BENCHMARKING LARGE LANGUAGE MODELS FOR INFORMATION EXTRACTION FROM JOB VACANCY DESCRIPTIONS

Abstract

The rapid development of online recruitment platforms has generated a large volume of unstructured job vacancy descriptions containing information about skills, qualifications, work experience, and employment conditions. Transforming such textual data into structured information is important for recruitment automation, labor market analysis, and job–candidate matching. Large language models (LLMs) have recently shown strong potential in natural language processing tasks due to their contextual understanding and generative capabilities. However, their effectiveness in extracting structured information from job vacancy descriptions remains a subject of discussion. This paper examines the role of LLMs in information extraction from job vacancy descriptions and compares their advantages and limitations with domain-adapted and traditional NLP approaches. Using a qualitative comparative methodology based on recent academic literature, the study analyzes key extraction tasks, including skill extraction, requirement identification, and structured parsing of vacancy texts. The findings suggest that LLMs demonstrate strong semantic interpretation and flexibility, especially in complex and implicit contexts, but they do not always outperform supervised domain-specific models in extraction accuracy. The paper concludes that hybrid benchmarking frameworks may provide the most effective approach for recruitment-oriented NLP systems.