analiza aiqbench
This research challenges the assumption that "thinking before retrieval" inherently improves knowledge-graph question answering by demonstrating it can degrade performance under open-weight controllers. The authors introduce TraceBound, a diagnostic protocol that isolates specific failure modes like misallocated exploration budgets and repeated calls without modifying core graph data or ranking metrics. Ultimately, the study reframes this optimization challenge as one of action selection control rather than simple prompt engineering adjustments.
When Thinking Before Retrieval Hurts: TraceBound Diagnostics for Adaptive Knowledge-Graph Retrieval
arXiv:2607.24800v1 Announce Type: cross Abstract: 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.
Abstract: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.
11 pages, 4 figures, 15 tables. Accepted at the Failure Modes in Agentic AI (FAGEN) Workshop at ICML 2026
Information Retrieval (cs.IR); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
arXiv:2607.24800 [cs.IR]
(or arXiv:2607.24800v1 [cs.IR] for this version)
https://doi.org/10.48550/arXiv.2607.24800 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 269 cuvinte
Subiecte conexe
Diagnosing Knowledge Gaps in LLM Tool Use: An Agentic Benchmark for Novel API Acquisition
arXiv · 3 iun. 2026
Presentation: From OTEL to SLMs: Distilling Frontier Model Behaviour from Production Telemetry
InfoQ · 17 iul. 2026
AdaThink-Med: Optimizing Inference-Time Compute for Medical Reasoning via Uncertainty Quantification
arXiv · 5 aug. 2026
Presentation: Building Evals for AI Adoption: from Principles to Practice
InfoQ · 29 mai 2026