Automatic MCQ with answer generation for Bangla medium SSC-level students

Abstract

It has almost become a tradition for Bangladeshi students to join coaching centers before exams only to take part in Multiple Choice Question model tests. Since these coaching centers charge a substantial amount of fees for these MCQ exams, students not only waste their money but also their valuable time attending them. This widespread practice of pre-exam testing is specially noticeable in Bengali medium students. However, there has not been much research done in this regard to solve this issue. Our paper aims at solving this problem by developing a system that will not only automatically generate MCQs but also be able to predict the answers of inputted MCQs. The proposed framework in this paper incorporates natural language processing (NLP) functions for extracting and cleaning academic Bengali textual data from NCTB verified text books and utilizes a hybrid form of Graph-based Retrieval Augmented Generation (GraphRAG) to produce appropriate multiple choice questions along with the capability to predict answer of a given MCQ. This would support Bengali medium SSC candidates in designing their own MCQ model tests. Our research result demonstrates the demand for high-quality Bengali embedding models as well as provides implementation strategies for any future RAG-based automated educational tool designed for Bengali language. The MCQ generation framework developed by our team, would be able to provide teachers and students with numerous practice MCQs whereas the answer prediction pipeline, could help students study more efficiently. Integration of our system can lead towards effective learning environments in Bengali medium institutions.