Method
Approximate nearest neighbors
HNSW, IVF, product quantization and LSH indexes for fast embedding search.
58 cases from 35 companies apply Approximate nearest neighbors. 25 were checked in an online A/B test; 20 report results in production. Includes variants: HNSW.
Typical tasks: Retrieve and rank from millions of items, Rank search results, Match entities, Learn user and item embeddings and 11 more. Companies: Pinterest, Mercari, LinkedIn, Instacart, Yandex, eBay and 29 more. Often combined with: Two-tower model, HNSW, Contrastive learning, In-batch negatives, Embedding fine-tuning and 105 more.






