Published Paper Details:

Manual preparation of question papers is often repetitive, time consuming and difficult to scale especially when personalization and fairness is required. This paper proposes a dynamic question paper generation system using transformer based natural language processing along with semantic similarity analysis for the automation and improvement of the assessment process. The system is designed to generate multiple choice questions from existing content while taking into consideration topic coverage, difficulty level and cognitive categories based on Bloom's taxonomy. We apply semantic similarity methods to identify and remove duplicate or close questions, to ensure quality and diversity of the generated papers. The system also randomizes the question sets for each student, reducing the repetition and increasing the fairness of the assessments. It also has a performance analysis module which analyses the responses of the students and points out the weaker areas for improvement. The overall approach lessens the manual effort required from educators while supporting more adaptive and data driven assessment practices. The results indicate that the system can generate consistent and reliable question papers with relevance and diversity.