Abstract
Personalized learning has become an essential paradigm in technology-enhanced education, yet most existing recommender systems are designed for older learners and rarely address the unique needs of toddlers. Early childhood learning requires approaches that account for developmental diversity, short attention spans, and the importance of engaging stimuli. This paper presents a hybrid LightFM-based recommendation framework that integrates toddler behavioral attributes from autism screening datasets with gamified content descriptors such as difficulty level, sensory load, and adaptive rewards. The goal is to deliver developmentally appropriate, motivating, and inclusive learning pathways for young children. A synthetic engagement simulation protocol was introduced to overcome the absence of large-scale toddler interaction data, enabling the generation of reward signals based on completion rates and time-on-task thresholds. Experimental evaluation compared the proposed framework against traditional baselines, including random, content-based, and supervised predictors, as well as ablated versions of LightFM. Results demonstrate that the proposed model achieved superior predictive accuracy (AUC 0.83, Precision@1 0.62, NDCG@5 0.71) while also maximizing cumulative rewards, minimizing regret, and ensuring higher diversity and faster convergence during simulated interactions. These findings validate the methodological soundness of the approach and highlight its potential to support personalized gamified learning in early childhood. Future work will involve real-world user studies, multimodal feature integration, and privacy-preserving deployment strategies to strengthen the practical applicability of the framework.

This publication has 16 references indexed in Scilit: