Predicting Employee Recommendations from Glassdoor Reviews: Ratings, Sentiment, and Firm-Level Variation
Analyzed a large Glassdoor job-review dataset to examine what predicts whether employees recommend their employer. Built a Python-based machine learning and statistical analysis pipeline using structured workplace ratings, review text, sentiment features, factor analysis, logistic regression, random forests, and model evaluation metrics. The project found that structured ratings such as senior management, culture and values, and career opportunities were strongly associated with employee recommendations, but also highly intercorrelated, suggesting a broader latent workplace-evaluation factor. Sentiment analysis of pros and cons text added interpretive value, while firm-level models showed variation in how negative review content related to recommendation outcomes across employers.
