Modeling of long history data suffers from long-context window attention dilution, system efficiency and catastrophic forgetting problems, where naive linear scaling approach like LastN would fail. We introduce Memento, a personalized retrieval-augmented framework that treats historical user engagements as a document corpus and ad requests as queries, retrieving relevant interactions via Maximal Marginal Relevance (MMR) to balance similarity with diversity. We identify two complementary applications: Representation Memento, which retrieves historical embeddings for feature augmentation, and Data Memento, which retrieves past training examples for multipass training. Through infrastructure co-design -- temporal chunking, INT8 quantization, and asynchronous serving -- Memento achieves 5-10× resource efficiency over linear scaling. Memento processes daily requests with sub-10ms latency, yielding 0.25-0.3% Normalized Entropy gain on both click-through and conversion prediction. In production, Memento delivers a 1% CTR lift on Facebook Feed and Reels and a 1.2% CVR lift, scaling personalization to 365+ days of history.
@article{arxiv.2605.24051,
title = {Memento: Personalized RAG-Style Long-Retention Data Scaling for META Ads Recommendation},
author = {Xiaoyu Chen and Ruichen Wang and Jieming Di and Suofei Feng and Nafis Abrar and Lilly Kumari and Tony Tsui and Yilin Liu and Yu Lu and Sowmya Patapati and Junwei Xiong and Qiao Yang and Dorothy Sun and Yang Cao and Victor Chen and Pan Chen and Ramsundar Sundarkumar and Shivendra Pratap Singh and Arnold Overwijk and Ling Leng and Dinesh Ramasamy and Sri Reddy and Robert Malkin and Sandeep Pandey},
journal= {arXiv preprint arXiv:2605.24051},
year = {2026}
}