Dataset-aware multi-task learning approaches for biomedical named entity recognition.

Area of Science:

• Biomedical text mining
• Natural Language Processing
• Machine Learning

Background:

• Biomedical Named Entity Recognition (Bio-NER) is crucial for text mining.
• Deep learning models for Bio-NER are limited by data quantity and quality.
• Existing multi-task learning (MTL) methods for Bio-NER do not fully leverage diverse datasets.

Purpose of the Study:

• To propose novel dataset-aware MTL approaches for Bio-NER.
• To enhance Bio-NER performance by enabling models to learn from multiple datasets.
• To investigate the synergistic relationship between Bio-NER and biomedical Part-of-Speech (POS) tagging.

Main Methods:

• Developed two dataset-aware MTL frameworks for Bio-NER.
• Jointly trained models across numerous Bio-NER datasets.
• Incorporated Bio-NER and biomedical POS tagging datasets into the MTL framework.

Main Results:

• Achieved superior performance compared to state-of-the-art MTL methods on 14 out of 15 Bio-NER datasets.
• Demonstrated that incorporating POS tagging significantly enhances Bio-NER.
• Showcased the effectiveness of dataset-aware MTL in leveraging information from multiple sources.

Conclusions:

• Dataset-aware MTL approaches offer a significant improvement for Bio-NER.
• Joint training with related tasks like POS tagging boosts performance.
• The proposed methods effectively address data limitations in biomedical NLP.