Improving precision in concept normalization.

Area of Science:

• Computational linguistics
• Bioinformatics
• Natural Language Processing

Background:

• Natural language processing (NLP) applications often face a precision-recall trade-off.
• Certain applications require prioritizing high precision over recall.
• Existing NLP methods may not sufficiently address the need for enhanced precision in specific contexts.

Purpose of the Study:

• To develop and evaluate a strategy for increasing precision in NLP applications.
• To explore the effectiveness of combining pre-processing and post-processing methods.
• To apply a hybrid rationalist/empiricist approach to biomedical concept normalization.

Main Methods:

• Utilized the Zipfian distribution of false positive results to guide precision enhancement.
• Implemented a combination of knowledge-based and frequentist language modeling techniques.
• Applied the strategy to a high-performance biomedical concept recognition pipeline using a manually annotated corpus.
• Evaluated the approach across eight different ontologies for concept normalization.

Main Results:

• The developed strategy demonstrated success in increasing precision for NLP tasks.
• The effectiveness of specific pre-processing and post-processing methods varied significantly across different ontologies.
• A hybrid rationalist/empiricist approach proved beneficial for concept normalization.

Conclusions:

• A novel strategy effectively enhances precision in natural language processing by exploiting the Zipfian distribution.
• The hybrid approach, integrating diverse linguistic modeling techniques, shows promise for improving biomedical concept normalization.
• The variability in performance across ontologies highlights the need for tailored NLP strategies.