NCTB-QA
收藏资源简介:
NCTB-QA是由孟加拉国国家课程与教科书委员会发布的孟加拉语教育问答数据集,包含87,805条问答对,源自1-10年级的50本教材。该数据集独特地平衡了可回答(57.25%)与不可回答问题(42.75%)的分布,并包含对抗性设计的干扰项问题。数据通过自动化流程从教材中提取文本段落,并基于Gemini模型生成问答对,最终以SQuAD格式存储。其覆盖科学、人文等17个学科,旨在推动低资源语言环境下教育问答系统的鲁棒性研究,特别是模型对无答案场景的识别能力。
NCTB-QA is a Bengali educational question-answering dataset released by the National Curriculum and Textbook Board (NCTB) of Bangladesh. It contains 87,805 question-answer pairs sourced from 50 textbooks for grades 1 through 10. This dataset uniquely balances the distribution of answerable questions (57.25%) and unanswerable questions (42.75%), and includes adversarially designed distractor questions. Text passages are extracted from the textbooks via an automated workflow, and question-answer pairs are generated based on the Gemini model, with the final dataset stored in the SQuAD format. It covers 17 academic disciplines including natural sciences, humanities and other fields. This dataset aims to advance robustness research of educational question-answering systems in low-resource language environments, with a particular focus on models' ability to identify unanswerable scenarios.

- 1NCTB-QA: A Large-Scale Bangla Educational Question Answering Dataset and Benchmarking Performance达卡大学·计算机科学与工程系 · 2026年



