Dual-Path Transformer: A Novel Model for Education Resource Analysis and Optimization

Abstract

Reasonable distribution of educational resources is an important foundation for realizing educational equity and improving educational quality. However, analyzing educational data effectively remains challenging due to its heterogeneous nature, comprising both numerical metrics (e.g., study time, attendance) and categorical features (e.g., school type, family income). This paper proposes the Dual-Path Transformer model, an innovative educational resource analysis framework that addresses these challenges through two key innovations: (1) A dual-path feature processing architecture that separately processes numerical and categorical features through dedicated paths, then intelligently fuses them using an adaptive fusion strategy to capture their intrinsic relationships; (2) A specialized multi-attention mechanism that models complex educational patterns from multiple perspectives, enabling the model to understand intricate relationships such as the coupling between family economic status and learning resource access. Experiments on public educational datasets demonstrate that our model achieves superior performance with 90% accuracy and 88% F1 score, significantly outperforming both traditional machine learning approaches (Logistic Regression, SVM, Random Forest) and classical deep learning models (Neural Network). The model's interpretable nature provides educators with actionable insights into key factors affecting resource allocation, enabling data-driven decisions for promoting educational equity. These results demonstrate the Dual-Path Transformer's effectiveness in educational data analysis and its practical value for optimizing resource allocation strategies.