Dynamic graph neural networks and evolutionary multi-objective optimization for adaptive quality evaluation in gamified preschool education

Play-based preschool education has emerged as a promising model for enhancing early childhood learning engagement and outcomes. However, traditional assessment models often fail to account for the dynamic and heterogeneous nature of learners, including cognitive differences, temporal interaction patterns, and individualized developmental trajectories. To address the neglect of learner dynamic heterogeneity in play-based preschool education, this paper proposed an adaptive assessment framework that integrated dynamic graph neural networks (GNNs) and evolutionary multi-objective optimization (EMO). The framework modeled curriculum–learner relationships by constructing heterogeneous interaction graphs, extracting temporal structural representations using GNNs, and balancing three pedagogical objectives—knowledge acquisition, engagement, and adaptability—through an EMO algorithm. A closed-loop feedback mechanism drove the co-evolution of both the model and the curriculum. Experimental results demonstrated that the proposed framework significantly improved post-test scores (by 0.2–0.7 points), learner engagement (correlation R² = 0.608), and individualized satisfaction, particularly among visual and kinesthetic learners. Comparative analyses further highlighted the robustness, scalability, and adaptability of the proposed method, establishing it as a computationally grounded and dynamically optimized intelligent curriculum design paradigm for early childhood education.