In my presentation, I will first give an overview of the “IAB-Kompetenz-Kompass”. This project aims to provide data from job advertisements and develop evaluation routines for extracting key information categories for labor market research. I will then focus on our approach to extracting hard skill requirements. This can be understood as a high-dimensional, multi-label classification problem. Multi-label classification problems are typically solved in one of the following ways: ontology-based methods, supervised machine learning, or, more recently, unsupervised classifications based on large language models. For ontology-based methods, completeness poses a significant challenge. The main problem with supervised machine learning is the need for high-quality training data with a sufficient number of observations for each label. Finally, a reliable and valid measurement of competency requirements using unsupervised approaches requires explicitly defined competency concepts that are supported by a sufficient number of formally and substantively correct training texts. Our current preferred solution is a dictionary-based approach, which we combine with a machine learning method to enrich our dictionary. In this presentation, I will describe our approach and discuss its advantages and disadvantages. The presentation will conclude with a discussion of possibilities for better integrating different approaches in the future.
COLLABORATION
Analyses, survey insights and a practical guide mark a key milestone in the Erasmus+ project supporting data-driven vocational training across five European countries.
