AI-Driven Automated Question Paper Generator Using Past Exam Datasets

Automatic Question Paper Generation is a crucial problem in current educational technology. The manual process of generating test papers is cumbersome, prone to bias in the distribution of difficulty, and rarely achieves consistency year after year. This article proposes a complete reviews existing works on automated question paper generation through AI and presents a new combined system that uses NLP, deep learning, and difficulty classification based on Bloom’s Taxonomy to develop balanced curriculum-compliant question papers using past exams. Five recent publications have been examined regarding rule-based systems, information retrieval, sequence-to-sequence deep learning, transformers, and LLMs. From the above literature review, we suggest a two-step process consisting of a (1) question bank generator using BERT to extract and de-duplicate questions and assign tags, and (2) paper assembler using GPT to assemble questions based on difficulty and topics. Experiments demonstrate our proposed system can achieve 91% question accuracy and takes less than 2 minutes compared to 1 hour previously, which is a 9-30% improvement.