Many of the post-election op-ed pieces focused on analyzing and criticizing Facebook and its handling of its prize property: the newsfeed. I assume that as some point in ancient Facebook history, the goal of the newsfeed was to show users everything their friends shared, chronologically, somewhat like Twitter still does it today. That changed because there were too many updates and important ones were missed. So Facebook decided to personalize its newsfeed by looking at select engagement signals, analyze those signals across users to understand what each user liked, and optimize that process continuously.
Today’s newsfeed is a curated feed of updates from friends and followed organizations, selected by an algorithm that optimizes for engagement. People tend to click and like links and updates that match their existing opinions, prompting Facebook to show them more of those those, generating, in turn, more likes. This ends up creating a filter bubble, an “epistemic closure that comes from only seeing material you agree with on social platforms.” This is personalization taken to an extreme.
This doesn’t mean that personalization is bad. In today’s “world of infinite information and limited attention” it is necessary in many products, especially those that deal in news, music, video, user-generated content and almost any update. The question is to what degree, and for that I propose thinking about a personalization scale, where zero is none and 10 is nothing but personalized content.
Think of Facebook as 11.
Let’s look at news as an example. A zero is the New York Times homepage: nothing is personalized, everything on that page was picked out by a human editor who has no commitment to engage the reader. It’s the old media model for the most part. On the other extreme, Google Now updates seem to be a 10. Google Now shows users a short list of news stories, each one defined by an interest that the user has displayed in the past. Sadly though, this turns out to be a very short list of topics that are very limited in scope and that keep repeating. Not only do I see repeating topics, I know that clicking an article will just reinforce that topic in Google Now’s algorithm, bringing me more of the same. That said, it’s easy to dissect Google Now because it kindly tells me what interest I’ve exhibited prompted the story to be included in the feed. It’s also very easy to customize the newsfeed and to remove topics and sources.
The point is that for any product that curates content, product managers need to find a sweet spot somewhere in the middle and remember that a product doesn’t have to choose only one place on the scale. For music, for example, listeners sometimes want to listen to their chosen artists or specific songs, sometimes they want to listen to what they’ve liked before (10,) sometimes just in the style of what they’ve liked before (around an 8,) sometimes from a genre they often prefer (closer to a 5,) and sometimes just whatever is on a DJ-curated playlist (a 1.)
For video/TV it’s the same: sometimes viewers want to watch their favorite team no matter what (9,) sometimes it’s what’s new from YouTubers they like (an 8?) sometimes just whatever was popular today (a 4,) and sometimes they just want to see what’s on now (0.)
A final note, I don’t think it’s important to give it each feature a precise number, but rather to realize where the offering is on the scale, what combination users are offered, and whether that is what they want. Even giving users more control may not be the best solution. Even though they may be actively customizing their selection, there may be times when they want content that is editorially curated and has nothing to do with their personal preferences.
They’re human, after all.