EGCForest
A structured agent memory framework for enterprise group-chat settings. It organizes message logs into three complementary views—channel forest, actor forest and artifact forest—and enables joint cross-view retrieval through a Placement Map and structural bridges, addressing the problem of distributed evidence recall.
Timeline (last 90 days)
EGCForest is a structured Agent memory framework for enterprise group chats, adopting a three-forest architecture (channel forest, role forest, artifact forest).
EGCForest establishes cross-view connections among the three forests through a Placement Map and sparse structural bridges.
EGCForest's retrieval strategy jointly leverages three types of signals: message-level relevance, hierarchical context from forest summaries, and cross-view links.
The paper conducted experiments on three enterprise group chat benchmarks, and EGCForest consistently outperformed RAG and memory system baselines.
In answer accuracy, EGCForest achieves up to a 20.85 percentage point improvement over MemOS.
EGCForest supports efficient online retrieval, still completing multi-view joint recall under real query latency constraints.
EGCForest's three-forest architecture enables the Agent to distinguish the temporal thread, speaker conditions, and deliverable status of information during retrieval.
All Facts (7)
EGCForest is a structured Agent memory framework for enterprise group chats, adopting a three-forest architecture (channel forest, role forest, artifact forest).
50%UnverifiedEGCForest establishes cross-view connections among the three forests through a Placement Map and sparse structural bridges.
50%UnverifiedEGCForest's retrieval strategy jointly leverages three types of signals: message-level relevance, hierarchical context from forest summaries, and cross-view links.
50%UnverifiedThe paper conducted experiments on three enterprise group chat benchmarks, and EGCForest consistently outperformed RAG and memory system baselines.
50%UnverifiedIn answer accuracy, EGCForest achieves up to a 20.85 percentage point improvement over MemOS.
50%UnverifiedEGCForest supports efficient online retrieval, still completing multi-view joint recall under real query latency constraints.
50%UnverifiedEGCForest's three-forest architecture enables the Agent to distinguish the temporal thread, speaker conditions, and deliverable status of information during retrieval.
50%