Elon Musk vs Spam Accounts: Using Networks to Identify Twitter’s Spam Accounts
Twitter vs. Musk: Musk’s Opposition to Expedited Proceedings
On problem set 1, we were asked to identify the spam account in a network of 14 nodes. What if we were asked the same question for a network of 100 nodes? 1,000 nodes? What about 330,000,000 nodes? This is the problem facing Twitter in their ongoing legal battle against Elon Musk.
In April, Twitter accepted Elon Musk’s offer to purchase the company for $54.20 per share, a $44 billion price tag that Elon originally thought to be fair given the data he was presented with. He has since challenged Twitter’s methods of identifying spam accounts and their claim that less than 5% of daily users are spam accounts; as a result, he is turning to the courts to back out of the deal. The official legal documentation mentions that Twitter’s spam identification process is conducted by human beings rather than AI or machine learning algorithms, and it consists of the random sampling of 100 accounts per day, which is just 0.00005% of the platform’s daily users.
Given what we know about networks and triadic closure, it is essential to analyze the edges between all the nodes to identify local bridges and triadic closure. While on the problem set, we were able to completely analyze the network visually, seeing that there was no triadic closure among any of the connections with the spam account and how the friends of the spam accounts had no common friends, this process becomes impossible for hundreds or thousands of nodes. To thoroughly screen for spam accounts would certainly require automation and a lot of computational power to cover as much of the 330 million-node network as possible. With the resources, technology, and capital available to Twitter, it is concerning that it turns to human random sampling instead of machine learning when it comes to something as integral as identifying bots among the network.
Being able to accurately analyze a network and identify spam accounts, in the case of Twitter, has multi-billion-dollar consequences. Certainly, if less than 5% of Twitter’s users are spam accounts, the $44 billion dollar price tag seems feasible given the potential advertising revenue generated by engagement from the other 95% of the user base that are actual human beings. If, however, the true number of spam accounts is closer to 20%, as Elon has publicly hypothesized, that will significantly lower the market value of Twitter. For now, we’ll have to observe as the lawsuit unfolds and the courts determine if Elon’s concerns are strong enough to halt the takeover.