原文对话块优于抽取的记忆条目：面向长篇大语言模型对话的记忆表示受控消融实验

中文速览

长对话记忆系统常把原始对话压缩成事实、决定和事件等结构化记忆，但这种“提炼后更好检索”的假设可能会丢失关键细节。研究者在同一套检索、重排和推理流程中，只替换存储内容，用大语言模型抽取的记忆条目与原文对话块进行严格对比，并测试语义图和多种干扰因素。结果显示，原文对话块在 LoCoMo 上以 43.9% 对 28.0%、在 LongMemEval-S 上以 67.4% 对 45.4% 大幅胜出，简单语义图也无法弥补差距，说明问题主要来自有损提炼。研究表明，结构化记忆更适合与原文并存而非取代原文：两者合并的效果可保持原文块水平，这为长对话记忆系统如何在信息完整性与结构化检索之间取舍提供了直接依据。

摘要

A growing class of conversational-memory systems compresses dialogue history into structured artifacts—extracted facts, decisions, or events—on the premise that distilled structure retrieves better than raw text. We test this premise with a controlled ablation: within one fixed retrieval–rerank–reasoning pipeline, we swap only the stored representation—LLM-extracted typed artifacts versus verbatim conversation chunks—holding the model, retriever, reranker, and judge constant. Verbatim chunks win by 15.9 points on LoCoMo (43.9% vs. 28.0%) and 22.0 points on LongMemEval-S (67.4% vs. 45.4%); a 1-hop semantic graph does not recover the gap, and five confound controls reproduce the effect. The mechanism is lossy distillation: extraction discards verbatim detail that chunks retain for free, and the extracted-artifact pipeline never beats naive RAG in overall accuracy. Concurrent positive results with near-verbatim, provenance-preserving units fit the same account: retrieval accuracy tracks how far the representation departs from the source. For the extraction designs we test, structured memory should augment verbatim text rather than replace it: a chunks $\cup$ artifacts union store matches chunks on both benchmarks while artifacts alone forfeit the gap. Code and data: https://github.com/tao-hpu/cog-canvas.

术语表

conversational-memory systems

对话记忆系统

structured artifacts

结构化记忆条目

LLM-extracted typed artifacts

LLM 抽取的类型化记忆条目

verbatim conversation chunks

原文对话块

retrieval–rerank–reasoning pipeline

检索–重排序–推理流水线

controlled ablation

受控消融实验

lossy distillation

有损蒸馏

semantic graph

语义图

source provenance

来源溯源

naive RAG

朴素 RAG

LoCoMo

LoCoMo

LongMemEval-S

LongMemEval-S

retrieval-augmented generation

检索增强生成

GraphRAG

GraphRAG

hybrid retriever

混合检索器

dense retrieval

稠密检索

lexical keyword matching

词法关键词匹配

embedding

嵌入

reranker

重排序器

bge-m3

bge-m3

bge-reranker-v2-m3

bge-reranker-v2-m3

memory-augmented LLMs

记忆增强型大语言模型

context window

上下文窗口

sparse attention

稀疏注意力

exact match

精确匹配

MemGPT

MemGPT

EverMemOS

EverMemOS

EMem

EMem

Mem0

Mem0