Online biomedical named entities recognition by data and knowledge-driven model.

Area of Science:

• Biomedical Natural Language Processing
• Artificial Intelligence in Healthcare

Background:

• Named Entity Recognition (NER) is crucial for processing biomedical text.
• Existing NER models often struggle with unstandardized online medical text due to errors and variations.
• Research on NER for unstandardized biomedical text remains limited.

Purpose of the Study:

• To develop a neural network method for effective entity recognition in unstandardized online medical/health text.
• To enhance the capacity of transformer models for processing online biomedical data.
• To improve the performance of NER models on noisy and polymorphic biomedical text.

Main Methods:

• A novel neural network approach for recognizing entities in unstandardized online medical/health text.
• A new pre-training scheme utilizing large-scale online question-answering pairs to boost transformer model capabilities.
• Integration of multi-channel knowledge labels from a knowledge base for enhanced entity representation, overcoming language-specific segmentation challenges.

Main Results:

• The proposed neural network method significantly outperforms baseline methods in experiments.
• The model achieves state-of-the-art results on a Chinese online medical entity recognition dataset.
• The approach demonstrates effectiveness in handling errors and polymorphisms in online biomedical text.

Conclusions:

• The developed neural network method offers a robust solution for NER in unstandardized online medical text.
• The pre-training scheme and multi-channel knowledge labels enhance model performance and generalizability.
• This work advances the field of biomedical NLP, particularly for real-world, unstandardized data sources.