Automatic In-Domain Exemplar Construction and LLM-Based Refinement of Multi-LLM Expansions for Query Expansion

Researchers developed an automated query expansion framework using multiple large language models that constructs domain-specific examples without manual intervention. The system uses a two-LLM ensemble approach where different models generate expansions that are then refined by a third LLM, showing significant improvements over traditional methods across multiple datasets.

• →New automated framework eliminates need for hand-crafted prompts and manual exemplar selection in query expansion systems.
• →Two-LLM ensemble approach with refinement stage delivers statistically significant improvements over baseline methods.
• →System uses BM25-MonoT5 pipeline to automatically harvest domain-relevant passages for training examples.
• →Framework tested successfully across TREC DL20, DBPedia, and SciFact datasets showing consistent performance gains.
• →Training-free cluster-based strategy enables stable in-context query expansion without supervision requirements.