Multi-task learning for Chinese clinical named entity recognition with external knowledge.

Area of Science:

• Natural Language Processing
• Medical Informatics
• Artificial Intelligence

Background:

• Named Entity Recognition (NER) on Chinese electronic medical records is crucial for healthcare applications.
• Existing data-driven NER methods require extensive labeled data, which is costly and struggles with rare entities.

Purpose of the Study:

• To develop a novel multi-task deep neural network for Chinese medical NER.
• To address limitations of data-driven approaches, including data scarcity and handling of rare/unseen entities.

Main Methods:

• Incorporation of dictionary features into neural networks.
• Utilizing a secondary named entity segmentation task as an auxiliary task to enhance primary NER performance.
• Development of a multi-task deep neural network architecture.

Main Results:

• The proposed model achieved a 91.07% average f-measure on two public datasets.
• The model obtained an 87.05% f-measure on a private dataset.
• Experimental results demonstrated superior performance compared to other popular methods and traditional statistical models.

Conclusions:

• The novel multi-task deep neural network model is effective for Chinese medical NER.
• The approach successfully overcomes challenges associated with data-driven methods, particularly for rare entities.
• The model shows significant improvements over traditional statistical models in medical NER tasks.