针对孟加拉语低资源问题，提出结构感知的问答生成框架。
BengaliMCQ: Automatic Generation and Answer Prediction of Academic Multiple-Choice Questions in a Low-Resource Language
- 将孟加拉语教材建模为分层图，用图神经网络检索相关段落
- 生成的多选题相关性提升，答案预测准确率达78.3%
- 适合需要低资源语言教育内容生成的研究者和开发者
传统检索增强生成框架在处理文档时忽略其层级结构，导致在低资源语言如孟加拉语中表现不佳。为此，我们提出一种结构感知的RAG框架，将孟加拉语教科书建模为分层图，并使用对比训练的图神经网络检索少量相关段落。这些段落为大语言模型提供聚焦上下文，支持主题特定的多选题生成与领域内答案预测。实验表明，该框架在检索指标上优于强基线，生成的多选题更相关，答案预测准确率达到78.3%。
原文摘要 · Abstract (English)
Traditional retrieval-augmented generation (RAG) frameworks process documents without attending to their hierarchical structure, leading to poor performance, especially in low-resource languages such as Bengali. To address this, we propose a structure-aware RAG framework that models Bengali textbooks as hierarchical graphs and uses a contrastively trained graph neural network to retrieve a small set of relevant passages. These passages provide focused context for a large language model, enabling topic-specific multiple-choice question (MCQ) generation and in-domain answer prediction. Experimental results demonstrate that our framework outperforms strong dense retrieval baselines across retrieval metrics, produces more relevant MCQs, and achieves superior answer prediction accuracy.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。