Reproduce “DNB-AI-Project at SemEval-2025 Task 5: An LLM-Ensemble Approach for Automated Subject Indexing” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Lisa Kluge, Maximilian Kähler
Researchers at the Deutsche Nationalbibliothek developed an LLM-ensemble approach for automated subject indexing that combines multiple models and prompting strategies. The system achieved a fourth-place ranking in average recall and the top qualitative evaluation by subject indexing experts in the SemEval-2025 Task 5 competition.
Key insights
- Combining multiple LLMs with diverse prompting strategies enhances automated subject indexing performance, outperforming single-model baselines without requiring fine-tuning.
- An ensemble approach effectively leverages the complementary strengths of different LLMs, where some models excel in precision and others in recall, contributing to overall improved results.
- Mapping LLM-generated keywords to controlled vocabularies via embedding similarity, combined with LLM-powered relevance ranking, allows for practical integration into library systems and adherence to formal indexing rules.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.