Company
VK
8 cases from VK (Social, Media streaming, Adtech), each digested from a public talk or engineering post and linked to the original. 5 were checked in an online A/B test; 1 report results in production.
Tasks: Learn user and item embeddings, Retrieve and rank from millions of items, Label data with LLMs, Decide who gets an offer, Recommend to new users and new items and 1 more. Methods: ALS, Gradient boosting, A/B test, Word2Vec, Pairwise ranking, InfoNCE and 23 more. Tools: BERT, CatBoost, XGBoost, Optuna, Apache Spark, EfficientNet and 11 more.
| Ranking VK Stories | social | 2025-05 | |
| Graph Methods in Recommendations: How We Handle Large Interaction Graphs | media-streaming | 2025-05 | |
| Developing Multimodal Content Models for Recommendations | social | 2025-05 | |
| LLMs vs. Annotators: How to Get High-Quality Text Labeling | adtech | 2024-11 | |
| Experts Are Shocked: A Brilliant Strategy for Finding an Audience for Clickbait Content | social | 2024-05 | |
| From Small to Great: Finding and Growing Promising New Authors on VK | social | 2024-05 | |
| Transforming Recommendations | social | 2024-05 | |
| How We Trained a Content Video Encoder Without Labeled Data | social | 2024-05 |