Semantic Matching and Heuristic Search for a Dynamic Tour Guide

erminerebelAI and Robotics

Nov 15, 2013 (3 years and 9 months ago)

104 views

Semantic Matching and Heuristic
Search for a Dynamic Tour
Guide

Dipl.
-
Inf. (FH) Ronny Kramer

Prof. Dr.
-
Ing. Klaus ten Hagen

Klaus@ten
-
Hagen.org


University of Applied Sciences

www.HS
-
ZiGr.de


ENTER 2005 Research Track

2

Agenda


Semantic matching


Ontology


Algorithm


Tour computation


Benchmarks


Screenshots


DTG


Scenario


Advantages


Mobile Devices


Architecture

ENTER 2005 Research Track

3

Dynamic Tour Guide


Advantages


My tour starts and ends
where and when I
want.


The tour adapts to my
personal behaviour.


I see what I am
interested in.


I control how much
information I get.


Scenario


I have 3 hours time.
What can I do?


The DTG knows my
personal interests.


It offers me an
individually computed
tour.


I can modify the tour.


I get audible navigation
hints.

ENTER 2005 Research Track

4

Mobile Device


Platform with an OS


e.g. GUI library,
WebServices stack


Internet access


via GPRS, UMTS or
WLAN


Localisation ~1m
precision,


e.g. GPS with
WAAS/EGNOS


Graphical display


At least. 320x280
(MDA III) (VGA
480x680)


Map, user interaction


Headset (earphones)


Navigation and
information via Audio


Personal Area
Network


e.g. bluetooth

ENTER 2005 Research Track

5

Architecture

UDDI
registry
GPS
Tourist
TBB
Webservice
Authoringtool
TourBuildingBlocks
Information
Transactions
Time
&
interests
Modelling
Position
,
time
&
interests
Tour
Current information
Search
TBBs
Publishing
DTG
Webservice
Semantic matching

Tour computation

UI
DTG Agent
Navigation

Requests
Profiles
Requests
Directives
ENTER 2005 Research Track

6

Interactions

1)
Localisation using e.g. a GPS receiver

2)
Dynamic discovery


-

DTG agent discovers appropriate DTG server


-

DTG server discovers available Tour Building
Blocks (TBB)

3)
Information provision by WebServices per TBB


-

Profiles based on a common ontology


-

Audiovisual context

4)

Routing by navigation software on the mobile
device

ENTER 2005 Research Track

7

Semantic Matching


Solution


Model all TBBs using
the same ontology


Express the interests
based on a common
ontology


Semantic matching as
the ranking algorithm


Challenges


Selection of attractions
for a tour based on
individual interests


Reusable interest
profile

ENTER 2005 Research Track

8

Ontology


In computer science, an ontology is the
attempt to formulate a […] rigorous
conceptual schema within a given domain


Typically a hierarchical data structure
containing all the relevant entities and their
relationships and rules (theorems,
regulations) within that domain


http://en.wikipedia.org/wiki/Ontology

ENTER 2005 Research Track

9

Elements of an ontology


Hierarchy of concepts using “is
-
a” / subtype /
subclass relation


e.g.
baroque

“is
-
an”
architecture style


Other semantic relations


“part
-
of”:
tower

is “part
-
of”
castle


“synonym to”:
historism

is a “synonym to”
wilhelminian style


Logical rules:

„If something‘s built in baroque style,






it belongs to the 17th century.“

ENTER 2005 Research Track

10

Semantic matching algorithm



Rating the TBBs


Example: Tourist is interested in the Mediaeval



25 (2nd generalisation)



50 (1st generalisation = superclass)



100 (exact match)



100 (specialization = subclass)

Building

Type

Architecture

Castle

Tower

..

Romanic

Gothic

Modern

Mediaeval

ENTER 2005 Research Track

11

Tour computation


Optimal Tour = Sequence of TBBs with maximal
Interest Matching Points (IMP) that can be visited
in the given timeframe


Tour computation is not trivial: e.g. 20 TBBs



6 * 10
16
tours


Patience of the mobil tourist is limited


run
time
< 5 seconds



Optimization by guided depth first search and
heuristic insertion mechanisms

ENTER 2005 Research Track

12

Benchmark as in Paper

IMP loss of the best tour after 5 sec of runtime
0.00%
5.00%
10.00%
15.00%
20.00%
25.00%
30.00%
35.00%
25
105
200
350
495
660
845
1125
1360
1520
#availTBB * #TBBinTour
IMP loss
w/o pruning
w/ pruning
ENTER 2005 Research Track

13

Benchmark as in Paper

IMP loss of the best tour after 5 sec of runtime
0.00%
5.00%
10.00%
15.00%
20.00%
25.00%
30.00%
35.00%
25
105
200
350
495
660
845
1125
1360
1520
#availTBB * #TBBinTour
IMP loss
w/o pruning
w/ pruning

100 TBBs in
Destination


10 TBB in tour

ENTER 2005 Research Track

14

Benchmark as in Paper

IMP loss of the best tour after 5 sec of runtime
0.00%
5.00%
10.00%
15.00%
20.00%
25.00%
30.00%
35.00%
25
105
200
350
495
660
845
1125
1360
1520
#availTBB * #TBBinTour
IMP loss
w/o pruning
w/ pruning
Tour has
-
3.6%
less IMP than
optimal


100 TBBs in
Destination


10 TBB in tour

ENTER 2005 Research Track

15

Benchmarks for an extended
range

IMP loss of the best tour after 5 sec. of runtime
-10.00%
0.00%
10.00%
20.00%
30.00%
40.00%
50.00%
60.00%
0
2000
4000
6000
8000
10000
12000
#availTBB * #TBBinTour
IMP loss
pruning
Area of the
original diagram


1000 TBBs in
Destination


10 TBB in tour

ENTER 2005 Research Track

16

Improved Version

IMP loss of the best tour after 5 sec. of runtime
-10.00%
0.00%
10.00%
20.00%
30.00%
40.00%
50.00%
60.00%
0
2000
4000
6000
8000
10000
12000
#availTBB * #TBBinTour
IMP loss
pruning
bucket

1000 TBBs in
Destination


10 TBB in tour

Tour has
-
4.9%
less IMP than
optimal

ENTER 2005 Research Track

17

Improved Version

IMP loss of the best tour after 5 sec. of runtime
-10.00%
0.00%
10.00%
20.00%
30.00%
40.00%
50.00%
60.00%
0
2000
4000
6000
8000
10000
12000
#availTBB * #TBBinTour
IMP loss
pruning
bucket

1000 TBBs in
Destination


10 TBB in tour

Tour has
-
4.9%
less IMP than
optimal

Sufficient



Reduce
response time
from 5 to 2 sec

ENTER 2005 Research Track

18

Screenshots

Interests

Time

Map

Information

ENTER 2005 Research Track

19

Future work:

Benchmarks for Semantic Matching


Sights of Goerlitz are modelled as TBBs


Real tourists specify their interests


Compare lists and determine differences


“different interests result in different tours !?“


Let tourists rate each single sight


Compare the ratings


“Algorithm selects like a human expert !?“

ENTER 2005 Research Track

20

Acknowledgements




Marcel Hermkes for his work on the tour

calculation




VESUV project:
www.Vesuv
-
Projekt.de




funded by the Federal Ministry of

Economy (BMWA)

ENTER 2005 Research Track

21

Thanks for your attention!

Appendix

ENTER 2005 Research Track

23

Architecture

TBB Server
TBB Server
Tourist
DTG Server
Semantic matching

Tour calculation

Request

Position

Interests

Time

Request

..

profile

profile

ontology

tourlist

MapPoint

TBB‘s

distance

UDDI
TBB‘s

GPS
ENTER 2005 Research Track

24

Semantic matching algorithm



Rating the TBBs


Example: Tourist is interested in the Middle Ages

architecture
gothic
building
middle
ages
modern
times
structure
shape
tower
castle
romanesque


100
(
exact match
)


100
(
specialization
=
subclass
)

50
(
1
st generalisation
=
superclass
)

25
(
2
nd generalisation
)