Q&A Generation for Flashcards Within a Transformer-Based Framework

Flashcards are the main tool used in the Spaced Repetition memorization method, yet there are not always available for many topics due to the high effort required to create them. The combination of Transformer-based models with a Recommender System (RS) can enable a dynamic model to auto generate flashcards recommendation for learners and serious game players in order to improve their skills in learning programming. In previous work we introduced an Intelligent Serious Games (ISG) model that combined Deep Knowledge Tracing (DKT) with a Transformer-based Recommender. The ISG aimed at predicting the outcomes of the next missions in gameplay and enabling flashcard recommendations to complete the missions successfully. This research extends previous work by introducing a novel architecture and specifications for a Transformer-based recommender. We introduce a novel Transformer-based framework tailored to three different NLP tasks to dynamically generate flashcards in the form of supporting paragraphs, questions, and answers. We fine-tuned GPT-2, GPT-Neo, BART and T5 models on three new programming skills datasets, and evaluated them using standard metrics that target coherence and semantics. Our findings revealed that the framework is capable of generating coherent flashcards in a fully automated process using a single input string as prompt.