Entity recognition in Chinese clinical text using attention-based CNN-LSTM-CRF

Area of Science:

• Natural Language Processing
• Computational Linguistics
• Biomedical Informatics

Background:

• Clinical entity recognition is crucial for processing clinical text.
• Most research focuses on English, with limited work on Chinese clinical text.
• Developing effective methods for Chinese clinical entity recognition is an emerging area.

Purpose of the Study:

• To propose a novel deep neural network for Chinese clinical entity recognition.
• To enhance the performance of existing models by incorporating CNN and attention mechanisms.

Main Methods:

• Proposed an attention-based CNN-LSTM-CRF model, extending LSTM-CRF.
• Integrated a Convolutional Neural Network (CNN) layer to capture local word context.
• Incorporated an attention layer to select relevant words within a sentence.

Main Results:

• The proposed attention-based CNN-LSTM-CRF model outperformed CRF and LSTM-CRF on two benchmark datasets.
• The model demonstrated effectiveness in recognizing both contiguous and discontiguous clinical entities.
• Both CNN and attention mechanisms individually improved performance, with attention having a greater impact.

Conclusions:

• The attention-based CNN-LSTM-CRF model is effective for Chinese clinical entity recognition.
• Attention mechanisms provide a greater performance boost than CNNs in this context.
• The proposed model advances the state-of-the-art in non-English clinical text processing.