A Greek Government Decisions Dataset for Public-Sector Analysis and Insight

Public transparency can be enhanced through machine-readable government data: The paper demonstrates that converting 1 million Greek government decisions into a structured, searchable format enables citizens and researchers to access public-sector information more effectively than traditional PDF documents, supporting the principle that government decisions should be genuinely accessible to the public.

AI systems can help citizens navigate complex bureaucratic information: The researchers successfully built a retrieval-augmented generation (RAG) system that can answer questions about government decisions, suggesting that AI-powered chat assistants could democratize access to public information by allowing citizens to ask natural language questions rather than manually searching through thousands of documents.

Large-scale public datasets are essential for developing trustworthy AI in government: By releasing both the dataset and the reproducible extraction pipeline openly, the authors enable independent researchers to audit, validate, and improve AI systems used in public administration—a critical practice for ensuring government AI systems remain accountable and transparent.

Specialized AI models for legal and governmental domains require quality domain-specific training data: The paper highlights that general-purpose language models may not adequately serve specialized sectors like government and law; high-quality, domain-specific datasets like this one are necessary to develop AI systems that accurately understand and reason about public-sector decisions.

Responsible data practices require addressing limitations upfront: The authors explicitly discuss limitations and outline future directions, modeling how researchers should transparently communicate both what their datasets enable and what they cannot do, helping prevent misuse or overconfidence in AI systems built on such resources.