Job postings allow us to get a glimpse of what goes on in the corporate engine room. Which formal skills and which other relevant competences do employers seek? And further, will these requirements improve an organization’s performance? A Danish research project is set to provide the answers through use of supercomputing.
A few years ago, Rentian Zhu, PhD fellow at the Department of Economics, Copenhagen Business School (CBS) got a bright idea. He wanted to use job postings as a source of societal information. Which qualifications do employers seek, how do the patterns change over time, and would the requirements translate into better performance of enterprises and organizations?
However, it soon transpired that moving from idea to realization would take hard work.
“The project was challenging due to the volume of unstructured text data spanning both Danish and English. Without structured tags or categories, these job postings required substantial data processing, far beyond what could be handled by a standard computer, to become useful,” says Rentian Zhu.
3 million postings
To address the complexities, Rentian Zhu approached DeiC, the national research and education network (NREN) of Denmark. The initiative DeiC Interactive HPC (high-performance computing) allows researchers to gain experience with supercomputing.
“I became aware of DeiC Interactive through a former colleague, a natural language processing enthusiast who highly recommended it.”
In an initial project, Rentian Zhu began analyzing 3 million job postings. He was granted 36,000 hours of supercomputing which he utilized to extract, classify, and quantify skill demands across industries and job roles.
In the process, he used the language model BERT to extract key features from the text and enhance skill recognition by using GPT-3.5 and GPT-4 with specialized prompt strategies tailored to labor market terminology. These strategies help the models understand nuanced skill requirements better.
How labor skills influence profits
In a next step, Rentian Zhu would create a hierarchical classification of skills by integrating outputs from BERT and GPT models using LangChain – a library that helps coordinate the flow of information between different AI models.
Further, the association between aggregated skill categories at the firm level and performance metrics such as revenue growth and profitability were analyzed applying advanced econometric techniques. By employing panel data models, he and his supervisors uncovered how categorized workforce skills are associated with organizational outcomes, shedding light on labor-driven factors that influence profitability.
To meet the heavy computational demands of his project, Rentian Zhu took a strategic approach, distributing tasks across GPU (graphic processing units) and CPU (central processing units), ensuring he made the most of DeiC Interactive HPC’s power.
Effects of flexible work arrangements
Building on his initial project, Rentian Zhu continues to broaden the scope of his research. In December 2024, he was awarded 26,000 core hours through a national DeiC call. This will enable him to combine his extensive data material with registry and time-use data, and to fine-tune the language models for more complex workflows.
The expanded setup will not only allow him to deepen his investigations into flexible work arrangements and their broader economic and social effects but also improve efficiency and reduce processing time compared to standard computing environments.
“With the additional national HPC resources, I can now venture into more detailed hypotheses, incorporate a richer variety of data, and apply more advanced modelling techniques. This will help me probe the mechanisms that shape work patterns and skill demands and hopefully lead to a more rigorous scientific understanding of Denmark’s evolving labor market.”
The text is inspired by the article “Assessing trends in the Danish labor market with DeiC Interactive HPC” by Anne Rahbek-Damm at the DeiC website.
Contributors
