Dialogue agents track user changes more accurately using predictive memory
PreCoMem: Predictive Cognitive Memory for Self-Evolving Long-Term Dialogue Agents
Artificial Intelligence
Summary
Dialogue agents that chat over long times often struggle to remember what users really mean versus temporary confusion or noise. The authors propose PreCoMem, a system that keeps track of user information by weighing how confident it is in each memory and deciding whether to keep, test more, or update what it knows. This helps the agent avoid jumping to conclusions based on noisy input and better adapt when users change their minds for real. Tests show PreCoMem improves accuracy and balances stability with flexibility in long conversations.
What this means in practice
- •For chatbot developers: Build dialogue agents that maintain personalized, stable memories and adapt to genuine user changes without overreacting to noise.
- •For customer service platform teams: Improve long-term customer interaction by filtering signal from noise in user profiles during extended conversations.
Authors
Jian Zhong, Zeyu Liu, Pingchuan Cao, Rongduo Han, Shunye Tang, Guohuan Xie, Mingyuan Qin, Ran Ji, Xiao Liang, Wei Wang, Haining Zhang
Abstract
Long-horizon dialogue agents need to keep user memories stable enough for personalization while adapting when the user genuinely changes. Existing memory systems often append every observation or overwrite old entries by heuristic rules, which makes transient noise hard to distinguish from real belief drift. We introduce PreCoMem, a memory consolidation framework inspired by predictive coding that represents memory as confidence-weighted beliefs about the user's state. Its core signal, Effective Surprisal, combines semantic deviation and contradiction evidence, then downweights this signal when retrieval is diffuse and unreliable. According to effective surprisal, a two-threshold gate maps each turn to one of three deterministic updates: MAINTAIN reinforces supported beliefs, PROFILE stores ambiguous cues as low-confidence hypotheses, and CORRECT softly decays reliably contradicted facts. The PROFILE stage acts as an explicit waiting room, preventing premature commitment to noisy evidence. Experiments on LoCoMo, LongMemEval-S, and PersonaMem-v2 show consistent state-of-the-art accuracy; our new DiMoBench confirms that PreCoMem adapts to non-stationary belief drift without over-reacting to noise. The code is available at https://github.com/ZhongWishing/PreCoMem.