Edge

Your news feed on Facebook is essentially a summary of a selection of the actions by your friends on Facebook. It also includes a selection of the actions on the pages that you follow/like. (Liking a page makes you a fan of that page).

The range of actions on Facebook includes status updates, posts, commenting on a status update, tagging a photo, joining a fan page, or replying to an event. When you take an action, it triggers what Facebook calls an “Edge”. Depending on how it ranks, relevant content on the action, could appear on your friends’ or fans’ news feed.

The ranking algorithm scores the content for each individual, and only the ones that rank on top for that individual are shown on his or her news feed.

EdgeRank

To get a sense of what ranks high, Socialbakers, a social media analytics company, conducted a study, reported by Business Insider, covering 4,445 Brand pages and more than 670,000 posts between October 2014 and February 2015. Findings of the study revealed that video was the “most effective way to reach users in the newsfeed, driving more than twice as much reach as photo posts”.

Only an average of 4 out of every 100 (3.7%) page fans got to see a photo post, compared with, videos that garnered 8.7% on average. Links and text-only (defined by Socialbakers as “status”) posts follow with organic reach averaging 5.3% and 5.8% respectively, though their ranking varied over the course of the study.

Considering that the use and popularity of videos has grown with the penetration of high-speed internet, it is not surprising that they have supplanted photos, which used to garner high organic reach some years back. Having said that, the rapidity of the growth suggests that Facebook may consciously be driving the use of video.

In 2010, Facebook spelt out three factors — affinity, content weight and decay — used by their algorithm to rank content, and a simplified version of the algorithm was presented as:

$$\sum_{edges} u_e w_e d_e $$

Where ue is user affinity, we is content weight and de is the decay.


This algorithm, which was called EdgeRank, is no longer in use since 2011. According to company sources, though the new algorithm is based on machine learning, and though it takes a large number of elements into account, the original three factors remain important. It remains useful, therefore, to understand their role in the context of the EdgeRank algorithm.

Affinity is a measure of the strength of the connection between the user and the content creator. For instance if you write frequently on a friend’s wall or regularly interact in other ways, and you have a large number of mutual friends, your affinity score is likely to be high.

Each interaction has a different weight. Commenting, for instance, carries greater weight than liking, which outweighs clicking.

Taking your social graph into consideration, the affinity score, in an iterative manner, depends also on your friends’ actions, and on their friends’ actions. The weights differ depending on the strength, frequency and recency of the interactions with the friends.

And to reflect recency, the interaction weight is adjusted by multiplying with the reciprocal of the time (1/t, where t is the time) since the interaction occurred.

Importantly, affinity score is one-way. It recognizes that the extent of your interaction with a friend is not the same as the extent that he or she interacts with you.

Weight: Each category of edges has a different default weight. Somewhat similar to the standard for computing affinity, commenting carries greater weight than liking, which outweighs clicking. Based on a number of studies, videos have higher weight than links, which weigh more than photos.

Decay: Stories are weighed by multiplying with the reciprocal of the time (1/t, where t is the time) since the action occurred. Decay is also dependent on the length of time since the user logged into Facebook and the frequency of logging in. For instance, if you are very active on Facebook, the decay would be faster, so that you do not keep seeing “old news”.

One improvement with machine learning algorithm is the inclusion of global interactions in addition to personal interactions. This means that if some content gains heavy traction globally, it would raise the ranking of the content even for users whose personal interactions with the content might not be strong.

Relationship settings also play a bigger role. Though algorithms are capable of accurately deducing affinity, Facebook gives weightage too to the relationship labels, i.e. “close friend” versus “acquaintance”, set by their users.

The weight for category of content is adjusted at the individual level as well now. So for instance, if your past behaviour suggests a greater propensity to comment on photos, you will have higher content weight for photos.

The “hide post” and “mark as spam” features introduce additional variables that contribute to content ranking. For these variables, a decay parameter is introduced so that choices made in the recent past have a bigger impact.

Interactions with Facebook ads contribute yet another stream of variables that feed into the machine learning algorithm, and contribute to the ranking of content.

Relying on a vast number of variables, Facebook’s machine learning algorithm works in a manner that ordinary people find hard to comprehend and, therefore, difficult to contest. It protects Facebook from the scrutiny of their numerous stakeholders with their diverse agendas.

Previous     Next

Note: To find content on MarketingMind type the acronym ‘MM’ followed by your query into the search bar. For example, if you enter ‘mm consumer analytics’ into Chrome’s search bar, relevant pages from MarketingMind will appear in Google’s result pages.