ConvMemory v2: 一种保留召回率的前10证据重排序器用于对话记忆检索
ConvMemory v2: A Recall-Preserving Top-10 Evidence Reranker for Conversational Memory Retrieval
AI总结：
提出ConvMemory v2，一种轻量级重排序器，在保留v1的Recall@10前提下，通过微调交叉编码器提升MRR和H@1，并分析其机制。
摘要：
我们描述了ConvMemory v2，一种可选的token证据重排序器，位于轻量级ConvMemory v1重排序器之后，仅对v1保护的前10候选集进行重排序。v2是一个微调的ms-marco-MiniLM-L-6-v2交叉编码器（22,713,601个参数，从发布的检查点测量），应用于v1已经选择的十个（查询，记忆）对；它不改变返回的十个记忆，因此Recall@10和Hit@10与v1相同，这是构造决定的，而非统计巧合。在LoCoMo对话记忆基准测试（5个种子，n = 4955个测试行）上，v2将FULL MRR从v1的0.5824提升到0.6560（配对bootstrap +0.0734，95% CI [+0.0645, +0.0827]），H@1从0.4440提升到0.5474。v2缩小了与更昂贵的全池交叉编码器参考（mxbai-rerank-large-v1在前500个上，MRR 0.6688）的大部分差距但未完全消除：在FULL MRR上，v2比mxbai_top500低0.013，但在两个raw-dense-hard切片上（v1保护的前10个比mxbai自己的前10个具有更高的召回率），v2超过了mxbai_top500。一项四臂负载消融实验表明，候选特定的记忆文本是机制：移除、打乱或替换它会使MRR崩溃到低于原始稠密检索。v2最好被理解为一种标准的保留召回率的级联模式，具有LoCoMo特定的微调、显式的抗捷径推理契约和严谨的负载分析；其相对于mxbai的优势是切片特定的，而非一般的优势声明。本报告扩展了v1技术报告（arXiv:2605.28062）。
英文摘要：
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).
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
