Education Technology & AI

The rapid evolution of professional domains necessitates timely updates to English for Specific Purposes (ESP) curricula, yet traditional manual syllabus development struggles to keep pace with industry demands. This study presents a novel framework integrating Graph Neural Networks (GNN), Enhanced PageRank, and community detection algorithms to automate professional terminology analysis and ESP syllabus generation. We constructed terminology co-occurrence networks from domain-specific corpora spanning Computer Science, Finance, and Art (8 million tokens, 12,000 terms), employing GNN-based multi-task learning to generate semantic embeddings through joint domain classification and link prediction. An Enhanced PageRank algorithm incorporating embedding similarity, domain specificity, and teaching suitability scores identifies pedagogically relevant terms, while Louvain community detection partitions networks into coherent thematic modules labeled via LDA topic modeling. Experimental results demonstrate substantial improvements over baseline methods: Industry Demand Cover-age increased from 52.1% to 76.2% (+46%), and expert evaluation scores rose from 3.1 to 4.2 out of 5.0 (+35%). The automated system reduces syllabus generation time from two weeks to ten minutes, a 2,000-fold efficiency gain. Ablation studies confirm that graph structure modeling outperforms isolated frequency statistics by 10.3% in classification accuracy, validating the necessity of rela-tional learning for terminology analysis. This work pioneers the application of graph neural networks to ESP pedagogy, offering a scalable, data-driven solution for curriculum development across diverse professional domains and establishing a foundation for adaptive, industry-aligned language education.