Shopping is at the core of Pinterest’s mission to help people create a life they love. Every day hundreds of millions of users come to Pinterest to find inspiration to decorate their home, to wear outfits on different occasions, host parties and various other things to create a better life. Shopping recommendation connects those inspiration content to actual products that users can buy and create a better life for themselves. Pinterest provides shopping recommendations across different surfaces, on closeup of a pin, on users’ boards, on search results etc.
In this talk, I’ll provide under the hood details of Shopping recommendations at Pinterest. Shopping recommenders at Pinterest are some of the most large scale recommender systems in the world. They provide most relevant and visually similar recommendations leveraging the giant pin-board graph, visual similarity, user profile and many other features. A multi-head deep neural network is used to score and select the most relevant recommendations in real-time from hundreds of millions of products. Then multiple different types of Shopping recommendation modules are put together using a multi-armed bandit framework. The audience can expect to gain a deeper understanding of how a large scale recommendation system works, how deep learning and graph convolutional network is used to determine product similarity and how multi-armed bandit is used to put together different shopping recommendation modules.