Employer segmentation based on employee reviews with machine learning = Munkáltatói szegmentáció munkavállalói értékelések alapján gépi tanulás segítségével

Absztrakt (kivonat)

In this paper, the question of how employers can be segmented based on employee satisfaction with the help of machine learning models is examined. The different methods on how people can be influenced based on reviews are briefly reviewed with a special focus on how these methods have changed with the appearance and spread of the internet. Afterwards, the job market is presented, highlighting the Hungarian’s specificities. The main data source for research is the Profession.hu, Hungary’s biggest job search portal, where employees can choose from more than 10.000 jobs. I have used the reviews for the biggest and frequently rated companies that are offering jobs in Hungary. Every company has evaluations based on different categories, such as wages and benefits, working hours, work-life balance and how monotonous the work is. First, the free text comments for the companies are examined with sentiment analysis models. Then descriptive statistics and correlation analysis is given for the measures evaluating the employers. For workplace segmentation, the Kohonen method is utilized, which implements self-organizing maps for unsupervised pattern recognition and dimension reduction. To refine the model, hierarchical clustering is applied on the coordinates of the reduced Kohonen map. Optimal number of clusters is selected based on a dendrogram.