Building an Intelligent Annotation and Personalized Learning Platform for Ancient Chinese Poetry by Integrating ERNIE—Stu and Transformer—XL

Ancient Chinese texts contain important records of cultural heritage, yet existing learning systems still exhibit insufficient semantic coherence, limited word-sense disambiguation, and weak personalized recommendation capability. This paper proposes an intelligent annotation and personalized learning platform for ancient poetry by integrating ERNIE-Stu and Transformer-XL. ERNIE-Stu is used for word embedding, entity recognition, and knowledgegraph-enhanced word-sense disambiguation, while Transformer-XL captures long-range semantic dependencies through segment-level recurrence. Based on syntactic dependency analysis, the platform generates hierarchical context -aware annotations and constructs learner representations from interaction features. Semantic similarity and a timedecay mechanism are further combined to support adaptive recommendation. The semantic fusion framework also provides methodological reference for heterogeneous information understanding and intelligent interpretation in electromagnetic information processing and knowledge-driven communication systems. Experimental results show F1- scores of 0.91 for content-word annotation and 0.87 for function-word annotation, a BERTScore-F1 of 0.798 for longtext discourse similarity, NDCG@5 above 0.70, and CTR above 0.45, confirming improvements in annotation completeness, learning adaptability, and digital cultural-knowledge dissemination.