Few-Shot Named Entity Recognition Based on Self-Descriptive Network and Knowledge Graph Enhancer

Named Entity Recognition (NER) is a critical task in Natural Language Processing (NLP), focusing on extracting entities from text and classifying them into predefined categories. However, traditional NER methods are limited by their reliance on large, annotated datasets, making it challenging in scenarios with limited data. Few-shot NER has emerged as a promising approach to address these challenges, but existing methods still face limitations in handling unseen entity types due to knowledge mismatch and information insufficiency. To address these gaps, this paper proposes a novel few-shot NER model that integrates self-descriptive mechanisms with knowledge graph enhancement. The self-descriptive mechanism provides consistent descriptions of entity types across schemas, while the knowledge graph dynamically incorporates domain-specific knowledge to improve the model's understanding of unseen or low-resource entities. The proposed model was evaluated on several benchmark datasets, demonstrating significant improvements in entity recognition performance across various domains. The model achieved a notable F1 score of 74.7 on the Movie2 dataset, outperforming existing models such as SDNet and TaGuProto. Ablation studies further confirmed the effectiveness of knowledge integration in enhancing the model's robustness and adaptability in few-shot scenarios. The research contributes to the field by offering a robust solution to the knowledge mismatch problem in few-shot NER, and by providing a model that is both adaptable and generalizable across diverse domains. The integration of knowledge graphs with self-descriptive mechanisms represents a significant advancement in NER, especially in low-resource settings.