From feedback to Insight: Uncovering gender bias in STEM evaluations through NLP

Gender bias in faculty evaluations threatens equity in higher education. Our study uses Natural Language Processing (NLP) to analyze 70,000 student evaluations from a Latin American STEM university, revealing statistically significant disparities in how students assess male and female professors and exposing implicit bias in qualitative feedback. To achieve this, we developed a framework that combines sentiment analysis, thematic extraction, and stratified sampling to control for demographic differences. Results show significant differences in language and sentiment between male and female faculty evaluations, with additional variation by contract type and department, highlighting the complex interaction between structural and perceptual factors in student feedback. While NLP offers powerful tools for analyzing large datasets, it may overlook subtle contextual details; therefore, we incorporate cross-disciplinary methods to enhance bias detection. Ultimately, our study provides data-driven insights to inform the design of more equitable faculty evaluation systems.