PayPal Presents SAGA: Unified User Embeddings Across Multi-Surface Actions
_reachsumit · x · 2026-08-18
PayPal introduces SAGA (Structure-Attended Generative Action Embedding), a model designed to capture cross-surface behavioral signals for recommender systems. Unlike prior models operating in homogeneous spaces, SAGA encodes interaction sequences across a financial ecosystem—checkout, P2P, in-app engagement, and email—into a unified user representation.
Core Mechanisms:
- Utilizes a per-field tokenization schema to decompose actions into field-level tokens (e.g., product, interaction, surface).
- Enables field-level attention and per-field training objectives.
Performance:
Offline ablation studies validate the design choices. Downstream models integrated with SAGA embeddings achieved significant lift in clicks and conversions across diverse touchpoints compared to alternative architectures.
More from Research
- 10 YouTube Channels Worth Bookmarking for Learning Generative AI — goyalshaliniuk · 2026-08-18
- Berkeley Lab Unveils AI Model for Realistic Earthquake Simulation — Scobleizer · 2026-08-18
- Invented or discovered? Are neural networks a fundamental pattern of reality? — drabarca_ai · 2026-08-18
- KDD Cup Winners Unify Recommendation Systems, Team Built Winning Code with DeepSeek — 量子位 · 2026-08-18
- Spellcaster Uses 6-Agent Loop to Fix 'Unplayable' AI-Generated Games — 量子位 · 2026-08-18
- HumanCLAW open-sourced: all 9 SOTA VLMs fail embodied benchmark, best hits only 16.8% — liuziwei7 · 2026-08-18