跳转至

文章背景与核心概要

长期记忆对于AI智能体提供个性化回复和进行长周期交互至关重要。传统的记忆管理方法通常依赖大语言模型（LLMs）对对话历史进行压缩或重写，以此作为检索证据。然而，这些传统方法存在两个根本性的瓶颈：一是信息丢失（由于在压缩过程中丢弃了细粒度细节），二是语义漂移（由于在重写过程中改变了原始语调和上下文）。

为了克服这些挑战，研究人员引入了 HERO（Human-profile Enhanced Retrieval Optimization，人类画像增强检索优化）框架。HERO通过两个核心机制运作：首先，它构建了一个可追溯的异构记忆 graph，将对话历史转换为丰富的图结构，保留原始对话文本作为推理的直接证据，从而有效防止信息丢失；其次，它在检索过程中提取初始查询锚点，并结合人类画像以迭代方式引导搜索过程，自适应地激活记忆图中与上下文最相关的区域。

在两个基准数据集上的实验评估表明，HERO在事实性和个性化推理任务中均显著优于强基线模型，同时保持了对原始对话证据的忠实访问。

HERO: Human-profile Enhanced Retrieval Optimization Framework for Long-term Agent Memory

HERO: Human-profile Enhanced Retrieval Optimization Framework for Long-term Agent Memory

Summary

Summary

长程记忆对于AI智能体提供个性化响应和参与长周期交互至关重要。传统方法通常采用大语言模型（LLMs）将对话历史压缩或重写为检索证据。然而，这些常规方法遭受两个根本性瓶颈的困扰：信息丢失（由压缩过程中丢弃细粒度细节引起）以及语义漂移（由重写过程中改变原始语调和上下文引起）。

Long-term memory is essential for AI agents to deliver personalized responses and engage in long-horizon interactions. Traditional methods typically employ Large Language Models (LLMs) to compress or rewrite dialogue histories into retrieval evidence. However, these conventional approaches suffer from two fundamental bottlenecks: information loss (caused by discarding fine-grained details during compression) and semantic drift (caused by altering the original tone and context during rewriting).

为了克服这些挑战，研究人员引入了 HERO（Human-profile Enhanced Retrieval Optimization），这是一个用于长程智能体记忆的新颖框架。HERO通过两个主要机制运行： 1. 可追溯的异构记忆图： 它将对话历史转换为丰富的图结构，保留原始对话文本作为推理的直接证据，从而有效地防止信息丢失。 2. 结合人类画像的迭代图遍历： 在检索过程中，HERO提取初始查询锚点并整合人类画像，以迭代方式指导搜索过程，自适应地激活记忆图中与上下文最相关的区域。

To overcome these challenges, researchers introduce HERO (Human-profile Enhanced Retrieval Optimization), a novel framework for long-term agent memory. HERO operates through two primary mechanisms: 1. Traceable Heterogeneous Memory Graph: It converts dialogue histories into a rich graph structure that retains raw dialogue text as direct evidence for reasoning, effectively preventing information loss. 2. Iterative Graph Traversal with Human Profiles: During retrieval, HERO extracts initial query anchors and integrates human profiles to guide the search process iteratively, adaptively activating the most contextually relevant regions of the memory graph.

两个基准数据集上的实验评估表明，HERO在事实性和个性化推理任务中均显著优于强基线模型，同时对原始对话证据保持了更忠实的访问。

Experimental evaluations across two benchmark datasets demonstrate that HERO significantly outperforms strong baselines in both factual and personalized reasoning tasks while maintaining faithful access to raw dialogue evidence.

Document Metadata

Document Metadata

字段 详情
arXiv ID arXiv:2608.22310 [cs.AI]
标题 HERO: Human-profile Enhanced Retrieval Optimization Framework for Long-term Agent Memory
作者 Yuanhua Lin, Yile Li, Zhiyuan Zhao, Jing Shang, Jian Sun
主要学科 人工智能 (cs.AI)
提交日期 2026年8月23日
DOI 10.48550/arXiv.2608.22310
Field Details
arXiv ID arXiv:2608.22310 [cs.AI]
Title HERO: Human-profile Enhanced Retrieval Optimization Framework for Long-term Agent Memory
Authors Yuanhua Lin, Yile Li, Zhiyuan Zhao, Jing Shang, Jian Sun
Primary Subject Artificial Intelligence (cs.AI)
Submission Date August 23, 2026
DOI 10.48550/arXiv.2608.22310

Abstract

Abstract

长程记忆对于个性化响应和长周期智能体交互至关重要。现有方法通常依赖大模型来压缩或重写对话历史，并将转换后的记忆用作检索证据。尽管在组织碎片化上下文方面取得了进展，但仍存在两个主要缺点：(1) 压缩带来的信息丢失，它会丢弃细粒度但在后期有用的细节；(2) 重写带来的语义漂移，它会侵蚀原始语调和情境上下文。在这项工作中，我们提出了一种用于长程智能体记忆的新型人类画像增强检索优化框架 (HERO)。具体而言，HERO将对话历史转换为可追溯的异构记忆图，该图将原始对话文本保留为推理证据，从而减轻信息丢失。在检索方面，HERO从当前查询中提取初始锚点，并通过迭代图遍历整合人类画像；这些锚点和画像提供了引导信号，可以自适应地激活图中信息最丰富的区域。在两个基准数据集上的实验表明，HERO在事实性和个性化推理方面均优于强基线，同时提供了对原始对话证据更忠实的访问。

Long-term memory is crucial for personalized responses and long-horizon agent interactions. Existing methods often rely on LLMs to compress or rewrite dialogue histories and use the transformed memories as retrieval evidence. Despite the progress in organizing fragmented contexts, two major drawbacks persist: (1) information loss from compression, which discards fine-grained but later useful details, and (2) semantic drift from rewriting, which erodes the original tone and situated context. In this work, we propose a novel Human-profile Enhanced Retrieval Optimization framework for long-term agent memory (HERO). Specifically, HERO converts the dialogue history into a traceable heterogeneous memory graph that preserves raw dialogue text as evidence for reasoning, thereby mitigating information loss. For retrieval, HERO extracts initial anchors from the current query and incorporates human profiles via an iterative graph traversal; these anchors and profiles provide guidance signals that adaptively activate the most informative regions of the graph. Experiments on two benchmark datasets show that HERO outperforms strong baselines on both factual and personalized reasoning, while providing more faithful access to raw dialogue evidence.