PersonaTree: Structured Lifecycle Memory for Person Understanding in LLM Agents

2026-06-03

AI Summary

This work addresses the limitation of existing large language model agents, whose memory mechanisms primarily focus on information storage and retrieval while lacking explicit modeling of how interaction evidence is abstracted into stable person understanding. The authors propose PersonaTree, a structured lifelong memory framework that formalizes person understanding as a schema formation process. PersonaTree employs a three-level hierarchical persona tree to progressively abstract concrete interaction evidence into reusable patterns and claims, preserving explicit support paths from evidence to claims. Integrating conservative write operations, confidence-guided integration, and conditional path retrieval, the method dynamically adapts the depth of evidence retrieval within limited context windows. Experiments across six benchmarks and three backbone models show that PersonaTree achieves top-1 performance in 12 out of 18 compact evaluation metrics and ranks within the top two in 16. Ablation studies confirm that the hierarchical structure enhances abstract understanding and that path-aware retrieval improves preference alignment.

Abstract

Persistent LLM agents require memory representations that make the formation of person understanding explicit across long term interaction. Existing agent memory methods emphasize information retention and retrieval, yet give limited account of how accumulated interaction evidence is abstracted into person understanding. We view this process as schema formation, where situated evidence is abstracted into reusable patterns and stable person level claims. We introduce PersonaTree, a structured lifecycle memory framework that realizes this view as a three level persona tree with explicit support paths from evidence to claims. PersonaTree maintains the tree through conservative writing, confidence guided consolidation, and query conditioned path retrieval, returning only the evidence depth required by each query. Across six person understanding and persistent memory benchmarks with three answer backbones, PersonaTree ranks first in 12 of 18 compact scores and reaches the top two in 16 settings. Ablations show that hierarchy improves abstract person understanding on KnowMe, while support path retrieval improves RealPref alignment under a comparable context budget.

Problem

• person understanding
• memory representation
• schema formation
• persistent agents
• LLM agents

Innovation

• PersonaTree
• structured memory
• schema formation
• evidence-to-claim path
• persistent LLM agents