3 papers
cs.IR2026
LLMs Need Encoders for Semantic IDs Too
Xiangyi Chen, Zelun Wang, Xinyi Li +3
Multimodal LLMs use dedicated encoders to bridge non-language modalities (vision encoders for images, depth models for audio codec tokens) because raw token embeddings alone cannot…
cs.LG2025
PinFM: Foundation Model for User Activity Sequences at a Billion-scale Visual Discovery Platform
Xiangyi Chen, Kousik Rajesh, Matthew Lawhon +9
User activity sequences have emerged as one of the most important signals in recommender systems. We present a foundational model, PinFM, for understanding user activity sequences…
cs.IR2025
Multi-Faceted Large Embedding Tables for Pinterest Ads Ranking
Runze Su, Jiayin Jin, Jiacheng Li +16
Large embedding tables are indispensable in modern recommendation systems, thanks to their ability to effectively capture and memorize intricate details of interactions among diver…