Mitigating Context Interference for Reliable and Efficient Search Agents

2026-08-11

AI Summary

This work addresses the challenge of contextual interference in multi-turn retrieval-augmented agents, where irrelevant documents retrieved in each turn degrade generation reliability and efficiency. The study reveals, for the first time, that such interference predominantly stems from the most recently retrieved results. To mitigate this, the authors propose a novel “refine-then-generate” paradigm, featuring a distillation-based dynamic context refinement mechanism that filters redundant information prior to generation and seamlessly integrates this process into a reinforcement learning framework. Experimental results demonstrate that the proposed approach substantially enhances agent performance, underscoring the critical role of context refinement in multi-turn tasks.

Abstract

Recent research empowers Large Language Models (LLMs) as multi-turn search agents to iteratively retrieve and generate outputs until complex tasks are solved. However, the contexts of multi-turn search agents are lengthy and complex. For example, the retrieved set of documents in each turn would inevitably introduce irrelevant information that distracts LLMs, referring to \textit{context interference}, potentially hindering the reliability and efficiency of search agents. Therefore, we conduct a systematic study on context interference in multi-turn search agents, focusing on investigating i) which parts of the context of search agents will contribute to the context interference, ii) how to refine the contexts of search agents to mitigate the interference, and iii) can incorporating context refinement into search agent training yield further improvements. We reveal that interference primarily arises from the latest retrieved documents. Based on the explored findings, we then introduce a distill-based context refiner to dynamically mitigate context interference for multi-turn search agents. Finally, we validate that incorporating context refinement into RL training pipelines of search agents can significantly enhance both reliability and efficiency. This study highlights the importance of mitigating context interference of search agents, inspiring a novel paradigm of ``refine context and then generate'' for AI agents.

Problem

• context interference
• multi-turn search agents
• retrieval
• large language models
• search reliability

Innovation

• context interference
• multi-turn search agents
• context refinement
• distill-based refiner
• reinforcement learning