The NVIDIA Merlin and KGMON team earned 1st place in the RecSys Challenge 2021 by effectively predicting the probability of user engagement within a dynamic environment and providing fair recommendations on a multi-million point dataset. Twitter sponsored the RecSys Challenge 2021, curated the challenge’s multi-goal optimization requirements to mirror the real world, and provided multi-million data points each day over the course of the challenge for the teams to work with. NVIDIA’s win for a second year in a row reaffirms NVIDIA’s continued commitment to democratize and streamline recommender workflows.
NVIDIA Merlin Team Weaves Industry Engagement and Learnings Into Software Product
Billions of people are online. Each moment online represents an opportunity for a person to engage with a recommender while reading news, streaming entertainment, shopping, or engaging with social media. Twitter, as a single social media platform, reports an average of 199 million monetizable daily active users that engage within its dynamic environment. The RecSys Challenge 2021 reflects how providing quality recommendations that span millions of data points is extremely challenging. The Merlin team builds open source software designed to help machine learning engineers and data scientists tackle these problems and more. The team also leveraged their skills and experience building Merlin software to win the RecSys Challenge 2021. The challenge also provided insights and opportunities that feed back into Merlin, helping to continuously improve the product. For example, operators used to win last year’s RecSys Challenge 2020 were woven into a product release. This is particularly impactful when working with the Kaggle Grandmasters Of NVIDIA (KGMON) team who are regular collaborators with the Merlin team on recommendation competitions and who bring insight from hundreds of kaggle competition wins. The Merlin team’s hands-on engagement coupled with feedback from Merlin’s early adopters, is vital for reaffirming NVIDIA’s commitment of democratizing the building and accelerating of recommenders.
For more information, download Merlin or sign up to meet team members at the upcoming DL Recommender Summit, “Develop and Optimize Deep Learning Recommender Systems“.
Editor’s note: Feature image includes just a few of NVIDIA team members that participated in the challenge (clockwise from upper left): Bo Liu, Benedikt Schifferer, Gilberto Titericz and Chris Deotte.
