When Boko Haram kidnapped a group of school girls last spring, Winchester said, dating profile fakers would claim to be there abroad as part of a US special forces mission.
In reality, they were Nigerian con artists, hoping to be sent money to pay for a flight they would never take.“The sad reality is that the most effective scammers will tend to be the human beings who build trusting relationships over a long period of time with their targets,” says Winchester.
Suddenly I had to stop doing new features and trying to acquire new users,” in order to keep up with squashing scammers.
There was no dedicated screening service at that time, Winchester says. Well, he did along with an acquaintance, Nick Tsinonis, who already had expertise using machine learning to help match dating site users based not on their expressed preference, but on behavior.
Someone starts chatting with you through an online dating service and you enjoy the long chats with them and eventually become interested.
The most popular con-man profile text in the UK, for example is “so please i want you to get back to me here with your email address so that i can send you my pictures so get back to me thanks.” Hard to imagine swiping right on that.
Quantity of text isn’t a great indicator, says Winchester, in part because of the growing popularity of bots.
“But the operator of the bot is collecting payments for generating downloads, without ever having to interact with the user themselves.”If someone’s going to fall for a fake profile, that’s about as innocuous a result as one can hope for.
The bigger danger comes from human interaction, where, as in those familiar scam email exchanges, the person behind the profile doesn’t want your heart; they just want your money. While the UK’s favored scammer line sounds ridiculous, the top spot in the US goes to “i am very easy going and laid back.” Okay, so it’s no Pablo Neruda.
The result, Scamalytics, is a company that’s able not only to identify a number of key profile traits—in the “low hundreds,” says Winchester—but to measure how they play against one another for a more complete picture of who’s real and who’s swindling.“Features that in isolation may not give you too much information, in combination become much more powerful,” says Winchester.