Method
Multimodal embeddings
Joint image-text or audio-text embeddings such as CLIP and SigLIP.
25 cases from 15 companies apply Multimodal embeddings. 5 were checked in an online A/B test; 7 report results in production.
Typical tasks: Retrieve and rank from millions of items, Learn user and item embeddings, Rank search results, Evaluate models and 9 more. Companies: Meta, Avito, Yandex, Netflix, Sber, eBay and 9 more. Often combined with: Embedding similarity matching, Contrastive learning, Two-tower model, Approximate nearest neighbors, Self-supervised learning and 60 more.



