Ivan Lobov
13 Oct 2018
17:55 - 18:40
Stream A

Billions-scale recommendations

At the age of web-scale we want to make good personalized recommendations for billions of users and billions of items. But most of the state-of-the-art collaborative filtering approaches either cannot handle this kind of data or require a non-trivial amount of computational resources. We propose a simple approach based on SVD of a user-item similarity matrix that both achieves comparable quality results and can easily be scalable to almost any size of the dataset on a commodity cluster nodes.