How Does Personalized Memory Shape LLM Behavior? Benchmarking Rational Preference Utilization in Personalized Assistants

TL;DR

This paper introduces RPEval, a benchmark for evaluating personalized memory in LLMs, revealing irrational personalization issues and proposing RP-Reasoner to improve memory utilization through pragmatic reasoning, enhancing user experience.

Contribution

The paper presents RPEval for assessing personalized memory in LLMs and introduces RP-Reasoner, a pragmatic reasoning approach that improves memory selection and reduces irrational personalization.

Findings

RPEval uncovers widespread irrational personalization in LLMs.

RP-Reasoner significantly outperforms baselines on RPEval.

The method resolves 80% of problematic cases in commercial assistants.

Abstract

Large language model (LLM)-powered assistants have recently integrated memory mechanisms that record user preferences, leading to more personalized and user-aligned responses. However, irrelevant personalized memories are often introduced into the context, interfering with the LLM's intent understanding. To comprehensively investigate the dual effects of personalization, we develop RPEval, a benchmark comprising a personalized intent reasoning dataset and a multi-granularity evaluation protocol. RPEval reveals the widespread phenomenon of irrational personalization in existing LLMs and, through error pattern analysis, illustrates its negative impact on user experience. Finally, we introduce RP-Reasoner, which treats memory utilization as a pragmatic reasoning process, enabling the selective integration of personalized information. Experimental results demonstrate that our method significantly outperforms carefully designed baselines on RPEval, and resolves 80% of the bad cases observed in a large-scale commercial personalized assistant, highlighting the potential of pragmatic reasoning to mitigate irrational personalization. Our benchmark is publicly available at https://github.com/XueyangFeng/RPEval.