Embed Size (px)
Citation preview
Deriving Tourist Mobility Patternsfrom Check-in Data
Linus W. Dietz, Daniel Herzog, Dr. Wolfgang WörndlTechnical University of MunichDepartment of InformaticsChair of Connected MobilityLos Angeles, California, USAFebruary 9, 2018
Introduction
ProblemRecommend composite trips of travel destinations
MotivationIndependent travel planning is complex,information is scattered, outdated, of uncertain quality
c© Created by FreepikLinus W. Dietz (TUM) 2
Introduction
Interactive Recommender Systems• User modeling• Recommendation algorithms• UI/UX aspects• Item characterization
Research questions
1. Which destinations to use in the travel knapsack (constrained by time & money)?2. How long to stay at a specific region?
Linus W. Dietz (TUM) 3
Solution
Deriving Tourist Mobility Patterns from Check-in Data1. Which countries are frequently visited (together)2. How long travelers typically stay at a specific region
Visualization of a trip c© 2017 Google
Linus W. Dietz (TUM) 4
Data
Global check-in data set• April 2012 to September 2013• 266,909 users⇒ 65,745 travelers• 3,680,126 venues• 77 countries• 415 cities
Source: https://sites.google.com/site/yangdingqi/home/foursquare-dataset
Linus W. Dietz (TUM) 5
Metrics
Trip duration
Check-in rate, check-in density
Transition time
7 11 15 20 23 25 31 35 40 43 44 46 52 64 68 69 74 65 78 80 97 84 88
Days Traveled
Fre
quen
cy
050
010
0015
0020
0025
0030
00
Linus W. Dietz (TUM) 6
Metrics
Trip duration
Check-in rate, check-in density
Transition time
0.0 0.2 0.4 0.6 0.8 1.0
0.0
0.2
0.4
0.6
0.8
1.0
Check−in Density
Rat
io T
rave
ls
Linus W. Dietz (TUM) 6
Metrics
Trip duration
Check-in rate, check-in density
Transition time
Home d0→3
→3
Block︷ ︸︸ ︷d3 Brazil d30︸ ︷︷ ︸
Trip
→1
→1 d31 Home
Linus W. Dietz (TUM) 6
Metrics
Trip duration
Check-in rate, check-in density
Transition time
Home d0→3
→3
Block︷ ︸︸ ︷d3 Brazil d30︸ ︷︷ ︸
Trip
→1
→1 d31 Home d100→1
→1
Block︷ ︸︸ ︷d101 Netherlands d101→0
Block︷ ︸︸ ︷d101 Germany d105→1
Block︷ ︸︸ ︷d106 France d118→3
Block︷ ︸︸ ︷d121 Germany d124→0
Block︷ ︸︸ ︷d124 Netherlands d124︸ ︷︷ ︸
Trip
→6
→6 d130 Home . . .
Linus W. Dietz (TUM) 6
Data Processing
266,909 users
65,745 travelers
23,218 travelers34,892 trips
23,418 trips
Filter out non-travelers
Filter out short trips (< 7 days)
Filter out low check-in density users (< 0.2)
Problem: high transition time (9.80 days)
Mean transition time: 3.39 days 3
Linus W. Dietz (TUM) 7
Results Country Co-occurrences
n Countries ObservationsMalaysia, Singapore 269
Canada, USA 1622 United Kingdom, France 156
France, Italy 79United Kingdom, USA 79
France, Italy, Spain 50France, Belgium, Netherlands 38
3 United Kingdom, France, Italy 34United Kingdom, France, Spain 29Thailand, Malaysia, Singapore 22
United Kingdom, France, Italy, Spain 21France, Belgium, Germany, Netherlands 17
4 Austria, Czech Rep., Germany, Hungary 13France, Belgium, Italy, Netherlands 10
France, Germany, Italy, Spain 10
Linus W. Dietz (TUM) 8
Results Durations of StayD
ays
Kuw
ait
Dom
inic
an R
ep.
Cyp
rus
Aze
rbai
jan
Cos
ta R
ica
Gha
naK
azak
hsta
nP
arag
uay
Aus
tral
iaB
elar
usS
audi
Ara
bia
Turk
eyM
artin
ique
Bra
zil
Uni
ted
Sta
tes
Rus
sia
Col
ombi
aM
exic
oP
hilip
pine
sC
anad
aN
ew Z
eala
ndIn
done
sia
Japa
nA
rgen
tina
Chi
naU
nite
d K
ingd
omM
alay
sia
Ven
ezue
laT
haila
ndS
pain
Sin
gapo
reU
n. A
rab
Em
irate
sIta
lyIr
elan
dG
erm
any
Sw
eden
Fra
nce
Cze
ch R
ep.
Fin
land
Net
herla
nds
Sw
itzer
land
Bel
gium
Den
mar
kH
unga
ryE
ston
ia
03
69
1215
1821
050
015
0025
0035
00O
bser
vatio
ns
mean days
observations
Linus W. Dietz (TUM) 9
Conclusions & Future Work
Check-in data⇒ tourist mobility patterns⇒ • durations of stay• destination co-occurrences
Metric-driven data mining approach
Data models (destinations, trips)
Limited by data availability
→ Evaluate metrics with more data
→ Fine-grained region model
→ Combine several data sources
Refine heuristics
Determine traveler types
Linus W. Dietz (TUM) 10
Conclusions & Future Work
Check-in data⇒ tourist mobility patterns⇒ • durations of stay• destination co-occurrences
Metric-driven data mining approach
Data models (destinations, trips)
Limited by data availability
→ Evaluate metrics with more data
→ Fine-grained region model
→ Combine several data sources
Refine heuristics
Determine traveler types
Linus W. Dietz (TUM) 10
Deriving Tourist Mobility Patternsfrom Check-in Data
Linus W. Dietz, Daniel Herzog, Dr. Wolfgang WörndlTechnical University of MunichDepartment of InformaticsChair of Connected MobilityLos Angeles, California, USAFebruary 9, 2018
Data Model
Region tree of the planet
Earth→ continents→ countries→ states→ counties
GeoTree model
Continents→ Continental sections→ Countries→ Regions→ More regions→ Cities→ Districts
Wikitravel model
Linus W. Dietz (TUM) 12
Data Model
Region tree of the planet
Earth→ continents→ countries→ states→ counties
GeoTree model
Continents→ Continental sections→ Countries→ Regions→ More regions→ Cities→ Districts
Wikitravel model
Current granularity on country level only!
Linus W. Dietz (TUM) 12
