Quantifying Bias in Hierarchical Category Systems.

Area of Science:

• Cognitive Science
• Information Science
• Sociology

Background:

• Categorization is fundamental to human cognition and societal structures.
• Category systems are not objective and can perpetuate harmful biases.
• Hierarchical category systems, common in libraries, require methods for bias assessment.

Purpose of the Study:

• To propose and demonstrate methods for quantifying biases in hierarchical category systems.
• To analyze biases in library classification systems, specifically focusing on Western concepts and male authors.
• To compare bias levels between the Dewey Decimal Classification and the Library of Congress Classification.

Main Methods:

• Development of novel methods to measure biases within hierarchical category structures.
• Analysis of a large-scale library dataset comprising over 3 million books.
• Quantitative assessment of Western bias in religion, literature, and history categories, and gender distribution of authors.

Main Results:

• Library classification systems exhibit significant Western bias, particularly in religion categories.
• Books by male authors are more broadly distributed across categories than those by female authors.
• The Dewey Decimal Classification demonstrates a higher degree of bias compared to the Library of Congress Classification.

Conclusions:

• The proposed methods effectively quantify biases in hierarchical category systems.
• Library classification systems reflect and can amplify societal biases related to culture and gender.
• The developed methodology is applicable to diverse natural and institutional category systems beyond libraries.