Build a recommender system service that will predict future purchases while withstanding the load
You are required to build a fully-fledged recommendation system. Participants need to develop a high load web service that predicts the client’s future purchase based on general information about the client and the client’s purchase history. The web server responds in the form of a ranked list of stock keeping units (SKUs for short) that the client most likely is going to buy in the next purchase.
Participants submit solutions in the form of an archived web service code. For this contest, the quality criterion is MNAP@30 (Mean Average Precision): an average precision calculated for all requests and normalized afterward. The solution should conform to technical limitations (see page “Data”). Solutions that cannot handle the required load or respond slower than expected are also discarded as failed. During the test, all requests are sent in strict chronological order.
Prize fund is 800’000 RUR. The winning team will receive 300’000 RUR; second place gets 150’000 RUR; third place receives 100’000 RUR; fourth and fifth places both get 75’000 RUR. There is also a nomination for the best-open-sourced solution with a prize of 50’000 RUR. Teams of up to 4 people from all over the world are allowed, regardless of actual physical location. You are entitled to participate in different teams in multiple Retail Hero contests.