The latest in AI & Machine Learning at Pinterest
With 300B human-curated ideas, Pinterest is the biggest image-rich data set ever assembled. This lets us do interesting things like analyze trends, understand intent and predict consumer behavior. And we’re just scratching the surface of what’s possible.
Tens of millions of people interact with Pinterest each day, browsing, searching and discovering ideas inspired by their tastes. To help make it all happen, Pinterest’s researchers and engineers build new computer vision models that “see” the content of each Pin, filtering abusive and misleading content, optimize ad placements and rank nearly 300 billion Pins daily.
The Ads Intelligence team at Pinterest is charged with building and maintaining machine learning and algorithm-driven recommendations and the Recommendations Ranker to provide advertisers with the best experience in reaching relevant Pinners and help them reach their campaign goals.
Pinterest is at the forefront of visual discovery and recommendations development. With the launch of Pinterest Lens, people use their camera phone to get recommendations inspired by objects they see in the real world. Our browser extension also gives them recommendations based on images they find on Pinterest, or around the web. So far, we’ve detected over a billion visual objects, and now serve 250M+ visual searches each month. But there are still many challenging problems yet to be solved in fine grained recognition, object-to-object visual search, large-scale visual search infrastructure, and a deeper understanding of mediums like video.
User modeling and recommendations
Pinterest is one of the largest-scale recommender systems around, serving users more than 10 billion recommendations every day. By combining the data we’ve amassed over the years with human curation, we’ve built human-centered personalization engines that can serve the right recommendation to the right person at the right moment, choosing from a pool of over 300 billion objects—all in real time.
To meet the changing needs of the 450M+ people who use Pinterest, we have to know both how they’re using Pinterest today, and how they’ll be using it in the future.
To do this, our data science team has created a systematic approach, which gives us trustworthy conclusions that are both reproducible and automatable. We built tools that democratize the data, so engineers, product managers and designers can easily visualize and explore trends in content and experimental results. We also track how the world is changing as people modify their behaviors, diets, health routines, homes and interests. This gives us the knowledge we need to develop the revolutionary data-driven products our Pinners will be looking for next.
At Pinterest, our PIndigenous Pinclusion Group led us in our Native American Heritage Month sessions and activations. In this blog article, you’ll meet two of our PIndigenous Global Leads and learn more about how we celebrated Native American Heritage Month.
Our engineers are building an inspired and inclusive platform for Pinterest users worldwide. In the Pinterest Engineering series, you’ll meet the people behind the product, learn about their work and why they chose to grow their career with the Pinterest team.