MemRerank: Preference Memory for Personalized Product Reranking

arXiv cs.CL / 4/1/2026

📰 NewsSignals & Early TrendsIdeas & Deep AnalysisModels & Research

共有:

Key Points

The paper introduces MemRerank, a preference-memory framework that converts users’ purchase histories into concise, query-independent signals for personalized product reranking.
It argues that simply appending raw long histories to LLM prompts is ineffective due to noise, prompt length constraints, and relevance mismatch.
To evaluate the approach, the authors build an end-to-end benchmark and evaluation framework focused on an LLM-based 1-in-5 selection task, assessing both memory quality and downstream reranking utility.
MemRerank’s memory extractor is trained using reinforcement learning, with downstream reranking performance used as the supervisory signal.
Experiments with two LLM-based rerankers show consistent gains over no-memory, raw-history, and off-the-shelf memory baselines, reaching up to +10.61 absolute points in 1-in-5 accuracy for personalization in agentic e-commerce systems.

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.