Mikio Braun, Data Scientist and co-founder of Streamdrill, a start-up focusing on using stream mining technology to tackle real-time big data problems. In this video he talks about “Real-time personalization and recommendation with stream mining”
Recommendation and personalization system usually use elaborate store and batch algorithms to periodically crunch user event data like views, ratings, or purchases to compute predictions. A downside of this approach is that recommendations do not reflect the current user behavior, leading to missed opportunities in making good recommendations, or out-dated recommendations, for example when the purchase has already been made.
Mikio discusses novel systems based on stream mining algorithms which accumulate statistics on user behavior in real-time in a streaming fashion, this way always reflecting the most recent user behavior. Comparing profiles across different time-scales, one is also able to classify recent behavior which deviates from the long-term trend and might be particularly interesting. Such algorithms have applications in ad targeting, recommendation, retail, monitoring, some of which will be discussed in more detail. (Via Berlin Buzzwords)