Extracting Places from Traces of Locations

1
Extracting Places from Traces of Locations
Jong Hee Kang, William Welbourne, Benjamin
Stewart, Gaetano Borriello, October 2004,
Proceedings of the 2nd ACM international workshop
on Wireless mobile applications and services on
WLAN hotspots
2005. 04. 26 Database Lab. M.S.1 Kim Ji-Young
2
Outline

• Introduction
• Related work
• Extracting places
• Experimental evaluation
• Conclusions and future work

3
Location vs. Place

• Locations
• Values measured by the underlying location
  
• sensing technologies
• Expressed in coordinates or landmarks
• e.g., coordinates (-122.124511, 47.653932)
• landmarks nearby cell tower 34
• Places
• Locales with semantic meanings to individual
  users
• e.g., home, office, coffee shop

4
Translating Locations into Place(1/2)

• Places are more useful to applications
• than locations
• Locations need to be translated into
• places

My Office
(-122.124511, 47.653932)
Location Sensing Technology
Application
Place Translator
Location
Place
5
Translating Locations into Place(2/2)

• The Place Translator needs mapping
• information between locations and places
• Goal is to generate the mapping information
• automatically from the users behavior
  (trace of locations)

6
Trace of Locations(1/2)

• Collected using Place Lab,
• a coordinate-based location system using
• a database of locations of WiFi hotspots
• e.g. User initially stays at place A,
• then moves to place B and stays there
• Important places are those where the user
• spends a significant amount of time and/or
• visits frequently

A
B
7
Trace of Locations(2/2)
Important Places
8
Applying Clustering Algorithms
9
Existing Clustering Algorithms
Problems

• Algorithms require the number of
• clusters as a parameter
• Clusters include unimportant locations
• (intermediate and transitory locations between
  truly significant places)
• Algorithms require a significant amount of
  computation

10
Time-based Clustering(1/2)

• Clustering locations along the time axis
• A new location is compared with previous
  locations
• If the new location is moving away, starts a new
  cluster
• Then, ignore the clusters with short time duration

Important Places
11
Time-based Clustering(2/2)
Time-based Clustering(1/2)
Figure2. Existing Clustering
Figure4. Time-based Clustering
12
Two Parameters

• Distance threshold (d )
• Determines the size of
• clusters
• Time threshold (t )
• Determines the number of
• significant places

t
d
13
Determining d and t (1/2)
14
Determining d and t (2/2)
15
Experimental Evaluation-Campus Trace
16
Experimental Evaluation-City Scale Trace
17
Conclusion and Future Work

• Our approach can automatically find
• significant places from the trace of
  locations
• Future work
• Connecting to the destination prediction system
• To predict a users destination from their
  current location and past observations of their
  movements
• (Record the arrival and leaving time to and
  from
• the extracted places)