[arXiv]score: 0.15
Segment Tree Memory Preserves Temporal Order for Long-Horizon Agents
June 4, 2026
SegTreeMem organizes conversation history as a temporally ordered segment tree with an online rightmost-frontier insertion rule, combining local semantic matching with hierarchical temporal context during retrieval. It outperforms flat retrieval, graph-structured, and tree-structured memory baselines across three long-horizon benchmarks and two LLM backbones.
cs.CLcs.AI
HOW THIS AFFECTS YOU
●
builderYou can swap this in as a drop-in memory architecture for long-horizon conversational agents where event ordering matters for correctness.
●
researcherThe rightmost-frontier update rule and score propagation mechanism offer a concrete method for preserving chronological structure without sacrificing semantic retrieval quality.