The world鈥檚 most popular video website is quietly transforming itself. YouTube鈥檚 massive archive of video is being automatically organised into niche channels that cover seemingly everything, from daredevil wingsuit flying to an odd style of finger dancing called 鈥溾��.
A redesign launched yesterday will make these automated channels 鈥� not the same as existing channels that belong to users 鈥� a core part of what people experience when they log on. It鈥檚 all part of parent company Google鈥檚 efforts to get you to watch more videos online. Put simply, Google wants YouTube to be the future of television.
鈥淚t鈥檚 somewhat unlikely that someone will go out there and program 24 hours a week of wingsuit flying,鈥� says Noam Lovinsky, director of product management at YouTube. 鈥淏ut we want to make sure there is a channel for that.鈥�
Advertisement
Algorithms build the automated channels by tracking user activity. When someone enters a term like 鈥溾�� into YouTube, for example, the site records how a user navigates the search results, which video they end up clicking, which clips they go on to after that, and so on. It then scans user comments for each video, looking for words that help describe the clip.
Software also watches as users program their own channels, gathering metadata which, when aggregated over millions of users, can be used to build an interesting channel for everyone.
Content analysed
Even the content of videos is analysed by machine vision and listening software to help assign them to channels (see 鈥淥peration Finding Bieber鈥�). Detecting laughter in a clip will help refine the automated comedy channels in the new YouTube, for instance. 鈥淲e can tell that there鈥檚 music in the video, or laughing, whether it鈥檚 shot outside or inside,鈥� says Cristos Goodrow, YouTube鈥檚 head of search and discovery.
YouTube will also be pushing its users to tag video using labels from a structured library called Freebase. The idea is ultimately to make it easier for newly uploaded videos to find their way into a channel. The company hopes this will compel users to stay longer and watch more video, so it can serve more advertisements.
To help spark interest in the new channels, YouTube is aping a concept that鈥檚 been around in traditional TV for years: the DVR box. As of 6 December, a personalised panel called 鈥淭he Guide鈥� has begun following users around on YouTube, gently pushing channels to them based on their activity, as well as helping keep track of new content from channels they already follow. A user who searches for videos about the US Open tennis tournament, then proceeds to watch nothing but Roger Federer clips, might be pushed to subscribe to the Roger Federer channel.
Ultimately, YouTube鈥檚 engineers are trying to create a new formula for building television channels, one that relies on the entire YouTube-watching community for programming advice. But first it needs to classify its vast amount of content.
鈥淒iscovery is the absolute number one challenge for YouTube,鈥� says Rich Greenfield of BTIG Research, the research arm of investment firm in New York City. 鈥淎utomated channels is their attempt to simplify it.鈥�
In many ways, the site is at the same point now as cable television was back in the early 90s, he adds. 鈥淚f you鈥檇 asked people back then whether cable television would ever have the same level of content as broadcast, they鈥檇 have laughed.鈥� His guess is that 鈥渋t won鈥檛 take YouTube 30 years鈥� to assert itself as a major competitor with today鈥檚 television channels.
Operation Finding Bieber
The X-Factor musical talent quest of the future may be judged by a machine. While interning at Google in Mountain View, California, of the University of Indiana in Bloomington developed a system that analyses the quality of the music in home videos uploaded to YouTube. The system analyses the audio for harmonious chords and tight rhythms, and is able to make a basic distinction between high and low quality music.
