DeepWeaver: Bridging the Evidence Synthesis Gap in Open-Ended Question Answering
DeepWeaver targets the step where retrieved evidence has to become a grounded, cited answer.
The paper says retrieval-heavy QA systems still lose detail, underuse evidence, and attach citations poorly during generation. DeepWeaver addresses that with “Thought Block Chains,” which organize claims, keywords, salient points, and supporting evidence before the final answer is written. The authors also introduce LoQA, a benchmark focused on dense evidence synthesis, and report gains across multiple LLMs on sufficiency, grounding, and citation quality. ArXiv · AI/CL/LG's note
The paper says retrieval-heavy QA systems still lose detail, underuse evidence, and attach citations poorly during generation. DeepWeaver addresses that with “Thought Block Chains,” which organize claims, keywords, salient points, and supporting evidence before the final answer is written. The authors also introduce LoQA, a benchmark focused on dense evidence synthesis, and report gains across multiple LLMs on sufficiency, grounding, and citation quality. ArXiv · AI/CL/LG's note
score 5