Personalized item views improve sequential recommendations over long histories
PRISM: Personalized Item Representations in User History
Information RetrievalArtificial Intelligence
Summary
Recommendation systems usually treat items the same way for every user, giving each item a single representation. This paper presents a way to make item representations personalized based on each user's unique relationship with the item. The authors build a system called PRISM that creates these personalized embeddings from user history, improving how well the system can predict what a user wants next. Their tests on a large real-world dataset show better results as more user history is included.
What this means in practice
- •For ecommerce platform teams: Enhance user purchase predictions by generating item representations tailored to user history, improving recommendation accuracy over long browsing sessions.$Commercial implications: Enables product recommendation features with personalized item views that can boost sales and engagement on ecommerce platforms.
- •For digital content providers: Improve sequential recommendations by integrating personalized item embeddings that adapt to individual user-content interactions within user histories.
Authors
Renzhi Wu, Zikun Cui, Junjie Yang, Li Sheng, Yongjun Xie, Jerry Fu, Nipun Mathur, Sri Reddy, Hong Li
Abstract
Sequential recommenders typically construct history inputs from item identifiers or content features using an embedding map shared across users. The same item therefore yields the same initial embedding representation, even when users have different relationships to the item. We propose personalized item representations in user history, where each item's embedding expresses its relationship to the user. Personalization becomes a property of the history inputs themselves, supplying richer evidence for later sequence modeling. We instantiate this idea with PRISM, which reuses upstream ranking computation to generate personalized item embeddings. An autoencoder jointly compresses all layer states from a user tower, a user--item tower, and same-request candidate mixing into one embedding per scored pair for downstream replay. Offline evaluation on Surface 1, a major recommendation surface at billion scale, reports an average downstream normalized entropy (NE) change of $-0.15\%$, compared with $-0.07\%$ for VISTA at similar history lengths. We further identify an empirical sequence-length scaling law: over history lengths of 26--240 events, NE gain follows a fitted power law with exponent 0.468, increasing from 0.06\% to 0.17\%. Therefore, personalized item representations in user history provide a new scaling dimension for improving recommendation quality.