Improving deep learning method for biomedical named entity recognition by using entity definition information.

Area of Science:

• Biomedical informatics
• Natural Language Processing
• Computational Biology

Background:

• Biomedical named entity recognition (NER) is crucial for extracting information from biomedical texts.
• Existing state-of-the-art methods often overlook the semantic meaning of different entity types.
• The PharmaCoNER challenge focused on recognizing pharmacological substances, compounds, and proteins in Spanish biomedical literature.

Purpose of the Study:

• To investigate methods for integrating entity type meaning into deep learning models for biomedical NER.
• To apply these enhanced models to the PharmaCoNER 2019 challenge.
• To evaluate the impact of using entity definition information on NER performance.

Main Methods:

• Two deep learning approaches were explored: SQuad-style machine reading comprehension (MRC) and Span-level one-pass (SOne) methods.
• Entity definition information was utilized as queries in MRC and to represent entity type meaning in SOne.
• Models were trained and evaluated on the PharmaCoNER 2019 corpus using strict micro-averaged precision, recall, and F1-score.

Main Results:

• The integration of entity definition information improved both SQuad-style MRC and SOne methods, yielding approximately a 0.003 increase in micro-averaged F1-score.
• The SQuad-style MRC model, using entity definitions as queries, achieved the highest performance with a micro-averaged F1-score of 0.9137.
• This model outperformed the best-performing system in the PharmaCoNER 2019 challenge and achieved a 1% F1-score improvement over state-of-the-art models without manually-crafted features.

Conclusions:

• Entity definition information positively impacts biomedical NER detection, enhancing deep learning model performance.
• The developed models achieved state-of-the-art micro-average F1 scores, demonstrating the utility of semantic information.
• Future work will explore incorporating entity definition information from knowledge graphs for further improvements.