Int
ern
at
i
onal
Journ
al of Ele
ctrical
an
d
Co
mput
er
En
gin
eeri
ng
(IJ
E
C
E)
Vo
l.
9
, No
.
6
,
Decem
ber
201
9
, p
p.
5463~
5470
IS
S
N: 20
88
-
8708
,
DOI: 10
.11
591/
ijece
.
v9
i
6
.
pp5463
-
54
70
5463
Journ
al h
om
e
page
:
http:
//
ia
es
core
.c
om/
journa
ls
/i
ndex.
ph
p/IJECE
A
p
erf
ormanc
e of
c
ompa
rative
s
t
ud
y f
or
s
emi
-
s
t
ructured
w
eb
da
t
a e
xtr
act
ion
m
od
el
Ily Amali
n
a A
hma
d
S
ab
ri
,
Musta
fa
Man
School
of
In
for
m
at
ic
s a
nd
Appl
ie
d
Ma
the
m
at
i
cs
,
Univer
si
ti
Malay
sia
T
ere
ngg
anu
,
Mal
a
y
s
ia
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
J
a
n
5
, 201
9
Re
vised
A
pr
18
, 2
01
9
Accepte
d
J
un
1
2
, 201
9
The
extrac
ti
on
o
f
informati
on
fr
om
m
ult
i
-
source
s
of
web
is
an
essenti
al
y
e
t
complic
a
te
d
ste
p
for
dat
a
ana
l
y
sis
in
m
ult
ipl
e
dom
ai
ns.
In
th
is
pape
r,
we
pre
sent
a
data
ex
tra
c
ti
on
m
odel
b
ase
d
on
visual
segm
ent
ation,
D
OM
tre
e
an
d
JS
ON
appr
oac
h
which
is
known
as
W
rap
per
Ext
r
ac
t
ion
of
Im
age
using
DOM
and
JS
ON
(W
EIDJ)
for
ext
r
ac
t
ing
sem
i
-
struct
u
red
da
ta
f
rom
biodi
ver
s
i
t
y
web.
The
la
rg
e
num
ber
of
informati
on
from
m
ult
i
ple
source
s
of
web
which
is
image’
s
informati
on
wi
ll
be
ext
ra
ct
ed
using
thre
e
d
iffe
r
ent
appr
oach
;
D
ocument
Objec
t
Model
(DO
M),
W
rap
per
image
using
Hy
br
id
DO
M
and
JS
ON
(W
HD
J)
and
W
rap
per
E
xtra
c
ti
on
of
Im
age
using
DO
M
and
JS
O
N
(W
EIDJ).
Experim
ent
s
were
c
onduct
ed
on
se
ver
al
b
iodi
ver
si
t
y
websi
te.
The
exp
eri
m
ent
result
s
show
th
at
W
EIDJ
appr
o
ac
h
prom
ising
r
e
sults
with
respe
ct
to
ti
m
e
ana
l
y
sis
v
al
ues.
W
EIDJ
wrappe
r
has
succ
essfull
y
ext
r
ac
t
ed
gre
ater
tha
n
100
images
of
dat
a
from
the
m
ult
i
-
source
web
bio
dive
rsit
y
of
over
15
d
iffe
r
ent websit
es.
Ke
yw
or
d
s
:
Do
c
um
ent
o
bje
ct
m
od
el
(D
O
M)
Web
d
at
a
e
xtra
ct
ion
Wr
a
pper
e
xtra
ct
ion
of
i
m
age
us
in
g DO
M
and JSO
N
(
W
E
I
DJ)
Wr
a
pper im
age u
s
i
ng
h
y
br
id
DO
M a
nd J
SON (WH
D
J)
Copyright
©
201
9
Instit
ut
e
o
f Ad
vanc
ed
Engi
n
ee
r
ing
and
S
cienc
e
.
Al
l
rights re
serv
ed
.
Corres
pond
in
g
Aut
h
or
:
Ily
A
m
al
ina A
hm
ad
Sabri
,
School
of In
for
m
at
ic
s
and
Applie
d
Ma
them
at
ic
s,
Un
i
ver
sit
i M
al
ay
sia
Tereng
ga
nu,
Ku
al
a
Tere
ngga
nu, Te
reng
ga
nu D
a
r
ul I
m
an,
Mal
ay
sia
.
Em
a
il
: ily
l
ina@ya
hoo.com
1.
INTROD
U
CTION
The
e
xtracti
on
of
t
he
in
form
ation
from
the
la
rg
e
databa
se
is
known
as
Know
le
dg
e
Disc
overy
(KD).
Me
anwhil
e
dat
a
m
ining
is
th
e
process
of
e
xtracti
ng
the
use
fu
l
a
nd
relev
ant
inf
or
m
at
ion
f
ro
m
the
datebase.
It
al
lows
use
r
to
a
naly
ze
data
from
diff
e
ren
t
view
s
a
nd
cat
egorizi
ng
it
,
pri
or
t
o
c
on
cl
ud
i
ng
the
relat
ionshi
ps
betwee
n
data.
The
e
xtracti
on
and
a
naly
sis
of
the
we
b
pag
e
is
an
interest
ing
resear
ch
a
r
ea
in
the
fiel
d
of
data
m
ining
a
nd
w
eb
m
ining
.
I
ntern
et
has
m
ade
the
World
W
i
de
We
b
as
the
m
ai
n
pool
f
or
t
he
c
o
ll
ect
ion
a
nd
distrib
ution o
f i
nfor
m
at
ion
to t
he
use
rs
.
The
re
ports
fro
m
In
te
rn
et
World
Stat
s
sta
te
s
that
there
are
now
m
or
e
than
4
bill
ion
of
pe
op
le
ar
ound
the
world
are
us
ing
the
int
ern
et
(S
ta
ts,
2018
).
The
la
te
st
data
sh
ows
that
Asia
has
beco
m
e
the
bigge
st
popula
ti
on
of
r
egio
n
c
on
t
rib
ut
ing
to
m
or
e
th
an
2
bill
ion
use
rs.
Fig
ur
e
1
s
hows
t
he
worl
d
inte
rn
et
us
a
ge
an
d
popula
ti
on
sta
ti
sti
cs.
The
nu
m
ber
of
i
nter
ne
t
us
er
s
in
earl
ie
r
pa
rt
of
2018
was
4,156,9
32,14
0
(S
ta
ts,
2018)
.
This
directl
y i
m
pl
ie
s an
d co
nt
ribu
te
s t
o
trem
endo
us
gro
wth o
f data
in
t
he Wo
rld
W
i
de Web
.
Web
data
e
xtr
act
ion
is
a
te
chnolo
gy
de
vel
op
e
d
ov
e
r
the
past
deca
de
and
e
nc
ounter
s
m
any
new
chall
enges.
It
has
been
disc
usse
d
from
different
per
s
pecti
ves
a
nd
view
s.
It
le
ver
a
ges
on
va
rio
us
sci
e
ntific
m
et
ho
ds
f
ro
m
var
i
ous
discipli
nes
[
1
]
.
La
end
e
r,
Ri
beir
o
-
Net
o
[
2]
P
ropo
s
ed
ta
xono
m
y
fo
r
data
e
xtracti
on
appr
oach
to
ge
ner
at
e
w
rapp
er.
Fig
ur
e
2
s
hows
the
s
ugge
ste
d
cl
assifi
c
at
ion
pro
po
se
d
by
th
e
re
sear
cher
s
.
This
cl
assifi
cat
ion
is
ver
y
im
po
rtant
and
help
fu
l
in
or
de
r
to
unde
rsta
nd
the
e
xisti
ng
ap
proac
hes
for
we
b
data ext
racti
on.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
9
, N
o.
6
,
Dece
m
ber
2
01
9
:
5463
-
5470
5464
Figure
1.
N
umber
of inte
rn
et
us
ers
’ for
w
or
l
d wide
we
b
Figure
2.
Data
extracti
on a
ppr
oach
e
s
by Lae
nd
e
r (20
02)
Lan
gu
a
ges
f
or
wr
a
pper
de
velop
m
ent,
TS
IM
MIS
[
3
-
5
]
,
We
bOQL
[
6
]
,
L
orel
[
7
]
a
nd
Mi
ne
rv
a
[
8
]
are
so
m
e o
f
the tech
ni
qu
e
s that em
plo
y
la
ng
ua
ge
f
or
wr
a
pp
e
r devel
op
m
ent. Th
ese ap
proac
hes
are u
se
d
to
ad
dr
e
ss
the
pr
ob
le
m
of
w
ra
pp
e
r’s
te
chn
i
qu
e
s.
Var
i
ou
s
gen
e
ral
la
ngua
ges
wer
e
desi
gn
e
d
t
o
const
ru
ct
w
ra
pp
e
r.
The
e
xam
ple of lan
guages
tha
t are
us
ed
b
y
pro
gr
am
m
ers
in
dev
el
op
i
ng wr
app
e
rs
a
re
Perl
and Phyt
on.
HTML
a
war
e
,
these
a
ppr
oac
hes
dep
e
nd
upon
the
str
uct
ural
featur
e
s
of
web
pa
ges
t
o
perform
bo
th
wr
a
pper
gen
e
r
at
ion
an
d
data
extracti
on.
It
is
perf
or
m
automa
ti
cally
witho
ut
la
bo
ur
ta
s
k.
X
WRAP,
W4
F
a
nd
RoadR
unne
r
[
9
]
are
exam
ples
for
this
te
chn
i
qu
e
.
Natu
ra
l
Languag
e
P
r
ocessin
g
(
NLP)
te
chn
i
qu
e
is
us
e
d
in
order
t
o
m
ine
facts
f
ro
m
free
te
xt.
Fact
i
s
ind
ic
at
ed
by
entit
ie
s
and
r
el
at
ion
sh
i
ps
be
tween
e
ntit
ie
s
[
10
]
.
RAPI
ER
[
11
]
,
W
H
ISK
[
12
]
and
SRV
[
13
]
are
s
om
e
of
the
a
ppro
ac
hes
of
NL
P.
Wr
a
pp
e
r
in
duct
ion,
thes
e
appr
oach
es
a
re
based
on
cert
ai
n
featu
res
s
uc
h
as
f
or
m
at
tin
g.
It
ca
n
de
fine
th
e
str
uctu
r
e
of
data
that
fou
nd.
The
e
xtracti
on
ru
le
s ca
n be
ge
ner
at
e
d
based
on traini
ng set.
SoftM
eal
y
[
1
4
]
and IE
PAD
[
15
]
.
Mod
el
li
ng
bas
ed,
t
hese
a
ppr
oach
e
s
try
to
f
ind
sect
ion
s
of
the
web
pa
ge
s
that
s
uit
with
pre
-
def
i
ne
d
structu
res.
N
oDo
SE
[
16
]
an
d
DEByE
[
2
]
ar
e
the
te
ch
niqu
es
for
this
cat
e
gory.
O
nt
olog
y
based,
this
c
at
egor
y
is
total
ly
diff
e
ren
t
with
pre
vi
ou
s
te
c
hn
i
qu
e
s.
This
is
bec
ause
it
rely
ing
directl
y
on
th
e
data.
T
he
obvi
ous
te
chn
iq
ues
for
this ap
proac
h
a
re
OD
E
[
17
]
and DIA
DEM
[
18
]
.
2.
WEID
J MO
D
EL
In
certai
n
ki
nd
of
web
pa
ges,
it
is
com
m
on
that
data
inf
or
m
at
ion
of
web
pag
e
s
are
buil
t
dynam
ic
al
l
y
accor
ding
to
s
pecific
te
m
plate.
Ty
pical
exa
m
ple
is
infor
m
at
ion
of
im
age,
wh
e
re
im
a
ge
detai
ls
ha
s
al
ways
the
sa
m
e
structur
e,
di
ff
e
rin
g
only
in
cont
ent
loade
d
usual
ly
fr
om
a
database
.
Thes
e
data
are
rendere
d
in
the
sim
il
ar
way.
T
her
e
ha
ve
been
disc
usse
d
i
n
f
orm
er
w
orks
[
19
-
23
]
about
how
t
o
e
xtract
data
from
web p
a
ges
.
Figure
3
s
how
s
basic
c
on
ce
pt
s
of
data
ext
r
act
ion
process.
In
prel
i
m
inary
ste
p,
use
r
ne
ed
to
know
w
hat
ty
pes
of
data
that
they
wan
t
t
o
ext
ract
ei
ther
te
xt,
im
age,
vid
e
o
or
ot
her
s.
The
n,
t
he
y
m
us
t
decide
wh
ic
h
data
need
to
be
extracte
d.
T
his
sel
ect
ion
m
us
t
be
done
earli
er
because
each
data
has
their
own
source
of
ty
pe.
Diff
e
re
nt
data
has
t
heir
own
s
ource
of
da
ta
and
rele
va
nt
m
et
ho
ds
.
Af
te
r
sel
ect
io
n
ty
pe
of
dat
a
done
,
nex
t
process
c
ou
l
d
be
proce
ed
to
a
bs
tract
and
t
ran
s
f
orm
sel
ect
ed
data
in
ta
bu
la
r
f
or
m
at
us
ing
s
pecifi
c
appr
oach
or
m
et
hods
.
First,
us
er
nee
d
to
unde
rstan
d
t
he
pro
of
of
c
on
ce
pt
f
or
Web
Da
ta
Extracti
on
be
fore
dev
el
op a
wr
a
pper
.
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
A p
erf
orma
nce
o
f c
omp
arati
ve stu
dy
for
se
m
i
-
stru
ct
ur
e
d
we
b da
t
a
ext
r
action
...
(
Ily A
m
alina
Ah
m
ad Sa
bri)
5465
Figure
3. Ba
sic
d
at
a e
xtracti
on
process
2.1.
Problem
f
orm
ulat
i
on
As
par
t
of
the
input
for
the
extracti
on,
we
suppose
that
the
use
r
has
a
num
ber
of
str
uctu
red
we
b
so
urces
,
de
no
t
ed
in
the
f
ollo
wing,
w
he
re
each
re
pr
ese
nts
a
set
of
web
pag
e
s
that
des
cribe
im
ages
ob
j
ect
s
,
wh
ic
h
co
uld
be
seen
as
relat
i
on
al
tu
ples
for
m
ed
by
th
ree
at
omi
c
t
ype
values;
li
nk
,
i
m
a
ge,
siz
e.
W
e
a
ssu
m
e
a
set
of
entit
y
(atom
ic
)
ty
pes,
wh
e
re
each
s
uch
ty
pe
repre
sent
an
at
om
ic
piece
of
i
nform
at
ion
.
W
e
c
onti
nue
by
de
fining
th
e
ty
pin
g
form
a
li
s
m
,
by
wh
ic
h
any
us
e
r
ca
n
s
pecify
w
hat
da
ta
sh
ould
be
ta
rg
et
e
d
an
d
e
xtr
act
e
d
from
w
eb page
. W
e
the
n desc
ribe
t
he
e
xtract
ion
pro
blem
.
2.1.1.
Types
and
ob
j
ect descrip
tio
n
In
WE
ID
J
,
it
al
lows
us
ers
t
o
de
scribe
at
om
ic
t
ypes
of
obj
ect
s
.
As
buil
ding
bl
ocks
f
or
desc
ribi
ng
data,
we
a
ssum
e
a
set
of
e
ntit
y
(atom
ic
)
ty
pes,
w
her
e
e
ach
s
u
ch
ty
pe
represe
nts
an
at
om
ic
piece
of
inf
or
m
at
ion
, e
xpresse
d
as
im
ages.
An
insta
nce
of
an
entit
y
ty
pe
is
any
im
ages
that
exist.
It
is
def
i
ned
strai
gh
tforwa
rd
in
a
t
op
-
dow
n
appr
oach
an
d
can
be
view
a
s
a
tree
st
ru
ct
ur
e
w
hose
int
ern
al
nodes
de
no
te
s
t
he
us
e
of
a
c
om
plex
ty
pe
const
ru
ct
or.
F
or
e
xam
ple,
im
age
obj
ect
s
c
ou
l
d
be
sp
eci
fied
as
a
tu
ple
t
ype
com
po
se
d
of
t
hr
ee
e
ntit
y
ty
pes:
path
of
im
age,
siz
e
of
i
m
age,
date
of
extrac
ti
on
an
d
tim
e
of
ext
racti
on.
The
first
tw
o
entit
y
t
ypes
would
be
associat
ed
t
o
prede
fine
d
rec
ognize
rs
(
for
pa
th
of
im
ages
and
siz
e
of
im
a
ges),
since
this
kind
of
i
nform
at
ion
has
easi
ly
r
ec
ogniza
ble r
e
pres
entat
ion
patte
r
ns
, w
hile t
he
la
st o
nes would
hav
e
an i
ns
ta
nc
e of r
ec
ognize
rs.
2.1.2.
The
ex
tra
c
tio
n p
r
ob
le
m
Fo
r
a
giv
e
n
WEI
DJ
an
d
so
ur
ce
,
a
te
m
plate
with
res
pect
t
o
an
d
desc
ribe
s
how
insta
nce
s
of
can
be
e
xtract
ed fr
om
pa
ges
.
Fo
r
eac
h
set
ty
pe
=
[
{
}
,
]
a
pp
ea
ri
ng
in
,
de
fine
s
a
se
par
at
or
string
,
it
deno
te
s
that
consecuti
ve
ins
ta
nces
of
will
be
se
par
at
e
d b
y t
his strin
g.
Fo
r
eac
h
tu
ple
ty
pe
=
{
1
,
…
,
}
,
def
ine
s
t
ot
al
i
m
ages
ove
r
the
c
ollec
ti
on
of
ty
pes
a
nd
a
seq
uen
ce
of
+
1
separa
tor
stri
ng
s
1
,
…
,
1
;
this
de
no
te
s
that
the
instances
of
t
he
ty
pes
for
m
ing
,
in the
sp
eci
fie
d im
ages, wil
l b
e d
el
im
i
te
d
by
these se
par
at
ors.
The
e
xtracti
on
pro
blem
can
be
d
esc
ribe
d
as
fo
ll
ow.
F
or
a
gi
ven
i
nput c
ons
ist
ing
of
a
n W
EID
J
an
d
a set o
f
s
ources
{
1
,
…
,
}
,
1.
set
up type
rec
ognizers
for al
l t
he
e
ntit
y t
ypes in
,
2.
for
eac
h
s
ource
,
a.
fin
d
a
nd ann
otate
en
ti
ty
ty
pe
instances
or im
ages
in pa
ges,
b.
infer a te
m
plate
(
,
)
base
d o
n
the
sam
ple,
c.
us
e
to e
xtract
al
l t
he
im
ages o
f
from
,
d.
sel
ect
i
m
ages th
at
w
a
nt to
sto
re in si
ngle
m
ultim
edia d
at
aba
se
A web
p
a
ge
w
is
r
e
pr
ese
nted
as
a triple:
w
=
(
b
,
S
,
R
)
(1)
Finit
e set o
f bl
ocks is
represe
nted
a
s:
b
=
{
1
,
2
,
.
.
.
,
}
(2)
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
9
, N
o.
6
,
Dece
m
ber
2
01
9
:
5463
-
5470
5466
All
these
bloc
ks
a
re
not
ov
erlap
ped.
Eac
h
blo
c
k
can
be
rec
ur
si
vely
vi
ewed
as
a
s
ub
we
b
pa
ge
associat
ed
wit
h
s
ub
str
uctu
r
e
ins
pire
d
f
r
om
the
whole
pag
e
str
uctu
re
.
Finit
e
set
of
sepa
rato
rs
s
uc
h
as
horizo
ntal an
d vertic
al
are
r
e
presente
d
as:
S
=
{
1
2
,
…
,
}
(3)
Ever
y
sep
arato
r
in
the
sam
e
S
has
sam
e
weigh
t.
W
ei
ght
for
each
sepa
rato
r
ind
ic
at
in
g
it
s
visibil
it
y.
R
Is
t
he rel
at
ion
s
hip
of e
ver
y t
w
o bloc
ks
in
b
. I
t
can
be
e
xpress
ed
as
R
=
b
×
b
→
S
∪
{
NUL
L
}
(4)
Fo
r
e
xam
ple,
su
pp
os
e
i
w
an
d
j
w
are
two
ob
j
ect
s
in
,
=
(
,
)
≠
in
dicat
es
that
an
d
are
e
xactl
y
se
par
at
e
d
by
t
he
se
par
at
or
(
,
)
.
I
n
oth
e
r
w
ords,
t
wo
obj
ect
s
are
ad
j
ace
nt
t
o
ea
ch
oth
e
r,
oth
e
rw
ise
t
here
are
ot
her
ob
j
ect
s
bet
ween
two
blo
c
ks
and
.
Fig
ure
4
s
hows
la
yo
ut
s
tructu
re
visu
al
segm
enta
ti
on
of
WW
F
web
pa
ge wh
il
e
.
Figure
4
.
Layo
ut str
uctu
re a
nd
visu
al
se
gm
e
ntati
on
of
WWF w
e
b pa
ge
3.
RESU
LT
S
A
ND
DI
SCUS
S
ION
S
A
key
el
em
ent,
path
of
re
quired
HTML
el
e
m
ent
is
i
m
po
rtant
w
hich
al
lo
ws
to
l
ocate
and
e
xtrac
t
inf
or
m
at
ion
.
T
his
sect
ion
dis
cuss
the
e
xp
e
r
i
m
entat
ion
of
data
extracti
on
for
m
ulti
-
un
if
or
m
reso
urce
l
ocato
r
(U
RL
).
T
his
e
xp
e
rim
ental
is
diff
e
re
nt
f
or
da
ta
extracti
on
f
r
om
su
rf
ace
of
web.
T
his
is
be
cause
wh
e
n
w
orkin
g
with
m
ulti
ple
web
of
URL,
us
er
nee
d
to
input
seve
ral
web
URL
that
con
ta
in
var
i
ous
inf
orm
ation
to
be
extracte
d f
ro
m
d
iffe
re
nt str
uctur
e
of
web p
a
ge
s.
Fi
gure
5
s
hows
inter
face
f
or m
ulti
-
ur
l.
The
w
ork
desc
r
ibe
d
in
this
e
xperim
ental
wor
k
us
es
t
he
sam
e
ap
proac
h
in
the
s
urface
we
b.
H
oweve
r,
it
inv
olv
es
m
or
e
pr
ocess
a
nd
ti
m
e
con
sum
ing
because
the
extracti
on
process
r
ec
ur
si
vely
traversi
ng
al
l
el
e
m
ent’s
par
e
nts
for
each
w
eb
pa
ge.
Ta
ble
1
and
Ta
ble
2
descr
ibe
s
the
ben
c
hm
ark
of
web
a
ddress
th
at
are
i
m
ple
m
ented
on
te
sti
ng
for
i
m
age
extracti
on
of
m
ulti
ple
web.
Ta
ble
1
c
on
sist
s
of
t
hr
e
e
gro
up
s
of
U
RL
su
c
h
as grou
p A,
B
and C. Eac
h g
r
oup has
sev
e
ra
l web pa
ges
as
sh
ow
n
in
Ta
ble 2
.
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
A p
erf
orma
nce
o
f c
omp
arati
ve stu
dy
for
se
m
i
-
stru
ct
ur
e
d
we
b da
t
a
ext
r
action
...
(
Ily A
m
alina
Ah
m
ad Sa
bri)
5467
Figure
5
.
Interf
ace f
or
m
ulti
-
so
urces
of
web
Table
1.
Desc
ription o
f be
nchm
ark
Ben
ch
m
a
rk
Grou
p
of
URL
Descripti
o
n
1
A
Gen
eral
Bio
d
iv
ersi
ty
and
E
n
d
an
g
ered
Species
I
n
f
o
r
m
ati
o
n
2
B
Databas
es an
d
Dat
asets
: Broad
Scop
e
3
C
Databas
es an
d
Dat
asets
: Nar
row Sco
p
e
Table
2
.
Detai
ls o
f benc
hm
ark
Ben
ch
m
a
rk
Un
if
o
r
m
R
eso
u
rce Lo
cato
r
(
URL)
1
h
ttp
://www.a
m
n
h
.
o
rg
h
ttp
://www.ocean
.
si.ed
u
h
ttp
://www.iucn
.or
g
h
ttp
://www.end
an
g
eredsp
eciesin
ternatio
n
al.org
h
ttp
://www.wwf
.or
g
.
m
y
2
h
ttp
://www.gb
if
.org
/
h
ttp
://www.un
ep
-
wc
m
c
.org
/
h
ttp
://www.natu
re
serv
e.org
/
h
ttp
://www.org
an
is
m
n
a
m
es
.co
m
/q
u
e
ry
.ht
m
l
h
ttp
://www.catalo
g
u
eo
f
li
f
e.org
/co
l/s
earc
h
/all
h
ttp
://an
i
m
ald
iv
ers
ity
.u
m
m
z
.u
m
ich
.e
d
u
/site/in
d
ex
.ht
m
l
h
ttp
://www.thep
lan
tlist.o
rg/
h
ttp
://www.iucn
re
d
list.o
rg/
h
ttp
://www.itis.go
v
/
h
ttp
://www.con
sb
io
.org
/
h
ttp
://b
u
g
g
u
id
e.net
/n
o
d
e/v
iew/1
5
7
4
0
3
h
ttp
://www.a
m
p
h
i
b
iaweb.o
rg/
h
ttp
://www.r
eef
b
a
se.o
rg/
h
ttp
://p
ri
m
atelit.lib
rary.w
isc
.edu
/
h
ttp
://b
u
g
g
u
id
e.net
/n
o
d
e/v
iew/1
5
7
4
0
Figure
6
sho
w
s
the
res
ult
f
or
m
ulti
-
so
ur
ce
s
of
we
b
pag
e
s
for
web
dat
a
extracti
on
usi
ng
D
OM,
WHDJ
a
nd
our
pr
opos
e
d
m
od
el
,
WEIDJ.
T
his
ex
pe
rim
ent
has
bee
n
te
ste
d
on
be
nch
m
ark
f
or
th
ree
differen
t
gro
up
of
URL
that
can
be
re
f
err
e
d
on
Ta
ble
2.
Tim
e
extra
ct
ion
is
re
ferr
e
d
t
o
the
ti
m
e
l
oad
i
ng
f
or
e
xtr
act
ion
process
sin
ce
the
first
im
age
un
ti
l
the
la
st
i
m
age.
The
ti
m
e
extracti
on
is
m
easur
ed
in
s
econds
.
The
ti
m
e
of
WEIDJ
r
et
rie
va
l on any
data
set
is sig
nifica
ntly
low
e
r
tha
n othe
r
m
et
ho
ds an
d ou
t
perfor
m
s ef
fici
ently
.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
9
, N
o.
6
,
Dece
m
ber
2
01
9
:
5463
-
5470
5468
Fi
gure
6
.
Perf
orm
ance f
or im
a
ge
e
xtracti
on
f
or m
ulti
-
un
ifo
r
m
r
esou
rce
loc
at
or
(U
RL
)
Table
3
(a,
b)
sh
ows
su
m
m
arize
resu
lt
s
of
extracti
ng
im
a
ges
on
f
our
di
ff
e
ren
t
m
od
el
of
w
rappe
r;
DO
M,
WHD
J,
W
EI
D
a
nd
WEIDJ
no
r
ules.
As
it
ca
n
be
see
n
f
ro
m
Table
3
(a,
b)
,
t
he
res
ult
sho
ws
that
the
am
ou
nt
of
i
m
age
fo
un
d
a
re
ave
rag
e
sim
il
arly
bu
t
the
am
ou
nt
i
m
age
t
hat
ha
ve
bee
n
retrieve
d
an
d
f
il
te
red
by
each
m
od
el
are
total
ly
diff
eren
t.
Im
ages
that
retrieve
d
usi
ng
DO
M
a
ppro
ac
h
are
m
any
com
par
ed
to
othe
r
m
od
el
s
becaus
e
the
ru
le
s
of
f
il
te
ring
in
f
or
m
at
ion
a
re
m
inim
al
.
In
a
ddit
io
n,
D
OM
a
pproach
do
e
s
no
t
c
on
si
der
the
re
dunda
nc
y.
It
will
ret
rieve
the
sim
i
la
r
f
il
ena
m
e
for
ea
ch
im
age
if
it
exist
in
t
he
m
ult
iple
we
b
a
ddre
ss
that
act
s
as
inp
ut
.
Wh
il
e
pa
ge
lo
ad
or
kn
own
as
tim
e
extracti
on
is
us
e
d
a
s
pr
im
ary
benchm
ark
to
co
m
par
e
the
pe
rfor
m
ance
for
eac
h
m
od
el
,
t
he
fi
nd
i
ng
s
sho
w
that
WHDJ
is
bette
r
that
D
OM
because
it
us
e
s
JSON
appr
oach
to
tr
ansfo
rm
data.
JSON
is
approved
that
can
de
creased
ti
m
e
bette
r.
Wh
il
e
WHDJ
is
bette
r
than
DO
M,
but
we
nee
d
to
ext
ra
ct
dat
a
in
high
s
pee
d.
S
o
we
propose
A
JAX
te
ch
nolo
gy
to
m
ake
s
ur
e
that
the
entire
we
b
pa
ge
do
e
s
no
t
hav
e
t
o
be
reloa
ded
ea
ch
tim
e
the
us
er
re
quest
s
a
change
durin
g
w
e
b
data ext
racti
on.
Table
3
(
a
)
.
Re
s
ults o
f
im
ages ex
tract
io
n for
m
ul
ti
ple w
eb
a
ddress
(
DO
M,
JSON)
Table
3
(
b
)
.
Re
su
lt
s of im
ages ex
tract
io
n f
or
m
ul
ti
ple w
eb
a
ddress
(
WEIDJ
, WEI
DJ (no
-
r
ules)
)
Grou
p
W
E
IDJ
W
E
IDJ
(no
-
rules
)
I
m
g
f
o
u
n
d
I
m
g
retr
iev
ed
I
m
g
f
iltered
Ti
m
e
extraction
I
m
g
retr
iev
ed
Ti
m
e
extraction
1
86
43
43
1
8
.15
1
9
86
1
4
.67
8
9
2
129
65
64
2
8
.43
1
3
143
2
5
.87
9
3
50
33
17
1
6
.27
5
7
50
1
5
.69
3
4
Table 4 sh
ows
the su
m
m
ary of
tim
e extracti
on in perce
nta
ge
. Th
e
pe
rcen
ta
ge
f
or each m
et
hod
s
hows
the
bigge
r
co
m
par
ison
val
ue
especial
ly
for
D
OM
ap
pro
ach.
J
SON
is
well
known
a
s
fastest
m
e
thod
[
24
]
.
That
is
the
m
ai
n
reas
on
w
hy
JSO
N
can
be
fastest
in
ex
tract
ing
in
f
orm
at
ion
com
par
ed
t
o
D
OM
m
et
ho
d.
Althou
gh
JS
O
N
is
go
od
i
n
pe
rfor
m
ance
of
extracti
ng
data
bu
t
t
he
ext
rac
te
d
inf
orm
ation
are
not
so
e
f
fici
ent
because
this
m
et
hod
will
retrieve
the
sam
e
f
il
e
i
m
a
ge.
So
that,
W
E
IDJ
is
pro
po
se
d
f
or
t
his
ex
per
im
entat
ion
.
WEIDJ is good
in ex
tract
in
g ben
e
fici
al
inf
or
m
at
ion
an
d
the
tim
e p
erf
orm
a
nce is d
eg
rad
e
.
W
E
I
DJ
an
d
WEIDJ
no
ru
le
s
is
the
sam
e
m
e
tho
d
wh
ic
h
ap
ply
co
m
bin
at
ion
of
D
OM
an
d
JS
ON
bu
t
the
diff
e
re
nce
bet
ween
WEIDJ
no
–
ru
le
s
an
d
WEIDJ
m
et
ho
d
is
WEIDJ
no
-
r
ules
will
no
t
filt
er
no
isy
i
m
ages.
It
will
retrieve
any
i
m
a
ges
just
li
ke
D
OM a
nd
JSON.
Grou
p
DOM
JSON
I
m
g
f
o
u
n
d
I
m
g
retr
iev
ed
I
m
g
f
iltered
Ti
m
e
ex
traction
I
m
g
f
o
u
n
d
I
m
g
retr
iev
ed
I
m
g
f
iltered
Ti
m
e
ex
traction
1
96
66
30
1
6
8
.8032
86
54
32
1
3
3
.2378
2
163
88
75
2
3
2
.9429
143
78
65
1
8
4
.783
3
94
27
67
1
2
1
.7585
50
27
23
8
1
.37
6
8
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
A p
erf
orma
nce
o
f c
omp
arati
ve stu
dy
for
se
m
i
-
stru
ct
ur
e
d
we
b da
t
a
ext
r
action
...
(
Ily A
m
alina
Ah
m
ad Sa
bri)
5469
Table
4.
Res
ults o
f
im
ages ex
t
racti
on in pe
rc
entage
f
or
m
ulti
ple w
e
b
a
ddre
ss (
ti
m
e extracti
on)
4.
CONCL
US
I
O
N
We
ha
ve
pro
pose
d
sem
i
-
struc
ture
d
web
data
extracti
on
m
od
el
,
WEIDJ
to
extract
data
accor
ding
t
o
the
pr
e
de
fine
d
and
sim
ple
data
extracti
on
ru
le
.
It
consi
sts
of
se
ver
al
pa
ram
et
ers:
l
ink
of
the
im
age
wh
ic
h
locat
es
the
data
that
has
bee
n
retriev
es
fro
m
web
pa
ge,
im
ages,
siz
e
of
i
m
ages
and
ti
m
e
extracti
on
for
eac
h
i
m
age.
W
E
IDJ
corres
ponds
not
only
f
or
sin
gle
URL
but
a
lso
to
m
ulti
ple
sources
of
we
b
pag
e
.
T
he
c
urren
t
stud
y
lim
it
s
it
s
scop
e
to
e
xtr
act
i
m
ages
fr
om
su
rf
ace
of
m
ul
ti
ple
web
URL.
F
uture
r
esearch
m
ay
c
on
si
der
i
m
age ex
tract
ion f
ro
m
d
eep
web of m
ulti
ple U
RL.
ACKN
OWLE
DGE
MENTS
I
since
rely
thank
al
l
th
os
e
w
ho
helpe
d
m
e
in
com
pleti
ng
this
ta
sk
es
pec
ia
ll
y
Biasisw
a
Un
iv
ersit
i
Malaysi
a
Te
re
ngganu (B
UM
T)
.
REFERE
NCE
S
[1]
S.
J.
Nasti,
e
t
al
.
,
“
A
Com
par
at
iv
e
Stud
y
on
W
eb
Data
Ex
tra
c
ti
on
Approa
che
s
,”
In
te
rnati
onal
Journal
of
Engi
ne
ering
Sc
i
enc
e
,
2016
.
[2]
A.
H.
L
ae
nde
r,
et
a
l
.
,
“
D
EBy
E
–
d
ata
ex
tra
c
t
ion
b
y
ex
ample
,”
Data
&
K
nowle
dge
Enginee
ring
,
vol.
4
0,
p
p.
121
-
154
,
20
02
.
[3]
J.
H
amm
er,
e
t
al
.
,
“
Ext
ra
ct
ing
Se
m
istruc
ture
d
In
f
orm
at
ion
from
t
he
W
eb
,”
1997
.
[4]
J.
Ham
m
er,
e
t
al
.
,
“
Te
m
pla
t
e
-
bas
ed
wrapp
ers
in
t
he
TSIMM
IS s
y
stem
,”
ACM
Sig
mod R
ec
ord
,
19
97.
[5]
J.
Ham
m
er,
e
t
al
.
,
“
Sem
istruc
tured Dat
a
:
The
TSI
MM
IS E
xper
ie
n
ce
,”
1997
.
[6]
G.
Aroce
na
an
d
A.
Mende
l
zo
ns
,
“
We
bOQ
L:
restructuring
docume
nts,
da
t
abases
and
we
bs
,”
I
nte
rna
ti
on
al
conf
ere
n
ce on
d
at
a
engi
n
ee
r
ing
,
IEE
E
Com
pute
r S
oci
ety
,
1998
.
[7]
S.
Abite
boul
,
e
t
al
.
,
“
The
Lor
el
Quer
y
La
ngu
age
for
Sem
i
-
Structured
Dat
a
,”
Inter
nati
onal
Journ
al
Digit
Libr
ary
,
vol.
1
,
p
p
.
68
-
88
,
1997
.
[8]
R.
Moone
y
,
“
Relati
onal
l
earning
of
patt
ern
-
matc
h
rules
for
inf
or
mation
ex
trac
ti
o
n
,”
Proce
edi
ngs
of
the
Sixte
en
th
Nati
ona
l
Confe
r
enc
e
on
Art
ifi
c
ial
Int
el
l
ige
nc
e
,
19
99.
[9]
V.
C
resc
enz
i
,
et
al
.,
“
Roadrunne
r
:
Towa
rds
aut
om
at
i
c
da
ta e
xtr
acti
on
from
la
rg
e
w
eb
sit
es
,”
VLDB
,
2001.
[10]
A.
H.
Laende
r
,
e
t
al
.
,
“
A bri
ef
surve
y
of
web
d
at
a
ext
r
ac
t
ion tools
,”
ACM
Sigmod
Re
cord
,
vol
.
31
,
p
p.
84
-
93
,
2002
.
[11]
M.
E.
Ca
li
ff
a
nd
R.
J.
Moone
y
,
“
Relati
on
al
Le
arn
ing
of
Pattern
-
Matc
h
R
ule
s
for
Inform
at
ion
Ex
tracti
on
,”
AA
AI
/IA
AI
,
199
9.
[12]
S.
Soderla
nd,
“
L
ea
rning
informat
ion
ext
ra
ct
ion
ru
le
s
for
sem
i
-
struct
ure
d
and
fr
ee
t
ext
,”
Mac
hin
e
lear
ning
,
vol.
34,
p
p.
233
-
272
,
19
99
.
[13]
D.
Freitag,
“
M
ac
hin
e
l
ea
rni
ng
fo
r
informat
io
n
ext
r
ac
t
ion
in
informal
dom
a
ins
,”
Ma
chi
ne
l
earning
,
vo
l.
3
9,
p
p.
169
-
202
,
20
00
.
[14]
C.
N.
Hs
u
and
M.
T.
Dung,
“
G
ene
ra
ti
ng
finite
-
stat
e
tr
ansduc
ers
for
sem
i
-
struct
ur
ed
dat
a
ex
tra
c
tion
from
the
W
eb
,”
Information
Syst
ems
,
vol
.
23
,
p
p
.
521
-
538
,
1998
.
[15]
C.
H.
Chang
an
d
S.
C.
Lui
,
“
IE
PA
D:
inf
orm
ati
o
n
ex
traction
based
on
patt
ern
di
scov
ery
,”
Proce
edi
ngs
of
the
10th
int
ern
at
ion
al
con
fer
ence
on
W
orl
d
W
ide
W
eb
,
20
01.
[16]
B.
Adelb
erg
,
“
NoDoSE
—
a
tool
for
semi
-
auto
matic
ally
ex
trac
ti
ng
structured
and
semistructu
red
da
ta
from
t
ext
docume
nts
,”
AC
M Sigm
od
Rec
o
rd
,
1998
.
[17]
W
.
Su,
et
al
.
,
“
OD
E:
On
tol
og
y
-
a
ss
iste
d
dat
a
extr
ac
t
ion
,”
ACM
Tr
ansacti
ons
on
D
atabase
Syste
ms
(
T
OD
S
)
,
vol.
34,
p
p.
12
,
2009
.
[18]
T.
Furche
,
et
al
.
,
“
DIADEM:
do
main
-
ce
ntri
c,
intell
ig
ent,
automat
ed
data
ex
trac
tion
methodol
ogy
,”
Proce
edi
ngs
of
the
21st
Int
ern
ati
onal
Conf
ere
n
ce on
W
orld
W
id
e W
eb
,
2012.
[19]
I.
A.
A.
Sabri
and
M.
Man,
“
M
ult
iple
Types
of
Semi
-
structured
Data
Ex
tract
ion
Us
ing
W
rapper
for
Ex
tract
ion
of
Image
Us
ing
D
OM
(
WEI
D
)
,
”
R
egi
ona
l
Confer
e
nce
on
Scie
n
ce,
Te
chno
log
y
and
Socia
l
Sci
ences
(RCS
TSS
2016
)
,
N.A.Y.
e. al., Edi
tor
,
Spr
ing
er
Na
t
ure
Singapor
e
Pt
e
L
td.
,
Singapo
r
e
,
p
p
.
67
-
76
,
201
8
.
[20]
I.
A.
A.
Sabri
and
M.
Man,
“
I
m
proving
per
form
anc
e
of
DO
M
in
sem
i
-
struct
ure
d
da
ta
ex
traction
using
W
EID
J
m
odel
,”
Indone
s
ian
Journal
of
Elec
tri
cal
Engi
n
eer
ing
and
Compu
te
r Sc
ie
nc
e
(
IJEECS)
,
vol. 9,
p
p.
752
-
763
,
2018
.
[21]
I.
A.
A.
Sabri
a
nd
M.
Man,
“
The
proposed
al
g
orit
hm
for
sem
i
-
struct
ure
d
d
at
a
int
egr
at
ion
:
Case
stud
y
of
Setiu
wetl
and
data
se
t
,”
Journal
of
Telecomm
unic
at
ion
El
e
ct
ronic
and
Computer
Engi
n
ee
ring
,
vol
.
9
,
p
p.
79
-
84
,
2017
.
[22]
I.
A.
A.
Sabri
a
nd
M.
Man
,
“
W
EIDJ
:
An
improvised
algori
th
m
for
image
e
xtr
act
ion
from
we
b
pages
,”
The
8
th
int
ern
at
ion
al
c
o
nfe
ren
c
e
on
inf
orm
at
ion
techno
log
y
,
Al
-
Z
a
y
too
nah
Univer
sit
y
of
Jorda
n
(ZUJ),
Am
m
an,
Jordan
,
IEE
E
Xplore
,
20
17.
Grou
p
DOM
JSON
W
E
IDJ
W
E
IDJ(no
-
rules
)
Ti
m
e
ex
traction
Percentag
e
100%
Ti
m
e
ex
traction
Percentag
e
100%
Ti
m
e
ex
traction
Percentag
e
100%
Ti
m
e
ex
traction
Percentag
e
100%
1
1
6
8
.8032
5
0
.41
1
3
3
.2378
3
9
.79
1
8
.15
1
9
5
.42
1
4
.67
8
9
4
.38
2
2
3
2
.9429
4
9
.35
1
8
4
.783
3
9
.15
2
8
.43
1
3
6
.02
2
5
.87
9
5
.48
3
1
2
1
.7585
5
1
.79
8
1
.37
6
8
3
4
.61
1
6
.27
5
7
6
.92
1
5
.69
3
4
6
.68
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
9
, N
o.
6
,
Dece
m
ber
2
01
9
:
5463
-
5470
5470
[23]
I.
A.
A.
Sabr
i
an
d
M.
Man,
“
Mul
ti
ple
ty
p
es
of
se
mi
-
structured
da
ta
extract
ion
usi
ng
WEI
D,
”
Regional
Confer
ence
on
Scie
nce
s,
Technol
og
y
and
Social
Scie
nc
es
(RCS
TSS
)
,
Univ
ersit
i
T
eknol
ogi
Mara
,
Pahang:
Copthorne
Hote
l
Cameron
Highlands
,
2016
.
[24]
Y.
N.
L
i,
et
a
l
.
,
“
Mison:
A
Fast
JSON
Parser
fo
r
Data
Anal
y
ti
cs
,”
Proce
ed
ings
of
the
Vldb
End
owm
ent
,
vol.
10
,
p
p.
1118
-
1129
,
2017
.
BIOGR
AP
H
I
ES
OF
A
UTH
ORS
Ily
Amalina
Ah
mad
Sab
ri
rec
e
ive
d
her
Dip
loma
Inform
at
ion
T
ec
hnolog
y
from
Pol
y
t
ec
hni
c
of
Sulta
n
Miz
a
n
Za
in
al
Abidi
n
(PS
MZA).
Af
te
r
th
at
she
enr
oll
ed
to
Univer
siti
Mal
a
y
s
ia
Te
ren
gg
anu
to
f
urthe
r
her
d
egr
e
e
studie
s
in
Sof
twar
e
Eng
ine
e
ri
ng,
which
was
obta
in
ed
in
2009.
She
con
tinued
h
er
m
ast
er
degr
e
e
in
Mast
er
of
Sci
ence
(C
om
pute
r
Scie
n
ce)
in
th
e
sam
e
unive
rsit
y
and
gra
duated
in
20
14.
During
m
as
te
rs’s
degr
e
e,
h
er
rese
arc
h
was
in
d
ec
ision
support
s
y
stem,
foc
using
on
FA
HP
in
dec
ision
m
aki
ng
for
tour
i
sm
desti
nat
ion
.
Now
,
she
is
a
postgrad
uate
student
pursuing
he
r
Doctor
of
Philosoph
y
(Com
pute
r
Scie
nc
e),
al
so
in
Univer
siti
Malay
s
ia T
e
ren
g
ganu.
Her
cur
r
e
nt
ar
ea of int
er
est
is we
b
da
ta e
xt
rac
t
ion.
Mustafa
Man
i
s
an
A
ss
oci
at
e
Profess
or
in
School
of
Inform
at
ic
s
and
Applie
d
Mathe
m
at
i
cs
and
al
so
as
a
D
eputy
Dire
ct
or
at
Rese
arc
h
Ma
nage
m
ent
Innov
at
ion
C
ent
r
e
(R
MIC),
UM
T
.
He
start
ed
his
PhD
studie
s
in
Ju
l
y
2009
and
finis
hed
his
studie
s
in
Com
pute
r
S
ci
en
ce
from
UTM
in
2012.
He
has
recei
v
ed
Com
pute
r
Scie
n
ce
Diploma
,
Com
pute
r
Sci
enc
e
Degr
ee,
Master
s
Degre
e
from
U
PM
.
In
2
012,
he
has
bee
n
awa
rde
d
a
“
MIec
MO
S
Prestigi
ous
Aw
ard
s”
for
his
PhD
b
y
MIM
OS
Berha
d.
His
rese
arc
h
is
foc
used
on
th
e
d
eve
lopment
o
f
m
ult
ipl
e
t
y
pes
of
dat
ab
ase
s
integra
t
ion
m
odel
and
al
so
in
Aug
m
ent
ed
Reality
(AR),
andr
oid
b
ase
d,
and
IT
rel
a
te
d
int
o
acro
ss
dom
ai
n
pla
tfo
rm
.
Evaluation Warning : The document was created with Spire.PDF for Python.