Se esta mandando tu mensaje...

Case Study

How do you talk to your algorithms?

Most of us interact with recommendation algorithms every day, whether we’re exploring curated playlists, or looking for something new to watch. Video and music platforms like Netflix, YouTube, and Spotify each employ their own proprietary algorithms, designed specifically to sort out the chaff and leave us the good stuff. And as algorithms’ relevance grows in the world of media, so too does the complexity of their relationship to users.
 
We decided to dig deeper into this connection between individual and algorithm given its increasingly prevalent role in content production and consumption. We conducted a small study in which we talked to 25 people who regularly use services where they interface with algorithms for content recommendations. Our research centered on how individuals engage with and attempt to improve their algorithms – for example, by clicking on lots of content to “teach” the algorithm, or moderating their behavior in other ways to curate their identity in an algorithm’s (digital) eyes.
 
These “conversations” with algorithms and the interviews we conducted about them reveal a lot – principally about how algorithms could be improved, but also about content curation in general and the future of user-specific and user-centric programming. Flowing from the results of our study, we identified 3 user types that we describe below.
 

The Wanderer

 
Wanderers often watch or listen to content suggested by their algorithms. One participant in our study fit the bill perfectly:

“I recently was on Netflix and they recommended that I watch Amy Poehler’s directed Netflix movie, Wine Country, based on the shows I had watched prior … It was a refreshing and beautiful movie about female friendships that I truthfully don’t think I would have gone out of my way to watch if it wasn’t recommended.”

This is the easy way of falling into conversation with your recommendation algorithms. It is characterized by relaxed browsing, capricious clicking, and a certain nonchalance. If you are a wanderer, you are not oblivious to recommendation algorithms’ presence, but rather indifferent to helping or deliberately experimenting with them. Frequently, however, you are impressed by their ability to make good suggestions, leading you to discover worthwhile content otherwise off your radar.
 
The Organizer

 
Organizers spend a lot of time thinking about their algorithms, and want or need their recommendations to reflect their self-image, like this calculating user:

“Most of my algorithms are deliberately tailored. I’ve gone into profile settings on Amazon, Netflix, Hulu, Facebook etc. and selected uncommon but enjoyable interest codes, aggressively downrated, unliked, blocked or skipped videos I wasn’t interested in in order to curate my viewing experience.”

Because Organizers see their recommended videos and songs as direct readouts of their personal identities, they only make deliberate content choices that will add depth to their thoughtfully curated personal libraries. If you are an Organizer, you may have multiple accounts for separate people to ensure that others’ choices do not impact your own algorithms. You also find it concerning when recommendation algorithms suggest content you do not like or do not want to like. To combat this personal affront, you watch or listen to your favorites to get the algorithm back to “normal.”
 
The Experimenter


 
Experimenters enjoy making erratic or unusual selections just to see what happens. They know what they’re doing, like this guy does:

“I know that when I play specific types of songs (with a particular bpm for electronic and house music, for example), the Spotify algorithm will ‘remember’ that and imprint that style on future recommendations. … I know I can take advantage of [the algorithm] to find new styles and even entire genres of music by searching for and playing different songs.”

Experimenters, as the name suggests, like to experiment with a platform’s recommendation algorithm. You may be an Experimenter if you click on diverse content just to observe how your suggestions change. You may even adjust your profile settings to see how factors other than consumed content impact the recommendations you get. As an Experimenter, your behavior is driven by a desire to discover new and different content, and by your genuine interest in algorithms.
 
Concluding thoughts
 
While this list of types is far from comprehensive, we think it’s a useful jumping off point to highlighting the diversity of ways in which users interact with their algorithms. The upshot? The products of these algorithms reflect our defining features as individuals – what we like and dislike – and thus hold a special role in our ever-evolving relationship with personalized technology. Understanding how users are (and want to be) “talking to” algorithms has important implications for web and app design, content creation, and more. In our followup piece, we outline 5 recommendations for making users even more excited about conversing and playing with your algorithms.