Context About Classifiers: How YouTube uses machines to help make decisions about videos

With 400 hours of new video uploaded to YouTube every minute, we need a way to evaluate videos at scale to ensure they are safe for viewers and advertisers. So we enlist the help of machine learning – but we don’t always get everything right all the time. Wanted to break to down how machine learning works for you.

Important: Our classifier work is an ongoing process.

(2:33): Example of a machine learning classifier loss – white noise
(3:04): Importance of appeals process
(4:00): Self-Certification experiment – Ask creators to give us as much info as possible about their videos to help better train the algorithms
(5:17): Even Creator Insider has been affected
(5:45): Why YouTube doesn’t 100% rely on humans for ratings
(7:30): Some classifiers are easier or harder than others
(8:17): Concentrations of areas where YouTube gets it wrong
(8:56): Even Tom gets frustrated!
(9:53): Overview of how classifiers work


🔎 Links Mentioned:
Self Certification Video:


Please help transcribe and/or translate this video so more Creators
can benefit here:

Thanks! Team Creator Insider


Creator Insider is an informal YouTube channel to share information from the YouTube Creator technical team with the wider Creator community. We will feature different people talking about the products they work on and changes we are making so you have more context. Please note this is not an official YouTube channel and is an experiment.

Leave a Reply

Your email address will not be published. Required fields are marked *