Applying Machine Learning Tools on Web Vacancies for Labour Market and Skill Analysis – Economic and Policy Implications of AI, Blog Post #5


The following is a summary of Applying Machine Learning Tools on Web Vacancies for Labour Market and Skill Analysis by Emilio Colombo, Fabio Mercorio, and Mario Mezzanzanica. This paper was presented at the Technology Policy Institute Conference on The Economics and Policy Implications of Artificial Intelligence, February 22, 2018.


Advances in artificial intelligence have brought concerns about what people are generally calling “the future of work.” That is, some worry that even if AI and automation increase efficiency and productivity, they may also change labor markets more quickly than people can adapt.


It is easy to envision such scenarios: long haul truck drivers may be pushed out by autonomous vehicles; artificial intelligence may substitute for humans on manufacturing assembly lines; algorithms may make quicker work of tax filings or customer service requests. But we do not have good methods for predicting what changes in labor markets are most likely.


In a unique approach, Emilio Colombo, Fabio Mercorio, and Mario Mezzanzanica use machine learning algorithms in their paper, Applying Machine Learning Tools on Web Vacancies for Labour Market and Skill Analysis, to identify how demand for skills is changing, and how new jobs are being described. In theory, determining how in-demand skills are changing can give us a starting point for the education and training of a digital-age workforce.


The authors focus their analysis on a dataset comprised of online job postings from Italy. They employ machine learning text analysis techniques to identify cross-compare, and categorize job types, job titles, and required skills. Skills are split into two groups, hard and soft. Hard skills are largely job-specific and can range from knowledge of particular equipment to coding prowess. In contrast, soft skills are more universal, generally referring to one’s ability to communicate with others. The authors attempt to determine which skills are being sought in which positions and thus map labor force changes over time.


Results suggest that both soft skills and hard skills, like programming or data analytics are negatively related to automatability. The negative correlation between soft skills and automatability is unsurprising. We would expect soft skills that facilitate interactions, like emotional intelligence or problem solving, to be more context-dependent and thus difficult to automate. The negative relationship between the degree of digital skills and automatability follows similar logic. Jobs that require high levels of digital skill can still be context-dependent. Data scientists or computer programmers may write and rewrite code to fit specific circumstances, even if they are not using soft skills to interact with a machine. So, just as jobs that require less digital skill – perhaps those that require instead soft skills, such as a caregiver or a mediator – are less likely to be automated, so too are those jobs that require a high degree of digital skill.


The results suggest that the jobs most at risk of automation require basic digital skills, like word processing, spreadsheet management, or social media use, and few to no soft skills, like administrative positions. In addition to highlighting the skills necessary for current labor market openings in Italy, the authors identified a new group of occupations – including a whole host of analyst positions – that have recently emerged that require both hard and soft skills.


Research of this kind can serve as a starting point for future-of-work conversations. Instead of speculating about potential effects of automation on different industries, a similar analysis on new data might identify points of intervention in the education and training of a modern workforce.