Negation-based transfer learning for improving biomedical Named Entity Recognition and Relation Extraction.

Area of Science:

• Biomedical Natural Language Processing (NLP)
• Machine Learning
• Computational Biology

Background:

• Named Entity Recognition (NER) and Relation Extraction (RE) are crucial for biomedical NLP.
• Developing advanced automatic systems requires accurate detection of entities and their relationships.
• Incorporating negation information is vital for precise biomedical text analysis.

Purpose of the Study:

• To explore transfer learning for integrating negation detection into NER and RE systems.
• To analyze the impact of detecting negated entities on biomedical entity and relationship identification.
• To evaluate the effectiveness of these techniques in both Spanish and English.

Main Methods:

• Proposed three neural architectures based on Bidirectional Long Short-Term Memory (Bi-LSTM) and Conditional Random Fields (CRFs).
• Developed a general architecture for negation detection (triggers and scopes).
• Integrated negation detection weights into specific biomedical NER and joint NER+RE models.

Main Results:

• Achieved performance improvements in NER: ~3.5% F-Measure in English and >7% in Spanish.
• Enhanced NER+RE task performance: >13% F-Measure for NER submodel and ~2% for RE submodel.
• Demonstrated consistent improvements across tasks and languages when using transfer learning.

Conclusions:

• Negation-based transfer learning is suitable for biomedical NER and RE.
• Negation detection significantly improves the identification of biomedical entities and relationships.
• The explored techniques exhibit robustness and maintain consistent improvements.