Reproduce “Data Fusion of Synthetic Query Variants With Generative Large Language Models” 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
Timo Breuer
This research evaluates using generative Large Language Models to create synthetic query variants for data fusion in information retrieval systems. Combining LLM-generated queries via Reciprocal Rank Fusion consistently enhances retrieval effectiveness over single-query baselines and pseudo-relevance feedback methods, with optimal results achieved when providing rich contextual information to the LLM.
Key insights
- Providing LLMs with comprehensive contextual information (title, description, narrative) is crucial for generating high-quality query variants that significantly improve retrieval effectiveness.
- Generative LLMs offer a cost-effective (approximately $22.50 for over 120,000 queries) and unsupervised method for producing diverse query variants, overcoming limitations of prior methods.
- Data fusion effectively aggregates information from multiple LLM-generated query variants, consistently outperforming single-query baselines and traditional pseudo-relevance feedback (BM25 + RM3).
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.