AgentMemBench:对话式 AI 代理长期记忆管理策略的系统性基准
原标题:AgentMemBench: A Systematic Benchmark for Evaluating Long-Term Memory Management Strategies in Conversational AI Agents
AI 摘要
AgentMemBench 是一个用于评估对话式 AI 代理长期记忆管理策略的统一基准,在相同条件下测试了五种策略(ICW、EKV、GEM、CBS、WAM),并使用三个公共数据集和 491 个标注问题进行了评估。结果显示,外部键值存储(EKV)在所有质量指标上表现最佳,尤其在长距离回忆方面显著优于其他策略,但代价是更高的内存占用。研究还评估了 MemGPT/Letta 和 HippoRAG 两个现有系统,并发布了完整代码和结果以支持可复现性。
正文节选
Computer Science > Computation and Language Title:AgentMemBench: A Systematic Benchmark for Evaluating Long-Term Memory Management Strategies in Conversational AI Agents View PDF HTML (experimental) Abstract:Long-term memory remains a critical bottleneck for conversational AI agents, whose finite context windows cannot support coherent recall across thousands of turns. We present AgentMemBench, a unified, reproducible benchmark evaluating five memory management strategies under ident