A Novel GNN-Transformer Hybrid Model for Evaluation Chinese Teaching Quality in Universities

The global expansion of Chinese language education demands advanced methodologies for precise teaching quality assessment in higher education. This research tackles the inherent subjectivity and inefficiency in conventional evaluation systems by introducing an enhanced Graph Neural Network (GNN) with Transformer-based attention mechanisms. The proposed model dynamically analyses multimodal educational data, including student feedback (textual and numerical), instructor self-assessments, classroom interaction metrics, and longitudinal academic performance records. By incorporating relational inductive biases through graph structures, the network effectively captures complex teacher-student-course interdependencies that traditional CNNs/RNNs overlook. The model is trained on a curated dataset of $\mathbf{1 5, 0 0 0 +}$ evaluation records from Chinese language programs across 12 universities, augmented with natural language processing (NLP)-processed open-ended comments. Experimental results demonstrate a 99.0% accuracy (5.8% improvement over baseline CNN models) with 98.7 % precision and 98.9 % recall in identifying underperforming courses. Key innovations include, temporal graph attention layers to track teaching quality evolution. Cross-modal feature fusion for holistic assessment. Explainability modules generating actionable improvement insights. This system enables real-time quality monitoring and personalized instructor feedback, significantly advancing beyond static evaluation frameworks. The methodology offers scalable applications for other language education systems seeking data-driven quality optimization.