2. Background
■ We have accumulated many
location data with the
popularity of location-based
services.
■ Coordinates, time stamps, and
comments
■ Transportation routine
prediction, location-based
activity recognition, location-
based social network
GPS data management system (P 1029, Zheng, et al 2010)
3. Purpose
■ Discover interesting locations given a certain activity
■ Discover possible activities given a location
■ Location recommendation / Activity recommendation
6. Data Input
Location-activity matrix (entry: frequency)
(P 1030, Zheng, et al 2010)
http://www.mobilegpsonline.com/
Raw GPS Data
Comments
Sparse Data
Over 2.5 years, we have 12,765 GPS trajectories, but
only 530 comments from 162 users
8. Stay region extraction
Stay point s: a geographical region where a user stayed over a
time thresholdTr within a distance threshold of Dr.A virtual
location characterized by a set of consecutive GPS points.
A user’s trajectory Traj: a sequence of time-stamped points
Stay region r:A geographic region which contains a set of stay points.
Interpreted as ‘location’
(P 1030, Zheng, et al 2010)
9. ■ Grid based clustering
– Consider the size of the clusters
– d2 is the maximum size of a location cluster
– A unassigned g + 8 neighbor grids
■ K-means algorithm
– Needs to be given k (number of clusters) values
– Doesn’t control the size of clusters
Stay region extraction
s
s s
s
s s
d/3
A stay region r
s: stay points