PLANTUCD: A DATASET OF SOFTWARE REQUIREMENTS AND CORRESPONDING PLANTUML-BASED CLASS DIAGRAMS

The automatic generation of UML class diagrams from natural language software requirements remains a challenging problem at the intersection of software engineering and natural language processing. Recent advances in neural machine translation and large language models have renewed interest in data-driven approaches for requirement-to-model transformation; however, progress is constrained by the lack of high-quality, structured, and publicly available datasets. This paper introduces PlantUCD, a curated dataset that pairs natural language software requirements with corresponding PlantUML class diagram representations and structured abstract syntax trees (ASTs). The dataset is designed to support research in requirement understanding, model-driven development, and neural code generation. We describe the dataset construction process, schema design, and potential research tasks. PlantUCD is released publicly under the MIT License to facilitate reproducible and extensible research.