Retrieval-Augmented Generation (RAG) systems face challenges with complex, multihop questions, and agentic frameworks such as Search-R1 (Jin et al., 2025), which operates iteratively, have been proposed to address these complexities. However, such approaches can introduce inefficiencies, including repetitive retrieval of previously processed information and challenges in contextualizing retrieved results effectively within the current generation prompt. Such issues can lead to unnecessary retrieval turns, suboptimal reasoning, inaccurate answers, and increased token consumption. In this paper, we investigate test-time modifications to the Search-R1 pipeline to mitigate these identified shortcomings. Specifically, we explore the integration of two components and their combination: a contextualization module to better integrate relevant information from retrieved documents into reasoning, and a de-duplication module that replaces previously retrieved documents with the next most relevant ones. We evaluate our approaches using the HotpotQA (Yang et al., 2018) and the Natural Questions (Kwiatkowski et al., 2019) datasets, reporting the exact match (EM) score, an LLM-as-a-Judge assessment of answer correctness, and the average number of turns. Our best-performing variant, utilizing GPT-4.1-mini for contextualization, achieves a 5.6% increase in EM score and reduces the number of turns by 10.5% compared to the Search-R1 baseline, demonstrating improved answer accuracy and retrieval efficiency.
Figures & tables
Figure 1: An illustration of the information flow for our proposed test-time strategies compared to the baseline during a single inference turn i . Baseline: This represents the standard Search-R1 framework, where the LLM sends a query ( qi ) to the retriever and directly receives the retrieved documents ( Di ) to continue its reasoning. Deduplication: This approach filters out previously seen content and returns only a set of novel documents ( Di′ ) to the LLM. Contextualization: This approach parses the retrieved documents ( Di ) and reformulates their content to improve integration into the LLM’s reasoning process, returning an enhanced set of information ( Di∗ ). Hybrid: This approach combines both modules sequentially.
Figure 2: Illustration that questions requiring more agentic turns are inherently more difficult, as shown by the downward trend in Exact Match (EM) score for both the Search-R1 baseline and our Contextualization module. While the Contextualization module achieves a slightly higher mean EM at some points, the overlapping 95% confidence intervals indicate that we are not seeing a statistically significant improvement or difference between the two compared approaches for any given search count.
Variant
Exact Match
LLM Match
avg. # searches
Qwen2.5-3b Search-R1
base (PPO)
0.292
0.356
1.410
instruct (GRPO)
0.310
0.396
2.054
Qwen2.5-7b Search R1
base (PPO)
0.464
0.538
2.392
base (PPO) w/ Cont (Ours)
0.490
0.574
2.142
Table 1: Performance comparison of our proposed modules against the Search-R1 baselines on our 500-question evaluation set. We report Exact Match, LLM Match, and the average number of turns. Our modules are applied as test-time enhancements to the Qwen2.5-7b base (PPO) model. Cont: Contextualization (Section 3.1 ). De-Dup: De-duplication (Section 3.2 ). Hybrid (Section 3.3 ). Best results for our approaches are in bold.
Agentic retrieval-augmented generation (RAG) systems enable large language models (LLMs) to solve complex tasks through multi-step interaction with external retrieval tools. However, such multi-step interaction often involves redundant search steps, incurring substantial computational cost and latency. Prior work limits search depth (i.e., the number of search steps) to reduce cost, but this often leads to underexploration of complex questions. To address this, we first investigate how search depth affects accuracy and find a minimal sufficient search depth that defines an accuracy-efficiency trade-off, jointly determined by question complexity and the agent's capability. Furthermore, we propose AutoSearch, a reinforcement learning (RL) framework that evaluates each search step via self-generated intermediate answers. By a self-answering mechanism, AutoSearch identifies the minimal sufficient search depth and promotes efficient search by rewarding its attainment while penalizing over-searching. In addition, reward mechanisms are introduced to stabilize search behavior and improve answer quality on complex questions. Extensive experiments on multiple benchmarks show that AutoSearch achieves a superior accuracy-efficiency trade-off, alleviating over-searching while preserving search quality.
Jingbo Sun, Wenyue Chong, Songjun Tu +7
SKL-MAIS, Institute of Automation, Chinese Academy of Sciences · School of Advanced Interdisciplinary Sciences, University of Chinese Academy of Sciences · School of Artificial Intelligence, University of Chinese Academy of Sciences +1
Agentic retrieval-augmented generation (RAG) systems combine iterative reasoning loops, query decomposition, and adaptive retrieval to tackle multi-hop question answering. However, the contribution of each component remains poorly understood, particularly under resource-constrained settings using only local language models. Many agentic designs add adaptive retrieval routing and deeper retrieval loops on the assumption that the added complexity helps. To test whether it does, we run a controlled ablation study of a full agentic RAG pipeline evaluated on 5,000 questions from the HotpotQA distractor development set using a local 7B parameter model (Qwen2.5-7B-Instruct). Our full pipeline achieves EM=53.2% and F1=61.6%, compared to a single-pass dense-retrieval baseline of EM=43.1% and F1=54.0%. Across eight ablation conditions, we find that: (1) fixed hybrid retrieval via reciprocal rank fusion consistently outperforms rule-based adaptive routing (+1.8 EM, +1.9 F1), as the routing heuristic over-routes to BM25 by firing on named entities present in nearly all multi-hop sub-questions; (2) two retrieval iterations over the decomposed sub-questions capture 95% of the gains of five, with no meaningful benefit from deeper loops; and (3) query decomposition and cross-encoder reranking each contribute statistically significant but smaller gains (p<0.01 and p<0.001 respectively). Taken together, on a fixed local-model budget, the simpler and fixed choices turn out to be competitive with or better than their adaptive versions: most of the gain comes from running a short retrieval loop, not from adaptive routing or from many iterations. We use no proprietary APIs or large-scale compute.
Retrieval-Augmented Generation (RAG) enhances large language models (LLMs) by incorporating external knowledge, yet traditional single-round retrieval struggles with complex multi-step reasoning. Agentic RAG addresses this by enabling LLMs to dynamically decide when and what to retrieve, but current RL-based training methods suffer from sparse outcome rewards that discard intermediate signals and low sample efficiency where failed samples contribute nothing. We propose Search-P1, a framework that introduces path-centric reward shaping for agentic RAG training, comprising two key components: (1) Path-Centric Reward, which evaluates the structural quality of reasoning trajectories through order-agnostic step coverage and soft scoring that extracts learning signals even from failed samples, and (2) Dual-Track Path Scoring with offline-generated reference planners that assesses paths from both self-consistency and reference-alignment perspectives. Experiments on multiple QA benchmarks demonstrate that Search-P1 achieves significant improvements over Search-R1 and other strong baselines, with an average accuracy gain of 7.7 points.