Parenthetically speaking: classifying the contents of parentheses for text mining

Area of Science:

• Biomedical Informatics
• Natural Language Processing
• Computational Linguistics

Background:

• Parenthesized text in biomedical literature contains valuable information for text mining.
• Accurate classification of parenthesized content is crucial for leveraging this data.
• Existing methods may lack the specificity required for diverse parenthetical information.

Purpose of the Study:

• To develop and evaluate an automated system for classifying parenthesized text into 20 distinct categories.
• To improve the utility of biomedical text mining by enabling better understanding of parenthetical content.
• To provide a robust tool for researchers working with large biomedical corpora.

Main Methods:

• Development of a classification system for parenthesized biomedical text.
• Evaluation using an annotated corpus with 20 predefined categories.
• Implementation available as a Java class and a Perl module.

Main Results:

• The system achieved a micro-averaged accuracy of 68%.
• A macro-averaged accuracy of 60% was obtained on the annotated corpus.
• Demonstrated feasibility of automated classification for diverse parenthetical content.

Conclusions:

• Automated classification of parenthesized text is achievable and beneficial for biomedical text mining.
• The developed system offers a practical solution for categorizing complex parenthetical information.
• The tool's availability in Java and Perl enhances its accessibility for researchers.