Machine Learning for Company Review Sentiment Analysis Interpretation

Employee satisfaction is key for a productive, pleasant, and efficient atmosphere, customer satisfaction, and generally a successful business. A standard way to check and improve employee satisfaction is by seeking feedback. In big companies, this feedback might be cumbersome to analyze due to the numbers and differences between workers. Thus, this work applies Natural language processing in combination with various machine learning algorithms to a publicly available dataset containing feedback given by employees. Additionally, the best models are evaluated by SHAP analysis, allowing for a deeper understanding of the presented issue, and the process of sorting the feedback. Results suggest the best approach for the particular task was using the multilayer perceptrons, as these models yielded the best results. The attained outcomes suggest that an accuracy of 0.967407 is attained by the best-performing model. Interpretation revealed that the best-suited models emphasized the keywords “benefits, management, opportunities” and, as expected, “good, great, smart, work.” These words are most closely linked with positive assessments of the work environment.