How Alibaba Boosted Double 11 Sales with Deep Reinforcement Learning

On November 12, Alibaba's Double 11 sales surpassed 100 billion yuan in 6 minutes 58 seconds, with a peak of 175,000 transactions per second and 120,000 payments per second, reaching 120.7 billion yuan in 24 hours.

Last year the Search division won Alibaba's top CEO Award for using personalized recommendation technology; this year the technology was upgraded with cutting‑edge artificial intelligence, specifically deep reinforcement learning and adaptive online learning, boosting click‑through rates by 10‑20%.

According to Wang Zhirong, head of search and recommendation, these methods continuously train models and optimize decision engines, analyzing massive user behavior and billions of product features in real time to match users with items more efficiently, greatly improving the shopping experience.

Deep reinforcement learning, a frontier AI research area derived from deep learning and reinforcement learning, powered AlphaGo’s victory over Lee Sedol and is considered the next major breakthrough after deep learning.

Applying this technology to search and recommendation helps capture both explicit and implicit user needs; adaptive online learning captures real‑time clicks to avoid missing any demand signals, while reinforcement learning iteratively refines recommendations.

Alibaba processes over 450 million active user purchase records and billions of multi‑dimensional product features on its internal real‑time computing platform “Porsche,” which supports online learning, deep learning, and reinforcement learning workloads.

Recent years show AI and data intelligence playing an increasingly vital role in Alibaba’s shift toward a technology‑driven model, reflected in its $2.2 billion R&D investment and recruitment of leading AI experts worldwide.