Method
t-SNE
Nonlinear neighbor-preserving 2D visualization of embeddings.
6 cases from 5 companies apply t-SNE. 2 report results in production.
Typical tasks: Learn user and item embeddings, Recommend to new users and new items, Catch fraud and abuse, Moderate content at scale and 1 more. Companies: MegaFon, Walmart, Yelp, CFT and 1 more. Often combined with: Gradient boosting, Embedding similarity matching, Threshold selection, Matrix factorization, UMAP and 14 more.
| Single AI View of Customer: A retailers guide to Know Your Customer Better using Customer Embeddings and Customer Persona | retail | 2025-03 | |
| Yelp’s AI pipeline for inappropriate language detection in reviews | marketplace | 2024-03 | |
| Please Don't Shout. ML vs Operators: 1-0CF | CF | general | 2023-05 |
| Seismic Data Through a ResNet18 LensGN | GN | manufacturing | 2023-05 |
MegaFon RecSys Workshop, part 2: a two-level recommender (LightFM + CatBoost) and DSSM models![]() | ![]() | media-streaming | 2022-06 |
MegaFon RecSys Workshop, part 1![]() | ![]() | media-streaming | 2022-06 |
