Title: Extracting Semantic Location from Outdoor Positioning Systems
1Extracting Semantic Location from Outdoor
Positioning Systems
- Juhong Liu, Ouri Wolfson, Huabei Yin
- University of Illinois at Chicago
2Introduction - 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?
3Introduction 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
4Introduction problem statement
- Physical location -gt semantic location
- The place the user stays
- Devices
- Outdoor positioning systems
- Internet access
5Outline
- Introduction
- Input and output
- Algorithm for determining Semantic Location
- Experimental results
- Related Works
6Main 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)
7Online 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
8Auxiliary 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
9Outline
- Introduction
- Data Model
- Algorithm for determining Semantic Location
- Experiment
- Related Works
10Algorithm
11Step1 - 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)
12Stay 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
13Step2 Street address candidates
- Reverse Geocoding
- Physical location (stay_position) -gt street
address - Traditional geocoding method
- Nearest street address
- Incorrect result
14Step2 Street address candidates(2)
Street address candidates the street addresses
within k meters (graph distance) from
stay_position.
15Step3-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
16At 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
17Step4- 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
18Step4.1- 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
19Step4.1- 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
20Step4.2-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.
21Step4.3 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
22Step5 - Semantic Location determination
- For each SL in SLC, weighted sum of three
utilities - Weight setting (WSC, WSA, WP)
- Equal weighting
- Rank weighting
- Ratio weighting
23Initialization
- 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
24Outline
- Introduction
- Data Model
- Determination of Semantic Location
- Experimental results
- Related Works
25Experimental 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
26Experimental Results
- 96 correctness for all stays
- 76 stays home, office
- Non-frequent stay 90
- Remove home/office stays
27Outline
- Introduction
- Data Model
- Determination of Semantic Location
- Experiment
- Related Works
28Related 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
29Conclusion