PTIT-CourseQA: A Vietnamese University Course-Material QA Benchmark, Annotations and KG-RAG Evaluation Outputs
Authors/Creators
- 1. Faculty of Artificial Intelligence, Posts and Telecommunications Institute of Technology
Description
PTIT-CourseQA is a Vietnamese question-answering benchmark over 11 university textbooks of the Posts and Telecommunications Institute of Technology (PTIT) in four fields (information security, information technology, electronics, marketing). It contains 1,000 test and 200 development questions of four types (single-hop, multi-hop within a book, cross-chapter/cross-book, unanswerable) with answers, verbatim evidence excerpts, chunk identifiers and PDF page numbers.
The archive also contains the human annotation labels, the hallucination labels, all predictions and retrieval outputs of six systems (no retrieval, BM25, dense, hybrid, LightRAG and a structure-framed KG-RAG) with five small language models, the metrics, significance tests, judge scores and cost measurements, together with a snapshot of the code.
The full text of the textbooks is not included: it is copyrighted by PTIT and is outside the CC BY 4.0 licence. The corpus is described by a chunk index (identifiers, pages and SHA-256 hashes) and can be rebuilt from the textbooks with the released code. For UIT-ViQuAD2.0 only the question identifiers of the evaluation subset are included.
Version 1.0.1 adds descriptive statistics of the S5 graph and of the benchmark questions (outputs/descriptive_stats.json, computed by work/scripts/11_descriptive_stats.py); all other files are unchanged.
Files
ptit-courseqa-kgrag-v1.0.1-code.zip
Additional details
Related works
- Is supplemented by
- Software: https://github.com/cuongphamduc/ptit-courseqa-kgrag (URL)
- Is variant form of
- Dataset: https://huggingface.co/datasets/cuongphamduc/PTIT-CourseQA (URL)