RVR: Retrieve-Verify-Retrieve for Comprehensive Question Answering
Deniz Qian, Hung-Ting Chen, Eunsol Choi
Abstract
Comprehensively retrieving diverse documents is crucial to address queries that admit a wide range of valid answers. We introduce retrieve-verify-retrieve (RVR), a multi-round retrieval framework designed to maximize answer coverage. Initially, a retriever takes the original query and returns a candidate document set, followed by a verifier that identifies a high-quality subset. For subsequent rounds, the query is au...
Summary
RVR (Retrieve-Verify-Retrieve) is a multi-round retrieval framework: a retriever returns candidates, a verifier selects a high-quality subset, then the query is augmented with verified docs to retrieve new uncovered answers in subsequent. This page includes benchmark evidence for RVR shows consistent MRecall@100 and Recall@100 gains on QUEST (N=1727) and WebQ on QUEST. Reproduction guidance focuses on implementation viability and concrete risk controls.
Key Contributions
- RVR (Retrieve-Verify-Retrieve) is a multi-round retrieval framework: a retriever returns candidates, a verifier selects a high-quality subset, then the query is augmented with verified docs to retrieve new uncovered.
- Qwen3-30B-A3B-Instruct-2507 is used as the verifier in main experiments, achieving the best recall among tested verifiers including GPT-5-nano on QAMPARI top-100 retrieved documents.
- Fine-tuning retrievers specifically for the RVR inference procedure (FT D_r) brings further gains beyond using off-the-shelf or in-domain fine-tuned (FT D_i) retrievers.
- RVR achieves at least 10% relative and 3% absolute gain in complete recall percentage over baselines including agentic search on the QAMPARI test set (N=1000).
- RVR shows consistent MRecall@100 and Recall@100 gains on QUEST (N=1727) and WebQuestionsSP (N=1639) test sets across different base retrievers, demonstrating out-of-domain generalization.
Reproducibility Notes
- Estimate is based on paper-only reproduction flow.
Results & Benchmarks
| Task | Dataset | Metric | Value |
|---|---|---|---|
| RVR shows consistent MRecall@100 and Recall@100 gains on QUEST (N=1727) and WebQ | QUEST | Recall | 100 |
Hardware Requirements
- Expect multi-day setup/compute for meaningful reproduction based on current guidance.
Best Implementation
Maintained implementation evidence is not confirmed for this paper yet.
Use the Implementation Status and Reproduction Path sections below for the current action plan.
Reproduction Path
Follow this baseline workflow to decide if this paper is worth immediate prototyping.
- 1
Use the paper and benchmark evidence to scope a baseline reproduction plan.
- 2
Track assumptions and missing details in an experiment log before coding.
Additional Implementations
No additional verified repositories beyond the primary recommendation.
Hugging Face Artifacts
No trustworthy direct or curated related Hugging Face artifacts were found yet.
Continue with targeted Hugging Face searches: