An Ad Impression Forecasting Tool Built With Apache Spark
Details
OVERVIEW
GumGum's digital advertising platform generates data from billions of ad impressions per day: for example, each time anyone views a webpage that could potentially display a GumGum ad, a log is created with information about viewer demographics, publishers, etc. The success of upcoming ad campaigns depends partly on knowing how many ad impressions will occur in the near future, given certain campaign-specific restrictions.
We present a tool for addressing this problem that harnesses the cluster computing framework Apache Spark. The tool has several components. First, we use a parallelized distinct-item sampling method to reduce the past year's impression logs from over a petabyte to under 50 gigabytes. After further processing, the data enters a Spark SQL database, and users can submit queries--which often involve hundreds of restrictions--through a REST interface provided by Spark Job Server. Finally, each query result constitutes a time series, and we use a modified ARIMA model to predict future ad impression trends.
The end result is a visualization of past and future behavior, available to the user within seconds.
SCHEDULE
7:00 pm: meet, greet & eat
7:30 (ish) pm: main presentation, followed by a healthy Q&A
8:00 pm: doors close to new-entrants - so make sure to arrive before this time!
SPEAKER
Michael Williams
https://a248.e.akamai.net/secure.meetupstatic.com/photos/event/c/9/a/6/600_457311622.jpeg
• After receiving a Ph.D. in Mathematics and working for several years in academia, Michael joined the digital advertising platform Gum Gum, Inc., where he has been a Data Scientist for over one year. His current projects include a Spark-based system to forecast ad impressions, and optimizing real-time bidding (RTB) auction performance.
PARKING
GumGum is happy to offer a parking lot for attendees of this meetup. The lot is located across the street from GumGum office at:
1337 7th St. Santa Monica
Please note that the open parking lot does not have any sign indicating it's association with GumGum.
Here is a Google Maps Street view of the lot: