Papers › From Facts to Insights: A Persona-Driven Dual Memory Framework and Dataset for...
From Facts to Insights: A Persona-Driven Dual Memory Framework and Dataset for Role-Playing Agents
Rongsheng Zhang, Ruofan Hu, Weijie Chen, Jiji Tang, Junnan Ren, Wanying Wu, Xunuoyan Chen, Tangjie Lv, Tao Jin, Zhou Zhao
Title, abstract, authors and date from arXiv's metadata (CC0); this paper is not in the Papers with Code archive (frozen 2025-07-28).
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.
In Syntology Open this paper in Syntology's Atlas, the map of the papers in Syntology's graph and their citations.
Code
Reachability, where shown, is from one Syntology probe window (2026-09-16 to 2026-09-18); repositories not probed show nothing. GitHub stars are not tracked.
Code Syntology ran Syntology
Syntology holds the repository link but has not harvested or run code from it.
Results from the paper
The Papers with Code archive ends with its 2025-07-28 snapshot. This paper's arXiv identifier, 2605.25693, was issued in May 2026, after that date, so the archive has no leaderboard rows for it.
Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections