HuffPo: Machine Learning Yields Personalized Video Streams
Posted November 10, 2016
IRIS.TV wants to change the way you watch video. Using a computer algorithm to analyze viewing habits and content, IRIS.TV will be able to tell you what kind of videos you will want to watch based on your watch history. It's like Pandora for video.
Most video recommendations today work by analyzing what video you watch and then cross referencing what videos other people who watch that watch as well. IRIS.TV, rather, wants to analyze the actual content and use machine learning to suggest similar videos.
“We ingest the archive from a publisher, look at the content and meta data on the content, structure and classify it so that the content is more easily discoverable over time,” Garthwaite says in an interview with Beet.TV. “We match the right video to the right viewer in real time.”
This translates to several hundred million video views through the IRIS.TV Video Programming Platform each month. “Some of our customers alone have over a million videos,” says Garthwaite.
For the average publisher, about 80% of its audience “will actually leave before the first video ends,” according to Garthwaite. But the other 20% is there to watch as much as they can. “They will stick around and watch another video and basically, like science, IRIS is able to consistently drive another four to eight videos for those kind of super users we call them.”
The company believes that while the majority of video viewing has been on social media, companies and marketers experience poor unit economics and lose control of their audiences. Not surprisingly, Facebook and YouTube are some of the only video players IRIS does not work with.