Tourist Knowledge
Graph Generation to
Automating Personal
Travel Bookings
Innovation Development Area
Giuseppe Rizzo
Tourist Knowledge Graph Generation to Automating Personal Travel Bookings @ Milano-Bicocca
Turning
data into
actionable
insights
INNODEV
http://innodev.ismb.it
2017-05-02
@giusepperizzo
2
RECIPE OF TODAY’S TALK
The alchemy toward automation
- data modeling
- data integration
- personalization
Using ML and AI
The Alchemist Discovering Phosphorus -- Joseph Wright of Derby
TODAY’S RECIPE:
A “journey” from data to
personalized automation
Going through:
- Knowledge Graph
- Big Data
- Data Integration
- Personalization
Tourist Knowledge Graph Generation to Automating Personal Travel Bookings @ Milano-Bicocca
Turning
data into
actionable
insights
INNODEV
http://innodev.ismb.it
2017-05-02
@giusepperizzo
3
KNOWLEDGE GRAPH
http://searchengineland.com/laymans-visual-guide-googles-knowledge-graph-search-api-241935
Tourist Knowledge Graph Generation to Automating Personal Travel Bookings @ Milano-Bicocca
Turning
data into
actionable
insights
INNODEV
http://innodev.ismb.it
2017-05-02
@giusepperizzo
4
dbo:Person
dbo:Agent
is a
yago:Artist109812338
is a
yago:Creator109614315
subClassOf
subClassOf
yago-res:wo
rdnet_artist_
109812338
equivalentClass
T-Box
A-Box
dbo:author
Tourist Knowledge Graph Generation to Automating Personal Travel Bookings @ Milano-Bicocca
Turning
data into
actionable
insights
INNODEV
http://innodev.ismb.it
2017-05-02
@giusepperizzo
5
KEY FEATURE OF A
KNOWLEDGE GRAPH
It stores entity-relation-entity
data where the semantics of
entities and relations is explicit
Tourist Knowledge Graph Generation to Automating Personal Travel Bookings @ Milano-Bicocca
Turning
data into
actionable
insights
INNODEV
http://innodev.ismb.it
2017-05-02
@giusepperizzo
6
A KNOWLEDGE GRAPH
ENABLES
Algorithms to efficiently work over
the semantics of data for a large
range of tasks such as retrieval,
extraction, classification,
clustering, ranking where
automation is put in the first place
Tourist Knowledge Graph Generation to Automating Personal Travel Bookings @ Milano-Bicocca
Turning
data into
actionable
insights
INNODEV
http://innodev.ismb.it
2017-05-02
@giusepperizzo
7
NUMEROUS KNOWLEDGE
GRAPHS OUT THERE
http://lod-cloud.net
Some can be dumped
locally as the ones
pictured by lod-cloud,
some others are
technology assets of
popular products such
as Google Graph,
Facebook Graph and
can be accessed via
dedicated web
interfaces
Tourist Knowledge Graph Generation to Automating Personal Travel Bookings @ Milano-Bicocca
Turning
data into
actionable
insights
INNODEV
http://innodev.ismb.it
2017-05-02
@giusepperizzo
8
THE BROADER DATA MOVEMENT
TALKS ABOUT 5Vs OUT THERE
The big
VALUE
VOLUME
VELOCITY
VERACITY VARIETY
Tourist Knowledge Graph Generation to Automating Personal Travel Bookings @ Milano-Bicocca
Turning
data into
actionable
insights
INNODEV
http://innodev.ismb.it
2017-05-02
@giusepperizzo
9
TOURIST AND CULTURAL
DATA DELUGE
Tourist Knowledge Graph Generation to Automating Personal Travel Bookings @ Milano-Bicocca
Turning
data into
actionable
insights
INNODEV
http://innodev.ismb.it
2017-05-02
@giusepperizzo
SILOED,
UNCONNECTED,
UNSTRUCTURED,
WEAKLY-STRUCTURED,
HIDDEN SEMANTICS
10
Tourist Knowledge Graph Generation to Automating Personal Travel Bookings @ Milano-Bicocca
Turning
data into
actionable
insights
INNODEV
http://innodev.ismb.it
2017-05-02
@giusepperizzo
11
Automating the creation of city
knowledge graphs composed of
better linked tourist and cultural data
to answer complex queries
AMBITION #1:
BETTER DATA 4 TRAVELER
Tourist Knowledge Graph Generation to Automating Personal Travel Bookings @ Milano-Bicocca
Turning
data into
actionable
insights
INNODEV
http://innodev.ismb.it
2017-05-02
@giusepperizzo
12
collect + link
tourist and cultural data such as
events, places, user-generated data
(reviews, ratings), transportation means
Tourist Knowledge Graph Generation to Automating Personal Travel Bookings @ Milano-Bicocca
Turning
data into
actionable
insights
INNODEV
http://innodev.ismb.it
2017-05-02
@giusepperizzo
13
Integration
DATA INTEGRATION
Tourist Knowledge Graph Generation to Automating Personal Travel Bookings @ Milano-Bicocca
Turning
data into
actionable
insights
INNODEV
http://innodev.ismb.it
2017-05-02
@giusepperizzo
14
entities semantically connected
Rizzo G. et al. (2015) The 3cixty Knowledge Base for Expo Milano 2015: Enabling Visitors to
Explore the City. In: 8th International Conference on Knowledge Capture (K-CAP), Best P&D
Tourist Knowledge Graph Generation to Automating Personal Travel Bookings @ Milano-Bicocca
Turning
data into
actionable
insights
INNODEV
http://innodev.ismb.it
2017-05-02
@giusepperizzo
16
DATA SAMPLING
Tourist Knowledge Graph Generation to Automating Personal Travel Bookings @ Milano-Bicocca
Turning
data into
actionable
insights
INNODEV
http://innodev.ismb.it
2017-05-02
@giusepperizzo
17
Sample
cell
by cell
Tourist Knowledge Graph Generation to Automating Personal Travel Bookings @ Milano-Bicocca
Turning
data into
actionable
insights
INNODEV
http://innodev.ismb.it
2017-05-02
@giusepperizzo
18
Algorithm implements an ensemble
entity matching classifier, where the
ensemble of weak-classifiers is
performed via stacking
ENTITY MATCHING
Palumbo E., Rizzo G., Troncy R. (2017) STEM: Stacked Threshold-based Entity Matching for
Knowledge Base Generation. Submitted to Semantic Web Journal
Rizzo G. et al. (2015) 3cixty@Expo Milano 2015: Enabling Visitors to Explore a Smart City. In: 14th
International Semantic Web Conference (ISWC), SWC Winner
Troncy R, Rizzo G. et al. (2017) 3cixty: Building Comprehensive Knowledge Bases For City
Exploration. Submitted to Journal of Web Semantics
Tourist Knowledge Graph Generation to Automating Personal Travel Bookings @ Milano-Bicocca
Turning
data into
actionable
insights
INNODEV
http://innodev.ismb.it
2017-05-02
@giusepperizzo
19
name address capacity ...
Pala
Alpitour
C.so
Sebastopoli
20000 spettatori
max
name address capacity ...
PalaIsozaki Sebastopoli, 123 20200
Pala Alpitour
PalaIsozaki
Tourist Knowledge Graph Generation to Automating Personal Travel Bookings @ Milano-Bicocca
Turning
data into
actionable
insights
INNODEV
http://innodev.ismb.it
2017-05-02
@giusepperizzo
20
d(name)
name address capacity ...
Pala Alpitour C.so Sebastopoli 20000
spettatori max
name address capacity ...
PalaIsozaki Sebastopoli, 123 20200
d(address) d(capacity) ...
s(name) s(address) s(capacity) ...
p1
=p(Match|s1
) p2
=p(Match|s2
) p3
=p(Match|s3
)
Tourist Knowledge Graph Generation to Automating Personal Travel Bookings @ Milano-Bicocca
Turning
data into
actionable
insights
INNODEV
http://innodev.ismb.it
2017-05-02
@giusepperizzo
21
BAYESIAN
THRESHOLD-BASED
CLASSIFIER
f(e1
, e2
, t) : if p > T then e1
= e2
p1
p2
… pN
p1
p2
.. pN
+ (1-p1
)(1-p2
)..(1-pN
)
p =
Tourist Knowledge Graph Generation to Automating Personal Travel Bookings @ Milano-Bicocca
Turning
data into
actionable
insights
INNODEV
http://innodev.ismb.it
2017-05-02
@giusepperizzo
22
ENSEMBLE OF ENTITY
MATCHING CLASSIFIER
f1
(e1
, e2
, t1
)
f2
(e1
, e2
, t2
)
fn
(e1
, e2
, tn
)
...
SVM Classifier
GS of manually
annotated matches/no
matches
Match/
No Match
Palumbo E., Rizzo G., Troncy R. (2016) An Ensemble Approach to Financial Entity Matching for the FEIII 2016
Challenge. In: (SIGMOD'16), Financial Entity Identification and Information Integration Challenge, San
Francisco, US.
Tourist Knowledge Graph Generation to Automating Personal Travel Bookings @ Milano-Bicocca
Turning
data into
actionable
insights
INNODEV
http://innodev.ismb.it
2017-05-02
@giusepperizzo
23
MANUAL ANNOTATIONS OF
PLACE-TYPE ENTITIES
Tourist Knowledge Graph Generation to Automating Personal Travel Bookings @ Milano-Bicocca
Turning
data into
actionable
insights
INNODEV
http://innodev.ismb.it
2017-05-02
@giusepperizzo
24
MANUAL ANNOTATIONS OF
EVENT-TYPE ENTITIES
Tourist Knowledge Graph Generation to Automating Personal Travel Bookings @ Milano-Bicocca
Turning
data into
actionable
insights
INNODEV
http://innodev.ismb.it
2017-05-02
@giusepperizzo
25
2014 - 2016 EIT Digital
funded, action line Digital
Cities
https://www.3cixty.com
Explore a city
Plan a visit
nearby places?
related events?
Will be looking for an Indian restaurant
reachable in 10 minutes on foot from
the Madison Square Garden
Multitude of sources
with unconnected
and heterogeneous
data
Rizzo G., Troncy R., Corcho O., Jameson A., Plu J., Ballesteros Hermida J.C., Assaf A., Barbu C.,
Spirescu A., Kuhn K., Celino I., Agarwal R., Nguyen C.K., Pathak A., Scanu C., Valla M., Haaker T.,
Verga E.S., Rossi M., Redondo Garcia J.L. (2015) 3cixty@Expo Milano 2015: Enabling Visitors to
Explore a Smart City. In: 14th International Semantic Web Conference (ISWC'15), Semantic
Web Challenge
3cixty @ Milano Expo 2015
4Vs: 20 DIFFERENT DATA SOURCES
Variety
Veracity
and
authority
Velocity
Volume
https://www.youtube.com/watch?v=K6_ylq1ufH8
3cixty: Knowledge Base and
Applications for Milan and Expo 2015
https://www.3cixty.com
Other validated pilots:
Nice, London, Singapore, Madeira,
Las Palmas, Amsterdam
Tourist Knowledge Graph Generation to Automating Personal Travel Bookings @ Milano-Bicocca
Turning
data into
actionable
insights
INNODEV
http://innodev.ismb.it
2017-05-02
@giusepperizzo
AMBITION #2:
BETTER OFFERING 4 TRAVELER
37
Automating the creation of
personalized travel packages
Tourist Knowledge Graph Generation to Automating Personal Travel Bookings @ Milano-Bicocca
Turning
data into
actionable
insights
INNODEV
http://innodev.ismb.it
2017-05-02
@giusepperizzo
38
package = flights + hotel +
restaurants + events + ...
= a new personalized
in-destination experience
Tourist Knowledge Graph Generation to Automating Personal Travel Bookings @ Milano-Bicocca
Turning
data into
actionable
insights
INNODEV
http://innodev.ismb.it
2017-05-02
@giusepperizzo
39
CONTEXTUAL AND EPISODIC DATA
Tourist Knowledge Graph Generation to Automating Personal Travel Bookings @ Milano-Bicocca
Turning
data into
actionable
insights
INNODEV
http://innodev.ismb.it
2017-05-02
@giusepperizzo
40
contextual, i.e.
encyclopedic
KNOWLEDGE GRAPH TYPES
episodic such as
transactions, user
purchases, user likes
?
Tourist Knowledge Graph Generation to Automating Personal Travel Bookings @ Milano-Bicocca
Turning
data into
actionable
insights
INNODEV
http://innodev.ismb.it
2017-05-02
@giusepperizzo
41
EPISODIC DATA OUT THERE
re.
SILOED,
UNCONNECTED,
UNSTRUCTURED,
WEAKLY-STRUCTURED,
HIDDEN SEMANTICS
LOCKED
+
privacy sensitive data
Tourist Knowledge Graph Generation to Automating Personal Travel Bookings @ Milano-Bicocca
Turning
data into
actionable
insights
INNODEV
http://innodev.ismb.it
2017-05-02
@giusepperizzo
42
TRAVELER DATA
i.e. Giuseppe, flight number, departure, destination,
….
materialized in a Knowledge Graph
complementing the contextual
knowledge of local offerings such as
restaurants, events
Tourist Knowledge Graph Generation to Automating Personal Travel Bookings @ Milano-Bicocca
Turning
data into
actionable
insights
INNODEV
http://innodev.ismb.it
2017-05-02
@giusepperizzo
43
Algorithm implements a machine
learning recommender blending
tourist and personal data
RECOMMENDATION
Palumbo E., Rizzo G., Troncy R (2017) Learning User-Item Relatedness from Knowledge Graphs for
Top-N Item Recommendation. Submitted to 11th ACM Recommender Systems Conference (RecSys)
Tourist Knowledge Graph Generation to Automating Personal Travel Bookings @ Milano-Bicocca
Turning
data into
actionable
insights
INNODEV
http://innodev.ismb.it
2017-05-02
@giusepperizzo
44
CONTEXTUAL AND EPISODIC
KNOWLEDGE GRAPH
u1
u2
u3
item1
item2item3
item4
item5
item6
item7
● User
● User-Item
(purchase)
● User-Item (like)
● Item
● Item-Item
● User-User
(relation)
Tourist Knowledge Graph Generation to Automating Personal Travel Bookings @ Milano-Bicocca
Turning
data into
actionable
insights
INNODEV
http://innodev.ismb.it
2017-05-02
@giusepperizzo
45
KNOWLEDGE
GRAPH
USER-ITEM
RELATEDNES
S
LEARN-TO-
RANK ITEMS
3-STEP PROCESS
create compute perform
Tourist Knowledge Graph Generation to Automating Personal Travel Bookings @ Milano-Bicocca
Turning
data into
actionable
insights
INNODEV
http://innodev.ismb.it
2017-05-02
@giusepperizzo
46
TURNING ENTITIES AND
RELATIONS TO FEATURE VECTORS
u1
i1
i6
purchase
purchase f(u1, i1) = rel1(u1, i1)
like f(u1, i7) = rel3(u1, i7)
i7
purchase
like
u1
u1
...
Graph
u2
Relatedness
scores
Feature
vectors
f(u1, i6) = rel2(u1, i6) u1
#i1
#i2
#i7
Tourist Knowledge Graph Generation to Automating Personal Travel Bookings @ Milano-Bicocca
Turning
data into
actionable
insights
INNODEV
http://innodev.ismb.it
2017-05-02
@giusepperizzo
47
purchase
like
u1
u1
...
u1
#i1
#i2
#i7
Lambda
MART
purchase
like
u1
u1
...
u1
#i2
#i7
#i1
new
ranking
RECOMMENDATION AS
LEARNING TO RANK TASK
Feature
vectors
Tourist Knowledge Graph Generation to Automating Personal Travel Bookings @ Milano-Bicocca
Turning
data into
actionable
insights
INNODEV
http://innodev.ismb.it
2017-05-02
@giusepperizzo
48
PERSONALIZATION DIMENSIONS
history Episodic, what I’ve already bought
collaborative Colleagues who I know have bought
willingness to pay Budget availability
interests Nature, culture, sports, nightlife
needs Small kids, driving licence, allergies, mobility restrictions,
animals
time When doing things, time pressure, aversion to queuing,
advance vs on site payment, life pace
travel preferences Wifi availability, meal likings, spas, noise aversion
payment method Credit card, phone, bitcoin, paypal, cash
...
Tourist Knowledge Graph Generation to Automating Personal Travel Bookings @ Milano-Bicocca
Turning
data into
actionable
insights
INNODEV
http://innodev.ismb.it
2017-05-02
@giusepperizzo
49
2017 EIT Digital funded,
action line Digital Cities
http://pas-time.org
Frederic, CEO of a company, is going to
give a pitch to investors in SF.
He flies often over there, where he likes
staying in a comfortable house rented via
airbnb next to the sea to enjoying surfing
during his free spots. While staying, he likes
watching the Golden State Warriors once
Frederic
TRADITIONAL LINKAGE
Flights
Commuting
Hotel
(Online)
Travel
Agencies
contacts
formulate
submit for
acceptance/change
2) There is
a bunch of
accommod
ations
you’d like
the most
Frederic PasTime
1) These
flights will
suit you the
best, pick
one
3) You’ll enjoy
these
restaurants
as they
prepares your
favourite dish
and are next
to ...
4) Golden
State Warriors
are playing in
town. Would
you like to
attend?
package
“NETFLIX” FOR TRAVELERS
generates
PasTime for French Regional Committee of
Tourism
Package Persona-A
Package Persona-B
Package Persona-C
Tourist Offices
Persona-A
Persona-B
Persona-C
Tourist Personas
Tourist offices distribute and record likes of tourists. The
recommendation algorithm will exploit those likes in a learning fashion
http://www.freepik.com/free-vector/businesswo
man-working-at-the-office_799834.htm
http://www.pas-time.org
Further investigated pilots:
French RCT, Amsterdam and Las
Palmas
On both
- Per-group packaging
- Per-traveler packaging
Tourist Knowledge Graph Generation to Automating Personal Travel Bookings @ Milano-Bicocca
Turning
data into
actionable
insights
INNODEV
http://innodev.ismb.it
2017-05-02
@giusepperizzo
55
RECAP: a “journey” from data to
personalized automation
Going through:
- Knowledge Graph
- Big Data
- Data Integration
- Personalization
Any questions?
https://www.slideshare.net/giusepperizzo
Tourist Knowledge Graph Generation to Automating Personal Travel Bookings @ Milano-Bicocca
Turning
data into
actionable
insights
INNODEV
http://innodev.ismb.it
2017-05-02
@giusepperizzo
56
ACKNOWLEDGE
PROJECTS
3cixty: An Innovative Platform for Building
City Apps
EIT Digital, Digital Cities (2014-2016)
PasTime: Packaging Personalised Tourist
Experiences
EIT Digital, Digital Cities (2017)
Tourist Knowledge Graph Generation to Automating Personal Travel Bookings @ Milano-Bicocca
Turning
data into
actionable
insights
INNODEV
http://innodev.ismb.it
2017-05-02
@giusepperizzo
57
ACKNOWLEDGE
AMAZING TEAM
Lots of contributions along the way: Raphael T., a mentor and
a great colleague; Enrico P. and Julien P., researchers
currently doing a PhD: “there is always something new to learn
from young padawans”.
Enrico F., Michele O., all ISMB BMPI jedis, all students I’ve
supervised, and all colleagues I’ve been working with

Tourist Knowledge Graph Creation to Automating Travel Bookings

  • 1.
    Tourist Knowledge Graph Generationto Automating Personal Travel Bookings Innovation Development Area Giuseppe Rizzo
  • 2.
    Tourist Knowledge GraphGeneration to Automating Personal Travel Bookings @ Milano-Bicocca Turning data into actionable insights INNODEV http://innodev.ismb.it 2017-05-02 @giusepperizzo 2 RECIPE OF TODAY’S TALK The alchemy toward automation - data modeling - data integration - personalization Using ML and AI The Alchemist Discovering Phosphorus -- Joseph Wright of Derby TODAY’S RECIPE: A “journey” from data to personalized automation Going through: - Knowledge Graph - Big Data - Data Integration - Personalization
  • 3.
    Tourist Knowledge GraphGeneration to Automating Personal Travel Bookings @ Milano-Bicocca Turning data into actionable insights INNODEV http://innodev.ismb.it 2017-05-02 @giusepperizzo 3 KNOWLEDGE GRAPH http://searchengineland.com/laymans-visual-guide-googles-knowledge-graph-search-api-241935
  • 4.
    Tourist Knowledge GraphGeneration to Automating Personal Travel Bookings @ Milano-Bicocca Turning data into actionable insights INNODEV http://innodev.ismb.it 2017-05-02 @giusepperizzo 4 dbo:Person dbo:Agent is a yago:Artist109812338 is a yago:Creator109614315 subClassOf subClassOf yago-res:wo rdnet_artist_ 109812338 equivalentClass T-Box A-Box dbo:author
  • 5.
    Tourist Knowledge GraphGeneration to Automating Personal Travel Bookings @ Milano-Bicocca Turning data into actionable insights INNODEV http://innodev.ismb.it 2017-05-02 @giusepperizzo 5 KEY FEATURE OF A KNOWLEDGE GRAPH It stores entity-relation-entity data where the semantics of entities and relations is explicit
  • 6.
    Tourist Knowledge GraphGeneration to Automating Personal Travel Bookings @ Milano-Bicocca Turning data into actionable insights INNODEV http://innodev.ismb.it 2017-05-02 @giusepperizzo 6 A KNOWLEDGE GRAPH ENABLES Algorithms to efficiently work over the semantics of data for a large range of tasks such as retrieval, extraction, classification, clustering, ranking where automation is put in the first place
  • 7.
    Tourist Knowledge GraphGeneration to Automating Personal Travel Bookings @ Milano-Bicocca Turning data into actionable insights INNODEV http://innodev.ismb.it 2017-05-02 @giusepperizzo 7 NUMEROUS KNOWLEDGE GRAPHS OUT THERE http://lod-cloud.net Some can be dumped locally as the ones pictured by lod-cloud, some others are technology assets of popular products such as Google Graph, Facebook Graph and can be accessed via dedicated web interfaces
  • 8.
    Tourist Knowledge GraphGeneration to Automating Personal Travel Bookings @ Milano-Bicocca Turning data into actionable insights INNODEV http://innodev.ismb.it 2017-05-02 @giusepperizzo 8 THE BROADER DATA MOVEMENT TALKS ABOUT 5Vs OUT THERE The big VALUE VOLUME VELOCITY VERACITY VARIETY
  • 9.
    Tourist Knowledge GraphGeneration to Automating Personal Travel Bookings @ Milano-Bicocca Turning data into actionable insights INNODEV http://innodev.ismb.it 2017-05-02 @giusepperizzo 9 TOURIST AND CULTURAL DATA DELUGE
  • 10.
    Tourist Knowledge GraphGeneration to Automating Personal Travel Bookings @ Milano-Bicocca Turning data into actionable insights INNODEV http://innodev.ismb.it 2017-05-02 @giusepperizzo SILOED, UNCONNECTED, UNSTRUCTURED, WEAKLY-STRUCTURED, HIDDEN SEMANTICS 10
  • 11.
    Tourist Knowledge GraphGeneration to Automating Personal Travel Bookings @ Milano-Bicocca Turning data into actionable insights INNODEV http://innodev.ismb.it 2017-05-02 @giusepperizzo 11 Automating the creation of city knowledge graphs composed of better linked tourist and cultural data to answer complex queries AMBITION #1: BETTER DATA 4 TRAVELER
  • 12.
    Tourist Knowledge GraphGeneration to Automating Personal Travel Bookings @ Milano-Bicocca Turning data into actionable insights INNODEV http://innodev.ismb.it 2017-05-02 @giusepperizzo 12 collect + link tourist and cultural data such as events, places, user-generated data (reviews, ratings), transportation means
  • 13.
    Tourist Knowledge GraphGeneration to Automating Personal Travel Bookings @ Milano-Bicocca Turning data into actionable insights INNODEV http://innodev.ismb.it 2017-05-02 @giusepperizzo 13 Integration DATA INTEGRATION
  • 14.
    Tourist Knowledge GraphGeneration to Automating Personal Travel Bookings @ Milano-Bicocca Turning data into actionable insights INNODEV http://innodev.ismb.it 2017-05-02 @giusepperizzo 14 entities semantically connected Rizzo G. et al. (2015) The 3cixty Knowledge Base for Expo Milano 2015: Enabling Visitors to Explore the City. In: 8th International Conference on Knowledge Capture (K-CAP), Best P&D
  • 16.
    Tourist Knowledge GraphGeneration to Automating Personal Travel Bookings @ Milano-Bicocca Turning data into actionable insights INNODEV http://innodev.ismb.it 2017-05-02 @giusepperizzo 16 DATA SAMPLING
  • 17.
    Tourist Knowledge GraphGeneration to Automating Personal Travel Bookings @ Milano-Bicocca Turning data into actionable insights INNODEV http://innodev.ismb.it 2017-05-02 @giusepperizzo 17 Sample cell by cell
  • 18.
    Tourist Knowledge GraphGeneration to Automating Personal Travel Bookings @ Milano-Bicocca Turning data into actionable insights INNODEV http://innodev.ismb.it 2017-05-02 @giusepperizzo 18 Algorithm implements an ensemble entity matching classifier, where the ensemble of weak-classifiers is performed via stacking ENTITY MATCHING Palumbo E., Rizzo G., Troncy R. (2017) STEM: Stacked Threshold-based Entity Matching for Knowledge Base Generation. Submitted to Semantic Web Journal Rizzo G. et al. (2015) 3cixty@Expo Milano 2015: Enabling Visitors to Explore a Smart City. In: 14th International Semantic Web Conference (ISWC), SWC Winner Troncy R, Rizzo G. et al. (2017) 3cixty: Building Comprehensive Knowledge Bases For City Exploration. Submitted to Journal of Web Semantics
  • 19.
    Tourist Knowledge GraphGeneration to Automating Personal Travel Bookings @ Milano-Bicocca Turning data into actionable insights INNODEV http://innodev.ismb.it 2017-05-02 @giusepperizzo 19 name address capacity ... Pala Alpitour C.so Sebastopoli 20000 spettatori max name address capacity ... PalaIsozaki Sebastopoli, 123 20200 Pala Alpitour PalaIsozaki
  • 20.
    Tourist Knowledge GraphGeneration to Automating Personal Travel Bookings @ Milano-Bicocca Turning data into actionable insights INNODEV http://innodev.ismb.it 2017-05-02 @giusepperizzo 20 d(name) name address capacity ... Pala Alpitour C.so Sebastopoli 20000 spettatori max name address capacity ... PalaIsozaki Sebastopoli, 123 20200 d(address) d(capacity) ... s(name) s(address) s(capacity) ... p1 =p(Match|s1 ) p2 =p(Match|s2 ) p3 =p(Match|s3 )
  • 21.
    Tourist Knowledge GraphGeneration to Automating Personal Travel Bookings @ Milano-Bicocca Turning data into actionable insights INNODEV http://innodev.ismb.it 2017-05-02 @giusepperizzo 21 BAYESIAN THRESHOLD-BASED CLASSIFIER f(e1 , e2 , t) : if p > T then e1 = e2 p1 p2 … pN p1 p2 .. pN + (1-p1 )(1-p2 )..(1-pN ) p =
  • 22.
    Tourist Knowledge GraphGeneration to Automating Personal Travel Bookings @ Milano-Bicocca Turning data into actionable insights INNODEV http://innodev.ismb.it 2017-05-02 @giusepperizzo 22 ENSEMBLE OF ENTITY MATCHING CLASSIFIER f1 (e1 , e2 , t1 ) f2 (e1 , e2 , t2 ) fn (e1 , e2 , tn ) ... SVM Classifier GS of manually annotated matches/no matches Match/ No Match Palumbo E., Rizzo G., Troncy R. (2016) An Ensemble Approach to Financial Entity Matching for the FEIII 2016 Challenge. In: (SIGMOD'16), Financial Entity Identification and Information Integration Challenge, San Francisco, US.
  • 23.
    Tourist Knowledge GraphGeneration to Automating Personal Travel Bookings @ Milano-Bicocca Turning data into actionable insights INNODEV http://innodev.ismb.it 2017-05-02 @giusepperizzo 23 MANUAL ANNOTATIONS OF PLACE-TYPE ENTITIES
  • 24.
    Tourist Knowledge GraphGeneration to Automating Personal Travel Bookings @ Milano-Bicocca Turning data into actionable insights INNODEV http://innodev.ismb.it 2017-05-02 @giusepperizzo 24 MANUAL ANNOTATIONS OF EVENT-TYPE ENTITIES
  • 25.
    Tourist Knowledge GraphGeneration to Automating Personal Travel Bookings @ Milano-Bicocca Turning data into actionable insights INNODEV http://innodev.ismb.it 2017-05-02 @giusepperizzo 25 2014 - 2016 EIT Digital funded, action line Digital Cities https://www.3cixty.com
  • 26.
  • 29.
  • 30.
  • 31.
    Will be lookingfor an Indian restaurant reachable in 10 minutes on foot from the Madison Square Garden
  • 32.
    Multitude of sources withunconnected and heterogeneous data
  • 33.
    Rizzo G., TroncyR., Corcho O., Jameson A., Plu J., Ballesteros Hermida J.C., Assaf A., Barbu C., Spirescu A., Kuhn K., Celino I., Agarwal R., Nguyen C.K., Pathak A., Scanu C., Valla M., Haaker T., Verga E.S., Rossi M., Redondo Garcia J.L. (2015) 3cixty@Expo Milano 2015: Enabling Visitors to Explore a Smart City. In: 14th International Semantic Web Conference (ISWC'15), Semantic Web Challenge 3cixty @ Milano Expo 2015
  • 34.
    4Vs: 20 DIFFERENTDATA SOURCES Variety Veracity and authority Velocity Volume
  • 35.
  • 36.
    https://www.3cixty.com Other validated pilots: Nice,London, Singapore, Madeira, Las Palmas, Amsterdam
  • 37.
    Tourist Knowledge GraphGeneration to Automating Personal Travel Bookings @ Milano-Bicocca Turning data into actionable insights INNODEV http://innodev.ismb.it 2017-05-02 @giusepperizzo AMBITION #2: BETTER OFFERING 4 TRAVELER 37 Automating the creation of personalized travel packages
  • 38.
    Tourist Knowledge GraphGeneration to Automating Personal Travel Bookings @ Milano-Bicocca Turning data into actionable insights INNODEV http://innodev.ismb.it 2017-05-02 @giusepperizzo 38 package = flights + hotel + restaurants + events + ... = a new personalized in-destination experience
  • 39.
    Tourist Knowledge GraphGeneration to Automating Personal Travel Bookings @ Milano-Bicocca Turning data into actionable insights INNODEV http://innodev.ismb.it 2017-05-02 @giusepperizzo 39 CONTEXTUAL AND EPISODIC DATA
  • 40.
    Tourist Knowledge GraphGeneration to Automating Personal Travel Bookings @ Milano-Bicocca Turning data into actionable insights INNODEV http://innodev.ismb.it 2017-05-02 @giusepperizzo 40 contextual, i.e. encyclopedic KNOWLEDGE GRAPH TYPES episodic such as transactions, user purchases, user likes ?
  • 41.
    Tourist Knowledge GraphGeneration to Automating Personal Travel Bookings @ Milano-Bicocca Turning data into actionable insights INNODEV http://innodev.ismb.it 2017-05-02 @giusepperizzo 41 EPISODIC DATA OUT THERE re. SILOED, UNCONNECTED, UNSTRUCTURED, WEAKLY-STRUCTURED, HIDDEN SEMANTICS LOCKED + privacy sensitive data
  • 42.
    Tourist Knowledge GraphGeneration to Automating Personal Travel Bookings @ Milano-Bicocca Turning data into actionable insights INNODEV http://innodev.ismb.it 2017-05-02 @giusepperizzo 42 TRAVELER DATA i.e. Giuseppe, flight number, departure, destination, …. materialized in a Knowledge Graph complementing the contextual knowledge of local offerings such as restaurants, events
  • 43.
    Tourist Knowledge GraphGeneration to Automating Personal Travel Bookings @ Milano-Bicocca Turning data into actionable insights INNODEV http://innodev.ismb.it 2017-05-02 @giusepperizzo 43 Algorithm implements a machine learning recommender blending tourist and personal data RECOMMENDATION Palumbo E., Rizzo G., Troncy R (2017) Learning User-Item Relatedness from Knowledge Graphs for Top-N Item Recommendation. Submitted to 11th ACM Recommender Systems Conference (RecSys)
  • 44.
    Tourist Knowledge GraphGeneration to Automating Personal Travel Bookings @ Milano-Bicocca Turning data into actionable insights INNODEV http://innodev.ismb.it 2017-05-02 @giusepperizzo 44 CONTEXTUAL AND EPISODIC KNOWLEDGE GRAPH u1 u2 u3 item1 item2item3 item4 item5 item6 item7 ● User ● User-Item (purchase) ● User-Item (like) ● Item ● Item-Item ● User-User (relation)
  • 45.
    Tourist Knowledge GraphGeneration to Automating Personal Travel Bookings @ Milano-Bicocca Turning data into actionable insights INNODEV http://innodev.ismb.it 2017-05-02 @giusepperizzo 45 KNOWLEDGE GRAPH USER-ITEM RELATEDNES S LEARN-TO- RANK ITEMS 3-STEP PROCESS create compute perform
  • 46.
    Tourist Knowledge GraphGeneration to Automating Personal Travel Bookings @ Milano-Bicocca Turning data into actionable insights INNODEV http://innodev.ismb.it 2017-05-02 @giusepperizzo 46 TURNING ENTITIES AND RELATIONS TO FEATURE VECTORS u1 i1 i6 purchase purchase f(u1, i1) = rel1(u1, i1) like f(u1, i7) = rel3(u1, i7) i7 purchase like u1 u1 ... Graph u2 Relatedness scores Feature vectors f(u1, i6) = rel2(u1, i6) u1 #i1 #i2 #i7
  • 47.
    Tourist Knowledge GraphGeneration to Automating Personal Travel Bookings @ Milano-Bicocca Turning data into actionable insights INNODEV http://innodev.ismb.it 2017-05-02 @giusepperizzo 47 purchase like u1 u1 ... u1 #i1 #i2 #i7 Lambda MART purchase like u1 u1 ... u1 #i2 #i7 #i1 new ranking RECOMMENDATION AS LEARNING TO RANK TASK Feature vectors
  • 48.
    Tourist Knowledge GraphGeneration to Automating Personal Travel Bookings @ Milano-Bicocca Turning data into actionable insights INNODEV http://innodev.ismb.it 2017-05-02 @giusepperizzo 48 PERSONALIZATION DIMENSIONS history Episodic, what I’ve already bought collaborative Colleagues who I know have bought willingness to pay Budget availability interests Nature, culture, sports, nightlife needs Small kids, driving licence, allergies, mobility restrictions, animals time When doing things, time pressure, aversion to queuing, advance vs on site payment, life pace travel preferences Wifi availability, meal likings, spas, noise aversion payment method Credit card, phone, bitcoin, paypal, cash ...
  • 49.
    Tourist Knowledge GraphGeneration to Automating Personal Travel Bookings @ Milano-Bicocca Turning data into actionable insights INNODEV http://innodev.ismb.it 2017-05-02 @giusepperizzo 49 2017 EIT Digital funded, action line Digital Cities http://pas-time.org
  • 50.
    Frederic, CEO ofa company, is going to give a pitch to investors in SF. He flies often over there, where he likes staying in a comfortable house rented via airbnb next to the sea to enjoying surfing during his free spots. While staying, he likes watching the Golden State Warriors once
  • 51.
  • 52.
    2) There is abunch of accommod ations you’d like the most Frederic PasTime 1) These flights will suit you the best, pick one 3) You’ll enjoy these restaurants as they prepares your favourite dish and are next to ... 4) Golden State Warriors are playing in town. Would you like to attend? package “NETFLIX” FOR TRAVELERS generates
  • 53.
    PasTime for FrenchRegional Committee of Tourism Package Persona-A Package Persona-B Package Persona-C Tourist Offices Persona-A Persona-B Persona-C Tourist Personas Tourist offices distribute and record likes of tourists. The recommendation algorithm will exploit those likes in a learning fashion http://www.freepik.com/free-vector/businesswo man-working-at-the-office_799834.htm
  • 54.
    http://www.pas-time.org Further investigated pilots: FrenchRCT, Amsterdam and Las Palmas On both - Per-group packaging - Per-traveler packaging
  • 55.
    Tourist Knowledge GraphGeneration to Automating Personal Travel Bookings @ Milano-Bicocca Turning data into actionable insights INNODEV http://innodev.ismb.it 2017-05-02 @giusepperizzo 55 RECAP: a “journey” from data to personalized automation Going through: - Knowledge Graph - Big Data - Data Integration - Personalization Any questions? https://www.slideshare.net/giusepperizzo
  • 56.
    Tourist Knowledge GraphGeneration to Automating Personal Travel Bookings @ Milano-Bicocca Turning data into actionable insights INNODEV http://innodev.ismb.it 2017-05-02 @giusepperizzo 56 ACKNOWLEDGE PROJECTS 3cixty: An Innovative Platform for Building City Apps EIT Digital, Digital Cities (2014-2016) PasTime: Packaging Personalised Tourist Experiences EIT Digital, Digital Cities (2017)
  • 57.
    Tourist Knowledge GraphGeneration to Automating Personal Travel Bookings @ Milano-Bicocca Turning data into actionable insights INNODEV http://innodev.ismb.it 2017-05-02 @giusepperizzo 57 ACKNOWLEDGE AMAZING TEAM Lots of contributions along the way: Raphael T., a mentor and a great colleague; Enrico P. and Julien P., researchers currently doing a PhD: “there is always something new to learn from young padawans”. Enrico F., Michele O., all ISMB BMPI jedis, all students I’ve supervised, and all colleagues I’ve been working with