PGMem: Tightly Coupled Persona-Memory Graph for Lifelong Personalized Agents

2026-08-03

AI Summary

This work addresses the semantic gap and retrieval bias arising from conventional memory systems that store personality as static profiles, decoupled from the supporting events. To bridge this gap, the authors propose a heterogeneous personality-memory graph structure that links event and personality nodes via typed edges, enabling traceability and dynamic updating of personality information. They further introduce a seed-expansion-based retrieval strategy coupled with an evidence validity ranking mechanism, optimized end-to-end with a small language model. This approach uniquely achieves tight coupling between personality signals and historical events, significantly outperforming state-of-the-art baselines—including summary-based, personality-aware, graph-based, and agent memory methods—across three benchmark tasks, with performance consistently improving as contextual length increases.

Abstract

Long-term personalized dialogue agents must track user preferences as their personas evolve. Existing memory systems organize past events well, but store personas as flat profiles detached from the events that justify them. This loose coupling leads to the memory-persona validity gap and the persona-aware retrieval gap. We propose PGMem, a heterogeneous persona-memory graph that connects event and persona nodes through typed provenance and evidence edges, keeping each persona signal traceable to the events that support or revise it. At retrieval time, PGMem expands from query-relevant seeds and ranks signals by evidential validity. Across three benchmarks with small language model backbones, PGMem consistently outperforms summary-based, persona-aware, graph-structured, and agentic memory baselines, and improves performance as the context grows. The source code of PGMem is available at https://github.com/wonjunchoi23/pgmem/

Problem

• personalized dialogue agents
• persona evolution
• memory-persona validity gap
• persona-aware retrieval gap
• long-term personalization

Innovation

• persona-memory graph
• evidential validity
• provenance edges
• lifelong personalization
• heterogeneous graph