PRIME: Large Language Model Personalization with Cognitive Dual-Memory and Personalized Thought Process

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Zhang, Xinliang Frederick, Beauchamp, Nick, Wang, Lu
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914058629808128
author Zhang, Xinliang Frederick
Beauchamp, Nick
Wang, Lu
author_facet Zhang, Xinliang Frederick
Beauchamp, Nick
Wang, Lu
contents Large language model (LLM) personalization aims to align model outputs with individuals' unique preferences and opinions. While recent efforts have implemented various personalization methods, a unified theoretical framework that can systematically understand the drivers of effective personalization is still lacking. In this work, we integrate the well-established cognitive dual-memory model into LLM personalization, by mirroring episodic memory to historical user engagements and semantic memory to long-term, evolving user beliefs. Specifically, we systematically investigate memory instantiations and introduce a unified framework, PRIME, using episodic and semantic memory mechanisms. We further augment PRIME with a novel personalized thinking capability inspired by the slow thinking strategy. Moreover, recognizing the absence of suitable benchmarks, we introduce a dataset using Change My View (CMV) from Reddit, specifically designed to evaluate long-context personalization. Extensive experiments validate PRIME's effectiveness across both long- and short-context scenarios. Further analysis confirms that PRIME effectively captures dynamic personalization beyond mere popularity biases.
format Preprint
id arxiv_https___arxiv_org_abs_2507_04607
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle PRIME: Large Language Model Personalization with Cognitive Dual-Memory and Personalized Thought Process
Zhang, Xinliang Frederick
Beauchamp, Nick
Wang, Lu
Computation and Language
Artificial Intelligence
Large language model (LLM) personalization aims to align model outputs with individuals' unique preferences and opinions. While recent efforts have implemented various personalization methods, a unified theoretical framework that can systematically understand the drivers of effective personalization is still lacking. In this work, we integrate the well-established cognitive dual-memory model into LLM personalization, by mirroring episodic memory to historical user engagements and semantic memory to long-term, evolving user beliefs. Specifically, we systematically investigate memory instantiations and introduce a unified framework, PRIME, using episodic and semantic memory mechanisms. We further augment PRIME with a novel personalized thinking capability inspired by the slow thinking strategy. Moreover, recognizing the absence of suitable benchmarks, we introduce a dataset using Change My View (CMV) from Reddit, specifically designed to evaluate long-context personalization. Extensive experiments validate PRIME's effectiveness across both long- and short-context scenarios. Further analysis confirms that PRIME effectively captures dynamic personalization beyond mere popularity biases.
title PRIME: Large Language Model Personalization with Cognitive Dual-Memory and Personalized Thought Process
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2507.04607