How Well Does Facebook “Know” Us?
With Facebook so casually engrained in our lives, it’s easy to overlook the sophisticated structure that silently makes millions of calculations to create the user experience we come to love. In particular, the algorithm behind Facebook’s friendship recommendation is a topic that relates heavily to Networks. The reality is, Facebook’s algorithm isn’t actually all that magical. At its core, the recommendation engine uses nodes and edges, similar to ones discussed in class, and heavy mathematics to comb through tons of data and to identify meaningful connections to create recommendations.
From the moment you create your Facebook account, Facebook is collecting and monitoring all accessible data such as emails and contacts. However, the information collection does not stop there. As you populate your profile descriptions, Facebook collects more information and creates a dataset that defines your existence. All this data helps Facebook create a personalized node on its social network to represent you. From there, Facebook is able to calculate the likelihood of a connection using variables such as distance, number of shared friends, and degree of separation. Ultimately, these variables help Facebook define whether or not a connection between two nodes has a significant enough statistical possibility to recommend one node to another. However, there are a few aspects that make Facebook’s algorithm very unique. First, Facebook’s algorithm can handle unimaginable amounts of calculation at a time. This is no surprise considering just how many coefficients play a role in estimating the possibilities of connections. Second, unlike Amazon’s recommendation network, Facebook’s focus on users has allowed it to evolve its recommendation engine to cater toward human to human interaction.
The algorithm, however, is not without faults. It’s important to realize that Facebook’s algorithm, despite its large collection of data, still has no idea who you are as a person. It simply sees you as a datapoint on a map and views you through a collection of numerical and categorical data values. As explained before, the recommendation is solely based on statistical probabilities. Therefore, every estimation carries a risk of wrongly recommending a node–otherwise known as type II error.
This description may very well be just the tip of the iceberg in terms of the actual complexity of the recommendation engine. After all, Facebook is constantly updating and improving its recommendation algorithm to make it more relevant to its users. Although the idea that machines are able to “understand” or predict our behavior more and more may come across as creepy to some people, other people will see it as an opportunity to make better and more meaningful connection across social networks.
For Reference: