German National Library

Benchmark study compares LLM-based generative methods vs. supervised XMLC for automated subject indexing

A new arXiv preprint benchmarks supervised Extreme Multi-Label Classification (XMLC) methods against LLM-based generative approaches for automated subject indexing of German scientific literature at the German National Library. Supervised transformer-based XMLC methods achieve better overall binary relevance metrics, but LLM-based generative methods outperform on graded relevance and long-tail vocabulary coverage. The study suggests LLM-based approaches are a promising alternative for production library cataloging workflows.