A Vietnamese Public-Administration Procedural Question-Answering Benchmark and Retrieval Corpus with Five-Type Administrative Entity Annotations

Dataset title: A Vietnamese Public-Administration Procedural Question-Answering Benchmark and Retrieval Corpus with Five-Type Administrative Entity Annotations

Related research article: When and Why Entity Grounding Helps Retrieval-Augmented Generation: A Diagnostic Evaluation Framework and an Expert-Verified Benchmark for Low-Resource Procedural Question Answering (submitted to Information Processing & Management). DOI to be added on acceptance.

Data Description

release/
├── qa/
│ ├── public_test_200.xlsx # Public split, 200 questions (canonical)
│ ├── public_test_200.csv # CSV export (same rows/columns)
│ ├── public_test_200_ner.xlsx # + q_entities / a_entities (silver 5-type tags)
│ ├── private_test_99.xlsx # Private (held-out) split, 99 questions (canonical)
│ ├── private_test_99.csv # CSV export
│ └── private_test_99_ner.xlsx # + q_entities / a_entities
├── corpus/
│ ├── corpus_parent.xlsx # 1,946 procedure records (1,942 distinct codes)
│ └── corpus_child.xlsx # 11,676 token-level chunks (indexable unit)
├── README.md # full datasheet (columns, join key, splits, scope)
└── metadata.json # machine-readable manifest + SHA-256 checksums + schemas