Automated content generation for intelligent tutoring systems

Abstract

The demand for eLearning systems has grown as education becomes a priority and the need for better technological assistance in learning increases. One type of software aimed at providing a more engaging and educational experience is the Intelligent Tutoring System (ITS). The ITS system featured in this research is called FAIE, which focuses on using Artificial Intelligence (AI) in learning. The blueprint-based approach allows educators to design content for students, but this raises two challenges. First, designing the content requires technical skills that can be time-consuming for educators with many tasks. Second, it is difficult to immediately codify an educator’s intrinsic knowledge. To tackle these challenges, a Blueprint-designer was created to generate content based on a text prompt from the educator containing key information. The content consists of problems to be answered with correct answers and distractors meant to challenge the student. The process of generating QA-Pairs involves determining a list of possible answers using a Named Entity Recognition (NER) model and passing the answers to a Question-Generation (QG) model. The QG model is based on the T5 Transformer model, which was trained to output questions based on the context and answer. Finally, distractors are generated by training a sequence-to-sequence model, BART, on the RACE dataset.