Lifelong conversational agents improve personalization with latent memory
ThinkFlow: Self-Evolving Probabilistic Latent Memory for Lifelong Conversational Agents
Artificial IntelligenceComputation and Language
Summary
Conversational agents, like chatbots, need good memory to hold meaningful and personal conversations over time. The authors found that existing systems lose important subtle details and cannot adapt to new user habits automatically. They created ThinkFlow, which stores conversations as compact hidden information and keeps improving its understanding without needing extra labels. This approach helps chatbots remember users better and respond more accurately across many sessions.
What this means in practice
- •For chatbot developers: Build chatbots that remember long-term user preferences and adapt automatically without manual feedback.$Commercial implications: Enables creation of personalized conversational AI products with continuous self-improvement for diverse users.
- •For customer support teams: Enhance support bots to maintain context over multiple interactions, improving response relevance and user satisfaction.
Authors
Cai Ke, Xin Liu, Han Zhang, Jiangyue Yan, Zike Yuan, Ling Deng, Yue Yu, Hui Wang, Ruifeng Xu
Abstract
Lifelong conversational agents rely on memory systems to maintain deep, context-aware interactions with users. However, existing explicit textual memory pipelines suffer from a severe information bottleneck, often losing subtle behavioral patterns and emotional shifts. Furthermore, being typically static post-deployment, they cannot autonomously adapt to personal habits and preferences without manual feedback. Cognitive science, however, suggests that humans maintain mental models purely in a latent space and continuously refine them through predictive coding. Inspired by this, we propose \textbf{ThinkFlow}, a novel end-to-end latent memory framework for lifelong conversational agents. ThinkFlow bypasses the text bottleneck by dynamically compressing conversational flows into probabilistic latent memory skills, autonomously consolidating complex user states into disentangled, continuous vectors without semantic interference. To break this barrier, we introduce a test-time evolution paradigm. By coupling teacher-guided latent alignment to bootstrap the initial state with a self-supervised next-user-utterance prediction task for continuous refinement, the framework successfully overcomes cold-start challenges and achieves label-free lifelong personalization. Extensive experiments on long-term conversation benchmarks demonstrate that ThinkFlow significantly outperforms prevailing memory systems, providing highly personalized and contextually accurate responses over extended multi-session interactions.