Subject Knowledge Entity Relationship Extraction Based on Multi-feature Fusion and Relation Specific Horns Tagging

As the big data era progresses and academic resources grow rapidly, challenges like information overload and difficulty in identifying specific knowledge points have become increasingly apparent, especially in the context of C++ academic knowledge. To tackle these issues, numerous scholars have proposed entity relation extraction models based on deep learning techniques and have built knowledge graphs. Yet, most current methods exhibit limited capabilities in feature extraction and often struggle to handle complex academic data with overlapping entity types. Addressing the challenges of entity relation extraction and the pain points of student learning, this paper initially utilizes RoBERTa and BiLSTM (Bidirectional Long Short-Term Memory) to extract semantic features of academic knowledge, and then employs GAT to derive the syntactic structure features of academic knowledge. Subsequently, by integrating semantic and syntactic features through a gating mechanism, the model’s feature extraction capability is enhanced. Finally, the model’s ability to extract academic data with overlapping entity types is improved through a specific relation tagging strategy. Through these components, a subject-specific entity relation extraction model, RGBRel (Roberta-GAT-BiLSTM-Relation), has been developed. Experiments conducted on the first C++ academic knowledge dataset, which incorporates knowledge from QT (a C++ application development framework), have demonstrated the model’s robust capability in feature extraction and in handling complex data within specific academic fields.