From Facts to Insights: A Persona-Driven Dual Memory Framework and Dataset for Role-Playing Agents
2026-05-25 • Computation and Language
Computation and LanguageDatabasesMultiagent Systems
AI summaryⓘ
The authors found that current chatbots can remember facts but have trouble keeping their personality consistent in long conversations. They created a new dataset called RoleMemo to test if chatbots can understand facts through their own persona. They discovered that existing methods which ignore persona do not perform well. To fix this, the authors developed DualMem, a system that separately processes facts and persona-based understanding, leading to better personality consistency. Their approach, trained with special learning techniques, beat existing models in maintaining persona over time.
role-playing agentspersona fidelityexternal memorypersona-agnostic summarizationRoleMemo datasetDualMemSupervised Fine-Tuning (SFT)Reinforcement Learning (RL)DeepSeek-V3.2context windows
Authors
Rongsheng Zhang, Ruofan Hu, Weijie Chen, Jiji Tang, Junnan Ren, Wanying Wu, Xunuoyan Chen, Tangjie Lv, Tao Jin, Zhou Zhao
Abstract
While role-playing agents excel in short-term interactions, long-term conversations overwhelm context windows, motivating external memory frameworks. Current systems typically rely on persona-agnostic summarization, which records facts without persona-specific interpretation, yielding generic responses that compromise persona fidelity. To bridge this gap, we introduce RoleMemo, a dataset featuring four reasoning tasks where the factual fragments must be interpreted through the persona to reach the correct answer. Evaluation on RoleMemo exposes critical limitations of persona-agnostic frameworks. We thus propose DualMem, which decouples memory into two streams: factual cognition and persona-conditioned insight. Trained through Supervised Fine-Tuning (SFT) and Reinforcement Learning (RL), our framework with a 4B-parameter model outperforms zero-shot persona-agnostic frameworks powered by DeepSeek-V3.2 for sustained persona fidelity. Our resources are available at https://github.com/role2026/rolememo.