Personalized language models use smaller user histories for better results
Less Is Personal: Learning Minimal Sufficient User Profiles for Personalized Language Models
Artificial Intelligence
Summary
Large language models often personalize responses by including parts of a user's past behavior, but adding too much history can sometimes hurt accuracy or waste resources. This paper explores how to find the smallest set of user data needed to keep good personalization without extra information getting in the way. The authors created a method called ENOUGH that carefully adds user records one by one and decides when to stop using more. They tested this approach on several tasks and found it works better and faster than common methods by using only the important parts of a user’s history.
personalized language modelsretrieval augmentationuser profilesbehavioral recordscontext lengthcounterfactual searchvalue controllerdistillationinferencetoken cost
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.