New MemRerank framework helps AI shopping agents personalize product recommendations by efficiently distilling purchase history into concise memory signals.

MemRerank distills user purchase history into query-independent signals to improve LLM-based product reranking, addressing issues with noisy and irrelevant raw history data

Researchers built an end-to-end benchmark using a 1-in-5 product selection task to measure both memory quality and downstream reranking performance

Memory extractor trained with reinforcement learning using reranking performance as supervision signal

MemRerank achieved up to +10.61 absolute points improvement in 1-in-5 accuracy compared to no-memory, raw-history, and baseline memory approaches