Database Reference
In-Depth Information
very noisy human decision process of going and buying a pair of shoes. So the
reason we took upon ourselves the fraud detection is because we need to
clean out the data streams as they come in. So that has been a big focus over
the last year or so.
On the communication side, we have to communicate with people outside
of our group. This is where we spend time on data visualization. When a
marketer ask you how you actually find your audiences, it is really, really hard
to tell them, “Oh trust me, it is a black box that builds models in 10 million
dimensions.” That does not really fly. So being able to communicate some of
what the model is doing is a big part of our work. Lately, one of the things
we have been doing is embedding the signal from the model into a geoloca-
tion, kind of marrying the desktop with the mobile world, and actually seeing
where people have the highest probability for purchase and projecting this
into a geographical region and making a graphic out of it that people can inter-
act with. This really helps with communication.
Finally, our team and I serve as ambassadors for our company and our work.
I teach a high-level overview course on data mining for the NYU Stern MBA
program to give people a good understanding of what the opportunities are
and how to manage them instead of really teaching them how to do it. So
that is a slightly different perspective than what you get in a computer sci-
ence department. We also publish and we write papers. Increasingly, my time
has been taken up by helping to organize the KDD 2014 Conference in New
York City.
Gutierrez:
What about this work is interesting and exciting for you?
Perlich:
I have always been fascinated by math puzzles and puzzles in general.
The work that I do is a real-world version of puzzles that life just presents.
Data is the footprint of real life in some form, and so it is always interesting. It
is like a detective game to figure out what is really going on. Most of my time
I am debugging data with a sense of finding out what is wrong with it or where
it disagrees with my assumption of what it was supposed to have meant. So
these are games that I am just inherently getting really excited about.
People laugh at me because I am not a GUI [Graphical Usert Interface] person.
Because of privacy, firewalls, and other data security measures, we cannot open
any windows to look at the raw data, so it is a moot point. This means I liter-
ally have ASCII files running on my screen. I look at them and look for things
like “these numbers either are ordered when they should not be” or “this is
supposed to be a continuous number and it has too many zeros.” Things like
that get me really, really excited to figure out what is going on there. And hey,
I'm getting paid for solving puzzles, so that is absolutely perfect.
