ProPerAgent: Toward Memory-Driven Proactive Assistance
作者: Shenglai Zeng, Xiao Yang, Tiffany Ma, Xinyuan Zhang, Nanshu Wang, Deren Lei, Kai Sun, Xilun Chen, Tammy Stark, Xianhui Zhu, Yuetai Li, Rulin Shao, Tianxin Wei, Rakesh Wanga, Anuj Kumar, Jiliang Tang, Wen-tau Yih, Xin Luna Dong
分类: cs.CL
发布日期: 2026-10-15
💡 一句话要点
提出ProPerAgent以解决记忆驱动的主动辅助问题
🎯 匹配领域: 支柱六:视频提取与匹配 (Video Extraction)
关键词: 主动辅助 记忆驱动 强化学习 图组织存储 个性化助手 长时间记忆 干预时机 多模态学习
📋 核心要点
- 现有方法在处理长时间个人记忆时,难以有效挖掘干预所需的证据,导致主动辅助性能低下。
- 论文提出ProPerAgent,通过图组织的存储系统提取记忆,并结合强化学习优化干预时机,提高主动辅助的准确性。
- 实验结果显示,ProPerAgent在未训练情况下准确率提升至37.5%,经过强化学习后达到42.4%,显著优于现有基线模型。
📝 摘要(中文)
随着助手从被动回答转向主动干预,记忆的角色发生了变化:它不仅提供查询答案,还决定何时发言。现有方法在处理长时间个人记忆时面临挑战,论文提出了ProPerMem基准,结合自我中心视频和丰富的用户记忆,评估内容和时机的联合得分。实验表明,现有基线模型表现不佳,最高仅为15%。为此,作者设计了ProPerAgent,通过图组织的存储系统提取记忆,并引入轻量级强化学习阶段,显著提高了准确率,未训练情况下准确率提升至37.5%。
🔬 方法详解
问题定义:本论文旨在解决记忆驱动的主动辅助问题,现有方法在长时间个人记忆中挖掘干预证据的能力不足,导致系统性能低下。
核心思路:提出ProPerAgent,通过图组织的存储系统提取用户记忆,并引入轻量级强化学习阶段,以优化干预时机和内容选择,提升主动辅助的有效性。
技术框架:整体架构包括记忆提取模块、干预决策模块和强化学习优化模块。记忆提取模块从图存储中获取相关信息,干预决策模块评估何时干预,强化学习模块则优化干预策略。
关键创新:最重要的创新在于结合了图组织的记忆存储与强化学习,能够有效地从长时间记忆中提取信息并优化干预时机,与现有方法相比,显著提升了系统的准确性和响应性。
关键设计:在设计中,采用了轻量级的强化学习策略,确保系统在不频繁干预的情况下,仍能保持较高的准确率。具体参数设置和损失函数设计未在摘要中详细说明,需参考完整论文。
🖼️ 关键图片
📊 实验亮点
实验结果显示,ProPerAgent在未训练情况下的准确率提升至37.5%,经过强化学习后达到42.4%。相比之下,现有基线模型的最高准确率仅为15%,显示出显著的性能提升。
🎯 应用场景
该研究的潜在应用领域包括智能助手、个性化推荐系统和人机交互等。通过提升主动辅助的准确性,ProPerAgent能够在多种场景中提供更为精准和及时的支持,具有重要的实际价值和未来影响。
📄 摘要(原文)
As assistants move from reactive answering toward proactive intervention, memory changes role: it no longer only supplies the answer to a query, it determines whether and when to speak at all. Prior work in this setting hands the system the relevant fact in context; we study the regime that makes the problem hard, where the evidence justifying an intervention must be mined from a long-horizon personal memory. We introduce ProPerMem, a benchmark for memory-driven proactive assistance (MPA), pairing egocentric video with year-long, distractor-rich user memories (~618K tokens on average), human-verified trigger windows, and a protocol that scores content and timing jointly. Every baseline we evaluate performs poorly (at most 15%), including frontier models given summarized or retrieval-augmented memory. A controlled study explains why: with oracle evidence delivered inside the trigger window GPT-5.3-Chat reaches 96%; letting it choose the timing drops it to 38%, and replacing oracle evidence with retrieval drops it to 15%. Content and timing are separately necessary, and missing either is catastrophic. Guided by this, we build ProPerAgent (Proactive Personalized Agent) with a plug-and-play harness that surfaces memory from a graph-organized store and gates it down to what is worth interrupting for, plus a lightweight RL stage that curbs early and repeated firing. Without any training, ProPerAgent lifts strict accuracy from below 10% to 37.5% on Qwen3.8-27B and 38% on GPT-5.3-Chat; RL post-training reaches 42.4%.