Centre Universitaire d’Informatique
Institute of Information Service Science
A	Simple	Tags	Categoriza2on	
Framework	Using	Spa2al	Coverage	
to	Discover	Geospa2al	Seman2cs		
Tardy,	C.,	Moccozet,	M.,	Falquet,	G.	
Workshop	OD4LS		-	WWW2016	Companion	
April	2016		Montréal,	Canada
Filling	the	gaps	
Tardy	et	al.,	University	of	Geneva	 2
Filling	the	gaps	
•  Maps	missing	geographic	features	characteris@cs	
•  VGI	sources	with	tags	
•  geographic	and	spa@al	axis		
•  Non	sta@s@cal	approach	
•  Must	work	on	places	with	small	amount	of	data	
Tardy	et	al.,	University	of	Geneva	 3
Model	
•  Using	tagged	photos	as	source	
•  Mul@-facets	model	
	
•  Algorithm	by	elimina@on	
•  Based	on	spa@al	coverage	
Tardy	et	al.,	University	of	Geneva	 4
Spa@al	Coverage	
•  A	combina@on	of	seman2c	and	geographic	
no@ons	that	describes	precisely	the	spa2o-	
temporal	region	shown	in	the	picture.		
•  dis@nc@on	between		
– the	geographic	and	spa@al	content	of	a	resource	
– and	its	actual	spa@al	coverage.		
Tardy	et	al.,	University	of	Geneva	 5
Categories	
Geographic	
Features	
Geographic	
Class	
Event	
Temporal	
Weather	
Actor	
Meta	
Color	
Tardy	et	al.,	University	of	Geneva	 6
Algorithm	
Geo	
	iden@fica@on	
Tag	in	coverage	
Yes	 No	
Tag	in	category	
No	
Yes	
Select	
No	
Select	
Yes	
Tardy	et	al.,	University	of	Geneva	 7
Geo	Iden@fica@on	
Local	Area	&	
Hierarchy	mapping	
Un@l	match		
Query	GeoNames	for	each	hierarchy	
Neighbourhood	
County	
Region	
Country	
World	
If	match	:	
•  Save	geonames	en@ty	in	database		
•  Associate	with	tag	
•  Save	hierarchy	level	
Tardy	et	al.,	University	of	Geneva	 8
Geo	Iden@fica@on	&	coverage	Example	
https://flic.kr/p/m9ZBPB 	
In	coverage	:		
	
à	Théâtre	Pitoëff	
Tardy	et	al.,	University	of	Geneva	 9
NonGeo	Tag	Iden@fica@on	
1. Tag	language	detec@on	using	Wik@onary	/	WordNet	
2. Tag	disambigua@on:	
1.  Tags	of	same	photo	as	set	
2.  Babelfy	and	BabelNet	iden@fica@on	grouped	by	language	
3.  If	needed	disambigua@on	Geo	–	nonGeo	comparing	both	
weights	
Tardy	et	al.,	University	of	Geneva	 10
NonGeo	Tag	Categorisa@on	
•  Using	BabelNet	categories	extracted	from	
Wikipedia		
•  or	by	Hand	
•  Associate	each	tag	to	a	category	
Event	
Temporal	Weather	
Actor	
Meta	
Color	
Tardy	et	al.,	University	of	Geneva	 11
Tag	Categorisa@on	Example	
Actor	
Event	
Temporal	
Color	
Meta	
The	rest	of	the	tags	are	:		
	
•  Ambiance	
•  Nightlife	
•  Music	/	Musique		
•  Concert	/	Gigs	
•  Shows	
•  Fes@val	
	
Tardy	et	al.,	University	of	Geneva	 12
Geo	feature	target	
•  Flickr	photo	loca@on	info	:	
<loca2on	la2tude="46.193959"	
longitude="6.143385”	accuracy="16"	context="0"	
place_id="EDcBbVFWWrj07WE”	woeid="782861">	
•  OpenStreetMap	
Tardy	et	al.,	University	of	Geneva	 13
Missing	/	Wrong	Characteris@cs	
Tardy	et	al.,	University	of	Geneva	 14
Finding	characteris@cs	
Tardy	et	al.,	University	of	Geneva	 15	
The	rest	of	the	tags	are	:		
	
•  Ambiance	
•  Nightlife	
•  Music	/	Musique		
•  Concert	/	Gigs	
•  Shows	
•  Fes@val	
	
In	coverage	:		
	
à	Théâtre	Pitoëff
Conclusion	
•  Works	on	geographic	zones	with	low	density	of	
resources	
•  Can	be	used	as	a	pre-treatment	to	sta@s@cal	
approach	
•  Enhancement	of	spa@al	descrip@ons	in	geo-
services	(Citygml,	OpenStreetMap)	
•  Precise	queries	in	search	engines	
Tardy	et	al.,	University	of	Geneva	 16

A simple tags categorization framework using spatial coverage to discover geospatial semantics