Measuring pro-social message in job postings using machine learning

Abstract

When searching for jobs, job applicants are not only motivated by monetary compensation alone, the meaning and social effects of the work also matter. Pro-social motivation, the desire to have a positive impact on other people or social collectives also play an important role in job searching. On the other hand, organizations also have many incentives to promote pro-social jobs during the recruiting processes and accordingly design pro-social characteristics in job postings. Using latest machine learning techniques, we could possibly quantify pro-social characteristics in massive amount of job postings and potentially predict pro-social messages advertised in online job postings. In this thesis, we take up the challenge of developing novel measures of pro-social that satisfactorily address the problems identified with existing measures of pro-social. We proposed implementations of two different machine learning approaches to quantitatively measure pro-social messages from over five million online job postings documentation and effectively predict pro-social jobs, with 79% and 94% prediction accuracy yield from methodology I and methodology II respectively. Based on those approaches, we evaluate the model performance and measure correlation of industries' use of pro-social messages in job postings to compare the effectiveness of two models on several metrics.