vjdatabase

The VJDatabase is a specialized dataset designed to support research in legal-domain text retrieval by offering a collection of 3,259 Japanese query–document relevance pairs derived from labor insurance legislation. Each instance comprises three forms of a legal query—question_full containing chapter and article context, question_short as the succinct target sentence, and question as the core query—paired with a list of relevant_articles that maps to specific statutory provisions. The dataset’s JSON structure enables straightforward loading via the Hugging Face datasets library, fostering reproducibility and ease of use. By providing explicit links between natural-language legal statements and their corresponding laws, VJDatabase facilitates development and benchmarking of IR models, legal-tech applications, and hybrid retrieval–QA systems tailored to Japanese legislative texts. Its open accessibility on the Hugging Face Hub offers a robust benchmark for evaluating retrieval effectiveness in the legal domain.