Evidence Retrieval

Evidence retrieval studies how systems find the passages, records, images, or other source material needed to answer questions and check claims. Common approaches combine lexical and dense search with reranking, while newer systems use query refinement, iterative agents, and graph or document-structure navigation to gather evidence across sources. Evaluation spans tasks such as multi-hop question answering, fact-checking, scientific literature search, and long-document or multimodal QA, often measuring evidence recall and citation quality alongside answer accuracy. A central challenge is that relevant evidence may still be incomplete, contradictory, or misinterpreted, so systems must balance coverage, source reliability, efficient search, and evidence sufficiency.