Named entity recognition for Chinese based on global pointer and adversarial training.

Area of Science:

• Natural Language Processing (NLP)
• Machine Learning
• Artificial Intelligence

Background:

• Named Entity Recognition (NER) is crucial for information extraction and knowledge graph construction.
• Existing NER methods like Conditional Random Fields (CRFs) face challenges with large tag sets and nested entities.
• Pointer networks, while addressing some limitations, struggle with global entity context and can suffer from local instability.

Purpose of the Study:

• To propose a novel NER model that overcomes the limitations of existing methods.
• To enhance the model's ability to capture global entity information and improve positional awareness.
• To increase the robustness and generalization capabilities of NER models through adversarial training.

Main Methods:

• A novel NER model incorporating a global pointer mechanism for entity decoding.
• Integration of rotary relative position information to improve the model's understanding of entity positions.
• Application of adversarial training to mitigate local instability and enhance model generalization.

Main Results:

• The proposed model demonstrated improved F1 scores on multiple public datasets (OntoNotes5, MSRA, Resume, Weibo).
• Significant performance gains were observed compared to existing mainstream NER models.
• The global pointer and adversarial training effectively addressed limitations of previous approaches.

Conclusions:

• The global pointer and adversarial training approach offers a robust solution for Named Entity Recognition.
• This model effectively handles challenges like large tag sets and nested entities.
• The proposed method shows superior performance and generalization across various benchmark datasets.