Personalized content generation in family education based on deep learning

The generation of personalized learning content in family environments has been a growing interest in recent years, with the development of new and advanced deep learning techniques. Most previous works use the rule-based system and collaborative filtering techniques for educational content recommendation, and they do not demonstrate much flexibility in learner context and family factors. In this study, an Adaptive Content Generation system for Family Education (DL-ACG-FE) based on Deep Learning (DL) is proposed, aiming to assist personalization based on data in controlled computational environment. The framework combines Transformer-based NLP models for the modeling of content in context, reinforcement learning for the optimization of policies as a function of the signals generated by the interaction of the learner, and sentiment-based weighting systems based on structured performance indicators. The curriculum-constrained sequencing is provided for pedagogical coherence in the predetermined learning pathways. Experimental results are provided on a publicly available structured dataset, where improvements measured over baseline models are shown. In particular, DL-ACG-FE improves the Accuracy by 15.7%, F1-score by 15.9%, and Adaptive Content Relevance Score (ACRS) by 16.9% over the best evaluated baseline. Furthermore, short horizon retention analysis over a time period of 6 weeks also shows an improvement in comparative stability of retention in the tested experimental setting. These results provide a proof of computational efficiency of adaptive deep learning approaches under ideal test settings. The findings provide proof of feasibility using the available dataset, and further work is required to deploy them in the real world, validate them over time, and test them for scalability.