Method

InfoNCE

Contrastive loss with in-batch negatives (NT-Xent).

13 cases from 7 companies apply InfoNCE. 6 were checked in an online A/B test; 3 report results in production.

Typical tasks: Retrieve and rank from millions of items, Learn user and item embeddings, Recommend from what the user just did, Recommend to new users and new items and 4 more. Companies: Yandex, LinkedIn, VK, Pinterest, Avito, eBay and 1 more. Often combined with: Two-tower model, Embedding fine-tuning, Sequential recommendation, Knowledge distillation, In-batch negatives and 35 more.