Generating quality word sense disambiguation test sets based on MeSH indexing

Area of Science:

• Biomedical Natural Language Processing
• Computational Linguistics
• Information Retrieval

Background:

• Accurate interpretation of biomedical text requires correct identification of word meanings.
• Word sense disambiguation (WSD) is crucial for biomedical natural language processing.
• Quality evaluation sets are essential for developing and comparing WSD programs.

Purpose of the Study:

• To develop an automatic method for creating quality test sets for biomedical WSD.
• To preferentially annotate terms that are topical to the text.

Main Methods:

• A MeSH-based automatic sense-tagging method was developed.
• The method annotates terms based on their topicality within the text.
• Cross-validation with multiple annotators was suggested for quality assurance.

Main Results:

• Preliminary results demonstrated the promise of the developed method.
• The study identified key issues for future biomedical WSD research.
• The method showed potential for efficient generation of WSD test sets.

Conclusions:

• The developed MeSH-based method can efficiently generate quality WSD test sets.
• Cross-validation with annotators enhances the reliability of the generated test sets.
• This work contributes to advancing biomedical WSD research and evaluation.