Knowledge Retrieval-Based Intelligent Question and Answer Generation Framework for Education

Abstract

The paper introduces a Knowledge Retrieval-Based Intelligent Question and Answer Generation Framework for Education, leveraging Retrieval Augmented Generation (RAG) to enhance Large Language Models (LLMs) in producing high-quality, contextually relevant examination questions and answers across subjects. The framework addresses challenges in ensuring comprehensive subject coverage, educational standards, and evaluation metrics. Key components include Optical Character Recognition (OCR), data chunking, vectorization using multilingual BERT, and custom retrieval functions. The RAG system's effectiveness is evaluated using the RAGAs metric, covering metrics like faithfulness, answer relevancy, context recall, and context precision. Comparative results reveal OpenAI models surpassing Google's Gemini in coherence and context relevance due to differing architectures and training methods. Concluding, the paper highlights potential for both large and small language models in tailored educational applications, noting limitations such as hallucinations, high resource demands, and contextual drift.