Reproduce “Combining Distantly Supervised Models with In Context Learning for Monolingual and Cross-Lingual Relation Extraction” 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
Vipul Rathore, Malik Hammad Faisal, Parag Singla, Mausam
A new hybrid framework, HYDRE, integrates distantly supervised models with large language model in-context learning to enhance sentence-level relation extraction. This approach yields up to 20 micro-F1 points improvement in English and facilitates robust cross-lingual extraction for low-resource Indic languages, for which a new benchmark is introduced.
Key insights
- Hybridizing the high-recall candidate generation of DSRE models with the fine-grained reasoning and disambiguation capabilities of LLMs significantly boosts sentence-level relation extraction.
- A dynamic, three-stage exemplar selection process, which integrates DSRE model confidence with semantic similarity, is critical for effectively guiding LLMs using noisy bag-level training data.
- The proposed approach enables smaller, open-source LLMs to achieve performance comparable to or exceeding larger models, especially in low-resource cross-lingual settings, by providing high-quality in-context examples.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.