MemRerank: Preference Memory for Personalized Product Reranking

2026-03-31

AI Summary

This work addresses the limitations of directly utilizing raw purchase histories in personalized reranking for large language model (LLM)-based shopping agents, which are often hindered by noise, excessive sequence length, and weak relevance. To overcome these issues, the authors propose MemRerank, a novel framework that introduces an explicit preference memory mechanism to distill user history into concise, query-agnostic preference memory signals. The memory extractor is trained end-to-end via reinforcement learning, using reranking performance as the supervisory signal. The study also establishes a dedicated evaluation benchmark featuring a 1-in-5 selection task. Experimental results demonstrate that MemRerank achieves up to a 10.61 percentage point improvement in 1-in-5 accuracy, significantly outperforming baselines that use no memory, raw historical sequences, or off-the-shelf memory representations.

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.

Problem

• personalized product reranking
• preference memory
• LLM-based shopping agents
• purchase history
• memory distillation

Innovation

• preference memory
• personalized reranking
• LLM-based shopping agents
• reinforcement learning
• 1-in-5 selection task