Engagement-driven curation can trap culture consumers in ‘overfamiliarity,’ leading to less long-term satisfaction

Sometimes you can be too good at your job. Take streaming services, like Spotify or Netflix. With engagement as their main goal, their recommendation algorithms typically identify the content a user engaged with most in the recent past and efficiently suggest more of the same. Content and genres that received limited engagement over a few weeks or months typically get dropped.

But as marketing professor Samsun Knight points out in new research, familiarity isn’t the same as taste, which can take longer to develop. Hip-hop took a couple of decades before it became a top seller, well beyond the timeframe streaming services allow for assessing what users like. The study, “Engagement-based curation and the evolution of taste,” appeared in the Journal of Cultural Economics.

It’s also possible to have too much of a good thing, as anyone who’s ever played the life out of a beloved song or gone to every action movie sequel knows. What once brought joy can become the last thing you want to hear, watch or read.

“The same optimization that maximizes engagement this quarter is the one that breeds staleness and churn—and eventual broader disengagement,” says Knight, an assistant professor of marketing at the University of Toronto’s Rotman School of Management. “There’s a real risk of being very good at a short-term metric while slowly degrading the long-term product.”

Knight isolates how that happens using a model of an engagement-based curation system, drawing on a longstanding economic theory that says a taste for something increases with exposure. His model shows consumer engagement grows with moderate exposure, peaks, then declines as exposure grinds on. A too-short exposure horizon is one reason. But another is that the algorithms don’t recognize that users’ past engagement has more to do with what the algorithm sent before than with users’ independently acquired tastes.

“If the platform treats people’s current tastes as a fixed fact about them, rather than something it helped create, it ends up confirming its own past choices,” says Knight. “It keeps serving the familiar, that familiar content keeps engaging well, and the data seems to vindicate the strategy.”

Another downside is that new artistic movements and experiments may die before they have a chance to be born, and users may be denied the chance to get to know and love something very different from what they’ve been accustomed to. Imagine a world without hip-hop or electronic dance music. Knight would rather not. He points out that throughout human history, stagnant artistic periods have been bad for everybody.

What’s a streaming platform to do?

Knight suggests intentionally exploring more aggressively than might seem optimal in short-term evaluations can help, over time, provide more exposure to less-familiar content and even lead to the next cultural phenomenon. While that’s happening, the content users enjoyed in the past gets a break by popping up less frequently, reducing the overexposure problem.

Real human recommendations, like staff playlists or picks, would help too. “These people will sometimes champion something difficult or unproven because they believe in it, not because the early numbers justify it,” says Knight. “That ‘irrational’ willingness to keep playing something is, in my framework, exactly the exploration the pure engagement optimizer won’t do.”

And who knows? Those oddball picks might just end up changing your world.

Who’s behind this story?

Sadie Harley

BSc Life Sciences & Ecology. Microbiology lab background with pharmaceutical news experience in oil, gas, and renewable industries.

Full profile →

Andrew Zinin

Master’s in physics with research experience. Long-time science news enthusiast. Plays key role in Science X’s editorial success.

Full profile →