Authors: Peter Lenz*, Dstillery
Topics: Spatial Analysis & Modeling, Economic Geography, Geographic Information Science and Systems
Keywords: GeoAI, Big Data, Economic Geography, Industry
Session Type: Paper
Start / End Time: 3:05 PM / 4:45 PM
Room: Washington 6, Marriott, Exhibition Level
Presentation File: No File Uploaded
The advent of ever more powerful GIS allows us to rethink how we model human behaviors. Instead of proxies and heuristics, we can instead look at real behaviors writ-large. By collecting at, and predicting on, the most atomic level available, we can later aggregate data upwards to create accurate predictions of human behaviors at arbitrary aggregations. This talk will outline a multi-petabyte-scale GIS developed at Dstillery Inc. over the last decade to do exactly that, using passively collected signals; namely, data from websites and cell phones from the programmatic advertising ecosystem. We'll trace how the system was originally designed to serve digital advertising, how it has gained quality control against low quality data from both a sensor network outside our control and from actively hostile actors polluting its data stream, and how we decided the best way to model the real world was to trick our models into thinking it's just an extension of the web.