Events Conference on Foundations and Advances of Machine Learning in Official Statistics, 3rd to 5th April, 2024

Domain adaptation of a BERT Model for analyzing job advertisements at the German Federal Employment Agency

Lars Fiedler¹, Barbara Hofmann^{* 1}, Tobias Scherl^{* 1}

Abstract

We present insights about the use of different machine learning approaches from a project at the German Federal Employment Agency. The task at hand is the prediction of different entities and classes related to job advertisements. We adapted a gBERT model to our specific text domain of job advertisement texts similar to Gnehm et al. (2022). We find the adapted model to outperform other approaches. These results are in line with earlier findings in the literature.

*: Speaker

1: Federal Employment Agency - Germany