Entity recognition in the biomedical domain using a hybrid approach.

Area of Science:

• Biomedical Informatics
• Natural Language Processing
• Computational Biology

Background:

• Accurate extraction of biomedical entities from scientific literature is crucial for knowledge discovery.
• Existing methods often face challenges in balancing recall and precision.
• Developing robust systems for biomedical entity recognition remains an active research area.

Purpose of the Study:

• To describe a high-recall, high-precision approach for biomedical entity extraction.
• To evaluate the performance of a two-stage pipeline combining dictionary-based recognition and machine learning.
• To compare Conditional Random Fields and Neural Networks for entity filtering.

Main Methods:

• A two-stage pipeline integrating the OGER entity recognizer (high recall) with the Distiller framework.
• OGER annotates terms from domain ontologies, serving as features for a machine learning classifier.
• Comparison of Conditional Random Fields and Neural Networks as machine learning classifiers for entity selection.

Main Results:

• The best performing system, combining dictionary-based candidate generation with Neural-Network-based filtering, achieved 86% precision at 60% recall for named entity recognition.
• For concept recognition, the system achieved 51% precision at 49% recall on the CRAFT corpus.
• Evaluation included recognition of chemicals, cell types, cellular components, biological processes, molecular functions, organisms, proteins, and biological sequences.

Conclusions:

• The developed approach achieves state-of-the-art performance for biomedical entity extraction on the CRAFT corpus.
• The combination of dictionary-based methods and neural networks offers a powerful strategy for this task.
• This work contributes to advancing automated information extraction from biomedical texts.