Real-time Machine Learning For Recommendations
Eugene Yan
Chip Huyen discusses real-time machine learning for recommendation systems, explaining when real-time approaches make sense versus batch processing and how companies in China and the US implement them differently. Real-time recommendations are most valuable for time-sensitive, mission-centric activities like shopping and movie selection, or when dealing with cold-start problems in customer acquisition phases. The article covers collaborative filtering and Alibaba's Swing algorithm as practical approaches, noting that batch recommendations remain sufficient for most use cases despite real-time recommendations' benefits in specific contexts.
Why it matters
Why real-time? How have China & US companies built them? How to design & build an MVP?
