Computational Modeling of Stereotype Content in Text.

Area of Science:

• Computational Social Science
• Social Psychology
• Natural Language Processing

Background:

• Stereotypes are pervasive in daily communication, media, and online platforms.
• Understanding how stereotypes are expressed computationally is crucial for social science research.
• Existing methods may not adequately capture the nuances of stereotype expression in large-scale text data.

Purpose of the Study:

• To develop and validate a computational method for identifying and analyzing stereotype content in text.
• To map perceived group attributes onto the dimensions of warmth and competence.
• To demonstrate the model's utility through real-world case studies involving social groups.

Main Methods:

• Utilized a computational approach to mine large text corpora for sentences expressing perceptions of social groups.
• Mapped identified sentences onto a two-dimensional plane representing perceived warmth and competence.
• Validated the framework against expert annotations and crowd-sourced stereotype data; applied to Twitter data for case studies.

Main Results:

• The computational model successfully identified and mapped stereotype content from diverse text sources.
• Case studies revealed distinct stereotype patterns for subgroups (e.g., Black women vs. white women) and variations based on group labels (e.g., 'old people' vs. 'senior citizens').
• Demonstrated that language used to describe groups is associated with differential stereotype content.

Conclusions:

• The proposed computational framework offers a robust method for studying stereotype expression in large text datasets.
• The model provides insights into the dimensionality and variability of social group perceptions.
• This approach can be adopted by researchers to investigate stereotype dynamics across various textual corpora.