Learning Universal User Representations Leveraging Cross-domain User Intent at Snapchat

2025-04-30

AI Summary

Existing platforms typically model user behavior in isolation across multiple interfaces, hindering the capture of cross-interface collaborative filtering signals and limiting the modeling of complex behavioral relationships. To address this, we propose a Unified User Modeling (UUM) framework tailored for Snapchat’s multi-surface ecosystem—encompassing video feeds, Lenses, and notifications—marking the first industrially deployable cross-domain unified representation learning solution. UUM jointly embeds behavioral sequences across surfaces, aligns cross-domain user intents, and employs a lightweight online update architecture to balance domain specificity with collaborative signal fusion. Crucially, it enables multi-task shared representations—not merely feature concatenation. A/B testing demonstrates significant improvements: +2.78% long-video open rate, +19.2% average watch time, +1.76% Lens playback duration, and +0.87% notification open rate.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Application Domains: Social Networks
• Computer Vision: Multi-modal Vision

Application Category

• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Responsible Web: Human-perceived consequences of algorithmic deployment on the web
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

The development of powerful user representations is a key factor in the success of recommender systems (RecSys). Online platforms employ a range of RecSys techniques to personalize user experience across diverse in-app surfaces. User representations are often learned individually through user's historical interactions within each surface and user representations across different surfaces can be shared post-hoc as auxiliary features or additional retrieval sources. While effective, such schemes cannot directly encode collaborative filtering signals across different surfaces, hindering its capacity to discover complex relationships between user behaviors and preferences across the whole platform. To bridge this gap at Snapchat, we seek to conduct universal user modeling (UUM) across different in-app surfaces, learning general-purpose user representations which encode behaviors across surfaces. Instead of replacing domain-specific representations, UUM representations capture cross-domain trends, enriching existing representations with complementary information. This work discusses our efforts in developing initial UUM versions, practical challenges, technical choices and modeling and research directions with promising offline performance. Following successful A/B testing, UUM representations have been launched in production, powering multiple use cases and demonstrating their value. UUM embedding has been incorporated into (i) Long-form Video embedding-based retrieval, leading to 2.78% increase in Long-form Video Open Rate, (ii) Long-form Video L2 ranking, with 19.2% increase in Long-form Video View Time sum, (iii) Lens L2 ranking, leading to 1.76% increase in Lens play time, and (iv) Notification L2 ranking, with 0.87% increase in Notification Open Rate.

Problem

• Develop universal user representations for cross-domain recommendations
• Enhance collaborative filtering signals across diverse app surfaces
• Improve personalized user experience with general-purpose behavior encoding

Innovation

• Universal user modeling across in-app surfaces
• Cross-domain trends capture for user representations
• Embedding-based retrieval and ranking performance improvements