MemRerank: Preference Memory for Personalized Product Reranking

TL;DR

MemRerank introduces a preference memory framework that distills user purchase history into concise signals, significantly improving personalized product reranking in e-commerce systems using LLMs.

Contribution

The paper presents MemRerank, a novel preference memory approach with reinforcement learning training, and an end-to-end benchmark for evaluating memory quality and reranking utility.

Findings

MemRerank outperforms baselines with up to +10.61 accuracy points.

Explicit preference memory improves personalization in LLM-based e-commerce agents.

The framework effectively distills noisy purchase histories into useful signals.

Abstract

LLM-based shopping agents increasingly rely on long purchase histories and multi-turn interactions for personalization, yet naively appending raw history to prompts is often ineffective due to noise, length, and relevance mismatch. We propose MemRerank, a preference memory framework that distills user purchase history into concise, query-independent signals for personalized product reranking. To study this problem, we build an end-to-end benchmark and evaluation framework centered on an LLM-based \textbf{1-in-5} selection task, which measures both memory quality and downstream reranking utility. We further train the memory extractor with reinforcement learning (RL), using downstream reranking performance as supervision. Experiments with two LLM-based rerankers show that MemRerank consistently outperforms no-memory, raw-history, and off-the-shelf memory baselines, yielding up to \textbf{+10.61} absolute points in 1-in-5 accuracy. These results suggest that explicit preference memory is a practical and effective building block for personalization in agentic e-commerce systems.