Jiwon's Alcove

User-Controlled Recommender Systems

Aug 4, 2022
Clusters of weighted nodes in a social network graph.

Algorithmic recommendations are a fact of life on the contemporary internet, from Google Searches to Youtube, Facebook, Instagram, Netflix, Amazon...the list goes on. With the omnipresence of recommender systems comes a widespread recognition of its downside.

Meta has recently faced backlash for swarming the feed of Facebook and Instagram users with accounts they do not follow. TikTok's recommendation systems funnel users into bigotry. The fabled Youtube Algorithm has also been subject to constant criticism.

So what if we just...gave users more control over recommender systems? Much of the complaint against algorithmic recommendations is the feeling that it is shoved down our throats. The same cookie-cutter experience is delivered to the entire userbase. Sure, some users may like an algorithmic Instagram feed, but what about the rest?

Although a poorly designed panel of knobs and buttons could be a UX disaster for non-technical users, I don't think it has to be. Youtube already gives users simple control over the topic of their feed. A toggle button to set the feed to "tried and true" or "something adventurous" mode which changes the exploration rate of the recommender system could be an intuitive feature. What if we could control the length of recommended videos? Their tone? The ratio of content recommended from various sources with wildy different content output?

In the far future, we may even be able to precisely specify our tastes using natural language. Imagine being able to say, "Hey Youtube, I would like to stop my addiction to politically divisive content", or "I don't like this Youtuber, not because of their topic matter, but because of their incessant clickbait."

Of course, none of these are conducive to profit motives, which aim to maximize advertisement revenue beyond any other measures. User-controlled recommendation systems may be developed and disseminated through alternative frontend projects like Invidious. Or perhaps there is a legal argument - which I am not qualified to make any definitive statements about - in favor of users' right to control recommendation systems. One can dream.