Нейроранжирование рекомендаций в Авито — Хныков Ярослав, Авито

Yaroslav Khnykov discussed the structure of recommendations on Avito's main page, their product role, and ML architecture. 00:00:05 Introduction and Context 00:00:46 Recommendation Architecture 00:02:13 CatBoost Development 00:04:12 Transition to Neural Networks 00:05:04 Feature Encoding 00:07:59 First Model and Experiments 00:09:50 Second Model and Improvements 00:12:06 Reintroducing Embeddings 00:12:49 Metric Analysis and Ranking Issues 00:13:48 Improving Metric Balance 00:14:46 Model Blending 00:16:13 Using User History 00:17:10 Details of the Transformer 00:20:03 Results of the Fifth AB Test 00:21:06 Summary and Conclusions 00:23:19 Q&A AvitoTech is a team of Avito engineers. Subscribe to our channel, social media, and blogs to learn more about Avito technologies 👇🏻 VK: https://vk.com/avitotech Telegram: https://clc.to/-YHGlg Habr: https://habr.com/ru/company/avito GitHub: https://github.com/avito-tech RuTube: https://rutube.ru/channel/30462632/ Zen: https://dzen.ru/avitotech Website: https://clc.to/l5ZqRA