Method

k-means

Centroid-based partitioning clustering into k groups.

17 cases from 14 companies apply k-means. 3 were checked in an online A/B test; 6 report results in production.

Typical tasks: Learn user and item embeddings, Balance relevance and diversity, Recommend from what the user just did, Analyze user behavior and 8 more. Companies: Expedia Group, Netflix, VK, VTB, Airbnb, Wildberries and 8 more. Often combined with: HDBSCAN, Gradient boosting, UMAP, ALS, Logistic regression and 52 more.