Supplementary Dataset: Designing Grammar-Guided LLM Outputs for Open Data Integration – A DSR Approach to IoT Data Platforms

This supplementary dataset supports the research presented in "Designing Grammar-Guided LLM Outputs for Open Data Integration – A DSR Approach to IoT Data Platforms." It contains outputs from an extensive evaluation assessing the effectiveness of applying grammar-guided Large Language Models (LLMs) to transform various open datasets into OGC SensorThings API (STA)-compliant JSON documents. The dataset comprises 720 outputs derived from six distinct open data sources, processed through three different smaller-scale LLMs (Qwen 2.5 Instruct, Llama 3.1 Instruct, and Phi-4). Each output set is generated under varying grammar complexities (long vs. short grammars) and input lengths (long vs. short contexts), designed to analyze the impact of grammar constraints and input token limitations on output quality and computational efficiency. All outputs have been verified for JSON validity. The dataset facilitates further analysis and validation of grammar-based generation methodologies aimed at enhancing interoperability and data quality in IoT data platforms.