The authors created ClariEdge, a dataset containing multi-turn conversations with optionally retrieved documents, annotated for whether the last user query is ambiguous and needs clarification.
Fine-tuning small models outperforms large ones for query clarification in RAG
A lightweight LoRA model achieves 96.6% accuracy on a new dataset of ambiguous queries, beating prompted larger baselines.
Big Tech
Lihong He · Lucian Popa · Marina Danilevsky
IBM Research
Research Digest··2 min read
The authors introduce ClariEdge, a dataset of about 3,900 examples for query clarification in retrieval-augmented generation (RAG).
§