Personalized language models learn minimal user profiles to save computing costs
Less Is Personal: Learning Minimal Sufficient User Profiles for Personalized Language Models
Artificial Intelligence
Summary
Language models can give better, more personalized answers by looking at a user’s past behavior, but not all past data helps. The researchers found a way to pick just enough relevant user history each time, instead of loading a fixed amount that might be too much or too little. Their method, called ENOUGH, learns when to stop adding more user records based on how useful it is for predicting the best answer. This approach makes personalized models faster and cheaper while still being accurate.
What this means in practice
- •For chatbot developers: Create chatbots that send only the necessary user data to the model, reducing response time and computational cost without losing personalization quality.
- •For recommendation system engineers: Optimize user data retrieval in recommendation engines by dynamically selecting minimal relevant user history for better efficiency and user-specific results.
Authors
Minghang Liu, Qiang Qiu, Yuanzhuo Wang, Huawei Shen, Xueqi Cheng
Abstract
Retrieval-augmented personalization enables large language models to produce more accurate and preference-aligned outputs using relevant records retrieved from user histories. Personalized language models typically prepend a fixed number of retrieved user records, even when additional history is redundant, harmful, or unrelated to a user's distinctive behavior. We study minimal sufficient personalization: constructing the least costly ordered profile for each input while preserving the utility achievable from a retrieved candidate pool. We introduce ENOUGH, a method that iteratively appends behavioral records or emits STOP to construct profiles with adaptive lengths. Offline, bounded counterfactual search evaluates profile prefixes by jointly considering downstream gains, user specificity, and token costs. The resulting long-horizon targets are distilled into a multi-head value controller with explicit ranking and stopping supervision. At inference, the controller selects and orders records through lightweight decisions, and the frozen generator is invoked once after stopping. Extensive experiments on six personalized tasks demonstrate that ENOUGH consistently outperforms strong heuristic and retrieval-augmented baselines in both effectiveness and efficiency, achieving minimal sufficient profiles that preserve personalization utility while reducing unnecessary context costs.