Toward a Natural Language Processing-based Model for Workplace Culture Scoring

Abstract

Workplace culture plays a key role in a firm’s success, impacting employee engagement, productivity, and overall performance. To this end, effectively — and in particular quantitatively — measuring culture provides several advantages to firms and those studying the behavior of firms. In this project, I propose a natural language processing (NLP) framework to generate culture scores for individual employee reviews, leveraging text data from Glassdoor. I combine topic modeling to identify interpretable cultural themes with self-supervised learning to generate and predict review-level scores from sentiment and topic relevance measures. The scores show moderate predictive accuracy against user-entered Glassdoor Culture & Values ratings and moderate correlation with external culture scores from Revelio Labs. I uncover some heterogeneity in model performance across countries, suggesting that employee perceptions of culture may differ across geographic contexts.