Employee Analytics through Sentiment Analysis.

People discuss and talk about the most diverse topics in social media platforms, including about their jobs. This results in a stream of employee-related data, and organizations are increasingly interested in making sense of this data on an ongoing basis in order to assess key factors such as employee engagement, retention and satisfaction. In this paper we propose to estimate such factors from dierent sentiments that are implicit in employee communications in social media platforms. We introduce sentiment analysis approaches that are based on learning vector representations for employee communications in an unsupervised way, and then these representations are given as input to a state-ofthe-art supervised regression algorithm which nally maps text/sentiment to employee factors. We collected a large set of employee communications in social platforms, survey data such as work/life balance, job culture and management, and also ocial data about retention and salary. Then, we performed a systematic set of experiments using the collected data, and our results show that learning representations leads to better sentiment analysis performance than engineering features based on the standard term-frequency and inverse-document-frequency numbers (i.e., TF-IDF weighting scheme).