DIGITALLY FILE EXTRACTION OPTIMISED WITH GPT-4O BASED MOBILE APPLICATION FOR RELEVANT EXERCISE PROBLEM GENERATION

This research studies the creation of an AI-driven question extraction system using the GPT-4o model to improve the accessibility and variety of practice questions for students. The study tackles the difficulties in sourcing relevant practice materials and aims to transform educational technology by integrating mobile learning. A mobile application was built with Dart and Flutter, designed to extract questions from PDF files. The system is capable of generating both multiple-choice and essay questions across different difficulty levels. The quality and relevance of the generated questions were assessed using ROUGE metrics. The results indicated strong performance for multiple-choice questions, especially in single-answer and true/false formats. However, the system encountered difficulties in producing complex essay questions, highlighting the need for further improvements in understanding intricate contextual relationships. Key findings reveal effective generation of multiple-choice questions with high precision and recall; inconsistent performance in essay question generation, with simpler questions yielding better results; and ROUGE-1 metrics surpassing ROUGE-2 and ROUGE-L, indicating a stronger ability to generate straightforward questions. The research concludes that while the developed system shows potential in enhancing educational resources, additional research is necessary to refine complex question generation. Recommendations include broadening the training dataset and creating specialized models for question generation tasks to enhance the effectiveness of AI-assisted learning tools.