Indonesian J
ournal of Ele
c
trical Engin
eering and
Computer Sci
e
nce
Vol. 2, No. 1,
April 201
6, pp. 194 ~ 20
4
DOI: 10.115
9
1
/ijeecs.v2.i1.pp19
4-2
0
4
194
Re
cei
v
ed Se
ptem
ber 21, 2015; Revi
se
d Jan
uary 24,
2016; Accept
ed Feb
r
ua
ry
10, 2016
Challenges Over Two Semantic Repositories - OWLIM
and Allegro Graph
Paria Tajabo
r
1
, Tara Raafat
2
1
Departme
n
t of Chemic
al an
d Process Eng
i
n
eeri
ng,
Surre
y
Univers
i
t
y
, Gui
l
d
ford, Unite
d
K
i
ng
dom
2
Mphasis, Un
iv
ersit
y
of Surr
e
y
, Avco S
y
stem
s Ltd
Univers
i
ty of Surre
y, Gu
ildfor
d
, United
King
dom
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: p.tajabor
@g
mail.com
A
b
st
r
a
ct
T
he purp
o
se o
f
this research
study is expl
ori
ng tw
o kind of semantic re
po
sitories w
i
th regards t
o
vario
u
s factors
to find
the
be
st appr
oach
e
s
that a
n
artifici
al ma
na
ger
c
a
n
us
e
to prod
uce ontol
ogy i
n
a
system
based
on their interac
t
ion, assoc
i
ation and res
ear
c
h
. To this end, as the best way to evaluate
eac
h
system a
nd co
mp
arin
g w
i
th others is a
naly
s
is, seve
ral b
e
n
ch
marki
ng ov
er these tw
o repos
itories w
e
r
e
exa
m
i
ned. T
h
e
s
e tw
o se
ma
nti
c
rep
o
sitori
es:
OW
LIM and A
l
l
egroGra
ph w
i
l
l
be th
e
main
co
re of th
is stu
d
y
.
T
he ge
nera
l
o
b
jectiv
e of this
study is to be
able to
cr
eate
an efficie
n
t a
nd cost-effecti
v
e mann
er rep
o
rts
w
h
ich is requ
ir
ed to sup
port d
e
cisio
n
maki
ng
in any lar
ge e
n
terpris
e
.
Ke
y
w
ords
: OWLIM,
Allegr
oGraph, RDF,
reason
ing, s
e
mantic re
pos
itor
y, sema
nti
c
-Web, SPARQL,
Ontology, Query
Copy
right
©
2016 In
stitu
t
e o
f
Ad
van
ced
En
g
i
n
eerin
g and
Scien
ce. All
rig
h
t
s reser
ve
d
.
1. Introduc
tion
It is necessa
ry to point out
that in hug
e
pr
oje
c
ts,
gov
ernm
ent p
r
ocesse
s; co
mp
anie
s
, IT
system
s and
so forth the
r
e
are differe
nt informat
io
n an
d databa
se
s. These have variou
s types
of
formats,
set
s
of form
s, d
o
cum
ents an
d re
po
sitori
e
s
, with
not v
e
ry go
od
ma
nagem
ent a
n
d
coo
r
din
a
tion.
It is obviou
s
t
hat ra
mpa
n
t redun
dan
cy wi
ll occu
r
whe
n
the
procedu
re of devel
opi
ng
or re
develo
p
i
ng in the system, u
s
in
g the sam
e
information
but with di
fferent forma
t
s,
architectu
re
s and model
s,
is being p
r
o
c
e
s
sed. Basi
cally, views
and attitude
s toward
s a wide
rang
e
of information
as we
ll as integ
r
ati
on of
a
ccu
rat
e
an
d
com
p
l
e
te un
de
rsta
nding
is difficult
and i
n
som
e
ca
se
s almo
st imp
o
ssibl
e
and
u
n
a
c
hie
v
able. On
th
e othe
r
han
d
develo
p
ing
this
informatio
n is really expen
sive and tim
e
con
s
umi
ng;
hence the p
r
oble
m
is ex
ace
r
bate
d
m
o
re
when this process
would take
place in lack of
interoperab
ility, inconsist
e
nt designs and
redu
nda
nt system
s. For th
is rea
s
o
n
, different ki
nd
s o
f
reposito
r
ie
s are provided
to replace this
loss of i
n
formation a
nd
data fra
g
me
ntation
wi
th
an
archite
c
ted ap
pro
a
ch to inte
grat
ing,
developin
g
a
nd mana
ging
informatio
n throug
hout the
system.
This
wo
rk d
e
f
ines two different
sema
nt
ic re
po
sitorie
s
an
d co
mpa
r
ing them i
n
term
s of
different pa
ra
meters and f
a
ctors. To thi
s
ai
m, besi
d
e the backg
round
s an
d exploring vari
o
u
s
theorie
s of rese
arche
s
, a
nalysi
s
throu
gh differ
ent
asp
e
ct of asse
ssm
ent
is
done. In fact
th
e
resea
r
ch obj
ectives
ca
n
be summa
ri
zed in th
e fu
lly assessm
e
nt of a
sele
cted
sem
anti
c
repos
i
tories
.
2. Backg
rou
nd and Moti
v
a
tion
Over the la
st
decade, a
s
sema
ntic
we
b ha
s
ra
pidly
develop
ed, the
effort of
system
develop
ers h
a
s
also
signifi
cantly in
crea
sed, e
s
pe
cially
in field
s
whe
r
e
no
waday
s
the imp
o
rtan
ce
of sem
antic
repo
sito
rie
s
i
s
e
qual to
HTTP se
rvers.
This ha
s le
d to succe
ssful numb
e
rs
of
ontology st
an
dard
s
a
nd ro
bust meta
dat
a and the
rol
e
of these
st
anda
rd
s is
a
k
in to the
rol
e
o
f
SQL to developing o
r
sp
re
ading
DBMS relation
s [1].
The ch
ara
c
te
ristics of a semantic repo
sito
ry are g
e
nerally simil
a
r to the data base
manag
eme
n
t system
s (DBMS) [2]. They handle
da
ta
function
s,
storin
g info
rmation, que
rying
and m
ana
gin
g
organi
ze
d
data. In fa
ct, as Dimit
r
ov
e in
his 20
1
0
conferen
ce
pointe
d
o
u
t, a
sema
ntic rep
o
sitory combi
nes featu
r
e
s
of
inferen
c
e
engin
e
s a
nd DBMS. However, the maj
o
r
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 25
02-4
752
IJEECS
Vol.
2, No. 1, April 2016 : 194 –
204
195
differen
c
e
b
e
t
ween sem
a
n
t
ic
rep
o
sito
ry and DBMS
is ontology, wh
ich i
s
just u
s
ed in
sema
ntic
repo
sito
rie
s
a
s
se
mantic
schema [3].
By developin
g
a sy
stem m
any co
nne
cti
ons
amon
g la
ngua
ge
s may
be lo
st, whi
c
h co
uld
lead to
erro
r
pron
e, la
ck of
integration,
complexi
ty an
d finally
syste
m
failures.
T
h
is i
s
sue
wou
l
d
be cata
strop
h
ic when the
r
e are co
mbi
nation
s
of
different sy
ste
m
s, multiple
contracto
r
s a
n
d
vendors with
several archit
ectural metho
d
s. “Th
e
frag
mentation of informatio
n is
profou
nd
” [4].
As previo
usly
mentioned, this proje
c
t ha
s bee
n ca
rrie
d
out to find two differe
nt semantic
repo
sito
rie
s
. As Kiryakov argu
ed, O
W
L
I
M as an i
n
fe
ren
c
e laye
r a
nd sto
r
ag
e fo
r Sesame
with
reliabl
e
p
e
rsi
s
ten
c
e strate
gy
and high
perfo
rman
ce
rea
s
oni
ng ha
s
the po
we
r
to
combin
e
O
W
L
DLP, RDFS
and OWL Ho
rst usi
ng TRREE
en
gine
[
5
].The follo
wi
ng
se
ction
wi
ll outline
O
W
LIM
in more d
e
ta
il afterward
next part
s
which
will
be
about the
ot
her
rep
o
sito
ry that is
call
e
d
allograph.
2.1. OWLIM:
A Pragmatic
Semantic Re
pository
for OWL
OWLIM
with the perfo
rma
n
ce of O
W
L
DLP re
asoni
ng is ba
se
d
on forward chainin
g
of
entitlement ru
les. Thi
s
me
a
n
s that the
g
oal is
re
acha
ble amo
ng
st
arting from a
v
ailable fa
cts or
data usi
ng in
feren
c
e rule
s for extractin
g
more
data
until that goal can
be a
c
hieved [6]. The
signifi
can
c
e o
f
OWLIM is it
s scala
b
ility over million
s
o
f
statements.
In addition it
can p
r
o
c
e
s
s a
kno
w
le
dge b
a
se of mo
re
than 10 milli
on explicit
st
atements whi
l
e,
by
using forward ch
aini
ng
rea
s
oni
ng, it has the pot
ential to extend ha
ndlin
g
of statemen
ts to arou
nd
19 million. The
importa
nt poi
nt is th
at a
c
cordin
g to th
e
size of
th
e se
mantic re
po
si
tory,
the spe
ed
of uplo
adi
ng
and
storin
g i
s
vari
ed fro
m
3000 to
180
00 state
m
ent
s pe
r
se
con
d
for a
small t
o
maximal
si
ze
repo
sito
ry. On the othe
r h
and, a
s
note
d
, OWLI
M
acts as an i
n
ference
st
rategy
, hence del
etio
n
pro
c
e
s
ses are not
chea
p a
nd it takes
a few mi
nute
s
. Another fa
ct t
o
be
noted i
s
that the amo
unt
of diverse queries will be as
sessed in m
illiseconds [7].
Although
O
W
LIM
come
s from the te
rm of OWL In-Mem
ory, a
c
cordi
ng to
Kiryakov,
Ognyan
ov a
nd Ma
nov “OWLIM i
s
th
e shor
t nam
e
of
the OWLMemSchem
aRe
p
o
s
itory SAIL
(Storag
e
an
d
Inferen
c
e L
a
y
er) for Se
sa
me, whi
c
h
su
pport
s
pa
rtial
rea
s
oni
ng o
v
er OWL DL
P”
.
As di
scu
ssed
by named
re
sea
r
che
r
s
all
conte
n
t of
this repo
sitory i
s
p
r
e
s
erve
d
and lo
ade
d from
the mai
n
me
mory, he
nce i
t
is
a
kind
of i
n
-mem
or
y re
aso
n
ing
which
is capa
ble
of
well
-o
rga
n
i
z
ed
query a
n
swe
r
ing a
nd re
covering,
sub
s
eq
uently
it has
stron
g
strategi
es f
o
r ba
ckup a
nd
per
sist
e
n
cy
.
The key featu
r
es of the
current rel
e
a
s
e o
f
OWLIM ca
n be sum
m
ari
z
ed as follo
ws:
The mo
st scalable sema
n
t
ic repo
sito
ry in t
he World
,
both in terms of the volume of RDF
data it can st
ore an
d the speed
with
whi
c
h it can lo
ad
and infere
nce.
Pure Java implementatio
n, ensu
r
ing e
a
se of deploym
ent and po
rta
b
ility.
Compatible
with Sesam
e
2,
whi
c
h
bri
ngs interoperability ben
efits
and support for all major
RDF
syntaxe
s
and q
uery l
angu
age
s [8].
Special geom
etrical quer
y constructions and SPARQL ex
tensions functions
[1].
High p
e
rfo
r
m
ance retractio
n
of stateme
n
ts and
thei
r
inferen
c
e
s
-so
inferen
c
e m
a
teriali
z
ation
spe
e
d
s
up ret
r
ieval, but without
delete p
e
rform
a
n
c
e d
egra
dation.
Powe
rful
and expre
ssive co
nsi
s
ten
c
y/int
egrity con
s
trai
nt che
cki
ng
mech
ani
sm
s.
RDF
Primin
g,
based upo
n activation sp
readin
g
,
allo
ws effici
ent dat
a sel
e
ctio
n a
nd context-
awa
r
e qu
ery
answe
ring fo
r handlin
g hug
e datasets [9]
The limitatio
ns of O
W
LI
M are
rel
a
te
d to its rea
s
oni
ng
strate
gy. In gene
ral, the
expre
ssivity
of the lan
gua
ge supp
orte
d
can
not
b
e
e
x
tended in
th
e De
scri
ption
Logi
c di
re
ction,
becau
se the
sema
ntics m
u
st be a
b
le
to be capture
d
in (Horn) rules. Th
e total materiali
z
a
t
ion
strategy h
a
s
dra
w
ba
cks
when
chan
ge
s to the ex
plicitly asse
rted
stat
ement
s o
c
cur frequ
ent
ly.
For exp
r
e
ssiv
e
sem
antics
and
certai
n o
n
tologie
s
, the
numbe
r of i
m
plicit state
m
ents
can
grow
quickly with the expe
cted
degradatio
n in perfo
rman
ce [10].
2.2. Allegro Graph
Allegro
Gra
p
h
is a
kin
d
o
f
databa
se u
s
ed
as
a fra
m
ewo
r
k for
makin
g
sema
ntic web
appli
c
ation
s
. Data and m
e
ta-data
can b
e
saved in it in the triples form and this tri
p
les
sea
r
ch (do
query
)
is possible
among
different types of qu
ery APIs such
as P
r
olog and SPARQL, along
with
the appli
c
atio
n of RDFS ++ rea
s
o
n
ing
with its bu
ilt-i
n rea
s
on
er. Allergo G
r
a
ph i
n
clu
d
e
s
su
pp
ort
for fede
ratio
n
,
so
cial n
e
twork an
alysi
s
and
gruff [11]
. Inter alia, Al
legro
G
r
aph
i
s
cha
r
a
c
teri
zed
by being mo
dern a
nd a
s
a databa
se
with a
powerful, stable
grap
hic fram
e 20 and hi
gh
Evaluation Warning : The document was created with Spire.PDF for Python.
IJEECS
ISSN:
2502-4
752
Chall
enge
s O
v
er T
w
o Sem
antic Repo
sit
o
rie
s
- O
W
LI
M and Allegro Grap
h
(Pari
a
Tajabo
r)
196
efficien
cy.
Compa
r
ed wit
h
a rational
databa
se,
a
grap
hic data
base can
ha
ve any num
b
e
r of
relation
shi
p
s
for a
n
y save
d sample.
Th
ese
relation
ships ap
pea
r i
n
the fo
rm
of
links
whi
c
h
take
the form of
a network o
r
gra
ph in
co
mbinati
on
wit
h
ea
ch oth
e
r. Combi
ned
with di
sk
ba
sed
stora
ge, Alle
gro G
r
ap
h efficiently make
s use of
the memory. Thi
s
cau
s
e
s
a bet
ter perfo
rma
n
c
e
and mainte
n
ance, while a
t
the same time; billions o
f
quads
can
be mea
s
u
r
ed
. Among a very
high number of c
lient applic
ation programs
,
AllegroGraph
s
u
pports
SPARQL,
RDFS ++
and
prolo
g
re
ason
ing [12].
Expressive a
nd po
we
rful queryin
g
an
d
rea
s
oni
ng is one of the
posit
ive poi
nts of this
approa
ch. In
fact, Allegro
Grap
h en
able
s
the mo
st ex
pan
sive set of array
s
for q
uery an
d a
c
cess
to information
in the RDF d
a
tastore. De
scriptio
n logi
cs or O
W
L
-
DL
reasone
rs
are
more
cap
abl
e
of man
agin
g
compl
e
x o
n
to
logie
s
. Th
ey
try to be
com
p
lete a
n
d
su
ccessful i
n
re
spo
ndin
g
to
all
the que
ries,
yet, on the other h
and, wh
en increa
si
n
g
the numbe
r
of triples to
millions; they
ac
t
compl
e
tely different and a
r
e unp
redi
cta
b
le in te
rms
of execution
time. Franze (201
0) beli
e
ves
that Allegro G
r
aph p
r
ovid
es a very hi
gh speed rea
s
on
e
r
and p
r
a
c
tica
l RDFS
++.
Some Featu
r
es of Allegro Grap
h are poi
nted belo
w
:
Social Netwo
r
kin
g
Analysi
s
[11]
Native data types a
nd Efficient rang
e qu
erie
s [13]
Fede
ration
[1
1]
Gruff
[14]
Reg
a
rdi
ng the figure 1
,
the stru
cture
of
Allegro grap
h ca
n be better und
ersto
od.
Acco
rdi
ng to t
he inferen
c
e
of Fran
z, it sh
ould
b
e
menti
oned th
at “All
egr
o
G
raph provides a
RES
T
proto
c
ol a
r
chi
t
ecture
whi
c
h
is essentially
a supe
rset of the sesame
HTTP cli
ent”.
Fran
z
staff di
rectly
cove
rs ada
pters fo
r di
fferent
lan
guag
es like
sesam
e
java,
se
same
Jen
a
, Python usin
g the se
same sig
natu
r
e and Li
sp.
Figure 1
.
Architecture of Allegro g
r
a
ph [11]
3. Methodol
og
y
In this part th
e ways that
has b
een u
s
ed in
this p
r
o
j
ect to ben
ch
mark rep
o
sit
o
rie
s
and
achi
eve som
e
rea
s
on
able
results, are o
u
tlined a
s
follows:
3.1. Information Sourcing
First
of all to
begin
n
ing t
he an
alysi
s
,
I neede
d h
u
ge am
ount o
f
ontologie
s
,
and the
rea
s
on
wa
s t
o
put the
s
e
ontologi
es to
the sel
e
cte
d
repo
sito
rie
s
for que
rying
and a
c
cordin
gly
ben
chma
rkin
g. To this en
d load
s of ontologie
s
we
re
taken from t
he main web
site of ontology
whi
c
h is (Wa
t
son.kmi.ope
n). In additio
n
, a clas
s of ontology fro
m
my point
of view has
been
made a
nd im
ported it into
one of ontol
o
g
ies, this
effort wa
s don
e to evaluate fa
ster, ea
sie
r
a
nd
for more accu
racy an
d co
m
p
letene
ss of queri
e
s.
3.2. Benchm
arking and
Analy
s
is
The
aim
of th
is p
a
rt i
s
be
n
c
hma
r
king
th
e sele
cted
se
mantic re
po
si
tories for
co
mpari
s
on
betwe
en thei
r different
com
pone
nts a
nd f
i
nally a
con
c
l
u
sio
n
which
will be
ba
sed
on the o
b
tain
ed
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 25
02-4
752
IJEECS
Vol.
2, No. 1, April 2016 : 194 –
204
197
data. The
s
e repo
sitorie
s
a
r
e con
s
id
era
b
l
y
different
in the amou
nt of their levels
and that mig
h
t
have influen
ce on the final system
s o
n
deploy
me
n
t
decisi
on. For exampl
e, load and q
u
e
ry
response tim
e
s,
scalabilit
y, s
upported query la
nguages, semantic expre
ssivi
ty or
reasoni
ng
capability. Therefore, to
achiev
e this goal
some qu
eries were
provided
to use for
searchi
ng
among t
r
iple
s and findi
ng t
he rel
e
vant result
s. Que
r
i
e
s a
r
e d
e
si
gn
ed from
simpl
e
till compl
e
x for
better asse
ssing in different situation
s
and they
were tested
sep
a
rately in
each repo
si
tory
rega
rdi
ng to
their diffe
rent
si
ze
s from
small to me
di
um si
ze
onto
l
ogie
s
. It will
be te
sting t
w
o
sema
ntic re
p
o
sitori
es in t
e
rm
s of their dat
aset Loa
d time, query result
s and
query exe
c
uti
o
n
spe
ed. By a
d
d
ing va
riou
s
ontologi
es in
variou
s
si
z
e
s into
s
e
lec
t
ed repos
i
tories
,
a total of four
different dataset si
zes were creat
ed to test reposit
o
ries in di
fferent situations which
will
be
clarifie
d more
in the prac
ti
cal part of proj
ect.
4.
Finding
and
Analy
s
is
4.1. Benchm
arking Fac
t
o
r
s
Some facto
r
s upon both re
posito
r
ie
s will
be briefly pre
s
ente
d
in the Table 1.
Table 1. Feat
ure
s
of rep
o
si
tories
Fac
t
ors
Allegrograph
BigO
WLIM
Storage fo
rm
Native-based
Memor
y
and nati
v
e-
based
Quer
y La
nguage
Support
SPARQ
L
, TWI
N
Q
L
,
SeRQL a
nd Prol
og
SPARQL and
SeRQ
L
Reasoner
Integration
Built-in, Jena,
Rac
e
rpro, S
e
s
a
me
Built-in, Sesame
RDF Up
date
API
API
Reasoning tactic
Backw
ard chaini
ng
For
w
a
r
d chaining
Client part
PL/SQL, Java an
d C
java
Platform maintained
Unix, Windo
w
s
,
Solaris and Mac
Unix, Windo
w
s
,
Solaris and Mac
RDF vie
w
suppor
t
Not definitive
No
Format of
Ser
i
alization
N-T
r
iples, RDF-X
ML
and N3
N-T
r
iples, RDF-
XML and N3
4.2. Test Des
c
ription
The next leve
l of doing this benchma
r
k i
s
testi
ng th
ese two ki
nd of
sema
ntic rep
o
sitori
es
in term
s of th
eir d
a
taset L
oad time,
qu
ery results a
n
d
qu
ery exe
c
ution
spee
d.
The b
a
seline
test
is ru
n on
sm
all and m
ediu
m
size ontol
o
g
ies. By addi
ng vario
u
s
o
n
tologie
s
in v
a
riou
s
sizes i
n
to
sele
cted
rep
o
s
itorie
s, a tot
a
l of four
different
data
s
et
sizes we
re
created
to test
rep
o
sito
rie
s
in
different situ
ations. In ot
her
words, t
h
is te
st
ing is perfo
rmed
with the follo
wing va
riatio
n in
ontologi
es-si
z
e con
d
ition
s
(Table 2
)
.
Table 2. Onto
logy sizes
cond
itio
ns
1
2
3
4
On
tolo
gies
or tr
iples sizes
10000
50000
70000
100000
Furthe
rmo
r
e
to better a
s
se
ss th
ese repo
sitori
e
s
i
n
different
si
tuations, 5
d
i
fferent
queri
e
s
we
re
cre
a
ted fro
m
simple to
compl
e
x,
and
then they were te
sted
se
parately in e
a
ch
repo
sito
ry reg
a
rdin
g to their different si
ze
s (Ta
b
le 3).
Table 3. Que
r
ies
Q
u
e
r
i
es
Q1
Q2
Q3
Q4
Q5
Evaluation Warning : The document was created with Spire.PDF for Python.
IJEECS
ISSN:
2502-4
752
Chall
enge
s O
v
er T
w
o Sem
antic Repo
sit
o
rie
s
- O
W
LI
M and Allegro Grap
h
(Pari
a
Tajabo
r)
198
Hen
c
e
given
these
4 ontol
ogy si
ze
s an
d 5 que
rie
s
,
a total of 20
test co
ndition
s have
been
pe
rform
ed o
n
2
different
rep
o
sito
ries
whi
c
h l
e
a
d
to 40
resul
t
s that
will b
e
illust
rated
i
n
following part
s
.
4.3. Data
se
t Load Timings
First
of all
l
oadin
g
a
pa
rticular o
n
tolo
gy
from
all f
o
ur data
s
et
sizes,
ba
sed
on
two
selected repositori
es has been done with t
he help of SPARQL statements
whi
c
h is „
‟
load
transpo
rtation
’
’
,
can easily cal
c
ulate lo
ad
time (Figure 2).
Figure 2. Tra
n
sp
ortation o
n
tology loadi
ng time
It can b
e
cl
ai
med that Allo
grap
h h
ad th
e be
st pe
rformance in te
rms of lo
adin
g
time and
they
are con
s
iste
nt
an
d scala
b
le agai
n
s
t
in
crea
sing
data
s
et
size
and
BigO
WLIM is the l
o
wer
repo
sito
ry be
cau
s
e it perfo
rms fo
rward
-
chainin
g
of facts and sto
r
e
s
them explicitl
y
[15].
4.4. Quer
y
R
esults a
nd Execu
tion Speed An
aly
s
is
By performin
g provide
d
qu
erie
s, ope
rati
on of sea
r
chi
n
g and findi
n
g corre
c
t data
will be
activated an
d
return
ed ba
ck to the bro
w
ser.
As mentio
ne
d befo
r
e to
evaluate b
e
tter, I mad
e
a
cla
s
s from
my point of
view an
d
importe
d it into the ontolo
g
y of transpo
rtation,
hen
ce co
ncept of all five queri
e
s are ba
se
d
on
t
h
is cla
s
s.
Quer
y
1:
This i
s
a rela
tively simple
query
whi
c
h
wa
s de
sign
e
d
to find a dif
f
erent ki
nd of
CityCar
as well as tho
s
e cl
asse
s which a
r
e di
sjoi
nted with it.
SELECT ?y ?x W
H
ERE
{ ?x rdfs:subClassOf “CityCar
”
.
?y owl:disjoint
W
i
t
h
“CityCar”
. }
Figure 3. Que
r
y 1 outcom
e
s
0
20
Data
‐
set01
D
ata
‐
set02
D
ata
‐
set03
D
ata
‐
set04
Time(second)
loading
time
Alle
g
r
oG
raph
Bi
gOW
L
IM
0
2000
Data
‐
set01
D
ata
‐
set02
D
ata
‐
set03
D
ata
‐
set04
TIME(SECOND)
DATA
SET
SIZES
Bi
g
Owlim
Alle
g
r
og
raph
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 25
02-4
752
IJEECS
Vol.
2, No. 1, April 2016 : 194 –
204
199
From the
Fig
u
re 3
ca
n be
claime
d that
this
que
ry g
enerally retu
rns relatively quickly,
rega
rdl
e
ss of
ontology
size. Both store
s
were
co
mp
etitive up to dataset
04,
w
i
th sub-s
e
cond
res
pon
se t
i
m
e
s.
Table 4. Re
sults of Que
r
y 1 in detail
Repos
itor
y
Data-set size
A
n
sw
er
Correct a
n
s
w
er
Min ti
me (
m
sec
)
Triple c
o
u
n
tin
g
Allegrograph
10000
yes
yes
160
137
Big Ow
lim
10000
yes
yes
73
107
Allegrograph
50000
yes
yes
253
67
Big Ow
lim
50000
yes
yes
97
85
Allegrograph
70000
yes
yes
512
124
Big Ow
lim
70000
yes
yes
214
97
Allegrograph
100000
yes
yes
1257
74
Big Ow
lim
100000
yes
yes
116
114
From the tabl
e 4 can b
e
se
en that all both store
s
yield
ed usable d
a
ta for this que
r
y.
Quer
y
2:
This qu
ery was de
sig
ned t
o
find any rel
e
vant informa
t
ion about sp
ecific
cu
stom
er, su
ch a
s
“pa
r
ia”.
SELECT ?x ?y ?
z
?
w
?p
?R ?N
W
H
ERE
?
z
rdfs:subClass
Of ?w
{ “Paria
”
rdfs:sub
ClassOf ?x
?R rdfs:subClassOf ?N
?x rdfs:subClassOf ?y
?p rdfs:subClassOf ?R
?y rdfs:subClassOf ?
z
?w
rdfs:subClassOf ?p .
}
Figure 4. Que
r
y 2 outcom
e
s
As it is sh
own in the Figu
re 4, this qu
ery
generally re
turns
relativel
y
quickly, reg
a
rdle
ss
of Ontolo
gy size. It seem
s
that AllegroG
raph
did
b
e
t
te
r
on
th
is
qu
er
y, w
i
th
r
e
s
p
on
s
e
s
b
e
l
ow
10
0
(m/se
c
) acro
ss the boa
rd.
OWLIM
wa
s clo
s
e be
hind.
Table 5. Re
sults of Que
r
y 2 in detail
Repos
itor
y
Data-set size
A
n
sw
er
Correct a
n
s
w
er
Min ti
me
(msec
)
Triple
coun
tin
g
Allegrograph
10000
yes yes 31
88
Big
Ow
lim
10000
yes yes 48
86
Allegrograph
50000
yes yes 46
107
Big
Ow
lim
50000
yes yes 72
115
Allegrograph
70000
yes yes 69
115
Big
Ow
lim
70000
yes yes 84
79
Allegrograph
100000
yes yes 78
132
Big
Ow
lim
100000
yes yes 97
112
From the Ta
b
l
e 5, it can be
seen that bot
h
store
s
yield
ed usable d
a
ta for this que
r
y.
0
100
Data
‐
set01
D
ata
‐
set02
D
ata
‐
set03
D
ata
‐
set04
TIME(SECOND)
DATA
SET
SIZES
Bi
g
Owlim
Alle
g
r
og
raph
Evaluation Warning : The document was created with Spire.PDF for Python.
IJEECS
ISSN:
2502-4
752
Chall
enge
s O
v
er T
w
o Sem
antic Repo
sit
o
rie
s
- O
W
LI
M and Allegro Grap
h
(Pari
a
Tajabo
r)
200
Quer
y
3:
This query
which is relatively co
mplex
attempts
to find different
k
i
nd
s
o
f
Sp
or
t Ca
r
p
l
us
SU
V
ca
r
.
After that, the query d
e
fine
s tho
s
e
cars
whi
c
h a
r
e
City Car that u
s
e Fuel a
nd re
turns
ba
ck th
eir
types
.
SELECT ?
x
?
y
?
z
?w
WHE
R
E
“FuelT
y
pe
” rdfs:s
ubClassOf ?z
{ ?x
rdfs:subClassOf “SUVCa
r”
?
y
rdfs:subClass
Of “Spo
rtsCar>
”
?z rdfs:subClassOf “Cit
yCa
r
”
?w
rdfs:subClassOf “FuelT
y
pe” .
}
Figure 5. Que
r
y 3 outcom
e
s
Allegro
g
ra
ph
had the
lo
west respon
se
times u
p
to
dataset04,
with O
W
LIM
a clo
s
e
se
con
d
. Ho
wever, their re
spon
se at data
s
et04
wa
s 5 times fa
ster th
an Allegrogra
ph (Fig
ure 5
)
.
Table 6. Re
sults of Que
r
y 3 in details
Repos
itor
y
Data-set size
A
n
sw
er
Correct a
n
s
w
er
Min ti
me (
m
sec
)
Triple c
o
u
n
tin
g
Allegrograph
10000
yes
yes
1495
90
Big Ow
lim
10000
yes
yes
48
112
Allegrograph
50000
yes
yes
4532
107
Big Ow
lim
50000
yes
yes
72
85
Allegrograph
70000
yes
yes
4376
115
Big Ow
lim
70000
yes
yes
84
112
Allegrograph
100000
yes
yes
174321
105
Big Ow
lim
100000
yes
yes
97
100
Quer
y
4:
This q
u
e
r
y has b
een p
r
ov
ided to sho
w
details of
sev
e
ral
cu
stome
r
s, it deals
wit
h
larg
e
string
s. T
he
result
set g
r
ows lin
early
with o
n
to
logy
si
ze. T
h
is q
uery
gene
rall
y returns fairly
quickly at sm
all ontology si
ze
s,
but slo
w
l
y
at larger si
zes.
SELECT ?y ?X ?
z
?
A
?B ?C
?
m
?
n
?o ?T ?
U
?V W
H
ERE
?y rdfs:subClassOf ?A
?C rdfs:subClassOf ?o
{“T
a
ra: rdfs:subC
lassOf
?X
?A rdfs:subClassOf ?
m
“Zoh
re
”
rdfs:subClassOf ?
z
?B rdfs:subClassOf ?n
“Paria
”
rdfs:subC
lassOf ?y
?m
rdfs:subClassOf ?T
?
z
rdfs:subClass
Of ?C
?n rdfs:subClassOf ?U
?X rdfs:subClassOf ?B
?o rdfs:subClassOf ?V .
}
Figure 6. Que
r
y 4 outcom
e
s
0
100000
200000
Data
‐
set01
D
ata
‐
set02
D
ata
‐
set03
D
ata
‐
set04
Bi
g
Owlim
Alle
g
r
og
raph
0
50000
100000
Data
‐
set01
D
ata
‐
set02
D
ata
‐
set03
D
ata
‐
set04
Bi
g
Owlim
Alle
g
r
og
raph
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 25
02-4
752
IJEECS
Vol.
2, No. 1, April 2016 : 194 –
204
201
It can be
se
e
that Allegro
G
rap
h
ha
d a
failure
at d
a
t
aset04 t
r
iple
s. And O
w
lim
yielded
usa
b
le data a
t
all ontology sizes (Figu
r
e
6).
Table 7. Re
sults of Que
r
y 4 in detail
Repos
itor
y
Data-set size
A
n
sw
er
Correct a
n
s
w
er
Min ti
me
(msec
)
Triple
coun
tin
g
Allegrograph
10000
yes
Y
e
s
113
113
Big Ow
lim
10000
yes
Y
e
s
116
110
Allegrograph
50000
yes
Y
e
s
583
95
Big Ow
lim
50000
yes
Y
e
s
451
116
Allegrograph
70000
yes
Y
e
s
10001.5
330
Big Ow
lim
70000
yes
Y
e
s
1948
115
Allegrograph
100000
No
-
-
-
Big Ow
lim
100000
yes
yes
91841
113
OWLIM p
e
rf
orme
d the b
e
st on this q
uery.
Alle
g
r
oG
r
a
ph
r
u
n
n
i
ng
th
e
s
l
ow
est a
t
th
e
highe
r o
n
tolo
gy si
ze
s. Allegro
G
raph
h
ad
some
of
t
he fa
ster tim
e
s
at Data
se
t01, 02 a
n
d
03;
however, it did not yield re
sult
s at data
s
et04 triple
s (T
able 7).
Quer
y
5:
This qu
ery is
desi
gne
d to test disjoi
nts. Act
ually this query trie
s to find all classes whi
c
h
are
disj
ointed
with Ve
hicl
e, as well a
s
th
at disj
ointed
cla
s
s which h
a
s Be
nz in it
s su
bcl
a
sse
s
a
nd
also tho
s
e cl
asse
s whi
c
h
have City Ca
r in their
su
b
c
la
sses an
d are di
sjointe
d
with BMW3. This
is the mo
st complex que
ry
in this proje
c
t.
SELECT ?x ?y ?
z
?
v
?w W
H
ERE
?v rdfs:subClassOf ?x
{ ?y owl:disjointW
i
th “Vehicle
”
“Ben
z
”
rdfs:subC
lassOf ?v
?
z
rdfs:subClass
Of “Vehicle
”
?w rdfs:subClassOf “CityCa
r
”
?x owl:disjoint
W
i
t
h
“Vehicle
”
“BM
W
3
”
owl:disjoint
W
i
th ?w . }
“CityCar”
rdfs:subClassOf
?
z
Figure 7. Que
r
y 5 outcom
e
s
This qu
ery a
s
is sh
own in the figure 7 gene
ra
lly retu
rns fairly qui
ckly at small ontology
sizes, but sl
owly at larg
er
si
ze
s. And also Alleg
r
oGra
ph did
not yield usable data a
b
o
ve
dataset01 si
zes.
Table 8. Re
sults of Que
r
y 5 in detail
Repos
itor
y
Data-set size
A
n
sw
er
Correct a
n
s
w
er
Min ti
me
(msec
)
Triple
coun
tin
g
Allegrograph
10000
yes
yes
4931
127
Big
Ow
lim
10000
yes yes 98
120
Allegrograph
50000
No
-
-
-
Big Ow
lim
50000
yes
yes
313
83
Allegrograph
70000
No
-
-
-
Big Ow
lim
70000
yes
yes
983
62
Allegrograph
100000
No
-
-
-
Big Ow
lim
100000
yes
yes
22509
113
0
20000
40000
Data
‐
set01
D
ata
‐
set02
D
ata
‐
set03
D
ata
‐
set04
Bi
g
Owlim
Alle
g
r
og
raph
Evaluation Warning : The document was created with Spire.PDF for Python.
IJEECS
ISSN:
2502-4
752
Chall
enge
s O
v
er T
w
o Sem
antic Repo
sit
o
rie
s
- O
W
LI
M and Allegro Grap
h
(Pari
a
Tajabo
r)
202
Allegro G
r
a
p
h
only ran
at dataset0
1
, b
u
t had re
sp
o
n
se time
s th
at were mu
ch highe
r
than the su
b-se
con
d
re
spo
n
se
s of OWLIM (Table 8
)
.
4.5. Summary
As ou
r be
nch
m
arking
over these
2 re
p
o
sitori
es
hav
e finish
ed, I woul
d like to
highlight
some
we
akn
e
sse
s
and
st
rength
s
of e
a
ch tri
p
le
sto
r
e rega
rdin
g these in
dicated re
sult
s a
s
a
summ
ary of this pa
rt of pro
j
ect.
1.
AllegroGraph was gen
erally slower than OWLIM, especially for larg
er ontology
sizes. It
co
uld not execute query 5 for larger ontologies.
This was probably the
hardest que
ry for both systems. In addition it
performed better than
OWLIM o
n
query 2.
2.
OWLIM did t
he best on query 3. It ha
d a close se
cond to Alle
groGraph o
n
query 2. A
part of these
presented p
o
int are brief
l
y provided in Table 9 as a short summary.
Table 9. Sum
m
ary of Que
r
y Analysis
Quer
y
Q1
Q2
Q3
Q4
Q5
Q6
A
lle
g
r
og
ra
ph
Dataset04
Dataset02-04
Big O
w
lim
Has correc
t
da
t
a
- do
ne
Did n
o
t ru
n or
ti
me o
u
t
Incorrec
t
da
ta
5. Conclusio
n
s and Fu
tur
e
Work
Based
on the
evaluation ef
fort, analysts
have
und
erst
ood that “the
curre
n
t state
of RDF
persiste
n
t sto
r
es, “tri
ple sto
r
es”,
is not at a sufficie
n
tly
mature level to justify reco
mmendi
ng thei
r
use in a p
r
od
uction
system
being u
s
ed o
n
a daily basi
s
” [9], [16].
5.1. Ev
aluation Summar
y
If we want t
o
summa
rize
the evalu
a
tion b
r
iefl
y, we should
say that it wa
s
felt that
OWLIM
pe
rfo
r
med
the
be
st. OWLIM
wa
s often
mu
ch
better
at the
lower
ontolo
g
y si
ze
s, up
to
dataset02.
Ho
wever, at
the
larg
er ontology
si
ze
s, dat
a
s
et
03 an
d dat
aset0
4
, OWLIM
‟
s
throug
hput d
r
opp
ed off dramati
c
ally in many ca
ses. Thi
s
assessment, wh
ich was b
a
sed
prima
r
ily on p
e
rform
a
n
c
e a
nd stability consi
derat
ion
s
, has been e
v
aluated in three mai
n
levels.
These levels
were; Gene
ra
l, Functional
and Perfo
r
ma
nce.
5.2. General
Ass
essme
n
t
In the following Table 1
0
, we captured
some im
porta
nt factors tha
t
we faced wi
th them
durin
g this p
r
oject an
d pla
y
a significa
nt role in asse
ssing e
a
ch system.
Table 10. Ge
neral a
s
se
ssment
Parameters
Feat
ures
A
lle
g
r
og
ra
ph
Big O
w
lim
Usabili
t
y
Easy
t
o
install
100%
25%
Easy
t
o
develop
75%
75%
O
v
er
all usability
75%
75%
Supp
ort
Detecting defect
s
15%
15%
Documentation
100%
100%
Overall Support
50%
80%
Licensi
ng
ho
w
w
o
rks
50%
50%
Generally, OWLIM got hi
gh ma
rks for usability a
nd support.
And licensi
ng of all
approa
che
s
i
s
in the middl
e level.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 25
02-4
752
IJEECS
Vol.
2, No. 1, April 2016 : 194 –
204
203
Allegro
G
ra
ph
got goo
d ma
rks for u
s
abili
ty, and
overal
l sup
port
due
to the willin
g
ness of
the Fran
z tea
m
to engage
and wo
rk on issue
s
[11
]. Howeve
r, they
do not have a defect tracki
ng
sy
st
em.
5.3. Functio
n
al Asse
ssm
ent
An overall qu
alitative asse
ssment of
perceptio
ns i
s
ca
ptured in T
abl
e 11.
Table 11. Fu
nction
al asse
ssment
Parameters
Feat
ures
A
l
l
e
gro
g
rap
h
Big
O
w
lim
Sy
st
e
m
Architecture and
overall view
60%
95%
Data i
m
p
o
rt
batch loading
40%
90%
data formats sup
ported
40%
90%
overall view
40%
90%
AP
I
SPARQ
L
90%
90%
overall view
50%
85%
Quer
y
i
ng
overall view
50%
90%
Infren
c
y
in
g
overall view
-
90%
Intero
perabi
lit
y
overall view
50%
50%
Oper
a
t
io
nal
overall view
-
40%
5.4. Perform
a
nce Asse
ss
ment
As 2
sele
cte
d
semanti
c
repo
sitorie
s
were
analy
z
ed
, it can
b
e
claimed th
at,
overall,
based o
n
pe
rforma
nce an
d usability, OWLIM
wa
s
deeme
d
the
best
rep
o
sit
o
rie
s
(ba
s
ed
on
Table 11
).
5.5. Future
Work
“Utilizi
ng sem
antic web
technolo
g
ies
in comm
ercial a
pplication
s
re
quire
s co
nfidence
by
the deci
s
io
n make
rs that the und
erlying
semanti
c
re
p
o
sitori
es
can
deliver the re
quire
d quality
o
f
servi
c
e while managi
ng the
overhea
d of pro
c
e
ssi
ng th
e metadata o
f
potentially huge amo
unt of
informatio
n o
r
gani
ze
d in
complex taxon
o
mies” [1
7]. This stu
d
y e
x
plore
s
the
a
nalyzin
g of t
w
o
sema
ntic
rep
o
sitori
es
and
my idea to
ben
chma
rkin
g better a
nd
more
accu
rat
e
for future a
nd
further
work i
s
expan
ding
factors of asse
ssm
ent.
This in
cludin
g
addin
g
more triples to billi
on
one
s, pe
rformance im
pa
ct of
simult
aneo
us u
s
e
r
s an
d tran
sa
ction-rel
a
te
d p
r
o
c
e
s
ses,
modificatio
n
test such a
s
in
se
rtion,
updat
e and
deletion ope
ration
s.
Fu
rtherm
o
re, so
me
addition
al su
pport
s
can b
e
provid
ed t
o
improve
th
e que
ry re
sp
onse times.
For exa
m
ple,
to
ensure
the
fresh
n
e
s
s of
d
a
ta, so
me
efficient
upd
ate
s
o
n
do
cume
nts
can
b
e
o
r
gani
ze
d. Also
query optimi
z
ation tech
niq
ues
can h
e
lp
to improve th
e query respo
n
se time.
Based
on
the
s
e m
ention
e
d
point
s I can
summ
ari
z
e
m
y
recomme
nd
ation for future work
in as follow:
deployme
nt of more re
posit
orie
s
developm
ent of more ontol
ogy
development of SPARQL syntax
in more c
o
mplex queries
more compl
e
x operation
s
in repo
sitori
e
s
to find out more accu
rate about re
spo
n
se time
su
ch a
s
upd
a
t
e, deletion, modificatio
n
and so on
And finally, in
future I
woul
d like
to exa
m
ine
oth
e
r
s
e
mantic
re
po
sitory
a
r
chite
c
ture
s to
explore m
o
re
about ea
ch di
fferentiation a
s
pe
cts.
Referen
ces
[1]
Seab
orn
e
, And
y
. SPARQL Qu
er
y
L
a
n
g
u
age f
o
r RDF
. [Online] janu
ar
y
28,
200
8. [Cited: Jul
y
28,
2011.] http://
w
w
w
.
w
3
.org/T
R/
rd
f-sparql-quer
y
/#grammar.
[2]
Bench
m
arkin
g
over a Se
ma
nti
c
Repos
itory.
Yadav, Pran
ju
l a
nd Sama
la, Vin
i
th. 2010, IEEE
, ICoAC,
pp. 51-5
9
.
[3]
Wiley
,
John and Sons
. Ontology
Evolution.
S
e
mantic W
eb T
e
chn
o
lo
gi
es.
onlin
e : W
ile
y
,
2
006, pp. 5
1
-
70.
[4]
Desig
n
i
ng a Se
ma
ntic Re
posit
ory.
Casan
a
ve,
Cor
y
. 2
007, M
ode
l Drive
n
Sol
u
tions, pp. 1-
5.
[5]
T
auberer, Josh
ua. W
hat is RD
F
and
w
hat is it
good for?
rdfabout.
[Onlin
e] Janu
ar
y
20
08. [Cited:
August 05, 2
0
1
1
.] http://
w
w
w
.
r
d
fabo
ut.com/in
t
ro/.
Evaluation Warning : The document was created with Spire.PDF for Python.