Representation Learning
At the core of Pinterest's innovation lies our Taste Graph, a dynamic network of over 80 billion connections that captures rich relationships between Pins, boards, users, queries, and products through hundreds of billions of first-party user actions and human curation signals.

We develop novel algorithms to mine this unique Taste Graph and represent every entity in the graph as powerful embeddings so that they can be used in any downstream applications at Pinterest. These high-quality vector representations are the foundational layer for all personalized experiences. Here are some of our latest works that leverage this foundation: