Biomedical named entity extraction: some issues of corpus compatibilities.

Area of Science:

• Biomedical Informatics
• Natural Language Processing
• Machine Learning

Background:

• Named Entity (NE) extraction is crucial for biomedical information extraction.
• Lack of consensus in NE annotation hinders system comparison and corpus integration.
• Existing systems face challenges due to corpus incompatibilities.

Purpose of the Study:

• To address corpus compatibility issues in biomedical NE extraction.
• To develop a robust NE extraction technique using classifier ensembles.
• To improve the performance and comparability of biomedical NE extraction systems.

Main Methods:

• Utilized a single objective optimization (SOO) based classifier ensemble technique.
• Employed a genetic algorithm (GA) for optimizing classifier combinations.
• Developed models using Conditional Random Field (CRF) and Support Vector Machine (SVM) frameworks.
• Extracted features without deep domain knowledge or external resources.

Main Results:

• The GA-based ensemble achieved approximately 2% performance improvement over individual classifiers.
• Experiments demonstrated the efficacy of the ensemble technique compared to existing approaches.
• Performance degradation on integrated corpora highlighted task difficulties.

Conclusions:

• The ensemble approach achieved state-of-the-art performance on three biomedical datasets.
• The use of diverse features and the GA-based ensemble technique contributed to improved performance.
• The method offers a promising solution for overcoming corpus incompatibilities in biomedical NE extraction.