Knowledge-based biomedical word sense disambiguation: an evaluation and application to clinical document...

Area of Science:

• Natural Language Processing
• Biomedical Informatics
• Computational Linguistics

Background:

• Word Sense Disambiguation (WSD) is crucial for text processing, enabling unambiguous concept assignment to terms.
• Existing WSD methods require evaluation for their contribution to specific tasks like clinical text classification.

Purpose of the Study:

• To develop and evaluate a knowledge-based WSD system utilizing semantic similarity measures from the Unified Medical Language System (UMLS).
• To assess the impact of this WSD system on improving clinical document classification performance.

Main Methods:

• Developed a knowledge-based WSD system employing UMLS semantic similarity measures.
• Evaluated the WSD system on standard biomedical WSD datasets.
• Assessed the system's contribution to clinical document classification using the 2007 Computational Medicine Challenge corpus.

Main Results:

• The developed WSD system demonstrated competitive performance compared to other knowledge-based approaches.
• Machine learning classifiers trained on disambiguated terms significantly outperformed those using all terms.

Conclusions:

• A novel WSD system achieving high accuracy on biomedical datasets was successfully developed.
• The WSD system demonstrably improves the performance of clinical document classification tasks.