JAM has the highest LLM-judge accuracy among systems evaluated with the same GPT-4o-mini protocol on LoCoMo, NarrativeQA and HotpotQA. Its gains over the strongest baseline in each benchmark are 12.54, 20.00 and 11.20 percentage points. Memory-R1 scores are quoted separately and are not directly comparable under this judging pipeline.

JAM has the highest LLM-judge accuracy among systems evaluated with the same GPT-4o-mini protocol on LoCoMo, NarrativeQA and HotpotQA. Its gains over the strongest baseline in each benchmark are 12.54, 20.00 and 11.20 percentage points. Memory-R1 scores are quoted separately and are not directly comparable under this judging pipeline. · CiteArk

测量结果

已评估 0/29 项

获得支持 0有争议或冲突 0尚未评估 29

保留同一结论下的完整结果组。已有评估排在前面；单项判断不替代整组结论。展开行查看条件、来源与历次证据。

每行是一项测量，分别显示论文报告、最近可用实测和累计证据状态。
指标与条件	论文报告	最近可用实测	证据判断
method: RAG · benchmark: LoCoMo · sourceBlock: Evaluated with our judge	58.25%	—	尚未评估
method: RAG · benchmark: NarrativeQA · sourceBlock: Evaluated with our judge	48%	—	尚未评估
method: RAG · benchmark: HotpotQA · sourceBlock: Evaluated with our judge	53.13%	—	尚未评估
method: A-MEM · benchmark: LoCoMo · sourceBlock: Evaluated with our judge	54.74%	—	尚未评估
method: A-MEM · benchmark: NarrativeQA · sourceBlock: Evaluated with our judge	45%	—	尚未评估
method: A-MEM · benchmark: HotpotQA · sourceBlock: Evaluated with our judge	35.68%	—	尚未评估
method: Mem0 · benchmark: LoCoMo · sourceBlock: Evaluated with our judge	46.23%	—	尚未评估
method: Mem0 · benchmark: NarrativeQA · sourceBlock: Evaluated with our judge	43%	—	尚未评估
method: Mem0 · benchmark: HotpotQA · sourceBlock: Evaluated with our judge	38.02%	—	尚未评估
method: MemoryOS · benchmark: LoCoMo · sourceBlock: Evaluated with our judge	47.66%	—	尚未评估
method: MemoryOS · benchmark: NarrativeQA · sourceBlock: Evaluated with our judge	40%	—	尚未评估
method: MemoryOS · benchmark: HotpotQA · sourceBlock: Evaluated with our judge	30.21%	—	尚未评估

报告指标与观测值

Assessment（0）

关联运行（0）

测量结果