DREAM: Event-Aware Memory Graph for Role-Playing Agents
A new structured memory framework named DREAM has been developed by researchers for role-playing agents (RPAs), utilizing large language models (LLMs) to improve character simulation. Drawing inspiration from the Activating Event-Belief-Consequence (ABC) cognitive model, this framework converts unstructured literary text into an Event-aware Memory Graph (EMG). This graph systematically arranges character experiences into a sequence of temporally ordered, causally linked events, allowing for the creation of dynamic character profiles that reflect both stable traits and changing states. This method addresses the shortcomings of current RPAs, which depend on static descriptions and unstructured memory, causing inconsistencies in narrative and personality. DREAM seeks to enhance the temporal coherence and causal grounding of role-playing actions. The research paper can be found on arXiv with the identifier 2608.05170.
Key facts
- DREAM is a structured memory framework for role-playing agents.
- It is inspired by the Activating Event-Belief-Consequence (ABC) cognitive model.
- DREAM transforms unstructured literary text into an Event-aware Memory Graph (EMG).
- The EMG organizes character experiences into temporally ordered and causally linked events.
- It enables dynamic, dual-granularity character profiles.
- Existing RPAs rely on static character descriptions and unstructured memory.
- DREAM aims to improve temporal consistency and causal grounding in role-playing.
- The paper is available on arXiv with identifier 2608.05170.
Entities
Institutions
- arXiv