Data Science

The faults of recommendation.

A few months ago I watched a video from reddit featuring Logan Paul - for those who don’t know, he is an odious ‘youtube celeb’ who sells cheap merchandise to children. Unfortunately I let the video run a bit too long, so now when I go to the youtube homepage, I’m greeted by an endless pile of shitty videos by him and his kind.

The reason for this is that I watched one video by an ultra-popular creator and now the algorithms can’t help suggesting other ultra-popular crap to me. This is partly the fault of Machine Learning - statistically speaking I’m highly unlikely to not enjoy these vids that are so massively popular.

But that’s kind of the whole point. I want to see videos that are unusual, unpopular, and unpromoted. I want to see interesting and new content, not just the stuff that algorithmically optimises youtube’s monetisation flow.

I think it’s pretty simple for youtube to add some functionality to fix this:

  • Give me the option to remove the “recommended” items from my homepage

  • Let me see instead “rising” items that aren’t hot enough to be trending, and that maybe fit within my interests

The problem is that this would probably cause a short term dip in revenues for the largest creators, and this in turn creates all kinds of headaches for youtube. But in the longer term it would enable more creators to take a share of the total ad revenue available.