Let the Agent Search: Autonomous Exploration Beats Rigid Workflows in Temporal Question Answering
让代理搜索:在时间问题回答中自主探索胜过固定工作流
机构 * National University of Defense Technology(国防科技大学) ; Tsinghua University(清华大学)
专题命中 后训练与偏好优化 :LLM(abstract);large language model(abstract);language model(abstract);post-training(abstract)
AI总结 本文提出AT2QA代理,通过自主探索和动态自我纠正提升时间知识图谱问题回答性能,实验表明其在三个基准测试中超越现有最佳基线。
Comments Revised version with three added authors and additional experiments