Enhancing Job Matching: Occupation, Skill and Qualification Linking with the ESCO and EQF taxonomies

ACL ARR 2025 May Submission1972 Authors

Abstract:

This study investigates the optimal utilization of Large Language Models (LLMs) for linking job vacancy texts to the ESCO taxonomy and the EQF classification. We demonstrate that an entity-linking methodology significantly outperforms traditional sentence similarity approaches, and we release our entity linker to facilitate further research. To advance beyond skill extraction, we introduce two novel datasets for evaluating occupation and qualification extraction. Furthermore, we explore optimal embedding strategies for ESCO nodes in a retrieval setting, revealing which
combination of fields is the most effective for occupations and which works best for skills. Finally, we achieve state-of-the-art results on an established dataset for job entity extraction.

Paper Type: Long

Research Area: Information Extraction

Research Area Keywords: entity linking, named entity recognition, zero/few-shot extraction

Contribution Types: NLP engineering experiment, Approaches low compute settings-efficiency, Publicly available software and/or pre-trained models, Data resources

Languages Studied: English

Submission Number: 1972