BengaliMCQ: Automatic Generation and Answer Prediction of Academic Multiple-Choice Questions in a Low-Resource Language

2026-08-16

AI Summary

This study addresses the limitations of conventional Retrieval-Augmented Generation (RAG) in low-resource Bengali, where ignoring textual hierarchy degrades multiple-choice question (MCQ) generation and answer prediction. We propose a Structure-Aware RAG framework that models textbooks as hierarchical graphs for the first time, leveraging contrastive learning-based Graph Neural Networks to enable precise paragraph retrieval and provide structured context for large language models. Experimental results demonstrate that this framework outperforms dense retrieval baselines across key metrics. Furthermore, it significantly enhances both the relevance of domain-specific MCQ generation and answer prediction accuracy. These findings confirm the method’s effectiveness in overcoming challenges associated with structured knowledge retrieval and generation in low-resource scenarios.

Abstract

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.

Problem

• Retrieval-Augmented Generation
• Low-Resource Language
• Multiple-Choice Question Generation
• Hierarchical Structure
• Bengali

Innovation

• Structure-aware RAG
• Hierarchical Graph
• Graph Neural Network
• Low-Resource Language
• MCQ Generation