Year: 2025
Type: conference-paper
Abstract: The automated classification of occupational titles is a pivotal component of labor market analysis, survey research, and administrative data processing.This paper explores the viability of mapping German job titles to the German Classification of Occupations (KldB) by employing conventional machine learning methodologies to examine the challenges and limitations inherent in the data itself.To this end, the present study leverages two complementary datasets-manually annotated survey data and a dataset of occupational synonyms-to assess the performance of established classifiers under varying levels of taxonomic granularity.The methodological challenges inherent to this study include class imbalance, semantic ambiguity, and linguistic variability, which are all characteristics of German job title expressions.The findings of the study suggest that while coarse-level classifications can be addressed with relatively simple models and text representations, finer-grained distinctions remain challenging to resolve using title-based features alone.The findings indicate that more expressive models and richer contextual information may be necessary for high-resolution occupational coding.
DOI: 10.15439/2025f4081
Authors: Ralf Dorau, Kristine Hein
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W4415420713