Dual-path memory improves personalized language models across sessions
DPPM: Dual-Path Parametric Memory for Personalized Language Models
Computation and Language
Summary
Personalizing language models to remember user preferences over a long time is tricky. The authors propose a method called Dual-Path Parametric Memory (DPPM) that helps the model keep track of user interactions over many sessions without forgetting earlier information. DPPM combines two ways of storing memory: one pools all past evidence, and the other updates changes sequentially. This combination produces better adapters that know the user’s history and recent updates, leading to improved performance on personalization tasks.
What this means in practice
- •For chatbot developers: Create chatbots that remember user preferences consistently across multiple conversations using improved memory techniques.
- •For customer support platform teams: Integrate language models that maintain personalized user interaction history to provide tailored responses over repeated sessions.$Commercial implications: Enables selling personalized support assistants that improve customer satisfaction by retaining user history effectively.
Authors
Yuhao Chen, Shuochen Liu, Jiayao Shi, Jian Hong, Chen Cheng, Xinyun Ding, Tao Wang, Ya Li, Quan Liu, Tong Xu
Abstract
Long-term personalization requires language models to use interaction history to track users' preferences across sessions. Parametric memory encodes this interaction history into model parameters or adapters, reducing the need to include it in the inference context. However, independent context compilation leaves cross-session integration unspecified, while recurrent updates can attenuate earlier evidence. To address these challenges, we propose Dual-Path Parametric Memory (DPPM). Its Evidence path directly pools representations of the interaction history to preserve earlier evidence, while its Delta path sequentially updates an associative state to capture changes. Fusing both outputs produces history-conditioned LoRA adapters that combine evidence accumulation with ordered revision. Across multiple backbones, DPPM outperforms the evaluated baselines, achieving 54.22% on PersonaMem-v2 and 86.79% on PrefEval. These results suggest that DPPM provides a simple and effective design choice for cross-session personalized parametric memory.