Using algorithms to stop spam on Facebook
Condliffe, Jamie. “Stopping Fake Accounts Is a Cat-and-Mouse Game. Can Facebook Win with AI?” Protocol, 4 Mar. 2020, www.protocol.com/facebook-machine-learning-fake-accounts.
Facebook devised a new machine learning system called Deep Entity Classification to detect fake accounts on their platform. One of the ways it determines whether or not an account is genuine includes studying over 20,000 features that belong to the acquaintances of the account. Other measures include measuring the volume of requests from a particularly suspicious IP address, as well as noticing patterns of initial account behavior that have been repeated by previously suspended accounts. The task has proven to be more difficult as time goes on – like a deadly virus, bad-practices mutate into practices that allow their activity to slip through some unbeknownst cracks. The quality of spam accounts have gone up, making the chase to catch them all a despicably onerous task. However, the push for Facebook to strengthen their boundaries against malicious activity remains strong at a peak time of social unrest across the world: not only are spam accounts linked to running scams, but they can also be agents for inciting violence and organizing terrorism.
For this reason, Facebook has gone to the next level: instead of studying the features of the account itself, it has taken to study “deep features” that are out of the account’s control. For example, it looks at branches on its social graph and studies those users, groups and pages, allowing for a much more thorough weeding process that combs through data gathered from all across the social network. Ultimately, it seems as though managing spam on social media will be an inevitable evil, just as spam email continues to make its rounds at an albeit lower profile than a decade ago. Such an issue is a direct example of how studying networks and social graphs is vital to improving digital democracy, and can be used to benefit and protect society as a whole.