What is cold start mitigation using embeddings?

Embeddings help cold start by leveraging item or user features instead of interaction history.

For new users or items, embeddings can be generated from metadata like text, category, or images. Pretrained models or content encoders map items into latent space, allowing similarity-based recommendations without historical interactions.