Research map / Label-efficient and robust learning / Recognition and long-tail
Evidence Aggregation for Answer Re-Ranking in Open-Domain Question Answering
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| combine strength- and coverage-based rerankers | R3 open-domain QA · Quasar-T test | F1: 41.7 -> 49.6 | Helped |
| combine strength- and coverage-based rerankers | R3 open-domain QA · Quasar-T test | EM: 35.3 -> 42.3 | Helped |
| combine strength- and coverage-based rerankers | R3 open-domain QA · SearchQA test | F1: 55.3 -> 63.2 | Helped |
| combine strength- and coverage-based rerankers | R3 open-domain QA · SearchQA test | EM: 49.0 -> 57.0 | Helped |
| combine strength- and coverage-based rerankers | R3 open-domain QA · TriviaQA open test | F1: 53.7 -> 57.3 | Helped |
| combine strength- and coverage-based rerankers | R3 open-domain QA · TriviaQA open test | EM: 47.3 -> 50.6 | Helped |
| use coverage-based reranking alone rather than the full combination (ablation) | R3 open-domain QA · SearchQA test | F1: 63.2 -> 61.4 | Hurt |
| use coverage-based reranking alone rather than the full combination (ablation) | R3 open-domain QA · Quasar-T test | F1: 49.6 -> 49.1 | Hurt |
| use coverage-based reranking alone rather than the full combination (ablation) | R3 open-domain QA · TriviaQA open test | F1: 57.3 -> 57.0 | Hurt |
| use BM25 to rerank aggregated passages instead of neural coverage reranking | R3 open-domain QA · Quasar-T test | EM: 40.6 -> 33.6 | Hurt |
| increase the coverage reranker's candidate set from top-5 to top-10 (ablation) | Coverage-Based Re-Ranker · Quasar-T dev | EM: 41.8 -> 41.3 | No clear effect |
| increase the coverage reranker's candidate set from top-3 to top-5 (ablation) | Coverage-Based Re-Ranker · Quasar-T dev | EM: 40.5 -> 41.8 | Helped |
Closest papers by meaning
- RINK: Reader-Inherited Evidence Reranker for Table-and-Text Open Domain Question Answering (AAAI 2023 · similarity 0.91)
- Open Question Answering over Tables and Text (ICLR 2021 · similarity 0.88)
- R3: Reinforced Ranker-Reader for Open-Domain Question Answering (AAAI 2018 · similarity 0.88)
- SuRe: Summarizing Retrievals using Answer Candidates for Open-domain QA of LLMs (ICLR 2024 · similarity 0.87)
- Autofocus Retrieval: An Effective Pipeline for Multi-Hop Question Answering With Semi-Structured Knowledge (NeurIPS 2026 · similarity 0.84)
- CofCA: A STEP-WISE Counterfactual Multi-hop QA benchmark (ICLR 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.