LELA enables zero-shot domain adaptation for entity linking using LLMs
August 7, 2026
LELA is a modular coarse-to-fine entity linking method that uses LLMs to map text mentions to knowledge bases without fine-tuning. It achieves competitive performance against fine-tuned models across different domains and knowledge bases.
HOW THIS AFFECTS YOU
●
builderYou can implement entity linking in new domains without the cost of domain-specific fine-tuning.
●
researcherThis modular approach provides a new baseline for zero-shot knowledge graph construction.