AFAIK from some conversations I had at a conference last year with a few Netflix engineers who work on this system, it's sort of realtime. Recommendations are build up using Apache Flink, which provides stateful stream compute.
So your click turns into an event on a Kafka topic which is eventually consumed by Flink, joined with whatever other data sources they need and then written out to a store somewhere. IIRC the various machine learning models exist inside their Flink architecture.
So your click turns into an event on a Kafka topic which is eventually consumed by Flink, joined with whatever other data sources they need and then written out to a store somewhere. IIRC the various machine learning models exist inside their Flink architecture.
https://ci.apache.org/projects/flink/flink-docs-release-1.4/...