I
n
t
e
r
n
at
ion
al
Jou
r
n
al
of
E
lec
t
r
ical
an
d
Com
p
u
t
e
r
E
n
gin
e
e
r
in
g
(
I
JE
CE
)
Vol.
15
,
No.
1
,
F
e
br
ua
r
y
20
25
,
pp.
836
~
845
I
S
S
N:
2088
-
8708
,
DO
I
:
10
.
11591/i
jec
e
.
v
15
i
1
.
pp
8
36
-
845
836
Jou
r
n
al
h
omepage
:
ht
tp:
//
ij
e
c
e
.
iaes
c
or
e
.
c
om
A
r
ab
ic
f
ak
e
n
e
w
s d
e
t
e
c
t
io
n
u
s
in
g h
yb
r
id
c
on
t
e
xt
u
al
f
e
a
t
u
r
e
s
Hu
s
s
ain
M
oh
am
m
e
d
T
u
r
k
i
1
,
E
s
s
am
Al
Daoud
1
,
Gh
as
s
an
S
am
ar
a
1
,
Rae
d
Alaz
aid
ah
1
,
M
ais
Haj
Qas
e
m
1
,
M
oh
am
m
ad
Al
j
aid
i
1
,
S
u
h
a
il
a
Abu
owai
d
a
2
,
Naw
af
Als
h
d
aif
a
t
3
1
D
e
pa
r
tm
e
nt
of
C
omput
e
r
S
c
ie
nc
e
, F
a
c
ul
ty
of
I
nf
or
ma
ti
on T
e
c
h
nol
ogy, Z
a
r
qa
U
ni
ve
r
s
it
y,
Z
a
r
qa
,
J
or
da
n
2
D
e
pa
r
tm
e
nt
of
C
omput
e
r
S
c
ie
nc
e
, F
a
c
ul
ty
of
I
nf
or
ma
ti
on T
e
c
h
nol
ogy, Al a
l
-
B
a
yt
U
ni
ve
r
s
it
y, M
a
f
r
a
q, J
or
da
n
3
F
a
c
ul
ty
of
I
nf
or
ma
ti
on T
e
c
hnol
ogy, Applied Scie
nc
e
P
r
iv
a
te
U
ni
ve
r
s
it
y, A
mm
a
n, J
or
da
n
Ar
t
icle
I
n
f
o
AB
S
T
RA
CT
A
r
ti
c
le
h
is
tor
y
:
R
e
c
e
ived
M
a
r
24,
2024
R
e
vis
e
d
Aug
16,
2024
Ac
c
e
pted
Oc
t
1,
2024
T
ech
n
o
l
o
g
y
h
a
s
ad
v
an
ce
d
an
d
s
o
c
i
al
med
i
a
u
s
ers
h
a
v
e
g
ro
w
n
d
ramat
i
cal
l
y
i
n
t
h
e
l
a
s
t
d
ecad
e.
Beca
u
s
e
s
o
ci
a
l
med
i
a
mak
e
s
i
n
fo
rma
t
i
o
n
eas
i
l
y
acces
s
i
b
l
e,
s
o
me
p
e
o
p
l
e
o
r
o
rg
a
n
i
za
t
i
o
n
s
d
i
s
t
r
i
b
u
t
e
fal
s
e
n
e
w
s
fo
r
p
o
l
i
t
i
cal
o
r
co
mmerci
a
l
g
a
i
n
.
T
h
i
s
n
e
w
s
may
i
n
f
l
u
e
n
ce
el
ec
t
i
o
n
s
an
d
at
t
i
t
u
d
es
.
E
v
en
t
h
o
u
g
h
E
n
g
l
i
s
h
fak
e
n
e
w
s
i
s
w
i
d
el
y
d
et
ec
t
e
d
an
d
l
i
mi
t
e
d
,
A
rab
i
c
fak
e
n
ew
s
i
s
h
ar
d
t
o
reco
g
n
i
ze
o
w
i
n
g
t
o
a
l
ack
o
f
s
t
u
d
y
an
d
d
at
a
co
l
l
ect
i
o
n
.
W
ara
A
rab
i
c
b
i
d
i
rect
i
o
n
al
e
n
co
d
er
rep
re
s
en
t
at
i
o
n
s
fro
m
t
ra
n
s
f
o
rmers
(W
araBE
R
T
)
,
a
h
y
b
ri
d
feat
u
re
ex
t
rac
t
i
o
n
ap
p
ro
ac
h
,
co
mb
i
n
e
s
w
o
rd
l
e
v
el
t
o
k
en
i
zat
i
o
n
w
i
t
h
t
w
o
A
rab
i
c
b
i
d
i
rect
i
o
n
al
e
n
co
d
er
r
ep
res
e
n
t
a
t
i
o
n
s
fro
m
t
ran
s
fo
rmer
s
(A
raBE
R
T
)
v
ar
i
an
t
s
t
o
p
ro
v
i
d
e
v
ar
i
ed
feat
u
res
.
T
h
e
s
t
u
d
y
al
s
o
d
i
s
cu
s
s
e
s
el
i
mi
n
at
i
n
g
s
t
o
p
w
o
r
d
s
,
p
u
n
ct
u
at
i
o
n
s
,
an
d
t
a
n
w
een
mark
i
n
g
s
fro
m
A
rab
i
c
d
at
a.
T
h
i
s
s
t
u
d
y
emp
l
o
y
e
d
t
w
o
d
at
a
s
et
s
.
T
h
e
fi
r
s
t
,
A
rab
i
c
fak
e
n
e
w
s
d
at
a
s
et
(
A
FN
D
)
,
h
a
s
6
0
6
,
9
1
2
reco
r
d
s
.
Sec
o
n
d
d
a
t
as
e
t
A
ra
b
i
c
n
ew
s
(
A
raN
e
w
s
)
h
as
1
2
3
,
2
1
9
en
t
ri
es
.
W
araB
E
RT
-
V
1
o
b
t
ai
n
ed
9
3
.
8
3
%
A
FN
D
accu
racy
u
s
i
n
g
t
h
e
b
i
d
i
rec
t
i
o
n
a
l
l
o
n
g
s
h
o
r
t
-
t
erm
memo
r
y
(Bi
L
ST
M)
mo
d
el
.
H
o
w
ev
er,
t
h
e
W
araBE
R
T
-
V
2
t
ech
n
i
q
u
e
o
b
t
ai
n
ed
8
1
.
2
5
%
,
i
mp
r
o
v
i
n
g
d
et
ec
t
i
o
n
accu
rac
y
ab
o
v
e
p
re
v
i
o
u
s
res
earc
h
ers
fo
r
t
h
e
A
raN
ew
s
d
at
a
s
et
.
T
h
e
s
e
fi
n
d
i
n
g
s
s
h
o
w
t
h
a
t
W
araB
E
RT
o
u
t
p
erf
o
rms
w
o
r
d
l
i
s
t
t
ec
h
n
i
q
u
es
(W
L
T
)
,
t
erm
freq
u
en
c
y
-
i
n
v
er
s
e
d
o
cu
me
n
t
fre
q
u
en
cy
(T
F
-
I
D
F)
,
an
d
A
raBE
R
T
o
n
b
o
t
h
d
a
t
as
e
t
s
.
K
e
y
w
o
r
d
s
:
Ar
a
B
E
R
T
Ar
a
Ne
ws
F
a
ke
ne
ws
T
oke
niza
ti
on
W
a
r
a
B
E
R
T
Th
i
s
i
s
a
n
o
p
en
a
c
ces
s
a
r
t
i
c
l
e
u
n
d
e
r
t
h
e
CC
B
Y
-
SA
l
i
ce
n
s
e.
C
or
r
e
s
pon
din
g
A
u
th
or
:
S
uha
il
a
Abuow
a
ida
De
pa
r
tm
e
nt
of
C
omput
e
r
S
c
ienc
e
,
F
a
c
ult
y
of
I
nf
o
r
mation
T
e
c
hnology,
Al
a
l
-
B
a
yt
Unive
r
s
it
y
M
a
f
r
a
q
25113,
J
or
da
n
E
mail:
s
a
buowe
uda
@z
u.
e
du.
jo
1.
I
NT
RODU
C
T
I
ON
W
it
h
the
wide
s
pr
e
a
d
of
the
int
e
r
ne
t
in
the
wor
ld
,
many
plat
f
or
ms
of
s
oc
ial
media
li
ke
I
ns
tagr
a
m,
F
a
c
e
book,
T
witt
e
r
,
a
nd
many
other
s
,
ha
ve
be
c
o
me
popular
a
mong
pe
ople
a
nd
they
a
r
e
us
e
d
da
il
y
a
s
they
f
a
c
il
it
a
te
ge
tt
ing
in
f
or
mation
a
nd
r
a
pidl
y
s
pr
e
a
din
g
the
ne
ws
.
T
he
number
o
f
s
oc
ial
media
us
e
r
s
a
r
ound
the
wor
ld
is
incr
e
a
s
ing
dr
a
matica
ll
y,
the
number
of
u
s
e
r
s
in
2017
wa
s
a
bout
2.
73
bil
li
on
pe
ople,
while
in
2019,
the
number
of
us
e
r
s
incr
e
a
s
e
d
to
3.
51
bil
li
on,
a
nd
in
2022
they
be
c
a
me
4
.
59
bil
li
on
pe
ople,
whic
h
c
o
ns
ti
tut
e
s
a
bout
58%
of
the
wo
r
ld's
population
[
1
]
.
W
it
h
the
incr
e
a
s
e
in
the
nu
mber
o
f
us
e
r
s
,
the
number
of
pe
ople
who
de
pe
nd
on
s
oc
ial
media
a
s
a
s
our
c
e
f
r
om
whi
c
h
they
take
ne
ws
incr
e
a
s
e
s
.
R
e
c
e
nt
s
tudi
e
s
ha
ve
be
e
n
c
onduc
ted
in
Ame
r
ica
a
bout
the
number
of
s
oc
ial
media
us
e
r
s
s
howing
that
71
%
o
f
thes
e
pe
ople
us
e
the
F
a
c
e
book
platf
or
m
,
a
nd
a
bout
52%
of
them
r
e
ly
o
n
F
a
c
e
book
a
s
a
s
our
c
e
o
f
ne
ws
,
whic
h
mea
ns
that
36%
of
Ame
r
ica
ns
ge
t
their
ne
ws
f
r
om
F
a
c
e
book
[
2]
.
All
of
thes
e
f
a
c
tor
s
made
f
a
ke
ne
ws
a
pa
r
t
of
our
onli
ne
da
il
y
r
outi
ne
s
.
As
f
a
ke
ne
ws
ha
d
a
n
e
nor
mous
ha
r
mf
ul
im
pa
c
t
,
it
ga
ined
popu
lar
it
y
a
mong
poli
ti
c
ians
,
jour
na
li
s
ts
,
r
e
s
e
a
r
c
he
r
s
,
a
nd
the
publi
c
a
t
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
E
lec
&
C
omp
E
ng
I
S
S
N:
2088
-
8708
A
r
abic
fake
ne
w
s
de
tec
ti
on
us
ing
hy
br
id
c
ontex
tual
featur
e
s
(
Hus
s
ain
M
ohamm
e
d
T
ur
k
i
)
837
lar
ge
.
I
t
is
pr
e
pa
r
e
d
or
dis
s
e
mi
na
ted
to
de
c
e
ive
t
he
publi
c
a
nd
ha
r
m
the
r
e
putation
of
a
n
e
nti
ty
,
p
e
r
s
on,
o
r
a
ge
nc
y,
f
or
poli
ti
c
a
l
or
f
inanc
ial
ga
in.
F
a
ke
ne
w
s
may
a
f
f
e
c
t
pe
ople's
opini
ons
ne
ga
ti
ve
ly
by
tar
ge
ti
ng
a
s
pe
c
if
ic
ins
ti
tut
ion
or
pe
r
s
ona
li
ty
[
3]
–
[
5
]
.
I
n
2016
,
M
a
r
k
Z
uc
ke
r
be
r
g
c
onf
e
s
s
e
d
that
f
a
ke
ne
ws
on
F
a
c
e
book
pos
ts
a
nd
s
tor
ies
s
uppor
ted
by
R
us
s
ians
,
with
poli
t
ica
l
goa
ls
pr
e
s
e
nted
to
ove
r
126
mi
ll
ion
pe
ople
in
Ame
r
ica
[
6]
.
De
e
p
l
e
a
r
ning
(
DL
)
c
a
n
p
lay
a
n
im
po
r
tant
r
o
le
in
r
e
duc
ing
the
s
pr
e
a
d
of
f
a
ke
ne
ws
by
pr
ovidi
ng
mor
e
s
ophis
ti
c
a
ted
tec
hniques
to
identif
y
inac
c
ur
a
te
or
mi
s
lea
ding
inf
or
mation
[
7]
.
As
DL
models
a
r
e
c
a
pa
ble
of
a
utom
a
ti
c
a
ll
y
lea
r
ning
a
nd
e
xtr
a
c
ti
ng
high
-
leve
l
f
e
a
tur
e
s
f
r
om
r
a
w
text
da
ta,
ther
e
f
o
r
e
us
ing
DL
c
a
n
be
be
tt
e
r
than
tr
a
dit
i
ona
l
mac
hine
lea
r
ning
(
M
L
)
methods
f
or
Ar
a
bic
f
a
ke
ne
ws
de
tec
ti
on.
T
he
main
goa
l
o
f
thi
s
pa
pe
r
is
to
d
is
ti
nguis
h
f
a
ke
ne
ws
f
r
om
r
e
a
l
ne
ws
,
in
a
ddit
ion
to
b
uil
ding
a
s
ys
tem
to
de
tec
t
thi
s
ne
ws
to
li
mi
t
it
s
s
pr
e
a
d
on
s
oc
ial
media
,
T
he
s
e
goa
ls
c
a
n
be
s
umm
a
r
ize
d
a
s
f
oll
ows
:
c
r
e
a
te
a
ne
w
hybr
id
f
e
a
tur
e
e
xtr
a
c
ti
on
method
by
c
onc
a
tena
ti
ng
dif
f
e
r
e
nt
f
e
a
tur
e
e
xtr
a
c
ti
on
methods
(
s
uc
h
a
s
wor
d
-
leve
l
tokeniz
a
ti
on
,
te
r
m
f
r
e
que
nc
y
-
inver
s
e
doc
ument
f
r
e
que
nc
y
(
T
F
-
I
DF)
,
a
nd
Ar
a
bic
bid
ir
e
c
ti
ona
l
e
nc
ode
r
r
e
pr
e
s
e
ntations
f
r
om
t
r
a
ns
f
or
mer
s
(
Ar
a
B
E
R
T
)
)
to
e
nha
nc
e
the
de
tec
ti
on
a
c
c
ur
a
c
y,
c
o
mpar
ing
Ar
a
B
E
R
T
ve
r
s
ions
(
V1
a
nd
V2)
in
the
f
ield
of
f
e
a
t
ur
e
e
xtr
a
c
ti
on
(
tokeniz
a
ti
on)
a
nd
de
ter
mi
ning
the
b
e
s
t
one
,
a
nd
f
inally
indi
c
a
te
the
e
f
f
e
c
t
o
f
a
pplyi
ng
da
ta
c
lea
ning
(
s
uc
h
a
s
r
e
movi
ng
s
top
wor
ds
,
r
e
movi
ng
pun
c
tuation,
a
nd
nor
malizing
text
)
on
the
da
tas
e
t.
2.
RE
L
AT
E
D
WORKS
Kha
li
l
e
t
al.
pr
ov
ided
a
lar
ge
,
labe
led,
a
nd
div
e
r
s
e
Ar
a
bic
f
a
ke
ne
ws
da
tas
e
t
(
A
F
ND
)
that
is
c
oll
e
c
ted
f
r
om
publi
c
Ar
a
bic
ne
ws
we
bs
it
e
s
.
AF
ND
c
ons
is
ts
of
606
,
912
publi
c
ne
ws
[
8
]
.
S
a
a
da
ny
e
t
al.
c
r
e
a
ted
a
da
tas
e
t
c
on
s
is
ti
ng
of
3,
185
a
r
ti
c
les
f
r
om
two
Ar
a
bic
s
a
ti
r
ica
l
ne
ws
we
bs
it
e
s
[
9]
.
T
he
y
p
r
o
pos
e
d
a
model
that
a
c
hieve
d
high
a
c
c
ur
a
c
y,
up
to
98
.
6%
,
in
identif
ying
s
a
ti
r
ica
l
f
a
ke
ne
ws
in
A
r
a
bic
.
On
t
he
other
ha
nd,
a
n
innovative
method
f
o
r
a
utom
a
ti
c
a
ll
y
c
r
e
a
ti
ng
Ar
a
bic
-
manipulate
d
ne
ws
s
tor
ies
wa
s
pr
o
pos
e
d
by
Na
goudi
e
t
al.
[
10]
.
T
he
y
ga
ther
e
d
a
nd
publi
s
he
d
a
ne
w
pa
r
t
-
of
-
s
pe
e
c
h
-
tagge
d
(
P
OS
-
tagge
d
)
Ar
a
bic
ne
ws
da
tas
e
t
known
a
s
A
r
a
bic
ne
ws
(
Ar
a
Ne
ws
)
.
Alkha
i
r
e
t
al
.
de
ve
loped
a
n
Ar
a
bic
c
or
pus
f
o
r
f
a
ke
ne
ws
a
na
lys
is
,
c
onc
e
ntr
a
ti
ng
on
the
mos
t
r
umor
e
d
c
onc
e
pts
.
As
M
L
c
las
s
if
ier
s
,
they
us
e
d
mul
ti
nomi
a
l
n
a
ïve
B
a
ye
s
(
M
NB
)
,
de
c
is
ion
tr
e
e
s
(
DT
)
,
a
nd
s
uppor
t
ve
c
tor
mac
hin
e
s
(
S
VM
)
[
11]
–
[
13
]
.
C
onve
r
s
e
ly,
Az
a
d
c
onc
e
ntr
a
ted
on
c
r
e
a
ti
ng
Ar
a
bic
f
a
ke
ne
ws
da
tas
e
ts
of
s
upe
r
ior
qua
li
ty
a
nd
c
r
e
a
ti
ng
a
n
a
c
c
ur
a
te
c
las
s
if
ica
ti
on
f
or
Ar
a
bic
f
a
ke
ne
ws
.
I
t
pr
ovides
ins
ight
s
a
nd
r
e
c
omm
e
nda
ti
ons
f
or
r
e
s
e
a
r
c
he
r
s
in
the
f
ield
of
f
a
ke
ne
ws
r
e
c
ognit
ion
in
Ar
a
bic
[
14
]
–
[
16]
.
Na
s
s
if
us
e
d
a
r
a
nge
of
c
ontex
tualize
d
Ar
a
bic
e
mbedding
a
lgo
r
it
hms
to
c
r
e
a
te
mul
ti
ple
tr
a
ns
f
or
mer
models
to
de
tec
t
f
a
ke
ne
ws
in
the
Ar
a
bic
langua
ge
.
E
ight
t
r
a
ns
f
or
mer
models
we
r
e
us
e
d
in
the
s
tudy
[
17]
.
whe
r
e
a
s
M
L
a
ppr
oa
c
he
s
we
r
e
us
e
d
to
t
r
a
in
the
model
by
Himdi
e
t
al
.
[
18]
.
Aw
a
jan
highl
ight
e
d
the
c
r
it
ica
l
is
s
ue
of
f
a
ke
ne
ws
on
s
oc
ial
media
,
e
s
pe
c
ially
on
T
witt
e
r
,
whe
r
e
f
a
ls
e
inf
or
mation
is
inc
r
e
a
s
ingl
y
be
ing
s
pr
e
a
d.
U
s
ing
the
T
witt
e
r
a
ppli
c
a
ti
on
pr
ogr
a
mm
ing
int
e
r
f
a
c
e
(
API
)
,
the
r
e
s
e
a
r
c
he
r
ga
ther
e
d
a
da
tas
e
t
c
ompr
is
ing
20
6,
080
twe
e
ts
[
19
]
.
S
or
our
a
nd
Abde
lkade
r
us
e
d
a
hybr
id
method
that
int
e
gr
a
tes
tr
a
dit
ional
ne
ur
a
l
ne
twor
k
s
(
NN
s
)
with
long
s
hor
t
-
ter
m
memor
y
methods
(
L
S
T
M
)
methods
.
S
his
ha
h
c
r
e
a
ted
a
nd
im
pleme
nted
joi
nt
B
E
R
T
,
a
n
innovative
tec
hnique
f
or
de
tec
ti
ng
f
a
k
e
ne
ws
in
Ar
a
bic
da
tas
e
ts
[
20]
.
E
xtens
ive
e
xpe
r
im
e
nts
we
r
e
c
a
r
r
ied
out
e
mpl
oying
a
c
tual
Ar
a
bic
f
a
ke
ne
ws
da
tas
e
ts
s
uc
h
a
s
c
or
ona
vir
us
di
s
e
a
s
e
2019
(
C
OV
I
D
-
19)
f
a
ke
s
,
S
a
ti
r
ica
l,
Ar
a
Ne
ws
,
a
nd
other
s
[
21]
.
Aw
a
j
a
n
e
t
al.
noted
that
us
ing
e
ns
e
mbl
e
models
a
c
hieve
d
be
tt
e
r
a
c
c
ur
a
c
y
than
us
ing
a
s
ingl
e
mac
hine
lea
r
ning
c
l
a
s
s
if
ier
.
An
onli
ne
r
e
pos
it
or
y
wa
s
include
d
to
p
r
ovide
c
onti
nuous
upda
tes
a
nd
r
e
s
our
c
e
s
r
e
late
d
to
f
a
ke
ne
ws
,
including
e
duc
a
ti
ona
l
pr
ogr
a
ms
,
publi
c
a
ti
ons
,
ne
w
methods
,
da
tas
e
ts
,
a
nd
other
r
e
leva
nt
r
e
s
our
c
e
s
[
19]
.
R
a
ha
b
e
t
al.
a
ddr
e
s
s
e
d
the
c
ha
ll
e
ng
e
of
f
a
ke
ne
ws
de
tec
ti
on
a
nd
e
li
mi
na
ti
on,
pa
r
ti
c
ular
ly
in
the
c
o
ntext
of
us
e
r
-
ge
ne
r
a
ted
c
ontent
on
s
oc
ial
media
platf
or
ms
[
22]
.
Alotaibi
identif
ied
A
r
a
bic
f
a
ke
ne
ws
twe
e
ts
r
e
late
d
to
the
C
OV
I
D
-
19
pa
nde
mi
c
a
nd
c
las
s
if
ied
them
int
o
s
ix
c
a
tegor
ies
e
nter
tainment,
he
a
lt
h,
poli
ti
c
s
,
r
e
li
gious
,
s
oc
ial,
a
nd
s
por
ts
[
23]
.
A
dis
inf
or
mation
de
tec
ti
on
f
r
a
mew
or
k
s
pe
c
if
ica
ll
y
de
s
igned
f
or
c
ombating
th
e
s
pr
e
a
d
of
f
a
ke
inf
o
r
mation
r
e
late
d
to
the
C
OV
I
D
-
19
pa
nde
mi
c
on
Ar
a
bic
s
oc
ial
media
platf
or
ms
wa
s
pr
o
pos
e
d
by
E
laz
iz
e
t
al
.
[
24
]
.
T
he
r
e
s
e
a
r
c
he
r
e
mpl
oye
d
a
c
ombi
na
ti
on
o
f
mul
ti
-
tas
k
lea
r
ning
(
M
T
L
)
,
a
pr
e
-
tr
a
ined
tr
a
ns
f
o
r
mer
-
ba
s
e
d
model
Ar
a
B
E
R
T
,
a
nd
a
n
opti
m
iza
ti
on
a
lgo
r
it
hm
f
i
r
e
ha
wk
opti
mi
z
e
r
in
their
f
r
a
mew
or
k
.
A
meur
a
nd
Aliane
a
ddr
e
s
s
e
d
the
i
s
s
ue
of
a
n
“
inf
ode
mi
c
”
whe
r
e
f
a
ls
e
a
nd
mi
s
lea
ding
inf
or
mation
a
bout
C
OV
I
D
-
19
ha
s
e
mer
ge
d
a
nd
c
ompl
ica
ted
r
e
s
pons
e
e
f
f
or
ts
.
T
he
r
e
s
e
a
r
c
he
r
s
highl
ight
the
r
ole
of
s
oc
ial
ne
twor
king
s
it
e
s
in
c
ontr
ibut
ing
to
the
s
pr
e
a
d
of
r
umor
s
,
c
ons
pir
a
c
y
theor
ies
,
ha
te
s
pe
e
c
h,
xe
nophobia,
r
a
c
is
m,
a
nd
pr
e
judi
c
e
[
25]
.
F
ou
r
types
of
f
e
a
tur
e
s
we
r
e
us
e
d:
c
ount
ve
c
tor
,
wor
d
-
leve
l
T
F
-
I
DF,
n
-
gr
a
m
-
leve
l
T
F
-
I
DF,
a
nd
c
ha
r
a
c
ter
-
leve
l
T
F
-
I
DF.
S
ix
c
las
s
if
ier
s
we
r
e
e
m
ployed
(
na
ïve
B
a
ye
s
,
logi
s
ti
c
r
e
gr
e
s
s
ion,
s
uppor
t
ve
c
tor
mac
hine,
mul
ti
laye
r
pe
r
c
e
ptr
on,
r
a
ndom
f
or
e
s
t
ba
g
ging,
a
nd
e
xt
r
e
me
gr
a
dient
boos
ti
ng
)
[
26]
–
[
28
]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2088
-
8708
I
nt
J
E
lec
&
C
omp
E
ng
,
Vol
.
15
,
No.
1
,
F
e
br
ua
r
y
20
25
:
836
-
845
838
3.
DA
T
A
COL
L
E
C
T
I
ON
I
n
thi
s
pa
pe
r
,
two
da
tas
e
ts
we
r
e
us
e
d
f
o
r
the
e
xpe
r
im
e
nts
:
t
he
f
ir
s
t
da
tas
e
t
wa
s
c
oll
e
c
ted
by
Kha
li
l
e
t
al.
[
8]
,
a
lar
ge
a
nd
diver
s
e
AFND
f
r
o
m
publi
c
Ar
a
bic
ne
ws
we
bs
it
e
s
.
I
t
c
ontains
a
bou
t
606,
912
publi
c
ne
ws
a
r
ti
c
les
ga
ther
e
d
ove
r
6
mon
ths
f
r
o
m
134
publi
c
ne
ws
we
bs
it
e
s
in
19
di
f
f
e
r
e
nt
Ar
a
b
c
ountr
ies
.
T
his
da
tas
e
t
wa
s
us
e
d
to
c
ompar
e
the
pr
opos
e
d
f
e
a
tur
e
e
xtr
a
c
ti
on
method
wi
th
other
typi
c
a
l
f
e
a
tur
e
e
xtr
a
c
ti
on
methods
.
T
he
s
e
c
ond
da
tas
e
t,
Ar
a
Ne
ws
,
a
s
igni
f
i
c
a
nt,
mul
ti
-
topi
c
,
a
nd
mu
lt
i
-
c
ountr
y
A
r
a
bic
ne
ws
da
tas
e
t,
wa
s
c
r
e
a
ted
by
[
10]
.
I
t
c
ons
is
ts
of
123,
219
r
e
c
o
r
ds
.
T
his
da
tas
e
t
wa
s
us
e
d
to
c
ompar
e
the
r
e
s
ult
s
of
the
pr
opos
e
d
model
with
othe
r
r
e
s
e
a
r
c
he
r
s
’
r
e
s
ult
s
.
Da
ta
c
lea
ning
is
a
n
e
s
s
e
nti
a
l
pr
oc
e
s
s
whe
n
wor
k
ing
on
Ar
a
bic
textua
l
da
ta,
thi
s
pr
oc
e
s
s
include
s
s
e
ve
r
a
l
s
teps
:
−
R
e
movi
ng
p
unc
tuations
:
li
ke
c
omm
a
s
,
que
s
ti
on
mar
ks
,
pe
r
iods
,
a
nd
many
other
s
whic
h
us
ua
ll
y
d
o
not
ha
ve
a
n
im
por
tant
s
e
mantic
va
lue.
−
R
e
movi
ng
s
top
wor
ds
:
s
top
wor
ds
a
r
e
popular
wo
r
ds
that
a
r
e
us
e
d
us
ua
ll
y
in
the
langua
ge
but
ha
ve
li
tt
le
or
no
va
lue
,
s
uc
h
a
s
pr
e
pos
it
ions
,
a
nd
c
onjunctions
.
−
R
e
movi
ng
tanw
e
e
n
mar
ks
.
4.
M
E
T
HO
D
De
s
pit
e
the
e
nor
mous
de
ve
lopm
e
nt
in
the
f
ield
of
a
r
ti
f
icia
l
in
telli
ge
nc
e
a
nd
the
a
va
il
a
bil
it
y
o
f
many
DL
a
lgor
it
hms
[
29]
–
[
32
]
,
na
tur
a
l
langua
ge
pr
oc
e
s
s
ing
(
NL
P
)
s
uf
f
e
r
s
f
r
om
the
lac
k
o
f
a
va
il
a
ble
f
e
a
tur
e
e
xtr
a
c
ti
on
tec
hniques
f
or
Ar
a
bic
text
.
T
he
r
e
f
or
e
,
it
wa
s
ne
c
e
s
s
a
r
y
to
de
ve
lop
a
hybr
id
f
e
a
tur
e
e
xtr
a
c
ti
on
method
that
c
omb
ines
mor
e
than
one
f
e
a
tur
e
e
xtr
a
c
ti
on
method
a
s
s
hown
in
F
igur
e
1.
T
his
ne
w
m
ode
l
will
he
lp
im
pr
ove
the
wor
k
o
f
thes
e
a
lgor
i
thm
s
.
F
igur
e
1.
Ge
ne
r
a
l
p
r
opos
e
d
f
r
a
mew
or
k
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
E
lec
&
C
omp
E
ng
I
S
S
N:
2088
-
8708
A
r
abic
fake
ne
w
s
de
tec
ti
on
us
ing
hy
br
id
c
ontex
tual
featur
e
s
(
Hus
s
ain
M
ohamm
e
d
T
ur
k
i
)
839
W
he
n
de
a
li
ng
with
textua
l
da
ta,
one
o
f
the
mos
t
im
por
tant
po
int
s
is
e
xt
r
a
c
ti
ng
f
e
a
tur
e
s
f
r
om
texts
a
f
ter
be
ing
c
onve
r
ted
int
o
numer
ica
l
da
ta.
T
he
r
e
f
or
e
,
the
r
e
s
ult
s
a
r
e
a
f
f
e
c
ted
by
the
tokeniz
a
ti
on
methods
.
T
a
ble
1
s
hows
f
our
models
wor
d
li
s
t
tec
hniques
(
W
L
T
)
,
T
F
-
I
DF
,
a
r
a
B
E
R
T
-
V1,
a
nd
a
r
a
B
E
R
T
-
V2
a
nd
it
s
tokeniz
a
ti
on
us
ing
a
s
a
mpl
e
s
e
ntenc
e
.
I
n
thi
s
r
e
s
e
a
r
c
h
pr
opos
e
s
ne
w
hyb
r
id
models
that
c
onc
a
tena
te
be
twe
e
n
thes
e
models
a
s
f
oll
owing:
−
W
a
r
a
B
E
R
T
:
t
his
model
c
ombi
ne
s
W
L
T
a
nd
a
r
a
B
E
R
T
models
.
T
he
voc
a
bular
y
s
ize
of
a
r
a
B
E
R
T
is
s
tatic
64
,
000,
while
the
voc
a
bular
y
s
ize
o
f
W
L
T
is
a
h
ype
r
pa
r
a
mete
r
.
T
he
maxi
mum
length
f
o
r
both
m
ode
ls
mus
t
be
de
ter
mi
ne
d
be
f
or
e
c
a
ll
ing
the
models
f
or
tokeniz
a
ti
on.
Af
ter
tokeniz
a
ti
on
a
nd
indexing
the
a
r
ti
c
les
us
ing
e
a
c
h
model
a
lone,
the
output
of
th
os
e
models
would
be
c
onc
a
tena
ted.
T
he
c
onc
a
te
na
ted
output
will
be
us
e
d
f
or
t
r
a
ini
ng
a
nd
tes
ti
ng.
An
d
s
o
a
s
not
to
ge
t
dupli
c
a
te
va
lues
,
the
va
lue
o
f
the
a
r
a
B
E
R
T
voc
a
bular
y
s
ize
64,
000
mus
t
be
a
dde
d
f
or
e
a
c
h
va
lue
in
the
W
L
T
.
T
hus
,
the
r
a
nge
of
a
r
a
B
E
R
T
f
r
om
0
to
639
,
99,
while
the
W
L
T
va
lues
s
tar
t
f
r
om
64
,
000
.
Nota
ble
that
W
a
r
a
B
E
R
T
ha
s
two
ve
r
s
ions
W
a
r
a
B
E
R
T
-
V1,
a
nd
W
a
r
a
B
E
R
T
-
V2
de
pe
nding
on
the
c
hos
e
n
ve
r
s
ion
of
the
a
r
a
B
E
R
T
tokeniz
e
r
.
−
Ar
a
T
F
I
DF:
i
t
r
e
f
e
r
s
to
the
c
ombi
na
ti
on
of
a
r
a
B
E
R
T
with
the
T
F
I
DF
model
.
I
t
ha
s
the
s
a
me
a
r
c
hit
e
c
tur
e
a
s
W
a
r
a
B
E
R
T
.
I
n
a
ddit
ion
,
it
a
ls
o
ha
s
two
ve
r
s
ions
Ar
a
T
F
I
DF
-
V1,
a
nd
A
r
a
T
F
I
D
F
-
V2
de
pe
nding
o
n
the
c
hos
e
n
ve
r
s
ion
of
the
a
r
a
B
E
R
T
tokeniz
e
r
.
−
T
oke
nT
F
I
DF:
t
he
pa
r
t
T
oke
n
r
e
f
e
r
s
to
the
wor
d
to
ke
niza
ti
on.
T
his
model
c
ombi
ne
s
W
L
T
with
T
F
-
I
DF.
I
t
a
ls
o
ha
s
the
s
a
me
pr
oc
e
dur
e
a
s
the
pr
e
vious
ly
p
r
op
os
e
d
models
.
T
a
ble
1.
AFND
ve
r
s
ions
R
e
movi
ng s
to
pw
or
ds
R
e
movi
ng punc
tu
a
ti
ons
R
e
movi
ng
ta
nw
e
e
n ma
r
ks
A
F
N
D
-
V1
No
No
No
A
F
N
D
-
V2
Y
e
s
Y
e
s
No
A
F
N
D
-
V3
Y
e
s
Y
e
s
Y
e
s
5.
E
XP
E
RI
M
E
NT
AL
RE
S
U
L
T
S
AN
D
DI
S
CU
S
S
I
ON
T
o
de
ter
mi
ne
the
mos
t
a
ppr
opr
iate
DL
model
f
or
Ar
a
bic
f
a
ke
ne
ws
de
tec
ti
on
,
the
mos
t
known
a
nd
powe
r
f
ul
f
our
DL
models
we
r
e
us
e
d
to
pe
r
f
or
m
t
he
tr
a
ini
ng
pr
oc
e
s
s
to
de
ter
mi
ne
the
mos
t
a
ppr
opr
iate
one
:
r
e
c
ur
r
e
nt
ne
ur
a
l
ne
twor
k
(
R
NN
)
,
long
s
hor
t
-
ter
m
memor
y
(
L
S
T
M
)
,
bidi
r
e
c
ti
ona
l
long
s
hor
t
-
ter
m
memor
y
(
B
iL
S
T
M
)
a
nd
c
onvolut
ional
ne
ur
a
l
ne
twor
k
-
bidi
r
e
c
ti
ona
l
long
s
hor
t
-
ter
m
memor
y
(
C
NN
-
B
iL
S
T
M
)
.
As
the
AFND
is
a
nove
l
a
nd
huge
da
tas
e
t
a
nd
no
e
xpe
r
i
ments
ha
ve
be
e
n
c
onduc
ted
on
it
by
other
r
e
s
e
a
r
c
he
r
s
,
thi
s
pa
pe
r
us
e
s
30,
000
s
a
mpl
e
s
of
the
da
tas
e
t
a
nd
a
p
pli
e
s
the
e
xpe
r
im
e
nts
on
thr
e
e
ve
r
s
ions
of
the
da
tas
e
t
s
e
e
T
a
ble
1.
−
AFND
-
V1:
T
his
ve
r
s
ion
c
ontains
the
or
igi
na
l
da
ta,
without
a
pplyi
ng
da
ta
c
lea
ning
(
r
e
m
oving
punc
tuations
,
s
topwor
ds
,
a
nd
tanw
e
e
n
mar
ks
)
.
−
AFND
-
V2:
T
his
ve
r
s
ion
include
s
the
da
ta
a
f
te
r
r
e
movi
ng
punc
tuations
a
nd
s
topwor
ds
.
−
AFND
-
V3:
I
n
thi
s
ve
r
s
ion,
a
ll
da
ta
c
lea
ning
pr
oc
e
s
s
e
s
ha
d
be
e
n
c
onduc
ted
on
the
or
igi
na
l
da
tas
e
t.
T
his
mea
ns
that
punc
tuation,
s
topwor
ds
,
a
nd
tanw
e
e
n
m
a
r
ks
a
r
e
r
e
moved
.
F
or
a
ll
th
r
e
e
ve
r
s
ions
of
AFND,
e
ight
f
e
a
tur
e
e
xtr
a
c
ti
on
methods
,
a
nd
f
our
DL
models
we
r
e
us
e
d
in
the
e
xpe
r
im
e
nts
.
T
he
r
e
a
r
e
thr
e
e
main
goa
ls
be
hind
thes
e
e
xpe
r
im
e
nts
:
the
f
ir
s
t
goa
l
is
to
unde
r
s
tand
the
e
f
f
e
c
t
o
f
r
e
movi
ng
punc
tuations
,
s
topwor
ds
,
a
nd
t
a
nwe
e
n
mar
ks
on
the
de
tec
ti
on
pr
oc
e
s
s
.
T
he
s
e
c
ond
goa
l
is
to
de
ter
mi
ne
the
be
s
t
f
e
a
tu
r
e
e
xtr
a
c
ti
on
method
,
in
a
ddit
ion
to
c
he
c
king
the
pr
opos
e
d
methods
.
T
he
l
a
s
t
goa
l
is
to
s
e
lec
t
the
mos
t
a
ppr
opr
iate
DL
model.
Not
a
ble
that
the
tr
a
ini
ng
wa
s
r
e
pe
a
ted
f
ive
ti
mes
to
a
c
hieve
a
c
c
ur
a
te
r
e
s
ult
s
.
F
or
a
ll
DL
models
,
the
voc
a
bular
y
s
ize
f
or
W
L
T
,
a
r
a
B
E
R
T
,
a
nd
T
F
-
I
DF
is
176
,
000
,
64,
000
,
a
nd
1,
000
c
ons
e
c
uti
ve
ly.
T
he
y
ha
ve
the
s
a
me
le
ngth
(
number
of
a
tt
r
ibut
e
s
)
whic
h
is
160.
Al
l
m
ode
ls
a
r
e
e
va
luate
d
ba
s
e
d
on
the
a
c
c
ur
a
c
y
metr
ic.
T
he
a
ve
r
a
ge
tes
t
a
c
c
ur
a
c
y
is
c
a
lcula
ted
us
ing
K
-
f
old,
whe
r
e
K=
5.
T
he
f
ir
s
t
a
r
c
hi
tec
tur
e
is
R
NN
,
whe
r
e
the
hype
r
pa
r
a
mete
r
s
that
a
r
e
s
e
lec
ted
to
tr
a
in
the
model
a
r
e
s
hown
in
T
a
ble
2
.
T
he
voc
a
bular
y
s
ize
o
f
the
f
e
a
t
ur
e
e
xtr
a
c
ti
on
method
r
e
pr
e
s
e
nts
the
input
dim
e
ns
ion,
a
nd
the
output
dim
e
ns
ion
f
o
r
a
ll
methods
is
150
ne
ur
ons
.
T
he
S
im
p
le
R
NN
laye
r
is
the
s
e
c
ond
la
ye
r
with
146
ne
ur
ons
with
a
r
e
c
ur
r
e
nt
dr
opout
r
a
te
of
0
.
12.
T
he
ne
xt
hidden
laye
r
s
a
r
e
f
ive
de
ns
e
laye
r
s
with
212
ne
ur
ons
a
nd
a
r
e
c
ti
f
ied
l
inea
r
unit
(
R
e
L
U)
a
c
t
ivation
f
unc
ti
on
f
or
e
a
c
h
laye
r
.
T
he
output
laye
r
with
one
ne
ur
on
us
e
s
the
s
igm
oid
a
c
ti
va
ti
on
f
unc
ti
on
0
o
r
1.
T
a
ble
3
de
s
c
r
ibes
R
NN
model
hype
r
pa
r
a
me
ter
s
a
nd
T
a
ble
3
s
hows
the
r
e
s
ult
s
us
ing
di
f
f
e
r
e
nt
f
e
a
tur
e
s
.
C
ompar
ing
the
r
e
s
ult
s
,
W
a
r
a
B
E
R
T
-
V1
a
c
hieve
s
h
igher
a
c
c
ur
a
c
y
on
AFND
-
V1
84.
83
%
,
AFND
-
V2
83.
35%
,
a
nd
AFND
-
V3
75
.
83%
.
W
a
r
a
B
E
R
T
-
V2
s
hows
s
im
il
a
r
tr
e
nds
to
W
a
r
a
B
E
R
T
-
V1
but
ge
ne
r
a
ll
y
ha
s
lowe
r
a
c
c
ur
a
c
ies
a
c
r
os
s
a
ll
AFND
ve
r
s
ions
.
T
a
ble
4
pr
e
s
e
nts
L
S
T
M
hype
r
pa
r
a
mete
r
s
that
a
r
e
s
e
lec
ted
to
tr
a
in
the
model.
T
he
f
i
r
s
t
laye
r
is
the
e
mbedding
laye
r
a
nd
the
voc
a
bular
y
s
ize
of
the
f
e
a
tur
e
e
xtr
a
c
ti
on
method
r
e
pr
e
s
e
nts
the
input
di
mens
ion,
a
nd
the
output
d
i
mens
ion
f
or
a
ll
methods
is
344
ne
u
r
ons
.
T
he
L
S
T
M
laye
r
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2088
-
8708
I
nt
J
E
lec
&
C
omp
E
ng
,
Vol
.
15
,
No.
1
,
F
e
br
ua
r
y
20
25
:
836
-
845
840
c
ontains
144
ne
ur
ons
a
nd
a
r
e
c
ur
r
e
nt
dr
opout
r
a
te
0.
2
.
T
he
ne
xt
hidden
laye
r
s
a
r
e
f
ou
r
de
ns
e
laye
r
s
c
ontaining
77
ne
ur
ons
with
a
d
r
opout
r
a
te
0.
17
a
n
d
a
R
e
L
U
a
c
ti
va
ti
on
f
unc
ti
on
f
or
e
a
c
h
laye
r
.
T
a
ble
5
s
hows
the
r
e
s
ult
s
of
L
S
T
M
us
ing
dif
f
e
r
e
nt
f
e
a
tur
e
s
.
T
a
ble
6
S
hows
B
i
-
L
S
T
M
hype
r
pa
r
a
mete
r
s
that
a
r
e
s
e
lec
ted
to
tr
a
in
thi
s
model,
whe
r
e
120
ne
ur
ons
a
nd
a
r
e
c
ur
r
e
nt
dr
opout
r
a
te
0.
12
,
f
ou
r
de
ns
e
laye
r
s
of
77
ne
ur
ons
with
a
dr
opout
r
a
te
0.
12
a
nd
a
R
e
L
U
a
c
ti
va
ti
on
f
unc
ti
on
f
or
e
a
c
h
laye
r
.
T
a
ble
7
s
hows
the
r
e
s
ult
s
of
B
i
-
L
S
T
M
us
ing
dif
f
e
r
e
nt
f
e
a
tur
e
s
.
W
a
r
a
B
E
R
T
-
V1
a
c
hieve
s
higher
a
c
c
ur
a
c
y
on
AFND
-
V1,
AFND
-
V2,
a
nd
AFND
-
V3
92
.
67%
,
92.
37%
,
a
nd
89
.
91%
c
ons
e
c
uti
ve
ly.
W
hil
e
W
a
r
a
B
E
R
T
-
V2
ha
s
s
im
il
a
r
pa
tt
e
r
ns
to
W
a
r
a
B
E
R
T
-
V1,
b
ut
of
ten
pe
r
f
or
ms
les
s
a
c
c
ur
a
tely
via
a
ll
other
AFND
v
e
r
s
ions
.
T
he
hype
r
pa
r
a
mete
r
s
of
the
f
our
th
a
r
c
hit
e
c
tur
e
C
NN
-
B
iL
S
T
M
a
r
e
s
hown
in
T
a
ble
8.
T
he
s
e
c
ond
laye
r
is
the
c
onvolut
ional
laye
r
with
10
f
il
ter
s
a
nd
the
length
of
f
il
ter
s
is
20,
the
B
i
-
L
S
T
M
laye
r
c
ontains
120
ne
ur
ons
a
nd
the
dr
opout
r
a
te
is
0.
22
,
the
de
n
s
e
laye
r
s
c
ontain
128
ne
ur
ons
with
a
d
r
opout
r
a
te
0.
23
a
nd
a
R
e
L
U
a
c
ti
va
ti
on
f
unc
ti
on
f
or
e
a
c
h
laye
r
.
T
a
ble
9
s
hows
C
NN
-
B
iL
S
T
M
r
e
s
ult
s
us
ing
dif
f
e
r
e
nt
f
e
a
tur
e
s
.
T
a
ble
2.
R
NN
tr
a
ini
ng
hype
r
pa
r
a
mete
r
s
R
N
N
ne
ur
ons
O
th
e
r
hi
dde
n l
a
ye
r
s
ne
ur
ons
B
a
tc
h s
iz
e
T
ot
a
l
e
poc
h
s
L1
L2
L3
L4
L5
146
212
212
212
212
212
337
6
T
a
ble
3.
R
NN
r
e
s
ult
s
us
ing
dif
f
e
r
e
nt
f
e
a
tu
r
e
s
T
a
ble
4.
L
S
T
M
tr
a
ini
ng
hype
r
pa
r
a
mete
r
s
E
mbe
ddi
ng ve
c
to
r
l
e
n.
L
S
T
M
ne
ur
ons
O
th
e
r
hi
dde
n l
a
ye
r
s
ne
ur
ons
B
a
tc
h s
iz
e
T
ot
a
l
e
poc
h
s
L1
L2
L3
L4
344
144
77
77
77
77
180
4
T
a
ble
5.
L
S
T
M
r
e
s
ult
s
us
ing
dif
f
e
r
e
nt
f
e
a
tur
e
s
T
a
ble
6.
B
i
-
L
S
T
M
tr
a
ini
ng
hype
r
pa
r
a
mete
r
s
E
mbe
ddi
ng ve
c
to
r
l
e
n.
B
iL
S
T
M
ne
ur
ons
O
th
e
r
hi
dde
n l
a
ye
r
s
ne
ur
ons
B
a
tc
h s
iz
e
T
ot
a
l
e
poc
h
s
L1
L2
L3
L4
400
120
128
128
128
128
144
4
M
ode
l
V
oc
a
bul
a
r
y s
iz
e
M
a
x l
e
ngt
h
T
r
a
in
a
bl
e
pa
r
a
me
te
r
s
T
e
s
t
a
c
c
ur
a
c
y
A
F
N
D
-
V1
A
F
N
D
-
V2
A
F
N
D
-
V3
W
L
T
176,000
160
26,655,000
75.79
68.86
64.47
a
r
a
B
E
R
T
-
V1
64,000
160
9,855,000
76.10
74.02
59.44
a
r
a
B
E
R
T
-
V2
64,000
160
9,855,000
81,48
74.19
60.05
W
a
r
a
B
E
R
T
-
V1
176,000 +
64,000
160 +
160
36,255,000
84.83
83.35
75.83
W
a
r
a
B
E
R
T
-
V2
176,000 +
64,000
160 +
160
36,255,000
83.40
79.53
73.11
T
oke
nT
F
I
D
F
176,000 +
1,000
160 +
160
26,805,000
51.24
50.01
49.13
A
r
a
T
F
I
D
F
-
V1
64,000 +
1,000
160 +
160
10,005,000
76.23
70.88
72.66
A
r
a
T
F
I
D
F
-
V2
64,000 +
1,000
160 +
160
10,005,000
75.58
77.14
71.14
W
L
T
176,000
320
26,655,000
59.71
54.19
52.97
a
r
a
B
E
R
T
-
V1
64,000
320
9,855,000
61.02
79.17
73.30
a
r
a
B
E
R
T
-
V2
64,000
320
9,855,000
66.85
71.23
55.62
M
ode
l
V
oc
a
bul
a
r
y s
iz
e
M
a
x l
e
ngt
h
T
r
a
in
a
bl
e
pa
r
a
me
te
r
s
T
e
s
t
a
c
c
ur
a
c
y
A
F
N
D
-
V1
A
F
N
D
-
V2
A
F
N
D
-
V3
W
L
T
176,000
160
60,854,000
85.02
84.30
70.13
a
r
a
B
E
R
T
-
V1
64,000
160
22,326,000
87.12
87.09
83.36
a
r
a
B
E
R
T
-
V2
64,000
160
22,326,000
86.27
86.99
83.35
W
a
r
a
B
E
R
T
-
V1
176,000 +
64,000
160 +
160
82,870,000
89.30
88.33
85.31
W
a
r
a
B
E
R
T
-
V2
176,000 +
64,000
160 +
160
82,870,000
88.52
87.95
84.36
T
oke
nT
F
I
D
F
176,000 +
1,000
160 +
160
61,158,000
60.08
60.33
54.17
A
r
a
T
F
I
D
F
-
V1
64,000 +
1,000
160 +
160
22,630,000
86.55
86.81
73.94
A
r
a
T
F
I
D
F
-
V2
64,000 +
1,000
160 +
160
22,630,000
85.09
82.21
71.18
W
L
T
176,000
320
60,854,000
59.18
56.62
54.12
a
r
a
B
E
R
T
-
V1
64,000
320
22,326,000
87.55
86.54
83.77
a
r
a
B
E
R
T
-
V2
64,000
320
22,326,000
84.68
84.02
82.50
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
E
lec
&
C
omp
E
ng
I
S
S
N:
2088
-
8708
A
r
abic
fake
ne
w
s
de
tec
ti
on
us
ing
hy
br
id
c
ontex
tual
featur
e
s
(
Hus
s
ain
M
ohamm
e
d
T
ur
k
i
)
841
T
a
ble
7.
B
i
-
L
S
T
M
r
e
s
ult
s
us
ing
dif
f
e
r
e
nt
f
e
a
tur
e
s
T
a
ble
8.
C
NN
-
B
iL
S
T
M
tr
a
ini
ng
hype
r
pa
r
a
mete
r
s
E
mbe
ddi
ng
ve
c
to
r
l
e
n.
N
o. of
f
il
te
r
s
F
il
te
r
le
ngt
h
B
iL
S
T
M
ne
ur
ons
O
th
e
r
hi
dde
n l
a
ye
r
s
ne
ur
ons
B
a
tc
h
s
iz
e
T
ot
a
l
e
poc
hs
L1
L2
L3
L4
400
10
20
120
128
128
128
128
144
4
T
a
ble
9.
C
NN
-
B
iL
S
T
M
r
e
s
ult
s
us
ing
dif
f
e
r
e
nt
f
e
a
t
ur
e
s
Although
W
a
r
a
B
E
R
T
-
V2
a
nd
T
oke
n
T
F
I
DF
a
c
hie
ve
d
high
a
c
c
ur
a
c
y
c
ompar
e
d
with
other
methods
,
W
a
r
a
B
E
R
T
-
V1
a
c
hieve
d
higher
a
c
c
ur
a
c
y
on
AFND
-
V1,
AFND
-
V2,
a
nd
AFND
-
V3
91.
96%
,
91.
6
4%
,
a
nd
89.
46%
c
ons
e
c
uti
ve
ly
a
s
s
hown
in
T
a
ble
9
.
T
o
a
na
lyzing
the
r
e
s
ult
s
of
a
ll
DL
models
a
c
c
or
ding
to
th
e
AFND
ve
r
s
ion
a
nd
the
e
f
f
e
c
t
of
r
e
movi
ng
s
topwor
ds
a
nd
tanw
e
e
n
mar
ks
,
a
s
we
ll
a
s
c
ompar
ing
the
pe
r
f
or
manc
e
of
tokeniz
a
ti
on
methods
,
we
c
a
n
make
the
f
oll
owing
obs
e
r
va
ti
ons
:
−
C
ompar
ing
the
p
r
opos
e
d
hybr
id
models
with
other
typi
c
a
l
models
:
W
a
r
a
B
E
R
T
-
V1
c
ons
is
tently
outper
f
or
ms
both
W
L
T
a
nd
a
r
a
B
E
R
T
methods
a
c
r
os
s
a
ll
AFND.
−
AFND
-
V1,
whic
h
r
e
pr
e
s
e
nts
the
or
igi
na
l
da
ta
wit
hout
r
e
movi
ng
s
topwor
ds
a
nd
tanw
e
e
n
mar
ks
,
a
c
h
ieve
s
the
highes
t
a
c
c
ur
a
c
y
in
mos
t
c
a
s
e
s
.
−
AFND
-
V2,
in
whic
h
the
s
topwor
ds
we
r
e
r
e
mov
e
d,
s
hows
a
s
li
ght
de
c
r
e
a
s
e
in
a
c
c
ur
a
c
y
c
ompar
e
d
to
AFND
-
V1.
−
AFND
-
V3,
in
whic
h
both
s
topwor
ds
a
nd
tanw
e
e
n
mar
ks
we
r
e
r
e
moved
,
e
xhibi
ts
the
lowe
s
t
a
c
c
ur
a
c
ies
,
indi
c
a
ti
ng
the
potential
los
s
of
im
po
r
tant
s
e
mantic
inf
or
mation.
−
R
e
moval
of
s
topwor
ds
a
nd
tanw
e
e
n
mar
ks
ge
ne
r
a
ll
y
lea
ds
to
a
de
c
r
e
a
s
e
in
a
c
c
ur
a
c
y
a
c
r
os
s
a
ll
models
,
highl
ight
ing
the
im
por
tanc
e
of
pr
e
s
e
r
ving
s
topwor
ds
a
nd
tanw
e
e
n
mar
ks
.
6.
HYP
E
RP
AR
AM
E
T
E
RS
F
I
NE
T
UN
I
NG
De
pe
nding
on
the
r
e
s
ult
s
of
the
pr
e
vious
e
xpe
r
im
e
nts
,
us
ing
W
a
r
a
B
E
R
T
tokeniz
a
ti
on
with
B
iL
S
T
M
f
or
the
tr
a
in
ing
pr
oc
e
s
s
a
c
hieve
s
the
be
s
t
a
c
c
ur
a
c
y.
T
he
r
e
f
or
e
,
a
ne
w
model
wa
s
buil
t
to
e
nha
nc
e
the
a
c
c
ur
a
c
y
a
nd
a
void
ove
r
f
it
ti
ng.
C
onc
e
r
ning
the
AFND,
only
punc
tuation
mar
ks
ha
d
be
e
n
r
e
moved
f
or
the
da
ta
s
e
t.
T
he
da
tas
e
t
wa
s
s
pli
t
int
o
75%
f
o
r
tr
a
ini
ng
a
nd
25%
f
or
tes
ti
ng.
T
a
ble
10
s
hows
the
ne
w
hype
r
pa
r
a
m
e
ter
s
.
T
o
a
void
ove
r
f
it
t
ing,
we
inc
r
e
a
s
e
d
the
c
ompl
e
xit
y
of
the
model
(
incr
e
a
s
ing
the
ne
ur
ons
o
f
the
lay
e
r
s
)
a
nd
incr
e
a
s
e
d
the
dr
opout
r
a
te
f
or
e
a
c
h
laye
r
.
T
he
voc
a
bular
y
s
ize
us
e
d
is
214,
000
(
150,
000
f
or
W
L
T
a
n
d
64,
000
f
or
a
r
a
B
E
R
T
)
.
T
he
ba
tch
s
ize
of
280
is
s
e
lec
ted.
Ac
c
or
ding
to
the
e
xpe
r
im
e
nts
,
the
ba
tch
s
ize
of
2
80
is
the
be
s
t
s
ince
the
ba
tch
s
ize
of
144
incr
e
a
s
e
s
the
num
be
r
of
it
e
r
a
ti
ons
f
or
e
ve
r
y
e
poc
h,
s
lowing
down
th
e
tr
a
ini
ng
M
ode
l
V
oc
a
bul
a
r
y s
iz
e
M
a
x l
e
ngt
h
T
r
a
in
a
bl
e
pa
r
a
me
te
r
s
T
e
s
t
a
c
c
ur
a
c
y
A
F
N
D
-
V1
A
F
N
D
-
V2
A
F
N
D
-
V3
W
L
T
176,000
160
70,980,000
91.09
91.02
88.76
a
r
a
B
E
R
T
-
V1
64,000
160
26,180,000
86.51
86.69
83.3
a
r
a
B
E
R
T
-
V2
64,000
160
26,180,000
85.39
84.48
81.45
W
a
r
a
B
E
R
T
-
V1
176,000 +
64,000
160 +
160
96,580,000
92.67
92.37
89.91
W
a
r
a
B
E
R
T
-
V2
176,000 +
64,000
160 +
160
96,580,000
92.14
92.09
89.42
T
oke
nT
F
I
D
F
176,000 +
1,000
160 +
160
71,380,000
91.08
90.93
88.17
A
r
a
T
F
I
D
F
-
V1
64,000 +
1,000
160 +
160
26,580,000
87.27
86.43
84.22
A
r
a
T
F
I
D
F
-
V2
64,000 +
1,000
160 +
160
26,580,000
84.21
83.18
82.37
W
L
T
176,000
320
70,980,000
90.69
89.47
88.77
a
r
a
B
E
R
T
-
V1
64,000
320
26,180,000
84.52
86.75
83.72
a
r
a
B
E
R
T
-
V2
64,000
320
26,180,000
84.30
85.52
82.65
M
ode
l
V
oc
a
bul
a
r
y s
iz
e
M
a
x l
e
ngt
h
T
r
a
in
a
bl
e
pa
r
a
me
te
r
s
T
e
s
t
a
c
c
ur
a
c
y
A
F
N
D
-
V1
A
F
N
D
-
V2
A
F
N
D
-
V3
W
L
T
176,000
160
38,970,000
91.02
90.23
88.71
a
r
a
B
E
R
T
-
V1
64,000
160
14,330,000
85.12
84.74
84.09
a
r
a
B
E
R
T
-
V2
64,000
160
14,330,000
85.37
84.87
81.26
W
a
r
a
B
E
R
T
-
V1
176,000 +
64,000
160 +
160
53,050,000
91.96
91.64
89.46
W
a
r
a
B
E
R
T
-
V2
176,000 +
64,000
160 +
160
53,050,000
91.74
91.24
89.03
T
oke
nT
F
I
D
F
176,000 +
1,000
160 +
160
39,190,000
91.55
91.22
88.74
A
r
a
T
F
I
D
F
-
V1
64,000 +
1,000
160 +
160
14,550,000
87.89
88.43
84.52
A
r
a
T
F
I
D
F
-
V2
64,000 +
1,000
160 +
160
14,550,000
87.83
87.63
82.66
W
L
T
176,000
320
38,970,000
90.78
90.54
88.22
a
r
a
B
E
R
T
-
V1
64,000
320
14,330,000
87.24
87.46
84.41
a
r
a
B
E
R
T
-
V2
64,000
320
14,330,000
86.1
86.43
82.83
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2088
-
8708
I
nt
J
E
lec
&
C
omp
E
ng
,
Vol
.
15
,
No.
1
,
F
e
br
ua
r
y
20
25
:
836
-
845
842
pr
oc
e
s
s
,
while
the
ba
tch
s
ize
of
444
ne
e
ds
mo
r
e
memor
y.
A
lea
r
ning
r
a
te
o
f
0
.
0001
wa
s
us
e
d
f
or
tr
a
ini
ng,
whic
h
ne
e
ds
a
lowe
r
lea
r
ning
r
a
te
f
or
f
ine
-
tuni
ng.
P
r
ove
n
by
the
e
xpe
r
i
ments
,
e
a
r
ly
s
toppi
ng
a
t
the
pa
ti
e
nc
e
of
1
is
be
ne
f
icia
l
f
o
r
a
voidi
ng
ove
r
f
it
t
ing
a
nd
p
r
ovi
ding
a
c
ha
nc
e
f
or
the
model
to
e
nha
nc
e
.
T
he
s
a
me
pr
oc
e
dur
e
s
us
e
d
in
the
AFND
we
r
e
im
pleme
nted
on
the
Ar
a
NE
W
S
da
tas
e
t,
W
a
r
a
B
E
R
T
-
V2
wa
s
us
e
d
f
or
the
f
e
a
tur
e
e
xtr
a
c
ti
on,
the
B
iL
S
T
M
model
wa
s
us
e
d
f
or
the
tr
a
ini
ng
pr
oc
e
s
s
,
a
nd
only
punc
tuation
mar
ks
we
r
e
r
e
moved
f
r
om
the
da
ta.
T
he
da
tas
e
t
wa
s
s
pli
t
int
o
80%
f
or
tr
a
i
ning
a
nd
20%
f
or
tes
ti
ng.
S
e
lec
ted
hype
r
pa
r
a
mete
r
s
a
r
e
s
hown
in
T
a
ble
11.
T
he
voc
a
bular
y
s
ize
us
e
d
is
224,
000
(
160,
000
f
or
W
L
T
a
nd
64,
000
f
or
a
r
a
B
E
R
T
-
V2)
.
A
ba
tch
s
ize
o
f
70
is
s
e
lec
ted.
A
lea
r
ning
r
a
te
o
f
0.
0001
wa
s
us
e
d
f
or
tr
a
ini
ng,
whic
h
ne
e
ds
a
lowe
r
lea
r
nin
g
r
a
te
f
or
f
ine
-
tuni
ng.
As
pr
ove
n
by
the
e
xpe
r
im
e
n
ts
,
e
a
r
ly
s
toppi
ng
a
t
the
pa
ti
e
nc
e
of
1
is
be
ne
f
icia
l
f
or
a
v
oidi
ng
ove
r
f
it
ti
ng
.
C
ompar
ing
the
r
e
s
ult
s
of
the
pr
opos
e
d
model
with
thos
e
a
c
hieve
d
by
other
r
e
s
e
a
r
c
he
r
s
,
W
a
r
a
B
E
R
T
-
V2
e
nha
nc
e
d
the
tes
t
a
c
c
ur
a
c
y
by
1.
25%
,
i
t
a
c
hieve
d
81.
25%
.
T
a
ble
10
.
Bi
-
L
S
T
M
T
une
d
hype
r
pa
r
a
mete
r
s
f
o
r
A
F
ND
da
tas
e
t
L
a
ye
r
P
a
r
a
me
te
r
s
E
mbe
ddi
ng l
a
ye
r
I
nput
di
me
ns
io
n
E
mbe
ddi
ng
ve
c
to
r
l
e
ngt
h
214,000
768
B
iL
S
T
M
la
ye
r
N
e
ur
ons
D
r
opout
A
c
ti
va
ti
on f
unc
ti
on
512
0.6
R
e
L
U
O
th
e
r
hi
dde
n
la
ye
r
s
L
a
ye
r
1
N
e
ur
ons
D
r
opout
A
c
ti
va
ti
on f
unc
ti
on
512
0.6
R
e
L
U
L
a
ye
r
2
N
e
ur
ons
D
r
opout
A
c
ti
va
ti
on f
unc
ti
on
512
0.6
R
e
L
U
L
a
ye
r
3
N
e
ur
ons
D
r
opout
A
c
ti
va
ti
on f
unc
ti
on
256
0.6
R
e
L
U
L
a
ye
r
4
N
e
ur
ons
D
r
opout
A
c
ti
va
ti
on f
unc
ti
on
256
0.5
R
e
L
U
O
ut
put
la
ye
r
N
e
ur
ons
A
c
ti
va
ti
on f
unc
ti
on
1
S
ig
moi
d
T
r
a
in
a
bl
e
pa
r
a
me
te
r
s
170,583,000
T
e
s
t
lo
s
s
0.16
T
e
s
t
a
c
c
ur
a
c
y
93.83
T
a
ble
11.
B
i
-
L
S
T
M
tuned
hype
r
pa
r
a
mete
r
s
f
or
Ar
a
Ne
ws
da
tas
e
t
L
a
ye
r
P
a
r
a
me
te
r
s
E
mbe
ddi
ng l
a
ye
r
I
nput
di
me
ns
io
n
E
mbe
ddi
ng ve
c
to
r
l
e
ngt
h
224,000
222
B
iL
S
T
M
la
ye
r
N
e
ur
ons
D
r
opout
A
c
ti
va
ti
on f
unc
ti
on
128
0.12
R
e
L
U
O
th
e
r
hi
dde
n l
a
ye
r
s
L
a
ye
r
1
N
e
ur
ons
D
r
opout
A
c
ti
va
ti
on f
unc
ti
on
128
0.17
R
e
L
U
L
a
ye
r
2
N
e
ur
ons
D
r
opout
A
c
ti
va
ti
on f
unc
ti
on
128
0.17
R
e
L
U
L
a
ye
r
3
N
e
ur
ons
D
r
opout
A
c
ti
va
ti
on
f
unc
ti
on
128
0.17
R
e
L
U
L
a
ye
r
4
N
e
ur
ons
D
r
opout
A
c
ti
va
ti
on f
unc
ti
on
128
0.17
R
e
L
U
L
a
ye
r
5
N
e
ur
ons
D
r
opout
A
c
ti
va
ti
on f
unc
ti
on
128
0.17
R
e
L
U
O
ut
put
la
ye
r
N
e
ur
ons
A
c
ti
va
ti
on f
unc
ti
on
1
S
ig
moi
d
T
r
a
in
a
bl
e
pa
r
a
me
te
r
s
50,186,000
T
e
s
t
a
c
c
ur
a
c
y
81.25
7.
CONC
L
USI
ON
I
n
s
umm
a
r
y,
thi
s
pa
pe
r
ha
s
th
r
e
e
pr
im
a
r
y
objec
ti
v
e
s
.
F
ir
s
tl
y,
it
a
im
s
to
de
ve
lop
a
ne
w
hybr
id
f
e
a
tur
e
e
xtr
a
c
ti
on
tec
hnique
that
s
ur
pa
s
s
e
s
e
xis
ti
ng
me
thods
.
S
e
c
ondly,
it
s
e
e
ks
to
inves
ti
ga
te
the
e
f
f
e
c
ts
of
e
li
mi
na
ti
ng
punc
tuation,
s
topwor
ds
,
a
nd
tanw
e
e
n
mar
ks
on
the
de
tec
ti
on
pr
oc
e
s
s
.
L
a
s
tl
y,
it
e
nde
a
vor
s
to
identif
y
the
mos
t
s
uit
a
ble
de
e
p
lea
r
ning
model
R
NN
,
L
S
T
M
,
B
iL
S
T
M
,
or
C
NN
-
B
iL
S
T
M
f
or
e
nha
nc
ing
Ar
a
bic
f
a
ke
ne
ws
de
tec
ti
on
.
T
o
a
c
hieve
thes
e
ob
jec
ti
ve
s
,
the
s
tudy
ut
il
ize
d
two
da
tas
e
ts
.
T
he
Ar
a
bic
f
a
ke
ne
ws
da
tas
e
t
wa
s
e
mpl
oye
d
to
c
ompar
e
the
pr
opo
s
e
d
f
e
a
tur
e
e
xt
r
a
c
ti
on
method
with
c
onve
nti
ona
l
one
s
a
nd
to
a
s
s
e
s
s
the
im
pa
c
t
of
da
ta
c
lea
ning
on
tr
a
ini
ng
.
T
he
A
r
a
Ne
ws
da
tas
e
t
wa
s
us
e
d
to
c
ompar
e
the
model's
pe
r
f
or
manc
e
with
that
o
f
pr
e
vious
r
e
s
e
a
r
c
h.
T
he
inves
ti
ga
ti
on
int
o
the
r
e
moval
of
s
topwor
ds
a
nd
tanw
e
e
n
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
E
lec
&
C
omp
E
ng
I
S
S
N:
2088
-
8708
A
r
abic
fake
ne
w
s
de
tec
ti
on
us
ing
hy
br
id
c
ontex
tual
featur
e
s
(
Hus
s
ain
M
ohamm
e
d
T
ur
k
i
)
843
mar
ks
uti
li
z
e
d
thr
e
e
ve
r
s
ions
of
AFND:
AFND
-
V1
or
igi
na
l
da
ta,
AFND
-
V2
s
topwor
ds
r
e
mo
ve
d,
a
nd
AFND
-
V3
both
s
topwor
ds
a
nd
tanw
e
e
n
mar
ks
r
e
moved.
AFND
-
V1
c
ons
is
tently
yielde
d
the
highes
t
a
c
c
ur
a
c
y,
while
AFND
-
V3
e
xhibi
ted
the
lowe
s
t,
indi
c
a
ti
ng
potential
los
s
of
c
r
uc
ial
s
e
mantic
inf
or
mation.
Ac
r
os
s
a
ll
models
,
r
e
movi
ng
s
topwor
ds
a
nd
tanw
e
e
n
mar
ks
ge
ne
r
a
ll
y
r
e
s
ult
e
d
in
de
c
r
e
a
s
e
d
a
c
c
ur
a
c
y.
S
e
ve
r
a
l
f
e
a
tur
e
e
xtr
a
c
ti
on
methods
we
r
e
pr
opos
e
d
a
nd
e
va
luate
d
a
c
r
os
s
the
thr
e
e
AFND
ve
r
s
ions
us
ing
f
our
DL
models
.
Among
thes
e
methods
,
W
a
r
a
B
E
R
T
-
V1
c
ons
is
tently
outper
f
or
med
other
s
,
including
W
L
T
a
nd
a
r
a
B
E
R
T
,
a
c
r
os
s
a
ll
ve
r
s
ions
of
A
F
ND
.
W
a
r
a
B
E
R
T
-
V2
dis
playe
d
s
im
il
a
r
pa
tt
e
r
ns
to
W
a
r
a
B
E
R
T
-
V1
but
tende
d
to
a
c
hieve
lowe
r
a
c
c
ur
a
c
y.
T
he
r
e
s
ult
s
unde
r
s
c
or
e
d
the
im
por
tanc
e
of
r
e
taining
s
topwo
r
ds
a
nd
tanw
e
e
n
mar
ks
f
or
a
c
c
ur
a
te
f
a
ke
ne
ws
de
tec
ti
on.
AC
KNOWL
E
DGE
M
E
NT
S
T
his
r
e
s
e
a
r
c
h
is
f
unde
d
by
the
de
a
ns
hip
of
s
c
ientif
i
c
r
e
s
e
a
r
c
h
in
Z
a
r
qa
Unive
r
s
it
y,
J
or
da
n.
RE
F
E
RE
NC
E
S
[
1]
S
.
J
.
D
ix
on,
“
N
umbe
r
of
s
oc
ia
l
me
di
a
u
s
e
r
s
w
or
ld
w
id
e
f
r
om
2017
to
2028,”
St
at
is
ta
,
2
022.
ht
tp
s
:/
/ww
w
.s
ta
ti
s
ta
.c
om/
s
ta
ti
s
ti
c
s
/2
78414/num
be
r
-
of
-
w
or
ld
w
id
e
-
s
oc
ia
l
-
ne
twor
k
-
us
e
r
s
/
(
a
c
c
e
s
s
e
d N
ov. 21, 2022)
.
[
2]
E
.
S
he
a
r
e
r
a
nd
E
.
G
r
ie
c
o,
“
A
me
r
ic
a
ns
a
r
e
w
a
r
y
of
th
e
r
ol
e
s
oc
ia
l
me
di
a
s
it
e
s
pl
a
y
in
de
li
ve
r
in
g
th
e
ne
w
s
,”
P
e
w
R
e
s
e
ar
c
h
C
e
nt
e
r
,
vol
. 2, pp. 1
–
23, 2019.
[
3]
M
.
A
lz
youd
e
t
al
.
,
“
D
ia
gno
s
in
g
di
a
be
t
e
s
m
e
ll
it
us
us
in
g
m
a
c
hi
n
e
le
a
r
ni
ng
te
c
hni
que
s
,”
I
nt
e
r
nat
io
nal
J
our
nal
of
D
at
a
and N
e
tw
or
k
Sc
ie
nc
e
, vol
. 8, no. 1, pp. 179
–
188, 2024, doi:
10.5267/j
.i
jd
ns
.2
023.10.006.
[
4]
M
.
H
a
j
Q
a
s
e
m,
M
.
A
lj
a
id
i,
G
.
S
a
ma
r
a
,
R
.
A
la
z
a
id
a
h,
A
.
A
ls
a
r
ha
n,
a
nd
M
.
A
ls
ha
mm
a
r
i,
“
A
n
in
te
ll
ig
e
nt
de
c
is
io
n
s
uppor
t
s
ys
te
m
ba
s
e
d
on
mul
ti
-
a
ge
nt
s
y
s
te
ms
f
or
bus
in
e
s
s
c
la
s
s
if
ic
a
ti
on
pr
obl
e
m,”
Sus
ta
in
abi
li
ty
,
vol
.
15,
no.
14,
pp.
1
–
14,
J
ul
.
2023,
doi
:
10.3390/s
u151410977.
[
5]
R
.
A
la
z
a
id
a
h,
G
.
S
a
ma
r
a
,
S
.
A
lm
a
ta
r
ne
h,
M
.
H
a
s
s
a
n,
M
.
A
lj
a
id
i,
a
nd
H
.
M
a
ns
ur
,
“
M
ul
ti
-
la
be
l
c
la
s
s
if
ic
a
ti
on
ba
s
e
d
on
a
s
s
oc
ia
ti
on
s
,”
A
ppl
ie
d Sc
ie
nc
e
s
, vol
. 13, no. 8, pp. 1
–
16, Apr
. 2
023, doi:
10.3390/app130850
81.
[
6]
A
. D
e
nt
on, “
F
a
ke
ne
w
s
:
th
e
l
e
ga
li
ty
of
t
he
R
us
s
ia
n 2016 F
a
c
e
b
ook inf
lu
e
nc
e
c
a
mpa
ig
n,”
B
U
I
nt
’
l
L
J
, vol
. 37, pp. 183
–
210, 2019.
[
7]
D
.
M
.
J
.
L
a
z
e
r
e
t
al
.
,
“
T
he
s
c
ie
nc
e
of
f
a
ke
ne
w
s
,”
Sc
ie
nc
e
,
vol
.
359,
no.
6380,
pp.
1094
–
1096,
M
a
r
.
2018,
doi
:
10.1126/s
c
ie
nc
e
.a
a
o2998.
[
8]
A
.
K
ha
li
l,
M
.
J
a
r
r
a
h,
M
.
A
ld
w
a
ir
i,
a
nd
M
.
J
a
r
a
da
t,
“
A
F
N
D
:
A
r
a
bi
c
f
a
ke
ne
w
s
da
ta
s
e
t
f
or
th
e
de
te
c
ti
on
a
nd
c
la
s
s
if
ic
a
ti
on
of
a
r
ti
c
le
s
c
r
e
di
bi
li
ty
,”
D
at
a i
n B
r
ie
f
, vol
. 42, pp. 1
–
7, J
un. 2022,
doi
:
10.1016/j
.di
b.2022.108141.
[
9]
H
.
S
a
a
da
ny,
E
. M
oha
me
d,
a
nd C
.
O
r
a
s
a
n,
“
F
a
k
e
or
r
e
a
l?
a
s
tu
d
y
of
A
r
a
bi
c
s
a
ti
r
ic
a
l
f
a
ke
n
e
w
s
,
”
ar
X
iv
pr
e
p
r
in
t
ar
X
iv
:
2011.00
452
,
pp. 1
–
11, 2020.
[
10]
E
.
M
.
B
.
N
a
goudi,
A
.
E
lm
a
da
ny,
M
.
A
bdul
-
M
a
ge
e
d, T
.
A
lh
in
di
,
a
nd
H
.
C
a
vus
ogl
u,
“
M
a
c
hi
ne
ge
ne
r
a
ti
on
a
nd
de
t
e
c
ti
on
of
A
r
a
bi
c
ma
ni
pul
a
te
d a
nd f
a
ke
ne
w
s
,
”
ar
X
iv
pr
e
pr
in
t
ar
X
iv
:
2011.03092
,
pp. 1
–
15, 2020.
[
11]
M
.
A
lk
ha
ir
,
K
.
M
e
f
to
uh,
K
.
S
ma
ïl
i,
a
nd
N
.
O
th
ma
n,
“
A
n
A
r
a
bi
c
c
or
pus
of
f
a
ke
ne
w
s
:
c
ol
le
c
ti
on,
a
na
ly
s
is
a
nd
c
la
s
s
if
ic
a
ti
on
,”
i
n
A
r
abi
c
L
anguage
P
r
oc
e
s
s
in
g:
F
r
om
T
he
or
y
to
P
r
ac
ti
c
e
:
7t
h
I
n
te
r
nat
io
nal
C
onf
e
r
e
nc
e
,
I
C
A
L
P
2019,
N
anc
y
,
F
r
anc
e
,
O
c
to
be
r
16
–
17, 2019, P
r
oc
e
e
di
ngs
7
, 2019, pp. 292
–
302, doi:
10.1007/978
-
3
-
030
-
32959
-
4_21.
[
12]
H
.
G
ha
r
a
ib
e
h
e
t
al
.
,
“
A
r
a
bi
c
s
e
nt
im
e
nt
a
na
ly
s
is
of
M
onk
e
yp
ox
us
in
g
de
e
p
ne
ur
a
l
ne
twor
k
a
nd
opt
im
iz
e
d
hype
r
pa
r
a
me
te
r
s
of
ma
c
hi
ne
l
e
a
r
ni
ng a
lg
or
it
hms
,”
Soc
ia
l
N
e
tw
o
r
k
A
nal
y
s
i
s
and M
in
in
g
, vol
. 14, no. 1, J
a
n. 2024, doi:
10.1007/s
13278
-
023
-
01188
-
4.
[
13]
R
.
A
la
z
a
id
a
h,
G
.
S
a
ma
r
a
,
M
.
A
lj
a
id
i,
M
.
H
a
j
Q
a
s
e
m,
A
.
A
ls
a
r
ha
n,
a
nd
M
.
A
ls
ha
mm
a
r
i,
“
P
ot
e
nt
ia
l
of
ma
c
hi
ne
le
a
r
ni
ng
f
or
pr
e
di
c
ti
ng
s
le
e
p
di
s
or
de
r
s
:
a
c
ompr
e
he
n
s
iv
e
a
na
ly
s
is
of
r
e
g
r
e
s
s
io
n
a
nd
c
l
a
s
s
if
ic
a
ti
on
mod
e
ls
,”
D
ia
gno
s
ti
c
s
,
vol
.
14,
no
.
1,
pp. 1
–
19, De
c
. 2023, doi:
10.3390/di
a
gnos
ti
c
s
14010027.
[
14]
R
.
A
z
a
d,
“
A
nove
l
ta
xonomy
f
or
A
r
a
bi
c
f
a
ke
ne
w
s
da
ta
s
e
t
s
,”
I
nt
e
r
nat
io
nal
J
our
nal
of
C
om
put
in
g
and
D
ig
it
al
Sy
s
te
m
s
,
vol
.
14,
no. 1, pp. 159
–
166, J
ul
. 2023, doi:
10.12785/i
jc
ds
/1
40115.
[
15]
A
.
K
.
J
.
S
a
uda
ga
r
a
nd
H
.
V
.
M
oha
mm
e
d,
“
A
r
a
bi
c
c
h
a
r
a
c
te
r
e
xt
r
a
c
ti
on
a
nd
r
e
c
ogni
ti
on
us
in
g
tr
a
ve
r
s
in
g
a
ppr
oa
c
h,”
T
he
I
nt
e
r
nat
io
nal
A
r
ab J
our
nal
of
I
nf
or
m
at
io
n T
e
c
hnol
ogy
, vol
. 15,
no. 3, pp. 360
–
368, 2018.
[
16]
F
.
H
a
na
nde
h,
M
.
A
kour
,
E
.
A
l
D
a
oud,
R
.
A
l
s
hor
ma
n,
a
nd
I
.
A
ls
ma
di
,
“
K
P
-
tr
ie
a
lg
or
it
hm
f
or
upda
te
a
nd
s
e
a
r
c
h
ope
r
a
ti
ons
,
”
I
nt
e
r
nat
io
nal
A
r
ab J
our
nal
of
I
nf
or
m
at
io
n T
e
c
hnol
ogy
, vol
. 13,
no. 6, pp. 1
–
10, 2016.
[
17]
A
.
B
.
N
a
s
s
if
,
A
.
E
ln
a
ga
r
,
O
.
E
lg
e
ndy,
a
nd
Y
.
A
f
a
da
r
,
“
A
r
a
bi
c
f
a
ke
ne
w
s
de
t
e
c
ti
on
ba
s
e
d
on
d
e
e
p
c
ont
e
xt
ua
li
z
e
d
e
mbe
dd
in
g
mode
ls
,”
N
e
u
r
al
C
om
put
in
g and A
ppl
ic
at
io
ns
, vol
. 34, no. 18, pp. 16019
–
16032, S
e
p. 2022, doi:
10.1007/s
00521
-
022
-
07206
-
4.
[
18]
H
.
H
im
di
,
G
.
W
e
ir
,
F
.
A
s
s
ir
i,
a
nd
H
.
A
l
-
B
a
r
ha
mt
os
hy,
“
A
r
a
bi
c
f
a
ke
ne
w
s
de
te
c
ti
on
ba
s
e
d
on
te
xt
ua
l
a
na
ly
s
is
,”
A
r
abi
an
J
ou
r
nal
fo
r
Sc
ie
nc
e
and E
ngi
ne
e
r
in
g
, vol
. 47, no. 8, pp. 10453
–
10469, Aug. 2022, d
oi
:
10.1007/s
13369
-
021
-
06449
-
y.
[
19]
A
.
A
w
a
ja
n,
M
.
A
la
z
a
b,
R
.
A
.
K
hur
ma
,
R
.
A
ls
a
a
de
h,
a
nd
M
.
W
e
dya
n,
“
F
a
ke
ne
w
s
de
te
c
ti
on
a
nd
pr
e
ve
nt
io
n
us
in
g
a
r
ti
f
i
c
ia
l
in
te
ll
ig
e
nc
e
t
e
c
hni
que
s
:
A
r
e
vi
e
w
of
a
de
c
a
de
of
r
e
s
e
a
r
c
h,”
I
nt
e
r
nat
io
nal
J
our
nal
of
C
om
put
e
r
I
nf
or
m
at
io
n Sy
s
te
m
s
and I
ndus
tr
ia
l
M
anage
m
e
nt
A
ppl
ic
at
io
ns
, vol
. 14, pp. 326
–
337, 2022.
[
20]
S
.
E
.
S
or
our
a
nd
H
.
E
.
A
bde
lk
a
d
e
r
,
“
A
F
N
D
:
A
r
a
bi
c
f
a
ke
ne
w
s
de
te
c
ti
on
w
it
h
a
n
e
n
s
e
mbl
e
de
e
p
C
N
N
-
L
S
T
M
mode
l,
”
J
ou
r
na
l
of
T
he
or
e
ti
c
al
and A
ppl
ie
d I
nf
or
m
at
io
n T
e
c
hnol
ogy
, vol
. 100, no. 14, pp. 5072
–
5086, 2022.
[
21]
W
.
S
hi
s
ha
h,
“
J
oi
nt
B
e
r
t
f
or
de
te
c
ti
ng
A
r
a
bi
c
f
a
ke
ne
w
s
,”
I
E
E
E
A
c
c
e
s
s
,
vol
.
10,
pp.
71951
–
71960,
2022,
doi
:
10.1109/AC
C
E
S
S
.2022.3185083.
[
22]
H
.
R
a
ha
b,
A
.
Z
it
ouni
,
a
nd
M
.
D
jo
udi
,
“
A
r
a
bi
c
f
a
ke
ne
w
s
a
n
d
s
pa
m
ha
ndl
in
g:
M
e
th
od
s
,
r
e
s
our
c
e
s
a
nd
oppor
tu
ni
ti
e
s
,”
in
2021
I
nt
e
r
nat
io
nal
C
onf
e
r
e
nc
e
on
A
r
ti
fi
c
ia
l
I
nt
e
ll
ig
e
nc
e
fo
r
C
y
be
r
Se
c
ur
it
y
Sy
s
te
m
s
and
P
r
iv
ac
y
(
A
I
-
C
SP
)
,
N
ov.
2021,
pp.
1
–
7,
doi
:
10.1109/AI
-
C
S
P
52968.2021.9671174.
[
23]
F
.
L
.
A
lo
ta
ib
i
a
nd
M
.
M
.
A
lh
a
mm
a
d,
“
U
s
in
g
a
r
ul
e
-
ba
s
e
d
mode
l
to
de
te
c
t
A
r
a
bi
c
f
a
k
e
ne
w
s
pr
opa
g
a
ti
on
dur
in
g
C
ovi
d
-
19,”
I
nt
e
r
nat
io
nal
J
our
nal
of
A
dv
anc
e
d
C
om
put
e
r
Sc
ie
nc
e
a
nd
A
ppl
ic
at
io
ns
,
vol
.
13,
no.
1,
pp.
112
–
119,
2022,
doi
:
10.14569/I
J
A
C
S
A
.2022.0130114.
[
24]
M
.
A
bd
E
l
a
z
i
z
,
A
.
D
a
h
ou
,
D
.
A
. O
r
a
bi
, S
.
A
l
s
h
a
th
r
i,
E
.
M
. S
o
li
ma
n
,
a
n
d
A
.
A
.
E
w
e
e
s
,
“
A
h
yb
r
i
d m
ul
t
it
a
s
k
l
e
a
r
n
in
g f
r
a
m
e
w
or
k
w
i
t
h
a
f
ir
e
h
a
w
k
o
pt
im
i
z
e
r
f
or
A
r
a
bi
c
f
a
k
e
n
e
w
s
d
e
t
e
c
ti
on
,
”
M
at
h
e
m
at
i
c
s
,
v
ol
.
11
,
no
.
2,
p
p.
1
–
1
5,
J
a
n
.
20
23
,
doi
:
10
.3
39
0/
ma
th
11
02
02
58.
[
25]
M
. S
. H
a
dj
A
me
ur
a
nd H
.
A
li
a
ne
, “
A
r
a
C
O
V
I
D
19
-
M
F
H
:
A
r
a
bi
c
C
ovi
d
-
19 mul
ti
-
la
be
l
f
a
ke
ne
w
s
& ha
te
s
pe
e
c
h de
te
c
ti
on da
ta
s
e
t,
”
P
r
oc
e
di
a C
om
put
e
r
S
c
ie
nc
e
, vol
. 189, pp. 232
–
241, 2021, doi:
10.1016/j
.pr
oc
s
.2021.05.086.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2088
-
8708
I
nt
J
E
lec
&
C
omp
E
ng
,
Vol
.
15
,
No.
1
,
F
e
br
ua
r
y
20
25
:
836
-
845
844
[
26]
A
. R
. M
a
hl
ous
a
nd A
. A
l
-
L
a
it
h, “
F
a
ke
ne
w
s
de
te
c
ti
on i
n A
r
a
bi
c
t
w
e
e
ts
dur
in
g t
he
C
O
V
I
D
-
19 pa
nde
mi
c
,”
I
nt
e
r
nat
io
nal
J
ou
r
n
al
of
A
dv
anc
e
d C
om
put
e
r
S
c
ie
nc
e
and A
ppl
ic
at
io
ns
, vol
. 12, no. 6, p
p. 778
–
788, 2021, doi:
10.14569/I
J
A
C
S
A
.2021.0120691.
[
27]
V
.
T
h
a
na
s
e
ka
r
a
n
a
nd
A
.
S
e
l
va
r
a
j,
“
L
o
w
di
m
e
n
s
i
on
a
l
mul
ti
c
la
s
s
s
te
g
a
n
a
l
y
s
i
s
of
s
pa
ti
a
l
L
S
B
b
a
s
e
d
s
t
e
g
o
im
a
ge
s
u
s
in
g
t
e
x
tu
r
a
l
f
e
a
t
ur
e
s
,
”
T
h
e
I
nt
e
r
n
at
io
na
l
A
r
ab
J
ou
r
na
l
of
I
n
fo
r
m
a
ti
o
n
T
e
c
hn
ol
o
g
y
,
v
ol
.
21
,
no
.
2,
p
p.
2
33
–
2
42
,
20
24
,
do
i:
10
.3
40
28
/i
a
j
it
/2
1/
2
/6
.
[
28]
M
.
A
lq
a
r
ni
,
“
E
mbe
ddi
ng
s
e
a
r
c
h
f
or
Q
ur
a
ni
c
te
xt
s
ba
s
e
d
o
n
la
r
ge
la
ngua
ge
mode
ls
,”
T
he
I
nt
e
r
nat
io
nal
A
r
ab
J
our
nal
of
I
nf
or
m
at
io
n T
e
c
hnol
ogy
, vol
. 21, no. 2, pp. 243
–
256, 2024, doi:
10.34028/i
a
ji
t/
21/
2/
7.
[
29]
H
.
A
bu
O
w
id
a
e
t
a
l.
,
“
T
he
pe
r
f
o
r
m
a
n
c
e
of
a
r
t
if
i
c
i
a
l
in
t
e
ll
i
g
e
n
c
e
in
pr
o
s
ta
t
e
ma
gn
e
ti
c
r
e
s
on
a
nc
e
i
ma
gi
ng
s
c
r
e
e
ni
ng
,”
I
nt
e
r
nat
io
n
al
J
o
u
r
n
al
of
E
l
e
c
t
r
ic
al
a
nd
C
o
m
p
ut
e
r
E
ng
in
e
e
r
i
ng
,
vol
.
14
,
no
.
2,
p
p
.
2
23
4
-
2
24
1
,
A
pr
.
2
02
4,
d
oi
:
10
.1
15
91
/i
j
e
c
e
.
v1
4i
2.
pp
22
34
-
22
41.
[
30]
A
. A
l
S
ha
r
a
h, H
. A
bu
O
w
id
a
, F
. A
ln
a
im
a
t,
a
nd
S
. A
buowa
id
a
,
“
A
ppl
ic
a
ti
on of
ma
c
hi
ne
l
e
a
r
ni
ng i
n c
he
mi
c
a
l
e
ngi
ne
e
r
in
g:
ou
tl
ook
a
nd
pe
r
s
pe
c
ti
ve
s
,
”
I
A
E
S
I
nt
e
r
nat
io
nal
J
our
nal
of
A
r
ti
fi
c
ia
l
I
nt
e
ll
ig
e
nc
e
,
vol
.
13,
no.
1,
p
p
.
619
-
630
,
M
a
r
.
2024,
doi
:
10.11591/i
ja
i.
v13.i1.pp619
-
630.
[
31]
O
.
T
a
r
a
w
n
e
h
e
t
al
.
,
“
T
he
e
f
f
e
c
t
of
pr
e
-
pr
oc
e
s
s
in
g
on
a
c
onvo
lu
ti
ona
l
ne
ur
a
l
ne
twor
k
mode
l
f
or
dor
s
a
l
ha
nd
ve
in
r
e
c
ogni
ti
o
n,”
I
nt
e
r
nat
io
nal
J
our
nal
of
A
dv
anc
e
d
C
om
put
e
r
Sc
ie
nc
e
an
d
A
ppl
ic
at
io
ns
,
vol
.
15,
no.
3,
pp.
1284
–
1289,
2024,
doi
:
10.14569/I
J
A
C
S
A
.2024.01503126.
[
32]
S
.
A
buowa
id
a
e
t
al
.
,
“
P
r
opos
e
d
e
nha
n
c
e
d
f
e
a
tu
r
e
e
xt
r
a
c
ti
on
f
or
mul
ti
-
f
ood
de
te
c
ti
on
me
th
od,”
J
our
nal
of
T
he
or
e
ti
c
al
and
A
pp
li
e
d
I
nf
or
m
at
io
n T
e
c
hnol
ogy
, vol
. 101, no. 24, pp. 8140
–
8146, 2023.
B
I
OG
RA
P
HI
E
S
OF
AU
T
HO
RS
Hus
s
a
i
n
M
o
h
a
m
m
ed
Turki
earn
e
d
h
i
s
b
ac
h
el
o
r's
d
eg
ree
i
n
co
m
p
u
t
er
s
c
i
en
ce
fro
m
t
h
e
U
n
i
v
er
s
i
t
y
o
f
A
n
b
ar
i
n
Iraq
a
n
d
h
i
s
MSc
i
n
c
o
mp
u
t
er
s
ci
e
n
ce
fro
m
Z
arq
a
U
n
i
v
er
s
i
t
y
i
n
J
o
rd
a
n
.
H
i
s
s
p
eci
a
l
i
za
t
i
o
n
s
are
art
i
f
i
ci
a
l
i
n
t
e
l
l
i
g
e
n
ce,
d
eep
l
ear
n
i
n
g
,
an
d
mach
i
n
e
l
ear
n
i
n
g
.
H
e
can
b
e
co
n
t
ac
t
ed
v
i
a
ema
i
l
at
2
0
2
1
9
1
0
9
@
zu
.
ed
u
.
j
o
.
Es
s
a
m
A
l
D
a
o
ud
earn
ed
h
i
s
P
h
.
D
.
i
n
co
m
p
u
t
er
s
c
i
en
ce
fr
o
m
U
n
i
v
er
s
i
t
i
P
u
t
ra
Mal
ay
s
i
a.
A
s
a
p
ro
fes
s
o
r
at
Z
arq
a
U
n
i
v
ers
i
t
y
,
h
e
ex
cel
s
i
n
res
earch
,
cu
rri
cu
l
u
m
d
e
v
el
o
p
me
n
t
,
an
d
t
eac
h
i
n
g
,
w
i
t
h
ex
p
ert
i
s
e
i
n
d
a
t
a
s
ci
e
n
ce,
cry
p
t
o
g
r
a
p
h
y
,
an
d
b
i
o
i
n
fo
rma
t
i
c
s
.
H
e
h
as
al
s
o
t
au
g
h
t
at
i
n
s
t
i
t
u
t
i
o
n
s
s
u
c
h
as
Mu
l
t
i
med
i
a
U
n
i
v
er
s
i
t
y
i
n
Mal
ay
s
i
a.
D
ri
v
e
n
b
y
a
p
as
s
i
o
n
fo
r
acad
emi
c
e
x
cel
l
en
ce,
h
e
c
o
n
t
i
n
u
e
s
t
o
co
n
t
r
i
b
u
t
e
s
i
g
n
i
f
i
can
t
l
y
t
o
h
i
s
fi
e
l
d
t
h
r
o
u
g
h
re
s
earch
,
t
each
i
n
g
,
an
d
l
ea
d
ers
h
i
p
ro
l
es
.
H
e
ca
n
b
e
co
n
t
ac
t
ed
v
i
a
e
mai
l
at
e
s
s
am
d
z@
zu
.
ed
u
.
j
o
.
Gha
s
s
a
n
S
a
m
a
ra
h
o
l
d
s
BSc.
a
n
d
MSc.
i
n
c
o
m
p
u
t
er
s
c
i
en
ce
a
n
d
P
h
.
D
.
in
co
mp
u
t
er
n
e
t
w
o
rk
s
.
H
e
o
b
t
a
i
n
e
d
h
i
s
Ph
.
D
.
fro
m
U
n
i
v
ers
i
t
i
Sai
n
s
Ma
l
ay
s
i
a
(U
SM)
i
n
2
0
1
2
.
H
i
s
s
p
eci
a
l
i
za
t
i
o
n
s
are
t
h
e
i
n
t
ern
e
t
o
f
t
h
i
n
g
s
,
cry
p
t
o
g
rap
h
y
,
v
e
h
i
c
u
l
ar
ad
h
o
c
n
e
t
w
o
rk
s
,
w
i
re
l
es
s
s
e
n
s
o
r
n
e
t
w
o
rk
s
,
i
n
t
ern
e
t
acce
s
s
p
ro
t
o
c
o
l
s
,
acci
d
en
t
p
re
v
en
t
i
o
n
,
al
arm
s
y
s
t
em
s
,
cl
o
u
d
co
mp
u
t
i
n
g
,
co
mp
u
t
at
i
o
n
al
co
m
p
l
e
x
i
t
y
,
co
mp
u
t
er
cri
me,
co
mp
u
t
er
n
e
t
w
o
rk
s
ec
u
ri
t
y
,
d
at
a
i
n
t
eg
r
i
t
y
,
d
at
a
p
ri
v
ac
y
,
d
el
ay
s
,
d
i
g
i
t
al
fo
ren
s
i
c
s
,
d
i
rect
e
d
g
rap
h
s
,
en
erg
y
co
n
s
er
v
at
i
o
n
,
en
erg
y
man
ag
eme
n
t
s
y
s
t
em
s
,
fo
g
co
m
p
u
t
i
n
g
,
g
rap
h
t
h
e
o
ry
,
h
o
me
au
t
o
m
at
i
o
n
,
i
n
t
e
l
l
i
g
e
n
t
r
o
b
o
t
s
,
an
d
i
n
t
el
l
i
g
en
t
t
ran
s
p
o
rt
a
t
i
o
n
s
y
s
t
ems
.
Cu
rren
t
l
y
,
D
r.
Samara
i
s
an
as
s
o
ci
a
t
e
p
ro
fe
s
s
o
r
an
d
v
i
ce
d
ean
o
f
t
h
e
Co
l
l
e
g
e
o
f
In
f
o
rmat
i
o
n
T
ech
n
o
l
o
g
y
a
t
Z
arq
a
U
n
i
v
er
s
i
t
y
,
J
o
r
d
an
.
H
e
can
b
e
co
n
t
act
e
d
v
i
a
ema
i
l
at
g
s
amara@
zu
.
e
d
u
.
j
o
.
R
a
ed
A
l
a
z
a
i
d
a
h
recei
v
ed
t
h
e
B.
Sc.
d
eg
rees
i
n
co
mp
u
t
er
i
n
f
o
rmat
i
o
n
s
y
s
t
em,
M.
Sc.
d
eg
rees
i
n
c
o
mp
u
t
er
s
c
i
en
ce
fr
o
m
A
l
al
-
Ba
y
t
U
n
i
v
er
s
i
t
y
,
J
o
r
d
an
,
i
n
2
0
1
2
an
d
2
0
1
5
,
res
p
ec
t
i
v
el
y
,
an
d
t
h
e
Ph
.
D
.
d
eg
r
ee
i
n
co
mp
u
t
er
s
c
i
e
n
ce
fro
m
U
n
i
v
ers
i
t
i
Sai
n
s
Mal
ay
s
i
a,
Mal
ay
s
i
a,
i
n
2
0
2
3
.
D
r.
A
b
u
o
w
ai
d
a
i
s
cu
rre
n
t
l
y
w
o
r
k
i
n
g
a
s
an
a
s
s
i
s
t
an
t
p
r
o
fes
s
o
r
i
n
t
h
e
Co
mp
u
t
er
Sci
en
ce
D
e
p
art
me
n
t
at
Z
arq
a
U
n
i
v
ers
i
t
y
,
J
o
rd
a
n
.
H
er
res
earch
i
n
t
ere
s
t
s
i
n
c
l
u
d
e
d
eep
l
ear
n
i
n
g
,
d
e
p
t
h
es
t
i
mat
i
o
n
,
p
o
i
n
t
c
l
o
u
d
p
r
o
ces
s
i
n
g
,
an
d
co
m
p
u
t
er
v
i
s
i
o
n
.
He
can
b
e
co
n
t
act
e
d
v
i
a
ema
i
l
at
raza
i
d
a
h
@
zu
.
ed
u
.
j
o
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
E
lec
&
C
omp
E
ng
I
S
S
N:
2088
-
8708
A
r
abic
fake
ne
w
s
de
tec
ti
on
us
ing
hy
br
id
c
ontex
tual
featur
e
s
(
Hus
s
ain
M
ohamm
e
d
T
ur
k
i
)
845
M
a
i
s
H
a
j
Q
a
s
e
m
recei
v
ed
h
i
s
b
ach
e
l
o
r’
s
d
eg
ree
i
n
co
mp
u
t
er
s
c
i
en
ce
fr
o
m
t
h
e
U
n
i
v
er
s
i
t
y
o
f
J
o
r
d
an
,
J
o
r
d
an
,
i
n
2
0
1
1
;
t
h
e
Mas
t
er’
s
d
eg
ree
i
n
co
m
p
u
t
er
s
ci
en
c
e
fro
m
t
h
e
U
n
i
v
er
s
i
t
y
o
f
J
o
rd
a
n
i
n
2
0
1
6
;
an
d
h
i
s
P
h
.
D
.
d
eg
re
e
i
n
co
mp
u
t
er
s
ci
e
n
ce
an
d
art
i
fi
c
i
a
l
i
n
t
el
l
i
g
en
ce
fro
m
t
h
e
U
n
i
v
er
s
i
t
y
o
f
J
o
r
d
an
i
n
2
0
2
0
.
Cu
rren
t
l
y
,
h
e
w
o
rk
s
as
an
as
s
i
s
t
a
n
t
p
ro
fe
s
s
o
r
i
n
t
h
e
D
a
t
a
Sci
e
n
ce
an
d
A
rt
i
fi
c
i
al
I
n
t
e
l
l
i
g
e
n
ce
D
ep
ar
t
men
t
at
Z
arq
a
U
n
i
v
ers
i
t
y
,
J
o
r
d
an
.
H
i
s
res
earc
h
i
n
t
eres
t
s
i
n
cl
u
d
e
mac
h
i
n
e
l
earn
i
n
g
an
d
mu
l
t
i
-
a
g
en
t
s
y
s
t
em
s
.
Sh
e
can
b
e
co
n
t
act
e
d
v
i
a
ema
i
l
at
m.
h
a
j
q
a
s
em@
zu
.
ed
u
.
j
o
.
M
o
ha
m
m
a
d
A
l
j
a
i
di
recei
v
ed
h
i
s
B.
Sc.
(w
i
t
h
d
i
s
t
i
n
ct
i
o
n
)
i
n
co
m
p
u
t
er
s
c
i
en
ce
fro
m
Z
arq
a
U
n
i
v
er
s
i
t
y
,
J
o
rd
a
n
,
i
n
2
0
1
4
;
t
h
e
M.
Sc.
(w
i
t
h
d
i
s
t
i
n
ct
i
o
n
)
i
n
c
o
mp
u
t
er
s
c
i
e
n
ce
fro
m
Z
arq
a
U
n
i
v
ers
i
t
y
,
J
o
r
d
an
,
i
n
2
0
1
7
;
an
d
t
h
e
Ph
.
D
.
d
e
g
r
ee
i
n
co
m
p
u
t
er
s
ci
e
n
ce
an
d
art
i
fi
c
i
al
i
n
t
el
l
i
g
en
ce
fro
m
t
h
e
U
n
i
v
er
s
i
t
y
o
f
N
o
r
t
h
u
mb
r
i
a
at
N
ew
cas
t
l
e,
U
K
,
i
n
2
0
2
2
.
D
r.
A
l
j
ai
d
i
i
s
cu
rren
t
l
y
w
o
r
k
i
n
g
as
an
a
s
s
i
s
t
an
t
p
ro
fe
s
s
o
r
w
i
t
h
t
h
e
Co
mp
u
t
er
Sci
e
n
ce
D
ep
ar
t
men
t
,
Z
arq
a
U
n
i
v
er
s
i
t
y
,
J
o
r
d
a
n
.
H
i
s
re
s
earch
i
n
t
eres
t
s
i
n
cl
u
d
e
b
u
t
are
n
o
t
l
i
mi
t
ed
t
o
,
E
V
s
ch
ar
g
i
n
g
man
ag
eme
n
t
a
n
d
d
ev
e
l
o
p
men
t
,
w
i
re
l
es
s
s
e
n
s
o
r
n
e
t
w
o
rk
s
(W
SN
s
),
A
I,
cy
b
er
s
ecu
r
i
t
y
,
o
p
t
i
mi
zat
i
o
n
,
rei
n
f
o
rcemen
t
l
earn
i
n
g
.
H
e
can
b
e
co
n
t
act
ed
at
emai
l
:
razai
d
ah
@
z
u
.
ed
u
.
j
o
.
Suha
i
l
a
A
buo
w
a
i
da
recei
v
e
d
t
h
e
B.
Sc.
d
e
g
rees
i
n
c
o
mp
u
t
er
i
n
f
o
rmat
i
o
n
s
y
s
t
em,
M.
Sc.
d
eg
rees
i
n
c
o
mp
u
t
er
s
c
i
en
ce
fr
o
m
A
l
al
-
Ba
y
t
U
n
i
v
er
s
i
t
y
,
J
o
r
d
an
,
i
n
2
0
1
2
an
d
2
0
1
5
,
res
p
ec
t
i
v
el
y
,
an
d
t
h
e
Ph
.
D
.
d
eg
r
ee
i
n
co
mp
u
t
er
s
c
i
e
n
ce
fro
m
U
n
i
v
ers
i
t
i
Sai
n
s
Mal
ay
s
i
a,
Mal
ay
s
i
a,
i
n
2
0
2
3
.
D
r.
A
b
u
o
w
ai
d
a
i
s
cu
rre
n
t
l
y
w
o
r
k
i
n
g
a
s
an
a
s
s
i
s
t
an
t
p
r
o
fes
s
o
r
i
n
t
h
e
Co
mp
u
t
er
Sci
en
ce
D
e
p
art
me
n
t
at
Z
arq
a
U
n
i
v
ers
i
t
y
,
J
o
rd
a
n
.
H
er
res
earch
i
n
t
ere
s
t
s
i
n
c
l
u
d
e
d
eep
l
ear
n
i
n
g
,
d
ep
t
h
es
t
i
ma
t
i
o
n
,
p
o
i
n
t
c
l
o
u
d
p
ro
ce
s
s
i
n
g
,
an
d
co
mp
u
t
er
v
i
s
i
o
n
.
Sh
e
ca
n
b
e
co
n
t
act
e
d
v
i
a
ema
i
l
at
s
ab
u
o
w
eu
d
a@
zu
.
ed
u
.
j
o
.
N
a
w
a
f
A
l
s
hda
i
fa
t
recei
v
ed
t
h
e
B.
Sc
d
eg
rees
i
n
co
mp
u
t
er
s
c
i
en
ce
fro
m
AL
al
-
Bay
t
U
n
i
v
er
s
i
t
y
an
d
M.
Sc.
d
eg
rees
i
n
co
m
p
u
t
er
s
ci
e
n
ce
fro
m
t
h
e
U
n
i
v
ers
i
t
y
o
f
J
o
r
d
an
,
J
o
r
d
an
,
i
n
2
0
0
2
an
d
2
0
1
1
,
res
p
ect
i
v
e
l
y
,
an
d
t
h
e
Ph
.
D
.
d
eg
ree
i
n
co
mp
u
t
er
s
c
i
en
ce
fr
o
m
U
n
i
v
er
s
i
t
i
Sai
n
s
Mal
ay
s
i
a,
Mal
a
y
s
i
a,
i
n
2
0
2
3
.
H
i
s
res
ea
rch
i
n
t
er
es
t
s
i
n
c
l
u
d
e
d
eep
l
ear
n
i
n
g
an
d
mach
i
n
e
l
earn
i
n
g
.
H
e
can
b
e
c
o
n
t
act
e
d
v
i
a
emai
l
a
t
n
_
a
l
s
h
d
ai
fa
t
@
as
u
.
ed
u
.
j
o
.
Evaluation Warning : The document was created with Spire.PDF for Python.