Tag Archives: Suggestion Engines

But I Don’t Like That

11 Jun

Do you ever look at the personalized recommendations you get online and wonder where the algorithm went wrong? Maybe Netflix suggested movies you think you’d never watch, or Pandora suggested a playlist that feels totally off the rails. And you think to yourself: but I don’t like that kind of thing. Why would they suggest this?

Well…there’s usually a reason. But sometimes you can’t spot it. Websites and apps suggest content based on previous actions. Sometimes they mess up and use the wrong signals to assume interest. Maybe you watched Bring It On to wax nostalgic with a junior high friend, but don’t actually want to see other movies from that time period or genre.

But what if sometimes you’re the one who’s wrong?

I’ve been reading a fascinating book by Tom Vanderbilt called You May Also Like: Taste in an Age of Endless Choice. Vanderbilt sets out to explain how we develop our tastes, how psychology affects preference and how the modern age of never-ending-content and recommendation engines could change all of that. He cites an impressively broad mix of experts, from developers to academic researchers to philosophers.

One of the core claims is that we know less about our tastes than we think we do. We want to believe there’s logic to what we like, but can’t truly explain all of our preferences. That’s because too many factors play into our choices, from social norms to preconceived biases to generalizations. For example: maybe you love Justin Bieber, but you’d never say so in public because you think people might judge you. Or maybe you hate opera but pretend to like it, because you think that makes you sound cultured.

Turns out, we sort of stink at describing why we like what we like. But data is the great revealer. Companies like Spotify and Netflix know what you actually like based on your consumption patterns. You may say you love foreign films, but you’ve only watched one in the last five years. Meanwhile, you swear you hate chick flicks, but watch five a week. You can say whatever you want to your friends…but Netflix knows what’s up.

Sometimes, you’re explicitly trying to save face or establish expertise. But sometimes you don’t even realize the gap between what you say you like, and what you actually like. Sophisticated data programs can spot patterns we can’t discern on our own. Netflix has a classifications system that’s way more granular than typical genres, finding detailed connections between the types of things you tend to watch, and what you might want to see next. So even if you don’t always spot a connection between what you think you like, and what they recommend—it’s probably there. Try listening to one of Spotify’s curated playlists without looking at the screen. You’ll find you like songs that fall into genres you think you hate, and you hate songs that fall into genres you think you like. You’re actually more likely to find content that suits your tastes when you ignore the labels.

Vanderbilt’s book was on my mind as I read this article about Pippa Middleton’s wedding. I almost didn’t click, because the headline talked about her wedding dress—and I didn’t think I cared. But then I figured, why not? Dresses are pretty and it’s fun to learn about them. Clicccck.

I’m glad I took the bait, because the article was more commentary than couture. It talked about how the NYT decided on its coverage for Pippa’s wedding. This paragraph made me laugh pretty hard:

NYTimes coverage

People dissed the NYT for their wedding coverage, those wedding articles actually performed really well. Of course, it’s possible there were different audiences clicking the wedding content and complaining about it. But that’s not very likely, given how media works these days. It’s just that people didn’t want to believe they were interested in wedding content. They wanted to believe they had higher class tastes than wedding fodder…even as they secretly clicked to learn more.

Creepy? Yes, of course. But also pretty helpful, in my opinion. Wouldn’t you rather have data help you find things you’ll like, rather than wading through things you hate?

%d bloggers like this: