Robust relational parsing over biomedical literature: extracting inhibit relations

Area of Science:

• Biomedical Informatics
• Computational Linguistics
• Natural Language Processing

Background:

• Extracting biomolecular relations from biomedical literature is crucial for understanding biological processes.
• Existing methods face challenges in accurately identifying complex relational information.
• Developing robust parsers is essential for advancing automated information extraction.

Purpose of the Study:

• To design and evaluate a robust parser for identifying and extracting biomolecular relations from biomedical literature.
• To improve the performance of relation extraction through specialized automata and anaphora resolution.
• To demonstrate the system's effectiveness on inhibition relations using a gold standard corpus.

Main Methods:

• Developed separate automata for nominal-based and verbal-based relational information extraction.
• Optimized grammars independently for each module to enhance performance.
• Implemented text-based anaphora resolution to improve argument binding in relation extraction.
• Evaluated the system on inhibition relations against an annotated gold standard corpus.

Main Results:

• Achieved a precision of 90% for inhibition relation extraction.
• Obtained a recall of 57% and a partial recall of 22%.
• Demonstrated significant improvement over previously published results on Medline data.

Conclusions:

• The developed parser offers a robust and effective method for extracting biomolecular relations.
• A corpus-based linguistic approach, combined with specialized automata and anaphora resolution, significantly enhances information extraction performance.
• The system's high precision and recall indicate its potential for advancing biomedical knowledge discovery.