Jul 2026· Annual International ACM SIGIR Conference on Research and Development in Information Retrieval· 0 citations· 34 references
Computer Science
TL;DR
This work conducts a systematic study of overthinking and overretrieval in search agents from a scaling perspective and proposes a lightweight post-hoc reflection framework that converts the proposed evaluation signals into explicit feedback rewards to guide agents' reasoning trajectories.
Abstract
Search agents enables large language models (LLMs) to iteratively interleave retrieval and reasoning, yielding strong performance on knowledge-intensive tasks. However, their multi-step autonomy also introduces substantial inefficiencies. In practice, search agents often exhibit overretrieval, where redundant or irrelevant documents are repeatedly fetched, and overthinking, where reasoning steps become excessive or unproductive. Both behaviors significantly inflate retrieval and inference cost, yet remain poorly understood, particularly under model scaling. In this work, we conduct a systematic study of overthinking and overretrieval in search agents from a scaling perspective. We formalize both phenomena at the trajectory level and propose fine-grained evaluation protocols that combine automatic statistics with LLM-based judgments. Through controlled experiments across search agents built on LLMs of varying sizes, we find that increasing model capacity generally alleviates both behaviors, but to markedly different extents. Building on these analysis results, we further propose a lightweight post-hoc reflection framework that converts the proposed evaluation signals into explicit feedback rewards to guide agents' reasoning trajectories. Our findings provide a principled foundation for diagnosing and controlling inefficiencies in search agents.
NIS-Agent is proposed, which applies context isolation at the two decision points most vulnerable to inertia bias: webpage triage and final-answer validation, and trains an 8B model to be intrinsically more resistant to inertia bias.
Xiangdong Zhang, Zhanwei Zhang, Zhihang Fu et al.· 0 citations
This work introduces SLEUTH, which makes an agent's investigative state explicit and actionable through a structured epistemic working memory, and shows that how an agent organizes its reasoning, not raw model capability, is the active ingredient for scaling multi-hop reasoning.
Although reinforcement learning with verifiable rewards (RLVR) has improved the performance of large language models (LLMs) across a variety of reasoning tasks, there is significant debate as to whether RLVR expands the reasoning capability boundary, or just improves sampling efficiency. In this paper, we investigate the nature of test-time exploration in RLVR-trained LLMs by employing controlled maze-solving experiments and extracting a tree structure from mathematical reasoning traces (BODHI-Trees) based on semantic equivalence. This helps us delineate between entropy arising from stylistic variations and genuine inferential branching. Our findings demonstrate that the policy entropy collapse observed in RLVR models is not merely syntactic, and is accompanied by a significant reduction in semantic branching entropy. While RLVR improves adherence to environmental constraints and backtracking capabilities, it constricts the space of continuations; we provide evidence suggesting that this might be responsible for the sample efficiency gains of RLVR, albeit at the cost of genuine rollout diversity.
Soumadeep Saha, Krish Sharma, Akshay Chaturvedi et al.· 1 citation
In this paper, we analyze the reasoning trajectories of a variety of DRAs and show that existing agents often suffer from reasoning stagnation: the majority of iterations contribute little or no improvement to final performance, while agents lack awareness of their trajectories and are therefore ineffective at adapting their search strategies or determining when to terminate. To address this issue, we introduce a set of unsupervised signals and a Retrieval-Aware Agent Controller (RAAC), which assists the agent in selecting optimal actions at each stage of the research process. RAAC incorporates key information retrieval principles, namely search novelty and information coverage, resulting in more effective reasoning trajectories that improve overall performance while reducing unnecessary iterations, and consequently cost and latency. Specifically on BrowseComp-Plus and across a large set of DRAs, adding RAAC reduces the number of search calls by an average of 14, significantly improves the best-performing DRA on recall and accuracy, and achieves an accuracy gain of up to 10% (3% on average).
Heydar Soudani, Elisabeth Lingg, Faegheh Hasibi et al.· 0 citations
Artificial intelligence agents increasingly perform journalism tasks autonomously, searching for sources, evaluating credibility, and producing news content with minimal human oversight. Yet research has largely treated AI as a monolithic category, leaving the effects of architectural design unexamined. Drawing on gatekeeping theory, this study presents the first systematic comparison of four agent architectures, monolithic (Claude), chain-based (LangChain), multi-agent collaborative (CrewAI), and autonomous iterative (AutoGPT), across 200 controlled experiments spanning 50 journalism tasks of graduated difficulty. All architectures used the same underlying language model and identical tools, isolating architectural effects. Results revealed significant effects on task duration (F(3, 196) = 24.54, p<.001, eta-squared = .27) and computational strategy (F(3, 196) = 305.63, p<.001, eta-squared = .82), with architecture explaining 82% of the variance in processing behavior. Multi-agent collaboration achieved the highest accuracy (84.7%) at roughly twice the time cost of other designs. Multistage analysis of the monolithic architecture documented a 71.7% source rejection rate, a quantitative parallel to classic human gatekeeping, while framework-based systems obscured their filtering inside abstraction layers. Transparency emerged as an architectural choice: framework designs excelled at structured attribution, whereas monolithic and iterative designs produced superior methodological documentation. Findings position architecture as a new structural level of gatekeeping and offer evidence-based guidance for newsrooms: chain-based designs for speed, multi-agent for accuracy, monolithic for versatility, and iterative for auditability.
Obada Kraishan, Kulsawasd Jitkajornwanich, Kerk F. Kee· 0 citations
A novel RAIE taxonomy along four scaling dimensions is proposed, which optimizes the entire thought process through search algorithms and self-verification, and introduces a task-oriented guideline for choosing the best TTS strategy.
Jiayu An, Zheng Chen, Yongcheng Jing et al.· 0 citations
