Database Reference
In-Depth Information
addresses off of the Internet? Did the agent go to Facebook profiles that were
public and take them? All of these different ways to acquire emails will show
up as different relationships between sender and receiver.
So what we need to do is determine what the relationship is between sender
and receiver to figure out if it looks like a spammer or non-spammer relation-
ship. Once again, because MailChimp, unlike other competitors, has a massive
global set of information about how emails are connected, what people are
reading, and how users have interacted with email addresses in the past, we
can bring all of that information to bear on the problem. So we can calculate
if the relationship between a user and the list they've constructed feels right
or not. We do that by building a lot of AI models around transactional data,
metadata, and email list data. So when a new user comes in, we can run them
and their list through the AI models and either permit them to go through or
shut them down immediately. The great thing about this approach is that we
can actually determine, before a user has even created content for an email,
whether we should shut them down because the relationship looks really
fishy. Of course, we communicate with them that we think the relationship is
complete bunk and that they're going to have to go elsewhere because we can
tell immediately that this looks like a spammer account.
Gutierrez:
How did you get involved with programming?
Foreman:
Back in college, I was a pure math major, and I really enjoyed it.
I thought abstract algebra was awesome. I also really enjoyed things I learned in
number theory. I just thought all of it was a blast.
But my advisor, for better or for worse, sat me down one day and said, “You
know what? You could go all the way and become a pretty average mathema-
tician in academia. You're never going to be great, but you could be one of
those guys that toils along as an average mathematician if you wanted to do
that.” I left that meeting thinking well, you know, he shot me straight. He told
me more or less where my future lies in academia, which is being average, and
that is a perfectly noble pursuit.
As I was mulling over that meeting and my future, I got a job working for
another professor in the math department modeling knot tying. Knot tying
focuses on how knots or curves that loop back on themselves interact, what
happens when you tighten them up, and how compact can they get. This
has applications in things like physical cosmology, protein folding, and other
important subjects. So we did a lot of 3D modeling of these sorts of problems,
as well as determining if you have a knot, how do you even understand what
type it is? There's a huge typing system in knots, so there was a lot of coding
necessary for this work.
