Application of Interactive Data Visualization Design Based on BERT Sentiment Analysis in Patriotism Education

Complex cultural metaphors in patriotism education texts make accurate sentiment analysis difficult, while existing visualization tools provide limited interactivity for instructional support. To address these limitations, this study develops an interactive data visualization framework driven by a domain-enhanced BERT model for high-metaphor and highcompliance educational scenarios. Domain-adaptive pre-training and Fast Gradient Method adversarial training are introduced to strengthen semantic representation learning, enabling the model to identify implicit rhetorical expressions rather than relying on surface lexical features. The optimized DE-BERT model recognizes domain-specific metaphors, euphemisms, and latent value orientations, providing interpretable semantic evidence for educational analysis. Guided by cognitive load theory, a three-level visualization architecture integrating smoothed Bézier sentiment curves, dynamic heatmap arrays, and repulsive bubble labels is constructed to improve information presentation and user interaction. The framework also offers methodological value for semantic perception and intelligent visualization in electromagnetic information processing and human– machine collaborative analysis. Experiments show a classification accuracy of 0.90 and macro F1-score of 0.88 under high-metaphor-density conditions, with response times of 783.4 ms, 186.3 ms, and 874.6 ms across teaching scenarios and a single-operation success rate above 97%.