Maintenance and Optimization of Teaching Resource Library for Knowledge Graph Transfer and Update Strategies Based on Transfer Learning

As a core component of education informatization, the level of intelligence of teaching resource base directly affects the quality and efficiency of education and teaching. This paper introduces the foundation of knowledge graph construction, including the annotation of teaching resources, entity recognition and relationship extraction. Subsequently, the selection and pre-training process of the transfer learning model is described, especially the application of Bidirectional Encoder Representations from Transformers (BERT) model and the domain adaptation strategy. In addition, this paper proposes a knowledge graph migration strategy, including data preprocessing, model fine-tuning and knowledge fusion techniques. The updating mechanism of the knowledge graph is realized through incremental learning and automated updating process, which ensures the timeliness and dynamics of the teaching resource base. The experimental results show that the transfer learning strategy significantly improves the coverage and accuracy of the knowledge graph while maintaining an efficient update capability. Specifically, the coverage of the knowledge graph reaches 81.4%–99.9% and the update accuracy is as high as 98.1%, while the update time is as low as 310 ms. Compared with traditional methods, the transfer learning strategy demonstrates significant advantages in terms of training time, data requirement, model accuracy, and computational resource consumption.