Extracting Semantic Location from Outdoor Positioning Systems

1
Extracting Semantic Location from Outdoor
Positioning Systems

• Juhong Liu, Ouri Wolfson, Huabei Yin
• University of Illinois at Chicago

2
Introduction - Context

• Context
• Environment in which a user operates
• Location info., environmental info. (weather),
  social info. (who is around), etc.
• Location information important aspect of context
• Reminders
• Physicians office request a prescription
• Movie theatre -gt turn off phone
• User interface
• Computer store apple (computer)
• Grocery store apple (fruit)
• Places Ive been (analogous to Stuff Ive Seen)
• Where was I on 8/15/02 at 2pm?
• When was the last time I saw my dietician?

3
Introduction location information

• Location information
• Physical location
• Provided by positioning systems
• GPS (122.39, 239.11, 1120am)
• Unreadable by users
• Semantic location
• Not directly provided by positioning systems
• Dominicks grocery store, 1340 S. Canal St.
• Dermatologists office
• Home
• Useful to users

4
Introduction problem statement

• Physical location -gt semantic location
• The place the user stays
• Devices
• Outdoor positioning systems
• Internet access

5
Outline

• Introduction
• Input and output
• Algorithm for determining Semantic Location
• Experimental results
• Related Works

6
Main Input and Output

• Input Trajectory T (x1, y1, t1), (x2, y2,
  t2), , (xn, yn, tn)
• Output 1 Semantic location
• Location name (BestBuy)
• Semantic category
• Business type (electronics store),
• office
• home
• Street address
• Output 2 Semantic location log file
• (date, begin_time, end_time, semantic location)

7
Online and offline versions

• Online determine the current location
• On mobile device
• Based on incomplete trip trajectory
• Offline Determine multiple past locations
• At pc
• Based on complete trip trajectory

8
Auxiliary inputs

• Profile
• Calendar (event date, semantic location)
• Address Book (phone number, semantic location)
• Phone Call List (calling date, semantic
  location)
• Web Page List - (visiting date, semantic
  location)
• Destination List (searching date, semantic
  location)
• Users Feedback
• Confirmed list
• Denied list

9
Outline

• Introduction
• Data Model
• Algorithm for determining Semantic Location
• Experiment
• Related Works

10
Algorithm
11
Step1 - Stay extraction

• Stay
• Loss of GPS signal
• To spend at least min_time in an area with the
  diameter no larger than d.
• (stay_position, date, stay_start, stay_end)

12
Stay extraction details (prior work)

• Extraction
• Stay generation
• Last min_time, the physical positions are within
  d.
• Stay_position center of these physical
  locations
• Stay_start, stay_end
• Stay extension and finish
• A physical position p following a stay
• Distance(stay_position, p) lt d/2 gt stay_end
  extended
• Distance(stay_position, p) gt d/2 gt current stay
  finishes

Min_time5, d as shown stay_postion
p4 Stay_start p3 Stay_end p8
13
Step2 Street address candidates

• Reverse Geocoding
• Physical location (stay_position) -gt street
  address
• Traditional geocoding method
• Nearest street address
• Incorrect result

14
Step2 Street address candidates(2)
Street address candidates the street addresses
within k meters (graph distance) from
stay_position.
15
Step3-semantic location candidates

• Street address candidates -gt
• semantic location candidates
• Yellow pages
• Such as switchboard
• Profile
• Calendar, Address Book, Phone Call List, Web Page
  List, Destination List, User's Feedback

16
At end of step 3 A set of Semantic Location
candidates

• Semantic location
• Location name (BestBuy)
• Semantic category
• Business type (electronics store theater),
• office
• home
• Street address

17
Step4- three utilities calculation

• For each semantic location SL in set of
  candidates compute
• Semantic category (SC) utility how likely is the
  semantic category given users history
• Street address (SA) utility how likely is the
  street address given the stay location
• Profile (P) utility How well SL matches the
  profile

18
Step4.1- Semantic category utility

• Assumption
• Users visit semantic categories habitually.
• Semantic category history
• Time information and semantic category
• format
• workday_or_weekend, T1
• start time of stay,
  T2
• length_of_time_spent_there, T3
• semantic category C
  
• Can be extracted from semantic log file

19
Step4.1- Semantic category utility for a semantic
location SL

• Probability of semantic category C for SL is P(C
  T1, T2, T3)
• Intuitively
• the probability that a stay with the
  correspondent time information visits semantic
  category C (e.g. a theater).
• P(C T1, T2, T3) is computed by Bayes Model using
  the semantic location log file
• C - a semantic category
• Ti - the time information
• Z for normalization

20
Step4.2-Street address Utility for SL

• For the stay_position (x, y) the street address
  of the projection point p on each street has the
  highest probability
• The utility of a street address is proportional
  to its smallest (route) distance from a
  projection point.

21
Step4.3 profile utility of SL

• SL in SLC has a higher profile utility, if
  matches
• Calendar
• Address book
• Phone Call List
• Web Page List
• Destination List
• Users feedback confirmed list
• SL in SLC has a lower profile utility, if
  matches
• Users feedback denied list

22
Step5 - Semantic Location determination

• For each SL in SLC, weighted sum of three
  utilities
• Weight setting (WSC, WSA, WP)
• Equal weighting
• Rank weighting
• Ratio weighting

23
Initialization

• At outset
• No semantic category history
• No feedback history
• An Initialization is necessary
• Several weeks
• Build initial SC history using credit card
  statement, with user corrections
• Build feedback history

24
Outline

• Introduction
• Data Model
• Determination of Semantic Location
• Experimental results
• Related Works

25
Experimental data and setting

• Data
• GPS
• The trip of a student for 4 months
• The student gives feed back every week
• Weight setting (WSC, WSA, WP)
• Equal weighting (1,1,1)
• Rank weighting (1,2,3), (1,3,2),(2,1,3),
  (2,3,1), (3,1,2), and (3,2,1)
• Initialization time
• 2 weeks, 3 weeks and 4 weeks
• New city simulation
• Remove the information in Users feedback

26
Experimental Results

• 96 correctness for all stays
• 76 stays home, office
• Non-frequent stay 90
• Remove home/office stays

27
Outline

• Introduction
• Data Model
• Determination of Semantic Location
• Experiment
• Related Works

28
Related Works

• Indoors
• Easyliving (determine meeting room, lab, etc)
• Outdoors
• Cyberguide
• Tour guide points of interest around the users
  location
• Current semantic location not extracted
• Commotion
• Significant locations pick up
• The user names the locations, gives to_do lists
• To_do lists come out, when at correspondent
  location
• Lachesis
• Stays pick up
• User provides semantic location for stay
• Markov model built to predict future stay

29
Conclusion