UBFFM at the GermEval-2025 LLMs4Subjects Task: What if we take "You are an expert in subject indexing" seriously?

Clara Wan Ching Ho

Keywords: subject classification, subject indexing, large language models, Integrated Authority File

Paper Type: System Description Paper

Track: LLMs4Subjects

Abstract: This paper presents two contributions in subject classification and subject indexing of the UBFFM team at the GermEval shared task LLMs4Subjects. In Subtask 1, a fine-tune multilingual classifier is developed to assign LinSearch subject domains, achieving consistent performance across record types. For Subtask 2, an innovative generative approach is introduced by prompting to produce GND-like subject labels enriched with metadata. The pseudo-subjects are mapped to official GND terms via embedding-based similarity matching.

Submission Number: 12