发现提前思考会降低知识图谱检索效果，提出诊断方法TraceBound
When Thinking Before Retrieval Hurts: TraceBound Diagnostics for Adaptive Knowledge-Graph Retrieval
- 用TraceBound诊断自适应检索中的行为模式
- 提前思考导致检索质量下降，尤其在开放权重控制器下
- 适合研究知识图谱检索与智能体决策的学者
自适应检索通过让控制器搜索、检查邻域、修正动作并在证据充分时停止，有望提升知识图谱问答的鲁棒性。本文引入TraceBound——一种轻量级的、基于查询轮廓和轨迹条件的诊断协议，用于ARKit风格的检索器在文本丰富的知识图谱上的分析。TraceBound在检索前输出紧凑查询轮廓，检测到失败症状后发出简短轨迹提示，并记录轨迹计数器，同时固定图数据、工具、真实标签和排序指标。在STaRK验证集和保留子集上，附加条件虽提升了可解释性，但一致降低了检索质量，尤其在开放权重控制器下。配对轨迹分析将性能下降归因于重复调用、无结果调用和探索预算分配错误；更严格的交互预算虽缩短了轨迹，却未能修复策略缺陷。结论表明，‘先思考再检索’应被视为动作选择的控制问题，而非提示格式的改进。
原文摘要 · Abstract (English)
Adaptive retrieval promises to make knowledge-graph question answering more robust by letting a controller search, inspect neighborhoods, revise actions, and stop when evidence is sufficient. We study this premise by introducing TraceBound, a lightweight profile- and trace-conditioned diagnostic protocol for an ARK-style retriever on text-rich knowledge graphs. TraceBound exposes a compact query profile before retrieval, issues short trace hints after observable failure symptoms, and logs trajectory counters, while keeping graph data, tools, gold labels, and ranking metrics fixed. Across STaRK validation and held-out subsets, the added conditioning improves inspectability but consistently reduces retrieval quality under open-weight controllers. Paired trajectory analysis localizes the degradation to repeated calls, zero-result calls, and misallocated exploration budget, while stricter interaction budgets shorten trajectories without repairing the policy. The result diagnoses the common failure mode in that "thinking before retrieval'' must be evaluated as a control problem over action selection, not as a prompt-format change.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。