SKORE: Skill-oriented extraction engine from job advertisements using large language models

The digital transformation of the labor market has increased the number of Online Job Advertisements, which offer valuable data on job requirements and skills. However, extracting skills from this data is challenging because it is unstructured and lacks standardized analysis. Lexicon-based and supervised methods use predefined, labeled skill sets, which limit adaptability. Unsupervised methods often lack robustness and semantic clarity. Large Language Models (LLMs) have opened new possibilities for text analysis, but their use is often ad hoc. To address this, we present the Skill-Oriented Extraction Engine (SKORE), a framework for extracting, normalizing, and categorizing skills from Online Job Advertisements. It combines the semantic power of LLMs with human-in-the-loop approaches. The framework is model-agnostic to support a wide range of domains and data sources. A proof-of-concept case study shows that it can capture both technical and behavioral skills from job-post data. This framework provides information for professionals who wish to enter a particular domain or transition careers. SKORE also assists in creating Online Job Advertisements that reflect the real needs of the market. Its applicability is envisioned for educational institutions (e.g., technical schools and universities) to respond to the market's dynamic nature on an ongoing basis.