A few days ago I was browsing Youtube when I discovered something odd about the videos in the Trending tab — I couldn’t find a trend. First I looked at the content: cutlery ad, Buzzfeed video, a 13-second clip of somebody’s car getting hit by a piece of ice, movie trailer, news clip. I gave up after the 6th video, so I decided to try the number of views: 20 million, 70 thousand, 2.5 million, 800 thousand views. The view count was relatively high, but no clear trend there either. Then I looked at the videos’ popularity in terms of the likes to dislikes ratio: First video: 190,000:41,000. Second video: 120,000:5000. Third video: 130:163, Fourth video: 13,000:36,000. No real correlation with popularity either.
The only possible explanation I came up with was the time at which the video was posted, since all trending videos had to have been posted within the last couple days. But in a statistic released several months ago, it’s estimated that around 300 hours of videos are uploaded to Youtube every minute. Now, I realize that the majority of this content would be filtered out of Trending for obvious reasons, but there are a lot of highly popular or emerging Youtubers who upload videos daily and are rarely ever seen on the trending tab. So what exactly is going on here?
Social Algorithms: A Very Brief History
Before we answer this question, it’s important to know a little bit about social media algorithms and how they came about. The story began in 2006 with Facebook, when they created their first recognizable newsfeed. This allowed users to endlessly scroll through posts made by their friends and liked pages (prior to this, Facebook consisted mostly of profiles and pages). The newsfeed began by displaying information in reverse chronological order — the newest posts first, and the oldest posts buried underneath. However, Facebook realized that this was not the best way of doing things, as users complained of having to weed out all the irrelevant posts to find the things they actually cared about. They tried tweaking the content manually to comply with the numerous complaints, but for the most part, nothing really helped the situation.
In 2007, they tried resolving the issue by implementing one of the first semi-successful social media algorithms: EdgeRank (adapted from Google’s PageRank). The idea behind EdgeRank was to order the content of a user’s newsfeed on several key principles:
- The interactions (comments, likes, etc.) between Facebook friends and pages. In explanation, the more I share, comment on, or like a particular person’s/page’s posts, the more that person’s/page’s activity will show up on my newsfeed.
- The type of content a Facebook user is likely to look at or post (like a picture, video, or a status update). For example, if I started to interact primarily with videos on facebook (via likes, comments, or shares) and not pictures or written posts, then my newsfeed would become dominated by more video content.
- How quickly Facebook posts become irrelevant to users. In other words, the decay rate at which the post is viewed, liked or shared over time.
From these three principles, the algorithm orders newsfeed content based on relevancy instead of time or popularity. The result was an accomplishment but not entirely a success. Many users continued to complain that their newsfeeds displayed irrelevant content, and Facebook received backlash for diverging so much from their old platform. Nevertheless, with many thousands of new people registering for an account on Facebook, it made sorting the endless sea of information easier than ever before. Since 2007, Facebook has dropped EdgeRank for a more powerful algorithm that considers tens of thousands of variables (though the original three are still considered important). The new algorithm essentially reviews all of the things you’ve ever done on Facebook and makes predictions about what you would like to see on your newsfeed.
Facebook has changed their algorithm several times to make the newsfeed more relevant to users with some success.
Similar stories can be told by Twitter and Instagram. Both social media platforms once had a newsfeed in reverse-chronological order like Facebook did pre-2006 and attempted to smoothly transition into the realm of algorithms. In March of 2016, Instagram said they would be implementing an algorithm similar to Facebook’s that sorted content by relevancy. Not surprisingly, many thousands of people, particularly people who required views on Instagram to maintain their outreach, did not like this change. Similarly, Twitter floated the idea that they may be switching to an algorithm as well, and were met with all kinds of backlash. However, unlike Instagram and Facebook, they gave the option for people to turn off this feature if they so pleased.
Everything is About Business
From face value, it may seem that Facebook, Instagram, and Twitter are changing their algorithms to increase user satisfaction, and this is true. But what’s often missed is that the majority of decisions made by social media sites are based on strategic business decisions. Like any other enterprise, social media sites want to increase revenue, and in order to do that, they need get as many people to see the ads displayed on their sites as possible. Advertisement companies and social media outlets work in a seemingly mutualistic relationship: social media gets people to see ads, ads pay social media sites for doing so. In reality, the relationship is far more complex, and when algorithms began to be implemented, the game changed completely. Businesses (particularly small businesses who can’t afford to fork out most of their savings to invest in social media sites) attempt to take advantage of the existing algorithm and gain new customers. Playing the game right can mean increasing outreach tenfold, but social media sites don’t care about how you play the game; they just care that you’re paying them to put ads on their site. Twitter has transformed the social business landscape even further by making vast amounts of user information a commodity that they can sell to enterprises (a process called data licensing). Twitter doubled their revenue in about one year using this approach.
So what does this all have to do with Youtube?
Since its launch, Youtube has slowly made their way to appearing like a social media outlet similar to Facebook, Instagram, and Twitter, equipped with their own algorithm. Recently, their algorithm changed from prioritizing videos that have been clicked a lot, to videos that have longer watch time (the time people spend watching the video).
This seems reasonable — you would want to reward videos that people actually watch instead of click-baiting. But not everyone is happy. According to popular Youtubers like PewDiePie (with over 50 million subscribers) who has made a multi-million dollar business from his videos, something else is going on. In particular, PewDiePie has shown that views for his video uploads within the last several months have dropped and that many other Youtubers are experiencing something similar. PewDiePie further claims that Youtube will not explain the reasons for this sudden drop in views.
Felix Arvid Ulf Kjellberg (known to his fans as PewDiePie) has repeatedly called out Youtube for having transparency issues with their algorithm that they are unwilling to address. (Image Soruce: Unilad)
One reason for this may be that “watch time” does not actually mean the amount of time watched, but a combination of several factors including views, the frequency of uploads, the duration of sessions (the amount of time a user spends on Youtube in one sitting), several other variables, and then, of course, actual watch time. Watch time also has an extreme bias for longer videos, meaning that any short video (even if it is watched all the way through) is in danger of gaining no popularity over time.
Another theory is that new restrictions may be playing a big role in the issue as well. In a video recently uploaded by Youtube’s own channel (with comments disabled), they explained how “volunteers” could police new videos in order to gain points and move up through levels in a project called Youtube Heroes. Such a project would involve flagging and reporting videos en mass that, in the opinions of those users, could cause potential damage to Youtube channels through eventual video demonetization or removal. Immediately, the Youtube community exploded in outrage and the video was subsequently flagged out of irony by a number of users. The last time I checked, the video’s dislikes outweighed the likes 32 to one, which is about eight times worse than the like-dislike ratio for Rebecca Black’s infamous Friday music video.
While it received a lot of criticism, the Youtube Heroes project did have the prospect of creating a community of people, rather than an algorithm, to decide which videos to report or flag. After all, social algorithms are never perfect and require constant adjustment even with the best intentions in mind.
In general, Youtube isn’t that transparent about how their algorithm works, but that doesn’t mean that they’re trying to cause panic on purpose. Like all social media outlets, it can be very difficult to give viewers what they want, especially when there are so many of them. As a site that now upholds a community of over a billion people, Youtube has the best interest of keeping those people on their site. This being said, a bit more transparency may help to gain back the trust from their users. Perhaps with the increase in Artificial Intelligence technology, social algorithms may transform into something that will make everyone happy. Until then, I will take the content on social media with a grain of salt, and scroll on.
