Chinese Clinical Named Entity Recognition Using Residual Dilated Convolutional Neural Network With Conditional Random...

Area of Science:

• Natural Language Processing
• Machine Learning
• Biomedical Informatics

Background:

• Clinical Named Entity Recognition (CNER) is vital for clinical and translational research.
• Deep learning methods have advanced CNER, but recurrent neural networks (RNNs) are computationally intensive, leading to long training times.
• Existing methods often struggle with efficient processing of sequential clinical data.

Purpose of the Study:

• To develop a more computationally efficient deep learning model for Chinese CNER.
• To improve the speed of training for CNER tasks without sacrificing performance.
• To introduce a novel architecture that addresses the limitations of RNNs in CNER.

Main Methods:

• Proposed a Residual Dilated Convolutional Neural Network with Conditional Random Field (RD-CNN-CRF) model.
• Utilized dense vector representations for Chinese characters and dictionary features.
• Employed a residual dilated convolutional neural network to extract contextual features.
• Integrated a conditional random field (CRF) to model tag dependencies and optimize tag sequences.

Main Results:

• The RD-CNN-CRF model demonstrated significantly faster training times compared to RNN-based methods.
• Achieved competitive performance against state-of-the-art RNN-based methods on the CCKS-2017 Task 2 benchmark dataset.
• The model effectively captures contextual features and dependencies between neighboring tags.

Conclusions:

• The RD-CNN-CRF offers a promising alternative for Chinese CNER, balancing computational efficiency and accuracy.
• This asynchronous computation approach dramatically speeds up model training.
• The findings suggest potential for broader application in clinical and translational research requiring efficient CNER.