CQELedu: Design and Implementation of a LangChain and GPT-4o Mini-Based Web Application for Custom Question Generation and Error-Based Learning in Education

Abstract

Evaluation and feedback are central to the learning process, making it essential to analyze learners’ incorrect answers and generate tailored questions or supplementary materials that address their weaknesses and improve learning outcomes. This paper presents CQELedu, a web application designed to generate questions based on learning materials and analyze incorrect responses to create error-driven follow-up questions. The proposed system leverages GPT-4o mini and LangChain to generate both subjective and objective questions from user-provided materials. The system analyzes incorrect answers and generates customized follow-up questions, helping learners identify and address gaps in understanding. The relevance, accuracy, and response time of the generated questions are assessed both quantitatively and qualitatively to evaluate system performance. Experimental results demonstrate high cosine similarity scores for relevance, validating the relevance of the generated questions to the learning materials. The accuracy of the system is reflected in the applicability of the questions derived from incorrect answers. Additionally, the system exhibits improved response time compared to existing methods, highlighting its efficiency. Thus, CQELedu is well-suited for supporting dynamic learning environments, particularly in standardized testing contexts such as certification exams and school assessments. Future research should explore the integration of learning materials from diverse fields, the implementation of multilingual support, and the scalability potential, aiming to expand the applicability of CQELedu across various educational domains.