At BuzzFeed, it’s our mission to keep our social media audiences maximally engaged, by leveraging the hundreds of original articles, videos, and graphics we produce daily. With our vast network of social media distribution channels, we have a team of human curators who handpick the content that is most relevant and interesting to each channel’s audience. Of course, as our content production scales, curation becomes increasingly difficult — more and more content slips through the cracks and don’t reach their full potential audience. To this end, we’ve built a curation service, powered by natural language processing and deep learning techniques. It learns to: 1. Find the most relevant content for each channel 2. Identify and resurface content with evergreen appeal 3. Determine the optimal times in the day to post each content I’ll discuss the implications of this use case, how we developed our methods, and how the service cohabitates with existing human curation.