantUML to confirm syntactic correctness. Each diagram is then assigned to one of nine standard UML types using a large language model (LLM)-based classifier, and processed by a parser-based tool that extracts per-type counts of 57 structural element kinds and 28 connection kinds. The collection comprises class (51.6%), sequence (24.6%), activity (6.8%), component (5.7%), use case (5.0%), state (3.0%), object (1.7%), deployment (1.4%), and timing (0.1%) diagrams. Each diagram is released as both PlantUML source code and a rendered Portable Network Graphics (PNG) image, accompanied by metadata that records the assigned diagram type, the structural element and connection counts, textual size metrics, and source repository attribution. The collection is published on Zenodo. Anticipated reuse scenarios include empirical analyses of UML usage in open-source software, training and evaluation of artificial-intelligence models for diagram understanding and generation, benchmarking of diagram-to-code and code-to-diagram translation tools, and the development of educational resources for software engineering.