Capturing and Validating Real World Movements of Millions of People
summary
Foursquare explains how owned apps power validation and produces 9 billion stops per month from 500 million devices, uses a simple provider-evaluation rule (stop is within 250 meters of a first‑party checkin = good) to reject bad supply, trains a snap-to-place algorithm combining proximity, venue popularity, personal history and Wi‑Fi signals, reconstructs venue shapes from checkins (Central Park blobs, Metropolitan Museum steps), builds a veracity score for providers, and normalizes third‑party volumes using a first‑party panel instead of static census data.
AI-written summary · about
chapters (23)
- 0:00Foursquare’s movement engine
- 1:55Location data challenges
- 3:20Consumer apps and data pillars
- 5:52Raw GPS errors and distortion
- 7:43Data acquisition and privacy
- 9:11Ground truth scarcity and bias
- 10:39Apps improving data quality
- 12:03Simplified visit workflow
- 13:29Evaluating GPS data providers
- 15:42Initial checks for providers
- 17:12Defining visits and stops
- 18:15Snap-to-place features
- 19:20Visit assignment logic
- 20:26Learning venue shapes
- 21:50Central Park reconstruction
- 22:10Met steps case study
- 23:17Normalizing visit trends
- 25:48Data quality quote
- 26:11Q&A: User recognition
- 28:17Q&A: Altitude in malls
- 30:05Q&A: No geolocation benchmark
- 31:56Q&A: Third vs first-party data
- 33:42Closing remarks
related talks
32mEnriching Locations with Real-World Consumer Behavior Data
28mDiscussion: Humans As Sensors
14mBuilding OpenStreetMap Validation Processes in the Open
07mGil Elbaz interviewed at Web 2.0 Summit 2011
07mStartup Showcase: PlaceIQ
06mExtending Coverage of Stationary Sensing Systems with Mobile Sensing Systems for Human Mobility ...
15mThe turf wars on data - can be a matter of life and death
43mHacking your local bus system for the greater good