Automatic Prediction of Youth Unemployment via Social Media Data

MARIANO G. BEIRO; CIRO CATTUTO; KYRIAKI KALIMERI; ALESSANDRO ROSINA; ANDREA BONANOMI

Tesalónica

Conferencia; Conference on Complex Systems 2018; 2018

This study aims at improving the current understanding of the factors driving active job search, placing the focal point on the unemployed population, which often at risk of social marginalisation [3]. The failure to tap into the economic aspirations limits not only the income and skill development but also the likelihood of later employability. The main aim of the study is twofold; firstly, (i) to automatically identify the unemployed population inferring from their online digital traces and secondly, (ii) to uncover digital behaviours of the unemployed community easily accessible from online social platforms, which can indicate the most privileged communication channels for unemployment or educational advertising campaigns.Taking advantage of the popularity of the social platforms we created an ad-hoc Facebook-hosted application, whose major function is to administer psychometric questionnaires and quizzes. Upon acquiring participants? informed consent, we gathered information regarding their public Facebook profile and their ?Likes? on Facebook Pages. The application acts an innovative data-collection tool of rich nonverbal cues for behavioural understanding and profiling along with validated information from self-reported psychological assessments. The informationgathered included (i) demographic questions (i.e. gender, employment status, residence etc.), (ii) personality traits (iii) moral foundations, (iv) questions about the participants? life satisfaction level, and (v) work-life balance status.The goal is to assess how much of the information gleaned by questionnaires is behaviourally observable, which in turns relates to the scalability of the approach since behavioural observation at scale is usually simpler and more cost-effective than large-scale survey campaigns. We employed classification design based on random forest models with 5-fold cross-validation to automatically identify the employment status and the gender of the participants. We inferred these two attributes based only on the Likes on the Facebook pages, the category of the page, and the total activity of the participant (in terms of likes).According to our preliminary findings, the employment state of the participants and the gender is predicted with accuracy 78% and 96% respectively as reported in Table 1. Figure 1 provides a preview of the application. Furthermore, we provide useful insights for the socio-cultural attributes that characterise the two communities in terms of psychometric attributes and interests. For instance, the unemployed were found to be more interested in discount and coupon pages while the employed in political satyr ones. These pages emerged as top indicators of the unemployed status from the internal decisional process of the classification algorithm. Statistically significant differences between the two communities were also noted in their personality traits and moral values. Here, the unemployed found to be more fond of family bonds, tradition and social binding values in general with respect to the employed participants.