在不改变召回结果的前提下，提升对话记忆检索的排序精度。
ConvMemory v2: A Recall-Preserving Top-10 Evidence Reranker for Conversational Memory Retrieval
- 在v1选出的前10条记忆中，用微调交叉编码器重新排序。
- 全量MRR从0.5824提升至0.6560，命中率提升至0.5474。
- 适合需要高精度排序且保持召回率的对话系统研究者。
我们介绍ConvMemory v2，一种可选的基于标记证据的重排序器，位于轻量级ConvMemory v1之后，仅对v1保护的前10个候选记忆进行重排序。v2是微调后的ms-marco-MiniLM-L-6-v2交叉编码器（22,713,601参数，来自发布检查点），应用于v1已选定的十个（查询，记忆）对；它不改变返回的十条记忆，因此召回率@10和命中率@10与v1完全相同，这是构造决定而非统计巧合。在LoCoMo对话记忆基准测试中（5个种子，n=4955测试行），v2将全量MRR从v1的0.5824提升至0.6560（配对Bootstrap +0.0734，95%置信区间[+0.0645, +0.0827]），命中率@1从0.4440提升至0.5474。v2几乎弥合但未完全消除与更昂贵的全池交叉编码器参考模型（mxbai-rerank-large-v1在前500条上，MRR 0.6688）之间的差距：在全量MRR上，v2比mxbai_top500低0.013，但在两个原始密集硬样本子集上（v1保护的前10条召回率高于mxbai自身前10条），v2超越了mxbai_top500。四臂负载承载消融实验表明，候选特定记忆文本是关键机制：移除、打乱或替换它会使MRR低于原始密集检索。v2应被理解为一种标准的召回保持级联模式，配合LoCoMo特化微调、明确的反捷径推理契约和严谨的负载承载分析；其优势是子集特定的，而非普遍性优势。本报告扩展了v1的技术报告（arXiv:2605.28062）。
原文摘要 · Abstract (English)
We describe ConvMemory v2, an opt-in token-evidence reranker that sits after the lightweight ConvMemory v1 reranker and reorders only v1's protected top-10 candidate set. v2 is a fine-tuned ms-marco-MiniLM-L-6-v2 cross-encoder (22,713,601 parameters, measured from the released checkpoint) applied to the ten (query, memory) pairs that v1 has already selected; it does not change which ten memories are returned, so Recall@10 and Hit@10 are identical to v1 by construction, not by statistical coincidence. On the LoCoMo conversational memory benchmark (5 seeds, n = 4955 test rows), v2 raises FULL MRR from v1's 0.5824 to 0.6560 (paired bootstrap +0.0734, 95% CI [+0.0645, +0.0827]) and H@1 from 0.4440 to 0.5474. v2 closes most but not all of the gap to a much more expensive full-pool cross-encoder reference (mxbai-rerank-large-v1 over the top-500, MRR 0.6688): on FULL MRR v2 sits 0.013 below mxbai_top500, but on two raw-dense-hard slices (where v1's protected top-10 has higher recall than mxbai's own top-10) v2 exceeds mxbai_top500. A four-arm load-bearing ablation shows candidate-specific memory text is the mechanism: removing, shuffling, or replacing it collapses MRR below raw dense retrieval. v2 is best understood as a standard recall-preserving cascade pattern with LoCoMo-specific fine-tuning, an explicit anti-shortcut inference contract, and disciplined load-bearing analysis; its advantage over mxbai is slice-specific rather than a general dominance claim. This report extends the v1 technical report (arXiv:2605.28062).
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。