In the upper left corner of the Esportz Network home page you will notice a new feature. Hitting play on the video will show a quick flame that gives way to one of the top highlights of the day from League of Legends or Fortnite. Powered by Sizzle.gg, the newly launched website is using AI to compile top clips from the biggest streamers in the world.
“At Sizzle we are building the SportsCenter of gaming,” said Vijay Koduri, the CEO of Sizzle.gg on the Esportz Network Podcast. “What we are doing is taking ten-hour streams from top streamers and bringing you the 20 minutes of highlights. We are building AI to detect the action, the funny, and the cool moments.”
The vast hours of streams have made this too daunting a task for other companies. By giving that task to AI, Sizzle has found a way to make combing through hundreds of hours of streams possible for a small team. But creating a custom AI comes with its own set of challenges.
How Twitch Chat Provides Data that Pinpoints Highlights
On launch, Sizzle.gg has two games available: Fortnite and League of Legends. The plan is to expand that roster further and further until it encompasses the top 20 titles on Twitch. By tackling two extremely different games first, Sizzle has made it easier to expand to other esports in the future.
“Adding more games get easier by an order of magnitude,” said Koduri. “Now that we have pretty good building blocks we can rearrange things a bit differently for each game. There will always be a few new blocks for each game of course, but it’s certainly easier than building from scratch. One constant is the audio and the chat. Streamers get excited and they get louder so we can pinpoint those moments.”
Chat is effectively an endless flow of data points. There are some emotes synonymous with good clips. MonkaS, Kappa and even Hi YouTube are great indicators of when a big moment happens within a stream. By searching for those specific terms, Sizzle is able to find highlights with greater accuracy.
“Chat provides thousands of data points,” said Daniel Yoo, the CMO of Sizzle on the podcast. “It becomes fairly predictable, when you are limited to a few emotes, you get some really quality data. As we process all this data we are learning new things.”
Sizzle.gg’s Challenge of Customizing Game to Game
“From genre to genre and even from game to game audiences are different,” Yoo explained. “The average player for League of Legends compared to a Dota 2 player is quite different – even though the two games are similar. Deciding on what types of highlights to show has been a challenge. But, as we improve the algorithm and add more games, our algorithm should do a better job of figuring out what people want to see.”
Twitch chat’s various emotes have taken on different meanings in different communities. The times people spam MonkaS in a Tyler1 stream could be different from MonkaS in a Doublelift stream which is different from MonkaS in a Tfue stream. Some players are soft-spoken even when they rack up 20 elims in Fortnite. Other streamers base their channels around being overly hyped up.
“There are just certain outliers within streamers that we have to take extra precautions,” said Yoo. “The top two that come to mind are LoLTyler1 and Dr. Disrespect. We have to do some tweaking because these two guys are so much more vocal and active than your average streamer. Compare the Doc to someone like Fortnite World Cup champion Bugha who will get 20 kills without raising his voice.”
As with any AI, one of the biggest benefits is the evolving nature of the platform. Given the absolutely absurd number of hours to comb through, AI is the only way to provide all the top highlights. As Sizzle continues to grow and expand to more esports, a slice of the Esportz Network page will grow as well.
To hear more from our new partners, listen in to the Esportz Network Podcast below. We talk about expanding to Valorant and Call of Duty, the technical challenges Sizzle has overcome and the company’s plans for the future.
Podcast and Article by Mitch Reames
Leave a Reply
You must be logged in to post a comment.