Knowledge map construction and intelligent reasoning model design for digital learning

This paper investigates the construction of a knowledge graph and the design of an intelligent reasoning model for digital learning environments, aiming to address information overload and the challenge of planning personalized learning paths amidst vast resources. Initially, the BERT-BiLSTM-CRF model is employed for entity recognition, combined with a Graph Attention Network (GAT) to extract relationships between entities, and the TransE model is utilized to transform knowledge into low-dimensional vector form, thus constructing the knowledge graph. Subsequently, a hybrid reasoning algorithm framework is proposed, integrating rule-based reasoning and statistical reasoning. This framework leverages Graph Neural Networks (GNN) to achieve joint embeddings of users and courses, and enhances the accuracy and reliability of the recommendation system through an optimized ranking loss function. The experimental results show that the proposed hybrid model is significantly better than the traditional method in AUC and Top5 accuracy of recommendation tasks, and also performs well in precision and F1 value of problem-solving tasks. In addition, the quality evaluation indicators of knowledge map, such as the accuracy of entity alignment and the F1 value of relationship conflict resolution, have reached a high level, providing a high-quality knowledge base for recommendation and question-and-answer tasks.