MOOM: Maintenance, Organization and Optimization of Memory in Ultra-Long Role-Playing Dialogues

2025-09-15

AI Summary

To address uncontrolled memory growth and degraded coherence in ultra-long role-playing dialogues, this paper proposes MOOM, a dual-branch memory plugin. Methodologically, it pioneers the integration of literary theory into dialogue modeling by establishing parallel branches for plot progression and character development; incorporates psychological competitive inhibition theory to design a dynamic forgetting mechanism; and synergistically combines multi-timescale summarization, user-specific feature extraction, and a lightweight scheduling strategy to jointly achieve controllable memory capacity and long-term coherence. Experiments on our newly constructed Chinese ultra-long dialogue dataset ZH-4O (average 600 turns) demonstrate that MOOM significantly reduces LLM invocation frequency while improving memory efficiency and dialogue consistency, outperforming existing baseline methods.

Technology Category

• Machine Learning: Mixture of Experts (MoE)
• Natural Language Processing: Conversational AI/Dialog Systems
• Cognitive Modeling & Cognitive Systems: Computational Creativity

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Search and Retrieval-Augmented AI: Large language models for search

Abstract

Memory extraction is crucial for maintaining coherent ultra-long dialogues in human-robot role-playing scenarios. However, existing methods often exhibit uncontrolled memory growth. To address this, we propose MOOM, the first dual-branch memory plugin that leverages literary theory by modeling plot development and character portrayal as core storytelling elements. Specifically, one branch summarizes plot conflicts across multiple time scales, while the other extracts the user's character profile. MOOM further integrates a forgetting mechanism, inspired by the ``competition-inhibition'' memory theory, to constrain memory capacity and mitigate uncontrolled growth. Furthermore, we present ZH-4O, a Chinese ultra-long dialogue dataset specifically designed for role-playing, featuring dialogues that average 600 turns and include manually annotated memory information. Experimental results demonstrate that MOOM outperforms all state-of-the-art memory extraction methods, requiring fewer large language model invocations while maintaining a controllable memory capacity.

Problem

• Managing uncontrolled memory growth in ultra-long dialogues
• Extracting plot and character elements for coherent storytelling
• Maintaining controllable memory capacity with fewer model calls

Innovation

• Dual-branch memory plugin design
• Plot conflict and character profile extraction
• Competition-inhibition forgetting mechanism integration