A Data Driven Approach to Understanding the Drivers of Employee Satisfaction Across Sectors Through the Glassdoor

Abstract

In a highly competitive labor market, one of the biggest challenges for companies is attracting and retaining talent to ensure positive continued growth and company success. This thesis will focus on the analysis of a company’s internal assets: employee satisfaction. Using data collected from Glassdoor, a website where current and former employees rate and review companies and their management teams, we have access to approximately 2.6 million reviews spanning 5,833 unique companies. In predicting sentiment across the 1.2 million reviews of companies based in the United States, six unique machine learning algorithms were applied to predict positive and negative sentiment within the freeform PRO and CON reviews within the text. These classifying algorithms included Original Naive Bayes, Multinomial Naive Bayes, Multivariate Bernoulli Naive Bayes, Logistic Regressions, Linear Support Vector Classifiers, and Stochastic Gradient Descent. This sentiment prediction was then used to perform a bag-of-words analysis to find the words that were most indicative of sentiment, whether positive or negative. These words were then weighted and categorized to understand which categories of the employee experience were most important in creating a positive work experience, resulting in high employee ratings of employers, and which did the opposite. This work explores what these drivers looked like for the ten sectors that Glassdoor grouped companies by and presents a more focused analysis of the Technology and Healthcare sectors in specific, because they were the two sectors that had the most and least success with overall ratings of employers, respectively.