Sunday, September 27, 2020

Multi-task Learning and Calibration for Utility-based Home Feed Ranking

Here is the blogpost that should to follow the video from the last blogpost. Click on the image to go to Pinterest's engineering blog. In this blogpost you will find how we utilized multi-task learning (MTL) and calibration towards a utility-based ranking experience for Pinners. Using such an MTL-based ranking framework, we were able to 1) provide pinners more relevant and engaging pins, 2) increase engineering velocity and 3) quickly update the ranking surface based on current business needs.