---
title: "ABSeeker：通过答案回溯信用分配训练长程搜索智能体"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2608.05102v1"
permalink: "https://feed.astralor.com/items/itm_73b8a3bfe95a48d177274700"
published: "2026-08-05T17:41:31.000Z"
score: 76
channel: "ai"
tags: ["ABSeeker","arXiv","Qwen","信用分配","强化学习","搜索智能体"]
---

# ABSeeker：通过答案回溯信用分配训练长程搜索智能体

## AI 点评

> 这篇论文最值得关注的是它把搜索智能体训练从整段轨迹的粗粒度奖惩推进到了步骤级别的精细归因，尤其能让失败轨迹里的有效步骤被识别出来。对于在做 Agent 推理对齐或者想用低成本小模型冲复杂搜索任务的团队来说，这套线索回溯的思路很实用，值得结合自己的流程试试。

## 速览

长程搜索智能体需要执行多个顺序动作来搜索、检索、验证和整合证据，但现有训练方法在监督微调和强化学习中均将所有步骤等同对待，无法区分有效与错误或冗余动作。本文提出答案回溯信用分配（ABC）框架，将稀疏的轨迹级结果转化为密集的步骤级监督，奖励有效动作并抑制错误或冗余动作。ABC 首先通过答案回溯线索恢复从答案反推解决问题所需的中介线索，再基于线索锚定步骤评分评估每个搜索步骤。基于此实现 ABC-SFT 与 ABC-GRPO，并训练了 ABSeeker（Qwen3.5-4B，仅 8.5k 样本）。ABSeeker 在 BrowseComp 上达到 37.3%，在 BrowseComp-ZH 上达到 39.1%，加入上下文管理后分别提升至 55.3% 和 52.9%，显著优于同规模 4B 智能体，甚至可匹配约 30B 的更大模型。

## 关键点

- 提出 ABC 框架，将稀疏的轨迹级结果转化为密集的步骤级奖励，可区分有效、错误和冗余动作。
- ABC-SFT 按步骤权重重加权损失，ABC-GRPO 将步骤级分数作为 GRPO 奖励，二者均基于答案回溯的线索。
- 仅用 8.5k 样本训练 4B 规模的 ABSeeker，BrowseComp 达 37.3%，加上下文管理后达 55.3%，接近 30B 模型。

## 全文

Long-horizon search agents must make multiple sequential actions (steps) to search, retrieve, verify, and integrate evidence to reach a final answer. However, existing methods for training these agents typically treat all steps within a trajectory uniformly during both supervised fine-tuning (SFT) and reinforcement learning (RL), failing to distinguish useful actions from erroneous or redundant ones. In this paper, we propose Answer-Backtracked Credit Assignment (ABC), a fine-grained credit assignment framework for training long-horizon search agents by converting sparse trajectory-level outcomes into dense step-level supervision that rewards useful actions (even in failed trajectories) while suppressing erroneous or redundant actions. Specifically, given a potentially obscure query and its corresponding ground-truth answer, ABC first performs Answer-Backtracked Clue Recovery, which traces back from the answer to recover intermediate clues required to solve the question. It then applies Clue-Anchored Step Scoring to evaluate each search step against these clues, converting sparse binary outcome supervision into dense step-level rewards. Based on these rewards, we develop ABC-SFT, which reweights the loss of each turn, and ABC-GRPO, which uses the step-level scores as rewards in GRPO. Building on this framework, we train ABSeeker based on Qwen3.5-4B with only 8.5k examples. ABSeeker achieves 37.3% on BrowseComp and 39.1% on BrowseComp-ZH. With context management, the scores further improve to 55.3% and 52.9%, respectively, significantly outperforming same-scale (4B) agents and even matching the performance of larger ones (approximately 30B). These results demonstrate the effectiveness of answer-backtracked step-level credit assignment for training long-horizon search agents.

---

引用前请核对原文;本文由星际流动整理:https://feed.astralor.com/items/itm_73b8a3bfe95a48d177274700