Personalized LLM Response Generation with Parameterized User Memory Injection

ACL ARR 2025 May Submission954 Authors

Abstract: Large Language Models (LLMs) have exhibited remarkable proficiency in comprehending and generating natural language. On the other hand, personalized LLM response generation holds the potential to offer substantial benefits for individuals. However, existing work struggles with efficiently incorporating user information for LLM personalization. In this study, we draw inspirations from real-world bionic memory mechanism to propose a novel parameterized \textbf{M}emory-\textbf{i}njected approach using parameter-efficient fine-tuning (PEFT), combined with a Bayesian Optimisation searching strategy to achieve \textbf{L}LM \textbf{P}ersonalization(\textbf{MiLP}). Our MiLP takes advantage from the alignment between real-world memory mechanism and the LLM's architecture. Extensive experiments have shown the superiority and effectiveness of MiLP.

Paper Type: Long

Research Area: Dialogue and Interactive Systems

Research Area Keywords: retrieval, human-in-the-loop, conversational modeling

Contribution Types: Model analysis & interpretability, NLP engineering experiment

Languages Studied: English

Keywords: retrieval, human-in-the-loop, conversational modeling

Submission Number: 954