Words matter: AI can predict salaries based on the text of online job postings

The career landscape in the United States is significantly shifting: The COVID-19 pandemic has redefined important function and moved employees out of the business office. New technologies are transforming the character of numerous occupations. Globalization continues to force careers to new locations. And weather improve considerations are adding careers in the alternate electrical power sector when cutting them from the fossil fuel business. 

Amid this office turmoil, personnel, as effectively as employers and policymakers, could benefit from knowing which job properties lead to larger wages and mobility, states Sarah Bana, a postdoctoral fellow at Stanford’s Electronic Economic system Lab, portion of the Stanford Institute for Human-Centered Synthetic Intelligence. And, she notes, there now exists a large dataset that could possibly assist offer that being familiar with: the text of millions of on-line career postings. 

“Online facts offers us with a huge opportunity to evaluate what issues,” she suggests.

Certainly, using synthetic intelligence (AI) and device studying, Bana not too long ago confirmed that the text used in a dataset of extra than a single million on the web job postings make clear 87% of the variation in salaries throughout a huge proportion of the labor current market. It’s the very first perform to use this sort of a big dataset of postings and to glimpse at the romantic relationship involving postings and salaries. 

Bana also experimented with injecting new text – adding a skill certification, for instance – into applicable position listings to see how these terms changed the salary prediction.

“It turns out that we can use the textual content of work listings to assess the income-relevant qualities of careers in near-to actual time,” Bana suggests. “This details could make making use of for careers far more transparent and improve our method to workforce training and education.”

An AI dataset of 1 million position postings 

To evaluate how the textual content of on line occupation postings relates to salaries, Bana obtained a lot more than one million pre-pandemic career postings from Greenwich.HR, which aggregates hundreds of thousands of task postings from on-line work board platforms. 

She then used BERT, one particular of the most superior all-natural language processing (NLP) designs accessible, to train an NLP design using the text of additional than 800,000 of the career postings and their linked wage info. When she analyzed the product using the remaining 200,000 job listings, it precisely predicted the linked salaries 87% of the time. By comparison, using only the task postings’ job titles and geographic destinations yielded precise predictions just 69% of the time.

In adhere to-up do the job, Bana will attempt to characterize the contribution of various words and phrases to the salary prediction. “Ideally, we will color terms inside of postings from pink to eco-friendly, where the darker purple phrases are connected with decrease wage and the darker eco-friendly are connected with greater wage,” she states. 

The value of upskilling: A text-injection experiment

To determine which expertise matter for income prediction, Bana made use of a textual content-injection solution: To certain pertinent work postings, she included limited phrases indicating the occupation calls for a distinct vocation certification, these kinds of as all those listed in In fact.com’s 10 In-Need Occupation Certifications (And How To Achieve Them). Getting these certifications can be high priced, with price ranges ranging from about $225 to about $2,000. But, right until now, there has been no way to figure out regardless of whether the expenditure is worthwhile from a income stage of check out. 

Bana’s experiment discovered that some certifications (these types of as the IIBA Agile Evaluation Certification) deliver significant wage gains promptly though other individuals (these types of as the Cisco Licensed Internetwork Expert) do so much more little by little – beneficial information for workers who would like to have greater details about how an investment in skills training will have an effect on their salaries and prospective buyers, Bana suggests.

Staff aren’t the only ones to reward from this details, Bana notes. Businesses can use these results to better devote in human capital, she says. If, for illustration, machine understanding styles reveal a gradual shift away from some jobs and toward many others, employers would have advance warning and could retrain specified workers.

And policymakers thinking about what task instruction courses to market would similarly advantage from comprehension which competencies are waxing or waning in financial value.

To that finish, Bana and her colleagues are presently working on a companion paper that identifies what tasks are disappearing from occupation listings in excess of time and what new jobs are showing up. 

In the future, Bana hopes that textual examination of position postings could produce a net-based mostly software where staff or businesses could investigation the price included by upskilling or by transferring to a new geographic site. 

“Currently there’s not a lot of clarity around a path to larger earnings,” Bana says. “Tools like these could enable occupation seekers enhance their occupation prospective clients, businesses create their workforces, and policymakers respond to rapid changes in the economic climate.”

Katharine Miller is a contributing author for the Stanford Institute for Human-Centered AI.

This story originally appeared on Hai.stanford.edu. Copyright 2022