Reproduce “DREQ: Document Re-Ranking Using Entity-based Query Understanding” 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
Shubham Chatterjee, Iain Mackie, Jeff Dalton
DREQ proposes a "retrieve-harness-rerank" paradigm that generates query-specific entity-centric document representations by dynamically weighting document entities based on their relevance to an incoming query. The model consistently outperforms state-of-the-art re-rankers on four large-scale benchmarks, showing up to a 36% nDCG@20 improvement over baselines and significantly enhancing performance for difficult queries.
Key insights
- Entities are integral and pivotal to DREQ's re-ranking effectiveness, as their removal drastically degrades performance.
- A carefully selected and specifically trained entity ranking system is crucial for accurately identifying and leveraging query-relevant entities.
- Query-specific, probabilistic weighting of entities is superior to uniform or reciprocal rank weighting, effectively capturing semantic nuance for precise relevance assessment.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.