Recommendations
Real-time neighbors for items, users, and sessions — with business rules in the same query.
The problem
A batch recommender is stale by lunch, and the business rules live in a second service that re-sorts a huge candidate list.
How retrieval closes it
Recommend from the vector of the item in front of the user, then filter stock, region, and brand while the graph is walked.
Three features that matter
- Recommend and discover APIs
- In-stock and region filters in-graph
- MMR when the top hits are near-duplicates