Features extraction from employee reviews mining and the content-based job satisfaction prediction

In this thesis, the employees' job satisfaction was predicting by detecting the latent dimensions that hide in the employees’ reviews. The dataset of employees’ job satisfaction reviews was scrapped from the public website Glassdoor. The topic model Latent Dirichlet Allocation (LDA) was utilized to extract the informative hidden dimensions in the reviews. The topic probability obtained from LDA was taken as the predictor features (variables), which are used for predicting job satisfaction ratings through regression models. The results show that these LDA features can significantly improve the accuracy of prediction. In addition, it also provides information about what topics significantly influence job satisfaction. Based on that, some meaningful suggestions can be offered to these companies on how to increase their employees' job satisfaction, so as to decrease employee turnover.