Can recommendation engines really work?

I often comment here about the fact that the emergence of the internet has enabled the production of (and the cheapening of) content. By content I mean writing, music, video, art etc. It used to be that if you wanted to hear a song you had to either buy the cd it was on, or listen to the radio and hope you heard it. Nowadays most songs can be found on Spotify, youtube, pirate sites etc. Additionally there are gazillions of content creators, myself included, posting all kinds of content on various sites like soundcloud, youtube, Noise Trade (which is now offering free books) etc.

For content consumers (e.g. most of us) this is great. Lots of choice, lots of free or cheap stuff. But there’s an obvious problem. Most content is shit. It’s actually beyond shit—it’s utterly amateurish prattling devoid of nuance or refinement. (My work is an obvious exception.) And plenty of other content is not shit, but not all that great either. Only a small percentage of content really hits the mark. So how do you weed out the crap?

One idea is by having people rank content. This is how Amazon reviews, youtube “thumbs up and thumbs down” buttons, Facebook likes and similar concepts work. But they’re somewhat problematic. It turns out there’s a lot of people out there with no taste, so you really can’t trust their opinion on anything. How do I know the person doing the ranking is the kind of person I can trust?

Amazon has kind of gotten around this with their recommendation engine. It basically follows the logic that “this guy liked a lot of stuff you liked so you’ll like this new thing he said he likes.” It’s the obvious idea that like-minded people like the same stuff.

It kind of works, I guess. But I’m starting to wonder about another issue. All these processes assume that whether we will like something is fairly static. I see a movie on Sunday afternoon and like it. The presumption is that had I seen the movie on Thursday evening, or Tuesday morning I would have liked it just the same. But what if our liking something is more flexible? What if our mood before we examined the content affects whether we like it? What if whether we just ate a good meal affects our liking it? Then it matters less what some guy who has liked things we’ve liked thought. Maybe he liked it because he just ate a delicious Fettuccine alfredo?

And, I suspect there’s some truth to this supposition. Sometimes no music or TV, no matter how good, is going to keep my interest. And there are other times when anything seems pretty amusing. I may also like something simply because I like the person making the recommendation. There’s a lot of x-factors at work that are hard to weed out of the process.

You might say that I’m saying appreciating content is subjective (e.g. it depends on the person.) But I’m really saying it’s beyond subjective. The person I will be Saturday may not like stuff the person I am today likes.

  1. No Comments