Residual Finite Group Element Dandelion-Informed Attention Network for Symbolic Knowledge Integration in AI-Driven Educational Intelligence

Recent advances in intelligent educational systems have emphasized the need for accurate interpretation of Computational Thinking (CT) behaviors. The contextual dependencies and symbolic reasoning necessary to comprehend CT in game-based learning environments are frequently difficult for traditional deep learning techniques to capture. This disparity draws attention to the issue of the current models of educational analytics’ poor generalization and interpretability. The current study presents a novel framework called the Residual Finite Group Element Dandelion Informed Attention Network (Res-FGE-DIAN) under the paradigm of AI in Education with Symbolic Knowledge Integration in order to address this challenge. The proposed framework analyzes CT behaviors using game-based interaction data named “Educational Testing Service (ETS) and Educational Knowledge” dataset, which consists of 427 player records from the AutoThinking learning game. The Masked Attention Mask Transformer (Mask2Former) was used to extract features for improved contextual encoding, and the Cumulative Curve Fitting Approximation (CCFA) algorithm was used for data pre-processing in order to normalize and refine the learning traces. The Res-FGE-DIAN combines a Residual Group Attention Network (ResGANet) and a Finite-Element-Informed Neural Network (FEI-NN), both of which are optimized using the Dandelion Optimizer (DO) to guarantee effective convergence and stability. Symbolic representation and decision accuracy are further improved by supporting architectures, such as the Residual Finite Group Element Dandelion Informed Attention Network (RES-FGE-DIAN). The suggested model predicts and interprets learners’ CT performance with 99.9% accuracy, according to experimental validation.