Advanced Subjective Question Bank Generation Using Retrieval Augmented Generation Architecture

Abstract

Large language models (LLMs) have shown great promise for various natural language processing tasks, including question generation. However, generating subjective questions that are relevant and informative remains a challenge. Existing approaches often rely on predefined question templates or manually crafted knowledge bases, which limit the diversity and quality of generated questions. In this paper, we propose a novel approach that leverages LLMs to generate subjective questions with the help of a custom knowledge base. Our knowledge base is constructed by extracting and embedding relevant information from a given text corpus. By combining the LLM's language generation capabilities with the domain-specific knowledge from the knowledge base, our system can generate more informative and contextually relevant subjective questions. Experimental results show that our approach beats the existing methods in terms of question quality and relevance. Specifically, the Google Gemini LLM achieved the highest score among the compared models, with an average rating of 3.92 out of 5 for question quality and an average relevance score of 0.90. Our approach has several advantages over existing methods. First, it does not rely on predefined question templates, which can limit the diversity of generated questions. Second, our custom knowledge base is constructed from a domain-specific text corpus, which ensures that the generated questions are relevant to the given domain. Third, our approach can be easily adapted to different domains by constructing a new knowledge base from the corresponding text corpus.