Joint segmentation and named entity recognition using dual decomposition in Chinese discharge summaries

Area of Science:

• Natural Language Processing
• Computational Linguistics
• Medical Informatics

Background:

• Accurate word segmentation and named entity recognition are crucial for processing Chinese clinical text.
• Existing methods often treat these tasks independently, potentially missing valuable interdependencies.

Purpose of the Study:

• To develop and evaluate a joint model for Chinese word segmentation and named entity recognition.
• To annotate a standard corpus of Chinese discharge summaries for these tasks.
• To improve the efficiency and effectiveness of these NLP tasks in the medical domain.

Main Methods:

• Developed two independent conditional random field (CRF) based systems for word segmentation and named entity recognition.
• Proposed a joint model using dual decomposition to exploit correlations between the two tasks.
• Designed specific features to highlight the advantages of the joint model over independent and incremental approaches.

Main Results:

• A gold standard corpus was created from 336 Chinese discharge summaries (71,355 words).
• The joint model achieved a 0.2% improvement in segmentation and a 1% improvement in recognition compared to individual task models.
• Performance was evaluated using micro-averaged precision, recall, and F-measure.

Conclusions:

• The proposed joint model is effective and efficient for both word segmentation and named entity recognition tasks.
• The results demonstrate the feasibility and advantages of a joint approach for these correlated NLP tasks in Chinese medical text.