I
AE
S In
t
er
na
t
io
na
l J
o
urna
l o
f
Art
if
icia
l In
t
ellig
ence
(
I
J
-
AI
)
Vo
l.
14
,
No
.
3
,
J
u
n
e
2
0
2
5
,
p
p
.
2258
~
2
2
6
9
I
SS
N:
2
2
5
2
-
8
9
3
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijai.v
14
.i
3
.
p
p
2
2
5
8
-
2
2
6
9
2258
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
i
.
ia
esco
r
e.
co
m
Enha
nced Arabic
-
la
ng
ua
g
e cyberb
ully
ing
det
ec
tion:
deep
embedding
and
tr
a
nsfo
rmer (
BE
R
T) appro
a
ches
E
bte
s
a
m
J
a
ber
Alj
o
ha
ni,
Wa
el
M
.
S.
Ya
f
o
o
z
D
e
p
a
r
t
me
n
t
o
f
C
o
mp
u
t
e
r
S
c
i
e
n
c
e
,
C
o
l
l
e
g
e
o
f
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
a
n
d
En
g
i
n
e
e
r
i
n
g
,
T
a
i
b
a
h
U
n
i
v
e
r
s
i
t
y
,
M
a
d
i
n
a
h
,
S
a
u
d
i
A
r
a
b
i
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Ap
r
4
,
2
0
2
4
R
ev
is
ed
J
an
2
8
,
2
0
2
5
Acc
ep
ted
Ma
r
1
5
,
2
0
2
5
Re
c
e
n
t
tec
h
n
o
l
o
g
ica
l
a
d
v
a
n
c
e
s
in
sm
a
rtp
h
o
n
e
s
a
n
d
c
o
m
m
u
n
ica
ti
o
n
s,
in
c
lu
d
in
g
th
e
g
r
o
wth
o
f
su
c
h
o
n
li
n
e
p
latfo
rm
s
a
s
m
a
ss
iv
e
so
c
i
a
l
m
e
d
ia
n
e
two
rk
s
s
u
c
h
a
s
X
(
f
o
rm
e
rly
k
n
o
wn
a
s
T
witt
e
r)
e
n
d
a
n
g
e
rs
y
o
u
n
g
p
e
o
p
le
a
n
d
t
h
e
ir
e
m
o
ti
o
n
a
l
we
ll
-
b
e
i
n
g
b
y
e
x
p
o
sin
g
t
h
e
m
to
c
y
b
e
rb
u
ll
y
in
g
,
tau
n
ti
n
g
,
a
n
d
b
u
ll
y
in
g
c
o
n
ten
t.
M
o
st
p
ro
p
o
se
d
a
p
p
r
o
a
c
h
e
s
fo
r
a
u
to
m
a
ti
c
a
ll
y
d
e
tec
ti
n
g
c
y
b
e
rb
u
ll
y
i
n
g
h
a
v
e
b
e
e
n
d
e
v
e
l
o
p
e
d
a
ro
u
n
d
th
e
E
n
g
li
sh
la
n
g
u
a
g
e
,
a
n
d
m
e
th
o
d
s
fo
r
d
e
tec
ti
n
g
Ara
b
ic
-
la
n
g
u
a
g
e
c
y
b
e
rb
u
ll
y
in
g
a
re
sc
a
rc
e
.
M
e
th
o
d
s
fo
r
d
e
tec
ti
n
g
Ara
b
ic
-
lan
g
u
a
g
e
c
y
b
e
rb
u
ll
y
i
n
g
a
re
e
sp
e
c
ially
sc
a
rc
e
.
Th
is
p
a
p
e
r
a
ims
to
e
n
h
a
n
c
e
th
e
e
ffe
c
ti
v
e
n
e
ss
o
f
m
e
th
o
d
s
fo
r
d
e
tec
ti
n
g
c
y
b
e
rb
u
ll
y
i
n
g
i
n
Ara
b
ic
-
lan
g
u
a
g
e
c
o
n
te
n
t.
We
a
ss
e
m
b
led
a
d
a
tas
e
t
o
f
1
0
,
6
6
2
X
p
o
sts,
p
re
-
p
r
o
c
e
ss
e
d
th
e
d
a
ta,
a
n
d
u
se
d
th
e
k
a
p
p
a
to
o
l
to
v
e
rify
a
n
d
e
n
h
a
n
c
e
th
e
q
u
a
li
ty
o
f
o
u
r
a
n
n
o
t
a
ti
o
n
s.
We
c
o
n
d
u
c
ted
f
o
u
r
e
x
p
e
r
ime
n
ts
to
tes
t
n
u
m
e
ro
u
s
d
e
e
p
lea
rn
i
n
g
m
o
d
e
ls
fo
r
a
u
t
o
m
a
ti
c
a
ll
y
d
e
tec
ti
n
g
Ara
b
ic
-
lan
g
u
a
g
e
c
y
b
e
rb
u
ll
y
in
g
.
We
first
tes
ted
a
lo
n
g
sh
o
r
t
-
term
m
e
m
o
ry
(LS
T
M
)
m
o
d
e
l
a
n
d
a
b
id
irec
ti
o
n
a
l
lo
n
g
s
h
o
rt
-
term
m
e
m
o
ry
(Bi
-
LS
TM
)
m
o
d
e
l
with
se
v
e
ra
l
e
x
p
e
rime
n
tal
wo
rd
e
m
b
e
d
d
in
g
s.
We
a
lso
tes
ted
th
e
L
S
TM
a
n
d
Bi
-
LS
TM
m
o
d
e
ls
with
a
n
o
v
e
l
p
re
-
trai
n
e
d
b
id
irec
ti
o
n
a
l
e
n
c
o
d
e
r
fro
m
re
p
re
se
n
tatio
n
s
(BERT
)
a
n
d
t
h
e
n
tes
ted
th
e
m
o
n
a
d
iffere
n
t
e
x
p
e
rime
n
tal
m
o
d
e
ls
BERT
a
g
a
in
.
LS
TM
-
B
E
RT
a
n
d
Bi
-
LS
TM
-
BERT
d
e
m
o
n
stra
ted
a
9
7
%
a
c
c
u
ra
c
y
.
Bi
-
L
S
TM
with
F
a
stTex
t
e
m
b
e
d
d
in
g
wo
r
d
p
e
rfo
rm
e
d
e
v
e
n
b
e
tt
e
r,
a
c
h
iev
i
n
g
9
8
%
a
c
c
u
ra
c
y
.
As
a
re
su
lt
,
th
e
o
u
tc
o
m
e
s a
re
g
e
n
e
ra
li
z
e
d
.
K
ey
w
o
r
d
s
:
Ar
ab
ic
cy
b
er
b
u
lly
in
g
B
E
R
T
Dee
p
lear
n
in
g
Fas
tTe
x
t
T
r
an
s
f
o
r
m
e
r
s
W
o
r
d
em
b
ed
d
in
g
T
h
is i
s
a
n
o
p
e
n
a
c
c
e
ss
a
rticle
u
n
d
e
r th
e
CC B
Y
-
SA
li
c
e
n
se
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
E
b
tesam
J
ab
er
Aljo
h
an
i
Dep
ar
tm
en
t o
f
C
o
m
p
u
ter
Scie
n
ce
,
C
o
lleg
e
o
f
C
o
m
p
u
ter
Sci
en
ce
an
d
E
n
g
in
ee
r
in
g
,
T
ai
b
ah
Un
iv
er
s
ity
Ma
d
in
ah
,
Sau
d
i A
r
a
b
ia
E
m
ail:
ee
to
wi@
taib
ah
u
.
ed
u
.
s
a
1.
I
NT
RO
D
UCT
I
O
N
C
y
b
er
b
u
lly
in
g
is
b
ec
o
m
in
g
in
cr
ea
s
in
g
ly
p
r
ev
alen
t
in
s
o
c
iety
.
Alth
o
u
g
h
it
ca
n
o
cc
u
r
o
n
m
an
y
d
if
f
er
en
t
o
n
lin
e
p
latf
o
r
m
s
an
d
ch
an
n
els,
cy
b
er
b
u
lly
in
g
is
m
o
s
t
p
r
ev
alen
t
o
n
s
o
cial
n
etwo
r
k
in
g
s
ites
.
B
u
lly
in
g
is
a
n
eg
ativ
e
b
eh
a
v
io
r
th
at
in
ten
tio
n
ally
an
d
r
ep
ea
ted
ly
tar
g
ets
a
p
ar
ticu
la
r
p
e
r
s
o
n
;
b
u
ll
y
in
g
h
ar
m
s
p
eo
p
le
p
h
y
s
ically
an
d
h
ar
m
s
s
o
ciety
.
B
u
lly
in
g
o
cc
u
r
s
in
m
y
r
iad
cu
ltu
r
al
co
n
tex
ts
,
an
d
b
u
llies
em
p
lo
y
a
v
ar
ie
d
v
o
ca
b
u
lar
y
th
at
in
clu
d
es
ter
m
s
lik
e
“d
is
g
u
s
tin
g
,
”
“r
etar
d
ed
,
”
an
d
“stu
p
i
d
”
[
1
]
,
[
2
]
.
B
u
lly
in
g
’
s
th
r
ee
m
ai
n
ch
ar
ac
ter
is
tics
ar
e
r
ep
etitio
n
,
in
ten
tio
n
to
h
ar
m
,
a
n
d
an
u
n
e
q
u
al
p
o
wer
d
y
n
am
ic
b
etwe
en
th
e
v
ictim
an
d
th
e
o
f
f
en
d
er
[
3
]
th
is
b
eh
av
io
r
ca
n
h
av
e
ter
r
ib
le
c
o
n
s
eq
u
e
n
ce
s
,
in
clu
d
in
g
th
e
d
ea
th
o
f
s
o
m
e
v
ictim
s
.
C
u
r
r
en
tly
,
p
eo
p
le
ca
n
d
ig
itally
co
n
n
ec
t
with
o
th
er
s
r
eg
ar
d
less
o
f
th
e
d
is
tan
ce
b
etwe
en
th
em
.
Si
n
ce
p
r
o
tectin
g
f
r
ee
s
p
ee
ch
m
ea
n
s
p
r
o
tectin
g
s
o
cial
m
ed
ia
u
s
er
s
f
r
o
m
ab
u
s
e,
h
ar
ass
m
en
t,
an
d
i
n
tim
id
atio
n
,
d
etec
tin
g
b
u
lly
in
g
au
to
m
atica
lly
is
d
if
f
icu
lt
[
4
]
,
d
o
i
n
g
s
o
,
esp
ec
ially
in
A
r
a
b
ic
-
lan
g
u
a
g
e
p
o
s
ts
an
d
m
ess
ag
es,
is
b
ec
o
m
in
g
in
cr
ea
s
in
g
ly
d
if
f
icu
lt a
s
s
o
cial
p
latf
o
r
m
s
p
r
o
life
r
ate
[
5
]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
E
n
h
a
n
ce
d
A
r
a
b
ic
-
l
a
n
g
u
a
g
e
cy
b
erb
u
llyin
g
d
etec
tio
n
:
d
ee
p
e
mb
ed
d
in
g
a
n
d
…
(
E
b
tesa
m
Ja
b
er A
l
jo
h
a
n
i
)
2259
User
s
o
f
X,
a
p
o
p
u
lar
in
ter
ac
tiv
e
s
o
cial
n
etwo
r
k
i
n
g
p
latf
o
r
m
,
h
av
e
r
ec
e
n
tly
lef
t
b
o
th
p
o
s
itiv
e
an
d
n
eg
ativ
e
r
em
a
r
k
s
.
So
cial
m
e
d
ia
u
s
er
s
u
s
u
ally
u
s
e
p
s
eu
d
o
n
y
m
s
in
o
r
d
er
t
o
h
in
d
er
ef
f
o
r
ts
to
m
o
n
ito
r
th
eir
b
eh
av
io
r
[
6
]
.
Pre
v
io
u
s
ly
,
r
esear
ch
er
s
d
id
n
o
t
r
e
g
ar
d
th
is
p
r
o
b
lem
as
a
r
esear
ch
co
n
ce
r
n
b
u
t
ar
e
n
o
w
tak
in
g
it
s
er
io
u
s
ly
[
7
]
.
Nu
m
er
o
u
s
s
tu
d
ies
o
n
cy
b
er
b
u
lly
in
g
h
av
e
r
es
ea
r
ch
ed
a
v
ar
iety
o
f
im
p
lem
e
n
tatio
n
tech
n
iq
u
es
[
8
]
–
[
1
0
]
.
T
h
e
y
h
av
e
ev
alu
ate
d
th
e
d
i
v
er
s
ity
o
f
m
ac
h
in
e
lear
n
in
g
im
p
lem
en
tatio
n
s
,
wh
ich
p
r
o
d
u
ce
d
g
o
o
d
r
esu
lts
f
o
r
E
n
g
lis
h
ter
m
s
[
1
1
]
–
[
1
3
]
a
n
d
s
o
m
e
Ar
ab
ic
ter
m
s
[
1
4
]
,
[
1
5
]
.
Oth
er
s
u
s
ed
d
ee
p
l
ea
r
n
in
g
m
o
d
el
c
lick
o
r
tap
h
er
e
to
en
ter
tex
t
[
9
]
,
an
d
th
ese
also
p
r
o
d
u
ce
d
g
o
o
d
r
esu
lts
f
o
r
E
n
g
lis
h
ter
m
s
.
So
m
e
s
tu
d
ies
h
av
e
co
m
b
in
ed
d
if
f
er
en
t
tech
n
iq
u
es
.
Fu
r
th
er
r
esear
ch
o
n
tex
ts
in
lan
g
u
ag
es
o
th
er
th
a
n
E
n
g
lis
h
i
s
r
eq
u
ir
ed
.
Stu
d
ies
th
at
h
av
e
u
s
ed
A
r
ab
ic
tex
ts
ar
e
p
ar
ticu
lar
ly
r
ar
e.
Sev
er
al
s
tu
d
ies
test
ed
m
ac
h
in
e
lear
n
in
g
a
n
d
o
th
e
r
alg
o
r
ith
m
s
in
th
is
f
ield
,
s
u
ch
as
“
d
ec
is
io
n
tr
ee
(
DT
)
,
B
ay
esian
,
s
u
p
p
o
r
t
v
ec
to
r
m
ac
h
in
e
(
SVM)
,
r
a
n
d
o
m
f
o
r
est
(
R
F),
K
-
n
ea
r
est
n
eig
h
b
o
r
(
KNN)
”
[
1
2
]
,
[
1
6
]
,
o
n
alm
o
s
t
all
E
n
g
lis
h
-
lan
g
u
ag
e
p
o
s
ts
o
n
s
o
cial
m
ed
ia.
W
e
s
u
m
m
ar
ized
th
em
b
ased
o
n
th
r
ee
d
if
f
er
en
t a
p
p
r
o
ac
h
es.
I
n
th
e
m
ac
h
in
e
lear
n
in
g
a
p
p
r
o
ac
h
,
m
u
ltip
le
s
y
s
tem
s
ar
e
av
ailab
le
th
at
ca
n
p
r
ec
is
ely
an
d
au
to
m
atica
lly
d
etec
t
c
y
b
er
b
u
l
ly
in
g
.
Ali
et
a
l
.
[
1
2
]
s
o
u
g
h
t
t
o
d
r
aw
atten
tio
n
t
o
ea
r
lier
s
tu
d
ies
an
d
s
u
g
g
est
a
m
eth
o
d
f
o
r
id
en
tify
in
g
s
ar
ca
s
m
in
cy
b
er
b
u
lly
i
n
g
p
o
s
ts
f
r
o
m
a
d
if
f
er
en
t
d
atasets
.
T
h
ey
d
eter
m
in
ed
th
at
th
e
SVM
class
if
ier
o
u
tp
er
f
o
r
m
ed
th
e
class
if
ier
s
s
tated
ea
r
lier
o
n
class
if
y
in
g
d
atasets
2
,
3
,
an
d
4
an
d
ac
h
ie
v
ed
9
2
%
ac
cu
r
ac
y
o
n
d
ataset
1
.
R
esear
ch
o
n
id
en
tify
in
g
cy
b
er
b
u
lly
in
g
is
ex
p
an
d
in
g
b
ec
a
u
s
e
o
f
cy
b
er
b
u
lly
in
g
’
s
s
ig
n
if
ican
t
im
p
ac
t
o
n
s
o
cial
m
ed
ia
u
s
er
s
,
p
ar
ticu
lar
ly
c
h
ild
r
en
an
d
teen
ag
er
s
.
W
h
ile
m
an
y
s
tu
d
ies
h
av
e
f
o
cu
s
ed
o
n
id
e
n
tify
in
g
s
im
ilar
p
atter
n
s
in
E
n
g
lis
h
tex
t,
n
o
t
en
o
u
g
h
m
ac
h
in
e
lear
n
in
g
s
tu
d
ies
h
av
e
ex
am
in
e
d
Ar
ab
ic
-
lan
g
u
a
g
e
cy
b
er
b
u
lly
i
n
g
.
So
m
e
s
tu
d
ies
o
n
d
ete
ctin
g
Ar
ab
ic
-
lan
g
u
ag
e
cy
b
e
r
b
u
lly
in
g
p
r
o
d
u
ce
d
ex
tr
em
ely
g
o
o
d
r
esu
lts
wh
il
e
o
th
er
s
p
r
o
d
u
ce
d
a
v
er
ag
e
r
esu
lts
.
B
y
u
s
in
g
s
ev
er
al
clas
s
if
icatio
n
m
eth
o
d
s
,
Kan
an
et
a
l.
[
1
4
]
p
r
o
p
o
s
ed
a
p
p
ly
in
g
m
ac
h
in
e
lear
n
in
g
to
d
etec
t
b
ad
tex
tu
al
ac
ts
in
Ar
a
b
ic.
Var
io
u
s
Ar
ab
ic
n
atu
r
al
lan
g
u
a
g
e
p
r
o
ce
s
s
in
g
(
NL
P)
tech
n
o
lo
g
ies
h
av
e
also
b
ee
n
u
s
ed
.
T
h
e
r
esu
lts
d
e
m
o
n
s
tr
ate
th
at
th
e
F1
-
m
ea
s
u
r
e
v
alu
es
p
r
o
d
u
ce
d
b
y
th
e
R
F
alg
o
r
ith
m
wer
e
th
e
h
ig
h
est.
T
h
e
R
F
alg
o
r
ith
m
’
s
9
4
.
1
in
d
icate
d
a
9
5
%
ac
cu
r
ac
y
o
n
a
d
ataset
o
f
X
p
o
s
ts
,
an
d
SVM
g
av
e
9
1
.
7
o
n
a
d
ataset
o
f
Face
b
o
o
k
p
o
s
ts
.
T
h
ese
o
u
tco
m
es
h
eld
tr
u
e
wh
en
n
eith
er
s
tem
m
in
g
n
o
r
s
to
p
-
wo
r
d
r
e
m
o
v
al
was
u
s
ed
.
Oth
er
m
ac
h
i
n
e
lear
n
i
n
g
m
eth
o
d
s
co
u
ld
im
p
r
o
v
e
r
esu
lts
.
Ald
u
ailaj
an
d
B
elg
h
ith
[
1
5
]
f
o
u
n
d
b
u
lly
i
n
g
in
Ar
ab
ic
wr
itin
g
s
in
th
is
s
ec
to
r
.
On
e
s
tu
d
y
ex
am
in
ed
th
e
ap
p
licatio
n
o
f
SVM
an
d
s
ev
er
al
s
ig
n
if
ican
t
d
ata
p
r
o
ce
s
s
in
g
f
ea
tu
r
es
to
3
0
,
0
0
0
co
m
m
en
ts
.
Dif
f
er
en
t sit
u
atio
n
s
wer
e
in
clu
d
ed
in
it: cle
an
ed
,
s
tem
m
ed
,
s
eg
m
en
ted
,
a
n
d
s
to
p
-
w
o
r
d
c
o
m
p
ar
is
o
n
u
s
in
g
ter
m
-
f
r
eq
u
e
n
cy
tim
es
in
v
er
s
e
d
o
cu
m
en
t
-
f
r
eq
u
en
cy
(TF
-
I
DF)
an
d
b
ag
o
f
wo
r
d
s
(B
o
W
)
.
T
h
e
r
esu
lts
s
h
o
wed
th
at
SVM,
with
an
ac
c
u
r
ac
y
o
f
9
5
.
7
4
2
%,
id
en
tifie
d
cy
b
er
b
u
ll
y
in
g
m
o
r
e
ac
c
u
r
ately
t
h
an
n
aï
v
e
B
ay
es
(
NB
)
d
id
.
Du
e
to
o
u
r
m
o
d
el’
s
h
ig
h
ac
cu
r
ac
y
,
u
s
er
s
will
b
e
b
etter
p
r
o
tecte
d
f
r
o
m
s
o
cial
n
etwo
r
k
b
u
llies
th
an
th
ey
cu
r
r
en
tly
ar
e.
Oth
er
to
o
ls
m
ay
in
cr
ea
s
e
ac
cu
r
ac
y
.
AlFar
ah
et
a
l.
[
1
7
]
u
s
ed
o
n
lin
e
s
o
ci
al
n
etwo
r
k
d
ata
to
id
en
tify
Ar
ab
ic
c
y
b
er
b
u
lly
in
g
.
T
h
ey
c
r
ea
ted
a
d
ataset
u
s
in
g
in
f
o
r
m
atio
n
f
o
u
n
d
o
n
X
an
d
Yo
u
T
u
b
e.
T
h
ey
m
an
u
ally
an
n
o
tated
th
e
in
f
o
r
m
atio
n
to
en
s
u
r
e
th
e
a
n
n
o
tati
o
n
was
o
f
th
e
d
esire
d
q
u
ality
.
T
h
ey
o
v
e
r
s
am
p
led
th
e
m
in
o
r
ity
class
in
o
r
d
er
t
o
ad
d
r
ess
th
e
im
b
alan
ce
d
d
ata
s
et
p
r
o
b
lem
.
I
n
o
r
d
e
r
to
co
m
p
ar
e
th
e
class
if
ier
s
’
p
er
f
o
r
m
an
ce
s
,
th
ey
u
s
ed
f
i
v
e
m
ac
h
in
e
lear
n
in
g
ap
p
r
o
a
ch
es.
T
h
ey
p
r
esen
ted
th
e
o
u
tco
m
es
f
o
r
ea
ch
p
er
f
o
r
m
an
ce
p
ar
a
m
eter
.
SVM
an
d
lo
g
is
tic
r
eg
r
ess
io
n
(L
R
)
b
o
th
ac
h
iev
ed
8
8
%,
wh
er
ea
s
NB
h
ad
an
ar
ea
u
n
d
er
cu
r
v
e
(
AUC)
s
co
r
e
o
f
8
9
%.
T
o
en
h
an
ce
th
e
ac
cu
r
ac
y
o
f
m
eth
o
d
s
f
o
r
d
etec
tin
g
Ar
ab
ic
-
lan
g
u
a
g
e
cy
b
er
b
u
lly
in
g
,
m
o
r
e
d
ata
s
h
o
u
ld
b
e
co
llected
,
b
alan
ce
d
,
m
ea
s
u
r
ed
,
an
d
test
ed
u
s
in
g
d
ee
p
le
ar
n
in
g
m
o
d
els.
S
e
v
e
r
a
l
e
n
h
a
n
c
e
d
h
a
v
e
c
r
e
a
t
ed
d
e
e
p
l
e
a
r
n
i
n
g
m
o
d
e
l
s
b
a
s
e
d
o
n
o
b
s
e
r
v
a
t
i
o
n
s
f
r
o
m
p
r
i
o
r
s
t
u
d
i
e
s
a
n
d
g
e
n
e
r
a
t
e
d
h
i
g
h
l
y
p
o
s
i
ti
v
e
r
esu
l
t
s
a
n
d
a
p
p
li
e
d
t
o
i
d
e
n
ti
f
y
A
r
a
b
i
c
-
l
a
n
g
u
a
g
e
c
y
b
e
r
b
u
l
l
y
i
n
g
.
M
o
s
t
s
t
u
d
i
es
h
a
v
e
e
m
p
l
o
y
e
d
m
o
d
e
l
s
c
r
e
a
t
e
d
f
o
r
o
t
h
e
r
l
a
n
g
u
a
g
e
s
,
e
s
p
ec
i
a
ll
y
E
n
g
l
i
s
h
a
n
d
l
i
tt
l
e
f
o
r
A
r
a
b
i
c
.
Al
-
A
jl
a
n
a
n
d
Y
k
h
l
e
f
[
9
]
ar
g
u
ed
th
at
cy
b
er
b
u
lly
in
g
’
s
h
ar
ass
m
en
t
an
d
h
ate
is
a
s
ig
n
if
ican
t
ch
allen
g
e
.
T
o
b
r
id
g
e
th
e
k
n
o
wled
g
e
g
ap
i
n
th
is
f
ield
,
th
e
r
esear
ch
e
r
s
p
r
o
p
o
s
ed
a
n
o
v
el
co
n
v
o
lu
tio
n
al
n
e
u
r
al
n
etwo
r
k
(
C
NN)
alg
o
r
ith
m
f
o
r
cy
b
er
b
u
lly
in
g
d
etec
tio
n
in
o
r
d
er
to
elim
in
ate
th
e
n
ee
d
f
o
r
f
ea
tu
r
e
en
g
in
ee
r
i
n
g
an
d
to
p
r
o
d
u
ce
p
r
ed
ictio
n
s
y
s
tem
s
b
etter
th
an
tr
ad
itio
n
al
cy
b
er
b
u
lly
in
g
d
et
ec
tio
n
ap
p
r
o
ac
h
es.
Fo
r
th
e
alg
o
r
ith
m
,
th
e
y
ad
ap
ted
t
h
e
co
n
ce
p
t
o
f
wo
r
d
em
b
ed
d
in
g
,
w
h
er
e
s
im
ilar
w
o
r
d
s
h
av
e
a
s
im
ilar
em
b
ed
d
in
g
.
T
h
ey
s
h
o
wed
th
at
b
u
lly
in
g
p
o
s
ts
h
av
e
s
im
ilar
r
ep
r
esen
tatio
n
s
—
a
r
esu
lt
th
at
co
u
ld
h
elp
ad
v
a
n
ce
d
etec
tio
n
ef
f
o
r
ts
.
T
h
e
y
f
o
u
n
d
t
h
at
th
e
C
NN
-
cy
b
er
b
u
lly
i
n
g
alg
o
r
ith
m
,
wh
ich
ac
h
iev
ed
an
ac
cu
r
ac
y
lev
el
o
f
9
5
%,
o
u
tp
er
f
o
r
m
ed
c
o
n
v
e
n
tio
n
al
co
n
ten
t
-
b
ased
cy
b
er
b
u
lly
in
g
d
etec
tio
n
.
Al
-
Ajlan
an
d
Yk
h
lef
[
1
8
]
p
r
o
p
o
s
e
d
u
s
in
g
d
ee
p
lear
n
in
g
to
o
p
ti
m
ize
cy
b
er
b
u
lly
i
n
g
d
etec
tio
n
o
n
X.
T
h
eir
n
o
v
e
l
ap
p
r
o
ac
h
ad
d
r
ess
es
th
e
ab
o
v
e
ch
allen
g
es.
Un
lik
e
p
r
e
v
io
u
s
s
tu
d
ies
[
9
]
,
Al
-
Ajlan
an
d
Yk
h
lef
’
s
ap
p
r
o
a
ch
r
ep
r
esen
ts
a
p
o
s
t a
s
a
s
et
o
f
wo
r
d
v
ec
to
r
s
r
ath
e
r
th
an
ex
t
r
a
ctin
g
f
ea
tu
r
es f
r
o
m
p
o
s
ts
an
d
f
ee
d
in
g
th
em
in
to
cl
ass
if
ier
s
.
T
h
eir
ap
p
r
o
ac
h
ac
h
i
ev
ed
g
o
o
d
r
esu
lts
.
T
h
ey
ca
n
a
p
p
ly
d
if
f
er
e
n
t
d
ee
p
m
o
th
ed
as
b
id
ir
ec
tio
n
al
-
lo
n
g
s
h
o
r
t
-
ter
m
m
em
o
r
y
(
B
i
-
L
ST
M)
m
ay
u
s
e
to
im
p
r
o
v
e
r
esu
lts
.
Ah
m
ed
et
a
l.
[
1
9
]
tr
ain
ed
d
ee
p
lear
n
in
g
m
o
d
els
o
n
Ar
ab
ic
tex
t.
T
h
ey
ex
am
in
e
d
f
iv
e
d
ee
p
class
if
ier
m
o
d
els:
C
NN,
lo
n
g
s
h
o
r
t
-
ter
m
m
em
o
r
y
(
L
STM
)
,
B
i
-
L
STM
g
ated
r
ec
u
r
r
e
n
t
u
n
it
(
GR
U
)
,
an
d
C
NN
-
B
i
L
STM
.
T
h
ey
co
m
p
iled
a
d
ataset
f
r
o
m
th
e
web
s
ite
o
f
Al
J
az
ee
r
a,
an
Ar
ab
ic
n
ews
o
r
g
a
n
izatio
n
.
T
h
ey
u
tili
ze
d
th
e
T
F
-
I
DF
ap
p
r
o
ac
h
to
r
ep
r
esen
t
im
p
o
r
tan
t
ter
m
s
an
d
co
n
d
u
cte
d
ex
p
e
r
im
en
ts
with
v
a
r
y
in
g
la
y
er
s
.
Of
th
e
f
iv
e
m
o
d
els,
L
ST
M
ac
h
iev
ed
t
h
e
b
est
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
14
,
No
.
3
,
J
u
n
e
2
0
2
5
:
2
2
5
8
-
2
2
6
9
2260
o
u
tco
m
e:
a
9
2
.
7
5
%
s
u
cc
ess
r
ate.
T
h
e
s
tu
d
y
p
r
o
d
u
ce
d
s
o
m
e
p
r
o
m
is
in
g
r
esu
lts
,
b
u
t
it
m
i
g
h
t
h
a
v
e
b
ee
n
b
etter
h
ad
Ah
m
ed
et
al.
u
s
ed
a
d
d
itio
n
al
em
b
ed
d
in
g
r
e
p
r
esen
tatio
n
v
ec
to
r
s
.
A
s
m
all
n
u
m
b
er
o
f
s
tu
d
ies
h
a
v
e
tr
ain
ed
m
o
d
els
u
s
in
g
h
y
b
r
i
d
tech
n
iq
u
es
th
at
c
o
m
b
in
e,
f
o
r
ex
am
p
le,
d
ee
p
an
d
m
ac
h
i
n
e
lear
n
in
g
o
r
o
th
er
.
Stu
d
ies
th
at
u
s
e
h
y
b
r
id
tech
n
i
q
u
es
s
tan
d
o
u
t
with
in
a
liter
atu
r
e
co
m
p
r
is
ed
lar
g
el
y
o
f
s
tu
d
ies
th
at
u
s
ed
o
n
l
y
o
n
e
m
eth
o
d
.
Al
o
taib
i
et
a
l.
[
6
]
p
r
o
p
o
s
ed
an
a
u
to
m
atic
m
eth
o
d
to
d
etec
t
ag
g
r
ess
iv
e
b
eh
a
v
io
r
u
s
in
g
a
co
n
s
o
lid
ated
d
ee
p
lear
n
in
g
m
o
d
el
with
a
tr
a
n
s
f
o
r
m
e
r
.
T
h
eir
tech
n
iq
u
e
u
s
es
m
u
ltich
an
n
el
d
ee
p
lear
n
in
g
b
ased
o
n
th
r
ee
m
o
d
els’
tr
an
s
f
o
r
m
er
b
lo
ck
s
,
a
b
id
ir
ec
tio
n
al
g
ated
r
ec
u
r
r
en
t
u
n
it
(
B
iGR
U)
,
an
d
a
C
NN
to
clas
s
if
y
X
co
m
m
en
ts
in
to
two
ca
teg
o
r
ies:
o
f
f
en
s
iv
e
(
b
u
lly
in
g
)
an
d
n
o
n
-
o
f
f
en
s
iv
e
(
n
o
n
-
b
u
lly
in
g
)
.
T
h
eir
n
o
v
el
cy
b
er
b
u
lly
in
g
d
etec
tio
n
tech
n
iq
u
e
ac
h
iev
e
d
p
r
o
m
is
in
g
r
e
s
u
lts
,
in
clu
d
in
g
an
ac
cu
r
ac
y
o
f
8
8
%.
Alo
taib
i
et
a
l.
’
s
wo
r
k
co
u
ld
b
e
im
p
r
o
v
ed
b
y
ap
p
ly
i
n
g
d
if
f
er
en
t d
ee
p
m
et
h
o
d
s
with
a
v
a
r
iety
o
f
em
b
ed
d
in
g
s
.
As
with
E
n
g
lis
h
,
r
esear
ch
er
s
h
a
v
e
r
ec
e
n
tly
a
ttem
p
ted
to
i
n
cr
ea
s
e
th
e
ac
c
u
r
ac
y
o
f
m
eth
o
d
s
f
o
r
d
etec
tin
g
Ar
ab
ic
-
lan
g
u
a
g
e
c
y
b
er
b
u
lly
in
g
b
y
co
m
b
in
in
g
m
u
ltip
le
ap
p
r
o
ac
h
es.
T
h
er
e
is
s
till
a
n
ee
d
f
o
r
d
ev
elo
p
m
e
n
t
an
d
a
d
d
itio
n
al
v
alid
atio
n
in
th
e
Ar
a
b
ic
NL
P
ta
s
k
.
Han
i
et
a
l.
[
1
6
]
p
r
o
p
o
s
ed
a
s
u
p
er
v
is
ed
h
y
b
r
id
m
ac
h
in
e
lear
n
in
g
ap
p
r
o
ac
h
f
o
r
d
etec
tin
g
an
d
p
r
ev
en
tin
g
cy
b
er
b
u
lly
in
g
wh
ile
u
s
in
g
m
an
y
class
if
ier
s
to
tr
ai
n
an
d
r
ec
o
g
n
ize
b
u
lly
in
g
ac
ti
o
n
s
.
T
h
ey
f
o
u
n
d
th
at
n
eu
r
a
l
n
etwo
r
k
s
h
ad
t
h
e
h
ig
h
est
p
er
f
o
r
m
an
ce
a
n
d
d
em
o
n
s
tr
ated
a
n
F1
-
s
co
r
e
o
f
9
1
.
9
%
an
d
an
ac
cu
r
ac
y
o
f
9
2
.
8
%.
T
h
e
SVM
ac
h
iev
e
d
9
0
.
3
%.
T
h
e
m
o
d
els
also
s
h
o
wed
th
at
d
ee
p
n
e
u
r
al
lear
n
i
n
g
m
o
d
els o
u
tp
er
f
o
r
m
ed
th
e
S
VM
in
s
ev
er
al
ex
p
er
im
en
ts
.
T
h
e
s
tu
d
y
’
s
f
in
d
in
g
s
d
em
o
n
s
tr
ated
th
at
en
f
o
r
ce
d
to
s
u
p
p
o
r
t
th
e
lear
n
er
m
o
d
el
b
y
o
b
tain
in
g
c
o
n
tex
tu
al
d
ata
f
r
o
m
v
ar
io
u
s
s
o
u
r
ce
s
.
Han
i
et
a
l
.
[
1
6
]
also
co
n
tr
asted
th
eir
s
u
g
g
ested
m
eth
o
d
o
lo
g
y
with
cu
ttin
g
-
ed
g
e
ap
p
r
o
ac
h
es
to
d
em
o
n
s
tr
ate
th
at,
in
m
o
s
t
in
s
tan
ce
s
,
th
eir
s
tr
ateg
y
g
r
ea
tly
s
u
r
p
ass
ed
th
e
latter
’
s
o
u
tco
m
es.
B
id
ir
ec
tio
n
al
en
co
d
er
f
r
o
m
r
ep
r
esen
tatio
n
s
(
B
E
R
T
)
is
a
d
ee
p
atten
tio
n
-
b
ased
la
n
g
u
a
g
e
m
o
d
el
th
at
ca
n
f
in
d
p
atter
n
s
i
n
len
g
t
h
y
an
d
n
o
is
y
b
o
d
ies
o
f
tex
t.
R
ez
v
a
n
i
an
d
B
eh
esh
ti
[
7
]
p
r
o
v
id
ed
a
c
o
n
te
m
p
o
r
ar
y
m
ac
h
in
e
lear
n
in
g
tech
n
iq
u
e
th
at
t
u
n
es
a
v
ar
ian
t
o
f
B
E
R
T
an
d
r
ep
o
r
ted
an
ac
cu
r
ac
y
o
f
8
6
%
f
o
r
an
E
n
g
lis
h
-
lan
g
u
a
g
e
d
ataset.
T
h
ey
f
o
u
n
d
th
at
a
L
STM
m
o
d
el
an
d
B
E
R
T
o
u
tp
er
f
o
r
m
ed
b
aselin
e
m
eth
o
d
s
o
n
al
l
m
etr
ics.
Few
s
tu
d
ies
h
av
e
em
p
lo
y
ed
a
h
y
b
r
id
m
eth
o
d
o
l
o
g
y
o
f
b
o
th
m
ac
h
in
e
an
d
d
ee
p
lear
n
in
g
.
T
h
e
Ar
ab
ic
co
n
ten
t
u
s
ed
to
p
r
ep
a
r
e
t
h
e
s
e
h
y
b
r
i
d
m
et
h
o
d
s
co
u
ld
b
e
im
p
r
o
v
ed
.
T
a
b
le
1
s
u
m
m
ar
izes
r
ec
en
t
s
elec
ted
s
tu
d
ies
o
n
d
etec
tin
g
c
y
b
er
b
u
lly
in
g
.
W
e
class
if
ied
th
e
s
tu
d
ies d
ep
en
d
in
g
o
n
w
h
eth
er
th
ey
u
s
ed
m
ac
h
in
e
lear
n
i
n
g
,
d
ee
p
lear
n
in
g
,
o
r
an
o
th
e
r
ap
p
r
o
ac
h
.
T
ab
le
1
.
C
o
m
p
a
r
is
o
n
o
f
s
tu
d
ie
s
St
u
d
y
A
p
p
ro
ac
h
es
Cl
a
s
s
/
met
h
o
d
s
D
a
t
a
s
e
t
W
o
r
d
r
ep
r
es
en
t
a
t
i
o
n
s
L
a
n
g
u
ag
e
F1
/
A
cc
u
rac
y
(
%
)
[6
]
D
ee
p
l
e
ar
n
i
n
g
an
d
t
ra
n
s
f
o
rme
r
G
RU
,
B
i
G
R
U
,
C
N
N
,
RF
,
t
ra
n
s
fo
rmer
-
b
l
o
c
k
X
W
o
r
d
e
mb
ed
d
i
n
g
E
n
g
l
i
s
h
88
[9
]
D
ee
p
l
e
ar
n
i
n
g
CN
N
X
W
o
r
d
e
mb
ed
d
i
n
g
E
n
g
l
i
s
h
95
[1
6
]
Mac
h
i
n
e
l
ear
n
i
n
g
an
d
d
ee
p
l
e
ar
n
i
n
g
NN
,
S
V
M
Fo
r
ms
p
r
i
n
g
W
o
r
d
e
mb
ed
d
i
n
g
E
n
g
l
i
s
h
9
2
.
8
[1
8
]
D
ee
p
l
e
ar
n
i
n
g
CN
N
X
G
l
o
V
e
E
n
g
l
i
s
h
96
[2
0
]
D
ee
p
l
e
ar
n
i
n
g
L
S
T
M
,
Bi
-
L
S
T
M
,
G
R
U
,
C
N
N
,
CN
N
-
Bi
L
S
T
M
A
l
j
a
zeer
a.
n
e
t
TF
-
I
D
F
A
ra
b
i
c
9
2
.
7
5
[2
1
]
D
ee
p
l
e
ar
n
i
n
g
Bi
-
L
S
T
M
,
G
R
U
,
L
S
T
M
,
R
N
N
Face
b
o
o
k
/
X
/
In
s
t
ag
r
am
-
E
n
g
l
i
s
h
8
2
.
1
8
[2
2
]
Mac
h
i
n
e
l
ear
n
i
n
g
NB
,
S
V
M
Face
b
o
o
k
,
X
-
E
n
g
l
i
s
h
,
A
ra
b
i
c
93
[2
3
]
D
ee
p
l
e
ar
n
i
n
g
CN
N
,
C
N
N
-
L
S
T
M
,
B
i
L
S
T
M
-
CN
N
X
W
o
r
d
e
mb
ed
d
i
n
g
s
A
ra
b
i
c
81
[2
4
]
Mac
h
i
n
e
l
ear
n
i
n
g
LR
,
L
G
BM
,
SG
D
,
RF
,
A
d
aB
o
o
s
t
(A
D
B)
,
NB
,
S
V
M
X
-
E
n
g
l
i
s
h
9
2
.
8
[2
5
]
Mac
h
i
n
e
l
ear
n
i
n
g
NB
A
ra
b
i
c
s
o
c
i
a
l
med
i
a
s
t
re
am
s
-
A
ra
b
i
c
96
T
h
e
tech
n
iq
u
es
o
f
th
e
s
tu
d
ie
s
s
u
m
m
ar
ized
in
T
ab
le
1
,
wer
e
g
ap
s
m
o
s
tly
d
esig
n
ed
to
ex
am
in
e
E
n
g
lis
h
wo
r
d
s
at
th
e
tex
t
lev
e
l
an
d
th
u
s
th
eir
r
esu
lts
ar
e
b
et
ter
wh
en
ap
p
lied
to
E
n
g
lis
h
-
l
an
g
u
ag
e
te
x
ts
th
an
wh
en
ap
p
lied
to
te
x
ts
in
o
th
e
r
lan
g
u
ag
es
as
Ar
ab
ic
ter
m
s
.
T
h
is
p
ap
er
’
s
m
ain
p
u
r
p
o
s
e
is
to
p
r
esen
t
way
s
to
en
h
an
ce
m
o
d
els
f
o
r
d
etec
tin
g
Ar
ab
ic
-
lan
g
u
a
g
e
cy
b
er
b
u
lly
in
g
o
n
s
o
cial
m
ed
ia
p
latf
o
r
m
s
,
e
s
p
ec
ially
X.
I
t
also
in
v
esti
g
ates
d
ee
p
lea
r
n
in
g
o
n
d
if
f
er
en
t
wo
r
d
em
b
ed
d
in
g
s
an
d
a
n
o
v
el
h
y
b
r
i
d
d
ee
p
t
r
an
s
f
o
r
m
er
.
I
t
co
m
p
ar
es
i
t
to
s
tu
d
ies
o
n
th
e
b
aselin
e
m
a
ch
in
e
lear
n
in
g
m
o
d
el
an
d
th
e
tr
an
s
f
o
r
m
e
r
s
(
B
E
R
T
)
m
o
d
el.
T
h
is
s
tu
d
y
m
a
k
es
n
u
m
er
o
u
s
co
n
t
r
ib
u
tio
n
s
t
o
th
e
liter
atu
r
e
o
n
d
ee
p
lear
n
in
g
an
d
cy
b
er
b
u
lly
in
g
d
etec
tio
n
.
I
t
b
u
ild
s
a
n
o
v
el
d
ataset
o
f
b
u
lly
in
g
an
d
n
o
n
-
b
u
lly
in
g
Ar
ab
ic
-
lan
g
u
ag
e
p
o
s
ts
f
r
o
m
X.
I
t
p
r
o
p
o
s
es
two
m
o
d
els
o
f
d
ee
p
lear
n
i
n
g
.
T
h
e
f
ir
s
t
m
o
d
el
co
m
p
r
is
es
L
STM
an
d
B
i
-
L
STM
d
ee
p
lear
n
in
g
m
o
d
els
o
f
d
if
f
e
r
en
t
ex
p
er
im
en
tal
wo
r
d
em
b
ed
d
in
g
s
(
i.e
.
,
T
F
-
I
DF
an
d
th
e
m
o
d
er
n
p
r
e
-
tr
ain
ed
m
o
d
els
Ar
awo
r
d
2
v
ec
,
Glo
Ve,
Fas
tTe
x
t,
an
d
B
E
R
T
)
.
T
h
e
s
ec
o
n
d
m
o
d
el
co
m
p
r
is
es
h
y
b
r
id
d
ee
p
tr
an
s
f
o
r
m
er
s
(
L
STM
+BERT
an
d
B
i
-
L
ST
M+
B
E
R
T
)
u
s
ed
to
class
if
y
Ar
ab
ic
-
lan
g
u
ag
e
u
s
er
co
m
m
en
ts
d
ep
e
n
d
in
g
o
n
wh
et
h
er
o
r
n
o
t
th
ey
co
n
tain
cy
b
e
r
b
u
lly
in
g
.
T
h
is
s
tu
d
y
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
E
n
h
a
n
ce
d
A
r
a
b
ic
-
l
a
n
g
u
a
g
e
cy
b
erb
u
llyin
g
d
etec
tio
n
:
d
ee
p
e
mb
ed
d
in
g
a
n
d
…
(
E
b
tesa
m
Ja
b
er A
l
jo
h
a
n
i
)
2261
ev
alu
ates
th
e
p
r
o
p
o
s
ed
m
o
d
el
s
’
p
er
f
o
r
m
an
ce
s
o
n
all
ex
p
e
r
i
m
en
ts
’
ap
p
r
o
ac
h
es
u
s
in
g
th
e
m
o
s
t
co
m
m
o
n
ly
u
s
ed
m
etr
ics as
ac
cu
r
ac
y
,
r
ec
all
,
a
n
d
p
r
ec
is
io
n
f
o
r
class
if
icatio
n
.
T
h
e
p
r
esen
tatio
n
i
n
th
is
p
a
p
er
is
o
r
g
an
ize
d
as
f
o
llo
ws.
I
n
s
ec
tio
n
2
,
t
h
e
p
ap
er
ex
p
lain
s
th
e
m
eth
o
d
o
l
o
g
y
we
u
tili
ze
d
to
cr
ea
te
th
e
p
r
o
p
o
s
ed
m
o
d
el
s
,
co
llect
d
ata,
clea
n
,
p
r
e
p
r
o
ce
s
s
d
ata,
co
n
d
u
ct
ex
p
er
im
en
ts
,
an
d
class
if
y
th
e
m
o
d
els.
I
n
s
ec
tio
n
3
,
it p
r
esen
ts
th
e
r
esu
lts
o
f
th
e
s
tu
d
y
,
an
al
y
ze
s
th
em
,
an
d
th
e
n
co
m
p
ar
es th
em
t
o
g
eth
er
.
I
n
s
e
ctio
n
4
,
it p
r
esen
ts
o
u
r
c
o
n
clu
s
io
n
s
an
d
p
o
ten
tial d
ir
ec
tio
n
s
f
o
r
f
u
tu
r
e
r
esear
ch
.
2.
M
E
T
H
O
D
T
h
i
s
s
t
u
d
y
’
s
m
e
t
h
o
d
o
l
o
g
y
i
n
v
o
l
v
e
d
a
s
e
r
i
e
s
o
f
s
t
e
p
s
t
h
a
t
b
e
g
a
n
w
i
t
h
a
s
e
t
o
f
i
n
p
u
t
s
a
n
d
e
n
d
e
d
w
i
t
h
t
h
e
e
x
p
e
c
t
e
d
m
o
d
e
l
o
u
t
p
u
t
s
.
D
a
t
a
c
o
l
l
e
c
t
i
o
n
a
n
d
a
n
n
o
t
a
t
i
o
n
—
t
h
e
c
o
r
e
o
f
t
h
e
p
r
o
j
e
c
t
—
w
a
s
t
h
e
s
t
u
d
y
’
s
t
h
i
r
d
p
h
a
s
e
.
I
n
t
h
e
f
o
u
r
t
h
s
t
a
g
e
,
w
e
p
r
o
c
e
s
s
e
d
t
h
e
d
a
t
a
a
n
d
c
l
a
s
s
i
f
i
e
d
i
t
u
s
i
n
g
s
e
v
e
r
a
l
a
n
n
o
t
a
t
o
r
s
.
W
e
t
h
e
n
a
p
p
l
i
e
d
t
h
e
s
u
g
g
e
s
t
e
d
e
x
p
l
o
r
a
t
i
o
n
t
e
c
h
n
i
q
u
e
s
a
n
d
a
s
e
r
i
e
s
o
f
a
l
g
o
r
i
t
h
m
s
t
o
t
h
e
c
l
a
s
s
i
f
i
e
d
d
a
t
a
.
F
i
g
u
r
e
1
s
h
o
w
s
o
u
r
m
e
t
h
o
d
o
l
o
g
y
.
Fig
u
r
e
1
.
Me
th
o
d
o
lo
g
y
f
o
r
s
tu
d
y
in
g
m
o
d
els f
o
r
d
etec
tin
g
A
r
ab
ic
-
lan
g
u
a
g
e
cy
b
e
r
b
u
lly
i
n
g
2
.
1
.
Da
t
a
c
o
llect
io
n
S
i
n
c
e
a
c
c
u
r
at
e
l
y
t
es
t
i
n
g
c
la
s
s
if
i
c
a
t
i
o
n
m
e
t
h
o
d
s
d
e
p
e
n
d
s
o
n
d
a
t
a
,
w
e
h
a
d
t
o
c
a
r
e
f
u
l
l
y
s
e
l
ec
t
o
u
r
d
a
t
a
s
o
u
r
c
e
(
s
)
a
n
d
t
h
e
s
i
z
e
o
f
o
u
r
d
a
t
a
s
et
.
A
m
o
n
g
t
h
e
o
b
s
t
a
c
l
es
t
h
a
t
w
e
f
a
c
e
d
w
e
r
e
X
’
s
l
i
m
i
t
s
o
n
t
h
e
a
m
o
u
n
t
o
f
d
a
t
a
p
e
o
p
l
e
a
r
e
a
l
l
o
w
e
d
t
o
t
a
k
e
t
h
r
o
u
g
h
i
t
s
A
P
I
a
n
d
t
h
e
n
e
e
d
t
o
c
r
e
a
t
e
a
n
e
w
a
c
c
o
u
n
t
a
n
d
p
a
y
f
o
r
a
n
e
w
s
u
b
s
c
r
i
p
t
i
o
n
e
a
c
h
t
i
m
e
we
c
o
l
l
ec
t
e
d
d
a
t
a
.
W
e
u
s
e
d
n
t
s
c
r
a
p
e
r
a
n
d
p
a
n
d
a
t
o
c
o
l
l
e
c
t
X
p
o
s
ts
t
h
at
c
o
n
t
a
i
n
e
d
t
e
r
m
s
t
h
at
c
o
u
l
d
i
n
d
i
c
a
t
e
s
a
r
c
as
m
o
r
b
u
l
l
y
i
n
g
(
e
.
g
.
,
ف
لخ
ت
م
ا
ي
,
ف
ر
ق
م
ا
ي
,
a
n
d
ي
ب
غ
)
.
So
m
e
o
f
o
u
r
k
e
y
t
e
r
m
s
a
r
e
l
is
t
e
d
i
n
T
a
b
l
e
2
.
W
e
m
an
u
ally
class
if
ied
th
e
p
o
s
ts
.
W
e
lab
eled
p
o
s
ts
th
at
co
n
t
ain
ed
b
u
ll
y
in
g
with
o
n
es
an
d
t
h
o
s
e
th
at
d
id
n
o
t
with
ze
r
o
es.
Usi
n
g
C
o
h
en
’
s
k
ap
p
a
ag
r
ee
m
e
n
t
in
SP
SS
[
2
6
]
,
we
f
o
u
n
d
a
k
ap
p
a
o
f
0
.
9
8
,
in
d
icatin
g
g
o
o
d
ag
r
ee
m
e
n
t
b
etwe
en
th
e
an
n
o
tato
r
s
.
T
h
e
r
esu
lts
o
f
th
i
s
an
aly
s
is
ar
e
s
h
o
wn
in
Fig
u
r
e
2
.
W
e
s
av
ed
th
e
d
ataset
as
a
C
SV
f
ile
a
n
d
th
en
s
ep
ar
ated
th
e
b
u
lly
in
g
a
n
d
n
o
n
-
b
u
lly
in
g
p
o
s
ts
.
T
h
e
d
ataset
co
n
tain
s
two
f
ield
s
,
o
n
e
f
o
r
co
m
m
en
ts
an
d
th
e
o
th
er
f
o
r
class
if
icatio
n
lab
els.
T
ab
le
2
.
Key
te
r
m
s
f
r
o
m
d
ataset
A
r
a
b
i
c
t
e
r
m
En
g
l
i
sh
t
r
a
n
sl
a
t
i
o
n
فل
خ
ت
م
A
t
a
x
y
-
s
p
a
z
فر
ق
م
N
a
st
y
ي
ب
غ
S
t
u
p
i
d
Fig
u
r
e
2
.
An
n
o
tatio
n
s
’
k
ap
p
a
ag
r
ee
m
en
t v
al
u
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
14
,
No
.
3
,
J
u
n
e
2
0
2
5
:
2
2
5
8
-
2
2
6
9
2262
2
.
2
.
Da
t
a
clea
nin
g
T
h
is
p
ar
t
s
h
o
ws
th
e
clea
n
d
ataset
m
eth
o
d
s
.
C
lean
in
g
a
n
d
r
em
o
v
in
g
n
o
is
e
a
d
d
r
ess
es
X
d
ata'
s
clu
tter
ed
n
atu
r
e
.
W
e
r
em
o
v
e
d
n
o
n
-
Ar
ab
ic
c
h
ar
ac
ter
s
an
d
k
ep
t
o
n
ly
Ar
ab
ic
letter
s
.
W
e
s
tr
ip
p
ed
o
u
t
an
d
r
ep
lace
d
n
o
n
-
alp
h
ab
etic
ch
ar
ac
ter
s
an
d
r
em
o
v
ed
em
o
jis
,
p
u
n
ctu
atio
n
m
ar
k
s
(
e.
g
.
,
#
,
/
/,
!
,
$
)
a
n
d
o
t
h
er
s
y
m
b
o
ls
[
2
7
]
.
W
e
also
r
em
o
v
e
d
Ar
ab
ic
n
u
m
b
e
r
s
an
d
r
ep
ea
te
d
ch
ar
ac
ter
s
an
d
p
o
s
t r
o
ws in
d
ataset.
So
,
f
o
r
th
at
u
s
in
g
s
o
m
e
co
m
m
an
d
(
r
e.
s
u
b
)
in
p
y
th
o
n
c
o
d
e
to
r
em
o
v
e
an
d
o
th
er
m
a
r
k
s
.
2
.
3
.
P
re
-
pro
ce
s
s
ing
T
h
is
s
ec
tio
n
ex
p
lain
s
h
o
w
we
clea
n
ed
an
d
p
r
e
-
p
r
o
ce
s
s
ed
th
e
d
ataset
to
n
o
r
m
alize
th
e
p
o
s
ts
’
lin
g
u
is
tic
f
o
r
m
.
Pre
-
p
r
o
ce
s
s
in
g
is
cr
u
cial
f
o
r
im
p
r
o
v
in
g
th
e
class
if
icatio
n
m
o
d
els’
e
f
f
e
ctiv
en
ess
.
T
ab
le
3
co
n
tain
s
ex
am
p
les o
f
h
o
w
we
p
r
e
-
p
r
o
ce
s
s
ed
an
d
clea
n
p
o
s
ts
.
W
e
f
o
llo
wed
th
ese
s
tep
s
:
‒
No
r
m
alize
(
tex
t)
:
i
n
o
u
r
d
ata
s
et,
we
p
r
esen
ted
ea
ch
Ar
ab
i
c
ch
ar
ac
ter
wh
o
s
e
f
o
r
m
ca
n
v
ar
y
i
n
a
s
in
g
le
co
n
s
is
ten
t
f
o
r
m
.
Fo
r
ex
am
p
le,
alep
h
ca
n
b
e
wr
itten
in
s
u
ch
f
o
r
m
s
as
أ
–
إ
–
آ
,
an
d
we
n
o
r
m
alize
d
it
to
th
e
f
o
r
m
ا
b
ased
o
n
t
h
e
r
ec
o
m
m
en
d
ed
n
o
r
m
[
2
8
]
.
W
e
also
r
em
o
v
ed
d
iac
r
itics
(
tash
d
id
,
f
ath
a,
tan
win
f
ath
,
d
am
m
a,
tan
win
d
am
m
,
k
asra
,
tan
win
k
asr,
s
u
k
u
n
,
an
d
tatwil/k
ash
id
a)
[
2
7
]
.
‒
R
em
o
v
e
Ar
ab
ic
s
to
p
w
o
r
d
s
(
tex
t)
:
w
e
to
k
en
ized
an
d
r
e
m
o
v
ed
s
to
p
wo
r
d
s
f
r
o
m
Ar
ab
ic
te
x
t.
Sto
p
wo
r
d
s
ar
e
wo
r
d
s
th
at
d
o
n
o
t
c
h
an
g
e
h
o
w
a
s
en
ten
ce
is
u
n
d
er
s
to
o
d
.
T
h
er
e
ar
e
m
an
y
s
to
p
w
o
r
d
s
in
Ar
ab
ic,
in
clu
d
in
g
خ
لإ
،
ن
ي
ذلا
يتلا
،
يذلا
،
ىلع
.
T
h
ese
w
o
r
d
s
’
p
r
esen
ce
in
s
en
ten
ce
s
ex
p
an
d
s
th
e
d
ataset’
s
lex
ico
n
an
d
m
ak
es c
lass
if
icatio
n
m
o
r
e
d
if
f
icu
lt
[
2
7
]
.
‒
Stem
(
tex
t)
:
w
e
u
s
ed
th
e
Sn
o
wb
all
Stem
m
er
p
o
r
ter
2
s
tem
m
in
g
alg
o
r
ith
m
to
s
tem
th
e
Ar
a
b
ic
wo
r
d
s
in
o
u
r
d
ataset.
W
e
to
k
en
ized
th
e
in
p
u
t te
x
t,
s
tem
m
ed
ea
ch
wo
r
d
,
a
n
d
r
ec
o
n
s
tr
u
cted
th
e
te
x
t
[
2
9
]
.
T
ab
le
3
.
Pre
-
p
r
o
ce
s
s
in
g
an
d
cl
ea
n
ex
am
p
les
O
r
i
g
i
n
a
l
A
r
a
b
i
c
p
o
s
t
s
A
f
t
e
r
p
r
e
-
p
r
o
c
e
ss
i
n
g
ر
ي
وص
ت
لا
لي
م
ج
ر
ي
وص
ت
لي
م
ج
هلك
ش
فر
ق
م
b
a
a
لك
ش
فر
ق
م
Af
ter
p
r
ep
r
o
ce
s
s
in
g
an
d
clea
n
th
e
d
ataset,
it
co
n
tain
e
d
1
0
,
6
6
2
Ar
ab
ic
-
lan
g
u
ag
e
co
m
m
e
n
ts
:
5
,
7
3
6
(
5
3
.
8
0
%)
b
u
lly
in
g
an
d
4
,
9
2
6
(
4
6
.
2
0
%)
n
o
n
-
b
u
lly
in
g
.
T
h
e
lo
n
g
est
p
o
s
t
co
n
tain
ed
1
,
9
9
2
wo
r
d
s
,
an
d
th
e
s
h
o
r
test
co
n
tain
ed
s
ix
wo
r
d
s
.
W
e
d
eter
m
in
ed
wh
ich
ter
m
s
ap
p
ea
r
ed
m
o
s
t
o
f
ten
in
th
e
b
u
lly
in
g
an
d
n
o
n
-
b
u
lly
in
g
p
o
s
ts
.
W
e
illu
s
tr
ated
th
e
f
r
eq
u
en
cies
with
wh
ich
th
e
m
o
s
t
co
m
m
o
n
wo
r
d
s
o
cc
u
r
r
ed
b
y
co
n
s
tr
u
ctin
g
wo
r
d
clo
u
d
s
with
Py
th
o
n
as sh
o
wn
in
Fig
u
r
es 3
an
d
4
.
Fig
u
r
e
3
.
Mo
s
t f
r
eq
u
e
n
tly
u
s
e
d
b
u
lly
in
g
wo
r
d
s
Fig
u
r
e
4
.
Mo
s
t f
r
eq
u
e
n
tly
u
s
e
d
n
o
n
-
b
u
lly
i
n
g
wo
r
d
s
2
.
4
.
F
e
a
t
ures e
x
t
ra
c
t
io
n
W
e
u
s
ed
T
F
-
I
DF,
Ar
ab
ic
wo
r
d
2
v
ec
f
iles
,
Glo
Ve,
an
d
Fas
tTe
x
t
tech
n
iq
u
es
to
co
n
v
er
t
wo
r
d
s
an
d
p
h
r
ases
in
to
lo
w
-
d
im
en
s
io
n
al
f
ea
tu
r
e
v
ec
to
r
s
th
at
a
u
to
m
ate
d
an
aly
tical
to
o
ls
ca
n
u
s
e
[
3
0
]
.
Ap
p
l
y
in
g
n
eu
r
al
n
etwo
r
k
s
,
p
r
o
b
ab
ilis
tic
m
o
d
el
s
[
3
1
]
,
an
d
m
eth
o
d
s
b
ased
o
n
d
im
e
n
s
io
n
al
r
e
d
u
ctio
n
to
a
wo
r
d
c
o
-
f
r
e
q
u
en
c
y
m
atr
ix
ar
e
ex
am
p
les
o
f
wo
r
d
em
b
ed
d
in
g
tech
n
iq
u
es.
T
h
is
m
eth
o
d
ef
f
icien
tly
co
n
v
er
ts
tex
t
in
to
a
n
u
m
e
r
ical
f
o
r
m
th
at
m
ac
h
i
n
e
lear
n
in
g
al
g
o
r
ith
m
s
ca
n
h
an
d
le
an
d
ex
a
m
in
e
with
ea
s
e.
W
e
u
s
ed
T
F
-
I
DF
in
th
e
m
ac
h
in
e
lear
n
in
g
ex
p
er
im
en
t
an
d
em
p
l
o
y
ed
T
F
-
I
DF,
Ar
ab
ic
wo
r
d
2
v
ec
f
ile,
Glo
Ve,
an
d
Fas
tTe
x
t
in
th
e
ex
p
er
im
e
n
t
to
f
ac
ilit
ate
th
e
u
s
e
o
f
d
ee
p
lear
n
in
g
m
o
d
els.
‒
Ap
p
ly
T
F
-
I
DF:
we
em
p
lo
y
e
d
th
e
T
F
-
I
DF
v
ec
to
r
izer
to
tr
a
n
s
f
o
r
m
th
e
r
aw
tex
t
d
ata
in
to
n
u
m
er
ical
f
ea
tu
r
e
v
ec
to
r
s
.
W
e
u
s
ed
th
e
s
cik
it
-
lear
n
T
F
-
I
DF
weig
h
tin
g
tech
n
iq
u
e.
T
F
-
I
DF
is
a
s
tatic
m
ea
s
u
r
em
en
t
th
at
ca
n
b
e
ac
q
u
ir
e
d
as in
(
1
)
an
d
(
2
)
[
2
7
]
.
TF
-
I
DF
(
t,d
)
=
(
,
)
×
(
,
)
(
1
)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
E
n
h
a
n
ce
d
A
r
a
b
ic
-
l
a
n
g
u
a
g
e
cy
b
erb
u
llyin
g
d
etec
tio
n
:
d
ee
p
e
mb
ed
d
in
g
a
n
d
…
(
E
b
tesa
m
Ja
b
er A
l
jo
h
a
n
i
)
2263
I
DF
(
t)
=
(
)
(
)
+
1
(
2
)
T
o
r
ed
u
ce
co
m
p
u
tatio
n
al
co
m
p
lex
ity
,
we
lim
ited
t
h
e
v
o
ca
b
u
lar
y
s
ize
to
1
0
0
f
ea
tu
r
es.
W
e
co
n
v
e
r
ted
th
e
TF
-
I
DF v
ec
to
r
s
to
s
p
ar
s
e
Nu
m
Py
ar
r
ay
s
f
o
r
f
u
r
th
er
p
r
o
ce
s
s
in
g
an
d
c
o
n
s
tr
u
cted
b
ig
r
am
s
f
r
o
m
th
e
p
o
s
ts
.
‒
Ap
p
ly
Ar
awo
r
d
2
Vec
:
we
u
s
ed
p
r
e
-
tr
ain
e
d
Ar
ab
ic
wo
r
d
2
v
ec
em
b
ed
d
in
g
s
to
co
n
v
er
t
th
e
tex
t
d
ata
in
to
d
en
s
e
v
ec
to
r
r
ep
r
esen
tatio
n
s
.
W
e
lo
ad
ed
th
e
wo
r
d
2
v
ec
m
o
d
el
f
r
o
m
a
p
r
e
-
tr
ain
e
d
f
ile
(
twee
ts
_
cb
o
w_
3
0
0
)
.
T
h
e
ty
p
e
ap
p
lies
co
n
tin
u
es
B
o
W
.
T
o
m
an
ag
e
m
em
o
r
y
u
s
a
g
e,
we
tr
u
n
ca
ted
th
e
s
eq
u
e
n
ce
s
to
a
m
ax
im
u
m
len
g
th
em
b
ed
d
in
g
d
im
in
u
tio
n
o
f
1
0
0
.
W
e
to
k
en
ized
th
e
te
x
t
d
ata
in
b
atch
es
an
d
co
n
v
er
ted
ea
ch
wo
r
d
in
to
its
co
r
r
esp
o
n
d
in
g
wo
r
d
2
v
ec
e
m
b
ed
d
in
g
.
‒
Ap
p
ly
Glo
Ve:
we
lo
ad
e
d
p
r
e
-
t
r
ain
ed
Glo
Ve
wo
r
d
em
b
ed
d
in
g
s
f
r
o
m
a
f
ile
(
/co
n
ten
t/v
ec
to
r
s
.
tx
t)
an
d
s
to
r
ed
th
em
in
a
d
ictio
n
ar
y
(
em
b
ed
d
i
n
g
s
_
in
d
ex
)
.
W
e
cr
ea
ted
an
em
b
ed
d
in
g
m
atr
ix
b
ased
o
n
th
e
d
ataset’
s
lex
ico
n
an
d
th
e
lo
a
d
ed
Glo
Ve
em
b
ed
d
in
g
s
.
‒
Ap
p
ly
Fas
tTe
x
t:
we
u
s
ed
p
r
e
-
tr
ain
ed
Fas
tTe
x
t
to
g
en
er
at
e
wo
r
d
e
m
b
ed
d
in
g
s
f
o
r
Ar
a
b
ic
wo
r
d
s
.
W
e
d
o
wn
lo
ad
e
d
a
n
d
lo
ad
ed
th
e
F
astTex
t
m
o
d
el
(
cc
.
ar
.
3
0
0
.
b
in
)
.
T
h
ey
u
tili
ze
d
th
is
m
o
d
el
t
o
c
o
n
v
er
t
th
e
wo
r
d
s
in
th
e
p
o
s
ts
in
to
wo
r
d
e
m
b
ed
d
in
g
s
.
T
h
e
p
r
ep
r
o
ce
s
s
_
d
ata
f
u
n
ctio
n
to
k
en
ized
an
d
p
a
d
d
ed
t
h
e
in
p
u
t
co
m
m
en
ts
u
s
in
g
th
e
Fas
tTe
x
t m
o
d
el
an
d
m
ax
_
len
g
th
_
co
m
m
en
ts
9
0
to
m
a
n
ag
e
m
em
o
r
y
u
s
ag
e.
2
.
5
.
M
o
del im
plem
ent
a
t
io
n
T
h
is
s
ec
tio
n
ex
p
lain
s
th
e
m
o
d
els
we
test
ed
in
o
u
r
ex
p
er
im
en
ts
.
T
h
e
Ar
ab
ic
lan
g
u
ag
e
h
as
a
co
m
p
licated
s
y
n
tactic
an
d
m
o
r
p
h
o
lo
g
ical
s
tr
u
ctu
r
e
t
h
at
m
ak
es
it
b
o
th
ch
allen
g
in
g
a
n
d
s
e
m
an
tically
r
ich
[
3
2
]
.
Su
p
er
v
is
ed
m
ac
h
i
n
e
lear
n
in
g
i
s
wid
ely
u
s
ed
f
o
r
class
if
icatio
n
.
A
m
o
d
el
o
r
lear
n
er
is
f
ir
s
t
t
r
ain
ed
with
lab
eled
d
ata.
T
h
e
m
o
d
el
is
th
en
test
ed
b
y
b
ein
g
m
ad
e
t
o
class
if
y
s
am
p
le
d
ata
[
3
3
]
.
Dee
p
lear
n
in
g
alg
o
r
ith
m
s
h
av
e
s
u
cc
ess
f
u
lly
u
s
ed
d
atasets
th
at
ar
e
lar
g
e,
d
im
en
s
io
n
al,
an
d
well
-
o
r
g
an
ized
.
I
n
th
is
s
tu
d
y
,
two
d
ee
p
lear
n
in
g
m
o
d
els
(
L
STM
a
n
d
B
i
-
L
STM
)
with
em
b
ed
d
in
g
a
n
d
h
y
b
r
id
d
ee
p
tr
an
s
f
o
r
m
er
lear
n
in
g
class
if
ier
s
in
a
d
ee
p
n
eu
r
al
n
etwo
r
k
class
if
ied
d
ata
u
s
in
g
t
h
e
v
a
r
ity
B
E
R
T
Ar
ab
i
c
-
b
ased
p
r
e
-
tr
ain
ed
m
o
d
el
f
r
o
m
th
e
h
u
g
g
in
g
f
ac
e
lib
r
ar
y
.
W
e
co
m
p
ar
ed
th
ese
m
o
d
els’
p
er
f
o
r
m
an
ce
s
with
th
o
s
e
o
f
b
aselin
e
m
ac
h
in
e
m
e
th
o
d
s
.
T
h
e
b
ass
lin
e
m
o
d
els
o
f
m
ac
h
in
e
lear
n
in
g
p
r
o
p
o
s
ed
u
s
in
g
th
e
b
est
-
p
er
f
o
r
m
in
g
m
o
d
els
we
f
o
u
n
d
in
t
h
e
r
elev
an
t
liter
atu
r
e
(
R
F,
KNN,
lin
ea
r
SVM,
L
R
,
an
d
DT
im
p
le
m
en
ted
in
t
h
e
Py
th
o
n
wo
r
k
in
g
e
n
v
ir
o
n
m
en
t)
.
B
y
ex
tr
ac
tin
g
f
ea
tu
r
es
f
r
o
m
th
e
Ar
ab
ic
-
la
n
g
u
ag
e
p
o
s
ts
with
T
f
id
f
Vec
to
r
i
ze
(
)
m
eth
o
d
a
n
d
co
n
s
tr
u
ctin
g
b
ig
r
am
s
.
Als
o
,
we
co
m
p
ar
e
p
r
o
p
o
s
ed
m
o
d
els
with
th
e
b
aselin
e
tr
an
s
f
o
r
m
er
m
o
d
el.
T
h
e
f
o
llo
win
g
s
ec
tio
n
s
p
r
o
v
id
e
an
o
v
e
r
v
iew
o
f
th
e
m
o
d
els.
2
.
5
.
1
.
L
o
ng
s
ho
rt
-
t
er
m m
e
mo
ry
Sp
ec
ial
k
in
d
s
o
f
r
ec
u
r
r
en
t
n
e
u
r
al
n
etwo
r
k
s
(
R
NN)
an
d
ti
m
e
-
R
NN
ca
n
p
r
o
ce
s
s
an
d
f
o
r
ec
ast
tim
e
s
er
ies
wh
ile
av
o
id
in
g
is
s
u
e
s
with
lo
n
g
-
ter
m
r
elian
ce
.
W
e
r
ep
lace
d
th
e
L
STM
’
s
b
u
r
ied
lay
er
n
eu
r
o
n
s
with
a
d
is
tin
ct
s
et
o
f
m
em
o
r
y
ce
lls
.
R
NN,
an
d
it
s
m
em
o
r
y
ce
ll
s
tatu
s
,
is
cr
u
cial.
T
h
e
L
S
T
M
m
o
d
el
f
ilter
s
in
f
o
r
m
atio
n
.
I
ts
d
o
o
r
s
tr
u
ctu
r
e
co
n
s
is
ts
o
f
o
u
t
p
u
t,
in
p
u
t,
a
n
d
f
o
r
g
o
tten
g
ates
[
3
4
]
.
Fig
u
r
e
5
ex
p
lain
s
th
e
g
ate
s
tr
u
ctu
r
e
u
s
ed
to
u
p
d
ate
a
n
d
p
r
eser
v
e
m
em
o
r
y
ce
ll st
ates.
2
.
5
.
2
.
B
idi
re
ct
io
na
l lo
ng
s
ho
rt
-
t
er
m m
e
m
o
ry
T
h
e
L
STM
m
o
d
el’
s
m
em
o
r
y
ce
lls
ar
e
lim
ited
to
u
s
in
g
in
f
o
r
m
atio
n
f
r
o
m
th
e
p
ast
an
d
c
an
n
o
t
u
s
e
in
f
o
r
m
atio
n
f
r
o
m
th
e
p
r
esen
t.
B
i
-
L
STM
p
r
o
v
id
es
a
s
o
lu
tio
n
to
th
is
is
s
u
e.
B
i
-
L
STM
ca
n
b
e
th
o
u
g
h
t
o
f
as
two
s
ep
ar
ate
L
STM
s
tr
u
ctu
r
es
wit
h
two
in
p
u
ts
an
d
two
h
id
d
e
n
l
ev
els.
B
o
th
o
f
th
e
d
is
tin
ct
in
p
u
ts
em
p
lo
y
th
e
s
am
e
s
eq
u
en
ce
,
b
u
t
o
n
e
r
u
n
s
f
o
r
wa
r
d
wh
ile
th
e
o
th
er
r
u
n
s
b
ac
k
wa
r
d
.
T
h
e
h
id
d
en
lay
e
r
s
co
n
n
ec
t th
e
two
in
p
u
ts
,
an
d
ea
ch
s
tag
e’
s
o
u
tp
u
t is m
er
g
ed
[
3
4
]
as sh
o
wn
in
Fig
u
r
e
6
.
Fig
u
r
e
5
.
L
STM
m
o
d
el
Fig
u
r
e
6
.
B
i
-
L
STM
m
o
d
el
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
14
,
No
.
3
,
J
u
n
e
2
0
2
5
:
2
2
5
8
-
2
2
6
9
2264
2
.
5
.
3
.
T
ra
ns
f
o
rm
er
(
B
E
RT
)
B
E
R
T
u
s
es
tr
an
s
f
o
r
m
e
r
s
a
r
c
h
it
ec
t
u
r
es
,
a
n
att
e
n
ti
o
n
m
ec
h
an
is
m
t
h
a
t
r
e
c
o
g
n
i
ze
s
c
o
n
te
x
t
u
al
r
el
ati
o
n
s
h
i
p
s
b
etw
ee
n
w
o
r
d
s
(
o
r
s
u
b
-
wo
r
d
s
)
in
a
te
x
t
it.
B
E
R
T
’
s
m
u
ltil
ay
e
r
b
id
ir
ec
t
io
n
a
l
t
r
a
n
s
f
o
r
m
e
r
e
n
c
o
d
e
r
ar
ch
ite
ct
u
r
e
r
ese
m
b
les
t
h
e
t
r
a
n
s
f
o
r
m
er
co
n
ce
p
t
.
B
E
R
T
B
A
SE
co
m
p
r
is
es
tw
el
v
e
l
ev
els
.
T
h
e
e
n
c
o
d
e
r
s
t
ac
k
in
i
tial
ly
r
e
ce
i
v
es
t
h
e
C
L
S
t
o
k
e
n
as
in
p
u
t
b
ef
o
r
e
r
ec
ei
v
i
n
g
a
s
tr
i
n
g
o
f
w
o
r
d
s
as
i
n
p
u
t
.
C
L
S
i
s
t
h
e
ca
te
g
o
r
iza
ti
o
n
to
k
e
n
.
T
h
e
i
n
p
u
t
is
th
e
n
p
ass
e
d
t
o
th
e
la
y
e
r
s
a
b
o
v
e
.
E
v
e
r
y
l
ay
er
e
m
p
l
o
y
s
s
el
f
-
att
en
ti
o
n
,
t
r
an
s
f
e
r
s
t
h
e
o
u
tc
o
m
e
v
ia
a
f
e
ed
f
o
r
wa
r
d
n
etw
o
r
k
,
an
d
t
h
e
n
tr
an
s
f
e
r
s
co
n
t
r
o
l
t
o
t
h
e
s
u
b
s
e
q
u
e
n
t
e
n
c
o
d
e
r
.
T
h
e
m
o
d
el
o
u
t
p
u
ts
a
h
i
d
d
e
n
-
s
ize
v
ec
to
r
(
7
6
8
f
o
r
B
E
R
T
B
AS
E
)
[
3
5
]
.
T
h
e
r
e
a
r
e
n
ew
m
o
d
els
f
o
r
Ar
a
b
ic
d
ial
ec
ts
an
d
p
o
s
ts
,
i
n
c
lu
d
i
n
g
au
b
m
i
n
d
la
b
/
b
e
r
t
-
b
ase
-
A
r
a
b
e
r
t
v
0
2
an
d
au
b
m
i
n
d
la
b
/
b
e
r
t
-
b
a
s
e
-
l
ar
g
e
-
A
r
ab
er
tv
0
2
.
T
h
e
r
e
a
r
e
als
o
o
th
er
p
r
e
-
tr
a
in
e
d
la
n
g
u
a
g
e
m
o
d
els
,
s
u
c
h
as
C
AM
e
L
f
o
r
d
ia
le
cta
l
A
r
a
b
i
c
(
D
A)
,
a
n
d
a
m
o
d
el
p
r
e
-
t
r
ai
n
e
d
o
n
a
c
o
m
b
in
ati
o
n
o
f
cl
ass
ic
al
A
r
ab
ic
,
m
o
d
e
r
n
s
tan
d
ar
d
A
r
ab
ic
,
a
n
d
d
ia
lec
tal
Ar
ab
ic
(
Mi
x
)
.
T
h
es
e
tw
o
m
o
d
els
ar
e
k
n
o
wn
as
C
AM
eL
-
L
a
b
/
b
e
r
t
-
b
ase
-
a
r
a
b
i
c
-
ca
m
el
b
e
r
t
-
d
a
a
n
d
A
r
a
b
i
c
c
am
e
lb
e
r
t
-
m
i
x
,
r
es
p
e
cti
v
el
y
.
F
ig
u
r
e
s
7
a
n
d
8
s
h
o
w
th
e
s
tr
u
ctu
r
es o
f
th
e
B
E
R
T
m
o
d
el
an
d
th
e
h
y
b
r
i
d
d
ee
p
tr
an
s
f
o
r
m
er
(
B
iLST
M
-
B
E
R
T
)
m
o
d
el.
Fig
u
r
e
7
.
B
E
R
T
m
o
d
el
Fig
u
r
e
8
.
Hy
b
r
id
d
ee
p
tr
a
n
s
f
o
r
m
er
(
B
iLST
M
-
B
E
R
T
)
m
o
d
el
2
.
6
.
P
er
f
o
r
m
a
nce
ev
a
lua
t
io
n
W
e
m
ea
s
u
r
ed
th
e
ac
cu
r
ac
y
o
f
o
u
r
m
o
d
els
an
d
v
e
r
if
ied
th
eir
ef
f
ec
tiv
en
ess
b
y
tr
ain
i
n
g
t
h
em
o
n
X
p
o
s
ts
an
d
th
en
h
av
i
n
g
th
em
f
il
ter
an
d
o
r
g
an
ize
test
d
ata.
W
e
u
s
ed
ac
cu
r
ac
y
,
r
ec
all,
co
n
f
u
s
io
n
m
atr
ix
,
a
n
d
th
e
F1
-
m
ea
s
u
r
e
to
ass
ess
ea
ch
class
if
ier
’
s
o
u
tp
u
t.
T
h
e
r
esu
lts
o
f
th
is
s
tu
d
y
ca
n
ea
s
ily
b
e
s
u
m
m
ar
ized
u
s
in
g
a
co
n
f
u
s
io
n
m
atr
ix
[
2
7
]
,
w
h
er
e
t
r
u
e
an
d
f
alse r
ep
r
esen
t r
elate
d
an
d
u
n
r
elate
d
,
T
P a
n
d
T
N
r
ep
r
esen
t tr
u
e
p
o
s
itiv
e
an
d
tr
u
e
n
eg
ativ
e,
an
d
FP
an
d
FN
r
ep
r
esen
t
f
alse
p
o
s
itiv
e
an
d
f
alse
n
eg
ativ
e.
In
(
3
)
,
(
4
)
,
an
d
(
5
)
ar
e
u
s
ed
to
ca
lcu
late
ac
cu
r
ac
y
,
p
r
ec
is
io
n
,
an
d
r
ec
all
[
2
7
]
.
A
c
c
ura
c
y
=
+
+
+
+
(
3
)
=
+
(
4
)
=
+
(
5
)
T
h
e
F1
-
s
co
r
e
[
3
1
]
is
ca
lcu
late
d
as th
e
(
weig
h
ted
)
h
ar
m
o
n
ic
av
er
ag
e
o
f
p
r
ec
is
io
n
an
d
r
ec
all
(
6
)
.
F1
−
s
c
or
e
=
2
×
×
+
(
6
)
3.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
T
h
is
s
ec
tio
n
p
r
esen
ts
th
e
ex
p
er
im
en
ts
’
s
ettin
g
s
an
d
th
e
ex
p
er
im
en
tal
r
esu
lts
o
f
m
ac
h
in
e
lear
n
in
g
,
d
ee
p
lear
n
in
g
m
o
d
els
(
d
ee
p
e
m
b
ed
d
in
g
s
an
d
h
y
b
r
id
d
ee
p
tr
an
s
f
o
r
m
er
s
)
,
an
d
tr
an
s
f
o
r
m
e
r
o
u
tco
m
es.
All
o
f
th
e
ex
p
er
im
en
ts
wer
e
ev
al
u
ated
u
s
in
g
ac
cu
r
ac
y
,
p
r
ec
is
io
n
,
a
n
d
r
ec
all.
3
.
1
.
E
x
perim
ent
s
et
t
ing
s
T
h
is
p
ar
t
ex
p
lain
s
o
u
r
ex
p
er
i
m
en
tal
s
etu
p
.
W
e
u
s
ed
Go
o
g
le
C
o
lab
Pro
,
Py
th
o
n
3
.
8
,
an
d
a
GPU
with
5
0
GB
o
f
R
AM
.
Mo
s
t
NL
P
task
s
r
eq
u
ir
e
v
ar
io
u
s
lib
r
ar
i
es,
s
u
ch
as
s
cik
it
-
lear
n
.
Mo
s
t
tr
an
s
f
o
r
m
er
-
b
ased
lear
n
in
g
ap
p
r
o
ac
h
es
h
av
e
em
p
lo
y
ed
d
i
f
f
er
en
t
B
E
R
T
-
b
ased
Ar
ab
ic
p
r
e
-
tr
ain
e
d
m
o
d
els
a
n
d
h
u
g
g
in
g
f
ac
e’
s
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
E
n
h
a
n
ce
d
A
r
a
b
ic
-
l
a
n
g
u
a
g
e
cy
b
erb
u
llyin
g
d
etec
tio
n
:
d
ee
p
e
mb
ed
d
in
g
a
n
d
…
(
E
b
tesa
m
Ja
b
er A
l
jo
h
a
n
i
)
2265
tr
an
s
f
o
r
m
er
lib
r
a
r
y
.
I
n
o
r
d
e
r
to
r
em
o
v
e
b
ias,
we
d
iv
id
ed
th
e
d
ataset,
allo
ca
tin
g
8
0
%
f
o
r
tr
ain
in
g
th
e
m
o
d
els
an
d
2
0
%
f
o
r
test
in
g
th
em
.
I
n
a
ll
o
f
th
e
ex
p
er
im
e
n
ts
,
we
co
n
s
tr
u
cted
th
e
m
o
d
els
u
s
in
g
Ker
a
s
s
eq
u
en
tial
m
o
d
el
an
d
co
m
p
iled
u
s
in
g
th
e
b
in
a
r
y
cr
o
s
s
-
en
tr
o
p
y
lo
s
s
f
u
n
ctio
n
an
d
th
e
A
d
am
o
p
tim
izer
.
W
e
em
p
lo
y
ed
s
o
m
e
lay
er
s
co
n
tain
in
g
th
e
r
ec
tifie
d
lin
ea
r
u
n
it
(
R
eL
U
)
ac
tiv
atio
n
f
u
n
ctio
n
in
ce
r
tain
m
o
d
els
B
i
-
L
STM
an
d
L
STM
.
W
e
m
o
n
ito
r
ed
th
e
tr
ai
n
in
g
p
r
o
ce
s
s
an
d
ev
alu
ated
th
e
m
o
d
els’
p
e
r
f
o
r
m
an
ce
in
ter
m
s
o
f
ac
c
u
r
ac
y
.
T
h
e
f
o
llo
win
g
s
ec
tio
n
s
d
etail
th
e
s
t
r
u
ctu
r
e
o
f
o
u
r
ex
p
er
im
en
ts
.
3
.
2
.
E
x
perim
ent
s
re
s
ults
T
h
e
r
esu
lt
o
f
th
e
e
x
p
er
im
en
t
o
f
m
o
s
t
co
m
m
o
n
m
ac
h
i
n
e
lea
r
n
in
g
class
if
ier
s
,
d
ee
p
lear
n
in
g
m
o
d
els,
tr
an
s
f
o
r
m
er
s
an
d
p
r
o
p
o
s
ed
m
o
d
els
h
av
e
b
ee
n
co
m
p
a
r
ed
in
o
r
d
er
to
ev
alu
ate
th
e
p
er
f
o
r
m
a
n
ce
o
f
th
e
p
r
o
p
o
s
ed
m
o
d
els.
T
ab
le
4
co
n
tain
th
e
d
etailed
r
esu
lts
f
r
o
m
all
f
o
u
r
e
x
p
er
im
en
ts
.
W
e
co
m
p
ar
ed
th
e
m
o
d
els’
p
er
f
o
r
m
an
ce
s
.
T
a
b
l
e
4
.
C
o
m
p
a
r
i
s
o
n
b
e
t
w
e
e
n
t
h
e
a
c
c
u
r
a
c
y
o
f
t
h
e
m
a
c
h
i
n
e
l
e
a
r
n
i
n
g
m
o
d
e
l
s
,
t
r
a
n
s
f
o
r
m
e
r
,
a
n
d
t
h
e
p
r
o
p
o
s
e
d
m
o
d
e
l
s
M
o
d
e
l
s
R
e
c
a
l
l
(
%)
P
r
e
c
i
s
i
o
n
(
%)
F1
-
sc
o
r
e
(
%)
A
c
c
u
r
a
c
y
(
%)
B
a
se
l
i
n
e
ma
c
h
i
n
e
l
e
a
r
n
i
n
g
S
V
M
8
6
.
5
8
7
.
6
8
8
.
0
87
LR
8
2
.
5
8
5
.
4
8
2
.
0
83
RF
7
5
.
9
8
0
.
9
7
4
.
6
76
K
N
N
8
5
.
7
8
6
.
2
8
5
.
5
86
DT
8
2
.
4
8
5
.
5
8
2
.
1
82
P
1
:
D
e
e
p
+
w
o
r
d
e
m
b
e
d
d
i
n
g
s
LSTM
-
TFI
D
F
9
8
.
0
5
7
.
4
7
2
.
4
60
B
LST
M
-
TFI
D
F
9
7
.
3
5
7
.
0
7
1
.
9
59
LSTM
-
G
l
o
V
e
8
2
.
6
8
3
.
2
8
2
.
9
82
B
LST
M
-
G
l
o
V
e
7
5
.
0
8
7
.
8
8
0
.
9
81
LSTM
-
A
r
W
o
r
2
V
e
c
6
9
.
6
9
7
.
4
8
1
.
2
83
B
LST
M
-
A
r
W
o
r
2
V
e
c
9
5
.
4
9
8
.
4
9
6
.
9
97
LSTM
-
F
a
s
t
t
e
x
t
8
4
.
5
8
8
.
7
8
6
.
6
86
Bi
-
LST
M
-
F
a
st
t
e
x
t
9
6
.
5
9
9
.
7
9
8
.
1
98
Tr
a
n
sf
o
r
mer (
B
ER
T)
B
ER
T
C
A
M
e
L
-
da
9
6
.
6
9
7
.
2
9
6
.
9
97
B
ER
T
C
A
M
e
L
-
mi
x
9
5
.
6
9
6
.
5
9
6
.
0
96
B
ER
T
A
r
a
b
e
r
t
v
0
2
9
5
.
4
9
5
.
0
9
5
.
2
95
B
ER
T
-
a
l
a
r
g
e
-
A
r
a
b
e
r
t
v
0
2
8
9
.
2
9
5
.
1
9
2
.
1
92
P
2
:
H
y
b
r
i
d
d
e
e
p
t
r
a
n
sf
o
r
m
e
r
LSTM
-
B
ER
T
C
A
M
e
L
-
da
9
6
.
3
9
7
.
2
9
6
.
8
97
Bi
-
LST
M
-
B
ER
T
C
A
M
e
L
-
da
9
6
.
8
9
7
.
1
9
7
.
0
97
LSTM
-
B
ER
T
A
r
a
b
e
r
t
v
0
2
9
5
.
6
9
5
.
2
9
5
.
4
95
Bi
-
LST
M
-
B
ER
TA
r
a
b
e
r
t
v
0
2
9
5
.
2
9
6
.
4
9
5
.
8
96
3
.
3
.
Resul
t
s
dis
cu
s
s
io
n a
nd
a
na
ly
s
is
T
h
e
r
esu
lts
o
f
th
e
b
aselin
e
m
ac
h
in
e
lear
n
in
g
m
o
d
els
a
r
e
s
h
o
wn
in
T
ab
le
4
.
T
f
id
f
Vec
to
r
ize(
)
co
n
s
tr
u
cted
b
ig
r
am
s
f
r
o
m
t
h
e
p
o
s
ts
f
o
r
t
h
e
m
o
d
els
with
ce
r
tain
s
ettin
g
s
to
p
r
o
d
u
ce
th
ese
r
esu
lts
.
T
h
e
lin
ea
r
s
u
p
p
o
r
t v
ec
to
r
class
if
icatio
n
(
SVC
)
m
o
d
el
p
er
f
o
r
m
e
d
th
e
b
e
s
t.
T
h
e
KNN,
L
R
,
an
d
DT
m
o
d
els h
ad
8
6
%,
8
3
%
,
an
d
8
2
%
ac
cu
r
ac
y
,
r
esp
ec
tiv
el
y
.
R
F
h
ad
th
e
lo
west
ac
cu
r
ac
y
r
ate
(
7
6
%).
T
h
e
b
est
r
esu
lt
was
o
b
tain
ed
f
o
r
th
e
lin
ea
r
SVC
m
o
d
el.
T
h
ese
r
esu
lts
wer
e
n
o
t
s
u
p
er
io
r
to
th
o
s
e
o
f
th
e
o
th
e
r
ex
p
er
im
e
n
ts
.
T
h
e
r
esu
lts
in
T
ab
le
4
s
h
o
w
th
at
th
e
d
ee
p
lear
n
in
g
m
o
d
els
c
o
m
b
in
ed
with
em
b
ed
d
in
g
r
ep
r
esen
tatio
n
s
o
f
th
e
tr
ain
in
g
wo
r
d
s
i
n
Ar
ab
ic
p
o
s
ts
h
ad
e
x
ce
llen
t
p
e
r
f
o
r
m
an
ce
.
T
h
e
L
STM
a
n
d
B
I
-
L
STM
m
o
d
els
th
at
u
s
ed
T
F
-
I
DF
d
id
n
o
t
p
r
o
d
u
ce
o
u
ts
tan
d
in
g
r
esu
lts
.
C
o
m
b
in
in
g
th
e
Ar
a
b
ic
em
b
ed
d
in
g
Ar
a
wo
r
d
2
Vec
with
B
i
-
L
STM
ac
h
iev
ed
a
n
ex
ce
llen
t
ac
cu
r
ac
y
r
ate
o
f
9
7
%,
an
d
c
o
m
b
in
in
g
it
with
L
STM
ac
h
iev
ed
a
g
o
o
d
r
esu
lt
o
f
8
3
%.
L
STM
co
m
b
in
ed
with
Glo
V
e
ac
h
iev
ed
8
2
%
ac
cu
r
ac
y
,
an
d
th
e
co
m
b
in
ed
B
i
-
L
ST
M
-
Glo
V
e
m
o
d
el
attain
ed
8
1
%,
wh
ich
was
n
o
t
as
g
o
o
d
as th
e
s
am
e
m
o
d
el
with
t
h
e
Ar
awo
r
d
2
Vec
em
b
ed
d
in
g
.
W
e
co
m
b
in
ed
th
e
m
o
d
els with
th
e
m
o
d
er
n
em
b
e
d
d
in
g
Fas
tTe
x
t,
wh
ich
d
if
f
e
r
s
f
r
o
m
Ar
aw
o
r
d
2
Vec
in
th
at
it
s
p
lit
s
wo
r
d
s
in
to
s
m
all
wo
r
d
s
o
r
n
-
g
r
a
m
s
.
T
h
e
B
i
-
L
S
T
M
m
o
d
el
with
Fas
tTe
x
t
ac
h
i
ev
ed
a
n
o
tab
ly
h
ig
h
ac
cu
r
ac
y
o
f
9
8
%
c
o
m
p
ar
i
n
g
with
p
r
ev
io
u
s
s
tu
d
y
as
[
2
0
]
,
[
2
3
]
.
T
h
e
L
STM
m
o
d
el
with
F
astTex
t
ac
h
iev
ed
an
ac
cu
r
ac
y
r
ate
o
f
8
6
%.
T
h
e
B
i
-
L
STM
m
o
d
el
o
u
tp
e
r
f
o
r
m
ed
o
f
th
e
L
STM
m
o
d
el
with
b
o
t
h
th
e
Ar
awo
r
d
2
Vec
an
d
Fas
tTe
x
t
em
b
ed
d
in
g
s
.
T
h
e
f
in
d
in
g
s
o
n
th
e
B
i
-
L
T
SM
m
o
d
el
with
Fas
tTe
x
t
a
r
e
d
etailed
in
th
e
class
if
icatio
n
r
ep
o
r
t sh
o
wn
in
Fig
u
r
e
9
an
d
th
e
c
o
n
f
u
s
io
n
m
a
tr
ix
s
h
o
wn
in
Fig
u
r
e
1
0
.
I
n
a
n
e
f
f
o
r
t
t
o
m
a
x
i
m
i
z
e
t
h
e
r
a
t
e
o
f
m
o
d
e
l
p
r
o
g
r
e
s
s
,
w
e
e
m
p
lo
y
e
d
a
t
r
a
n
s
f
o
r
m
e
r
m
o
d
e
l
p
r
e
-
t
r
a
i
n
e
d
o
n
A
r
a
b
i
c
-
l
a
n
g
u
a
g
e
X
p
o
s
ts
(
A
r
ab
b
e
r
t
v
0
2
)
a
n
d
a
n
o
t
h
e
r
p
r
e
-
t
r
a
i
n
e
d
o
n
a
v
a
r
i
e
t
y
o
f
d
i
a
l
e
ct
s
(
C
A
Me
L
-
d
a
)
i
n
t
r
i
als
c
o
n
d
u
c
t
e
d
w
i
t
h
B
E
R
T
t
r
a
n
s
f
o
r
m
e
r
m
o
d
e
l
s
.
T
h
e
m
o
d
e
l
w
it
h
C
A
Me
L
-
d
a
p
r
o
d
u
c
e
d
t
h
e
b
e
s
t
r
e
s
u
l
ts
i
n
t
h
i
s
e
x
p
e
r
i
m
e
n
t
:
9
7
%
a
c
c
u
r
a
c
y
(
c
o
m
p
a
r
a
b
l
e
t
o
t
h
e
B
i
-
L
S
T
M
-
A
r
w
o
d
2
v
e
c
m
o
d
e
l
’
s
a
c
c
u
r
a
c
y
)
.
I
n
t
h
e
f
i
n
a
l
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
14
,
No
.
3
,
J
u
n
e
2
0
2
5
:
2
2
5
8
-
2
2
6
9
2266
e
x
p
e
r
i
m
e
n
t
,
w
e
m
e
r
g
e
d
t
h
e
m
o
d
e
l
s
t
h
at
h
a
d
a
c
h
ie
v
e
d
h
i
g
h
ac
c
u
r
a
c
y
i
n
t
h
e
e
a
r
l
i
e
r
e
x
p
e
r
i
m
en
t
s
wi
t
h
L
S
T
M
a
n
d
w
i
t
h
B
i
-
L
S
T
M
.
T
h
e
e
x
p
e
r
i
m
en
t
s
d
e
m
o
n
s
t
r
a
t
e
d
t
h
at
t
h
e
m
o
d
e
l
s
w
i
t
h
B
i
-
L
S
T
M
m
a
y
p
e
r
f
o
r
m
b
e
t
t
e
r
t
h
a
n
o
t
h
e
r
s
,
p
a
r
t
i
c
u
l
a
r
l
y
w
h
e
n
B
i
-
L
S
T
M
is
c
o
m
b
i
n
e
d
w
i
t
h
p
r
e
-
t
r
a
i
n
e
d
e
m
b
e
d
d
i
n
g
s
(
i
.
e
.
,
t
r
a
n
s
f
o
r
m
e
r
s
)
.
B
i
-
L
S
T
M
-
B
E
R
T
p
r
o
d
u
c
e
d
e
x
c
e
l
l
e
n
t
r
e
s
u
l
ts
:
9
7
%
a
c
c
u
r
a
c
y
.
T
h
e
L
S
T
M
a
n
d
B
i
-
L
S
T
M
m
o
d
e
l
s
w
i
t
h
C
A
M
eL
-
d
a
a
c
h
i
e
v
e
d
9
7
%
i
n
c
l
u
s
i
o
n
a
n
d
a
n
F
1
-
s
c
o
r
e
o
f
9
6
.
8
%
.
T
r
a
n
s
f
e
r
l
e
a
r
n
i
n
g
e
n
h
a
n
c
e
d
t
h
e
L
S
T
M
m
o
d
e
l
’
s
p
e
r
f
o
r
m
a
n
c
e
s
u
c
h
t
h
a
t
it
b
e
c
a
m
e
c
o
m
p
a
r
a
b
l
e
t
o
t
h
a
t
o
f
th
e
B
i
-
L
S
T
M
m
o
d
e
l
wi
t
h
a
d
e
ep
l
e
a
r
n
i
n
g
m
o
d
e
l
a
n
d
e
m
b
e
d
d
i
n
g
s
.
W
e
d
e
te
r
m
i
n
e
d
t
h
e
A
UC
s
c
o
r
e
s
o
f
t
h
e
b
es
t
-
p
e
r
f
o
r
m
i
n
g
m
o
d
e
l
s
a
s
s
h
o
w
n
i
n
Fig
u
r
e
s
1
1
a
n
d
1
2
.
T
h
e
f
i
n
d
i
n
g
s
d
e
m
o
n
s
t
r
a
t
e
d
t
h
at
:
‒
Bi
-
L
STM
-
Fas
tTe
x
t
h
ad
th
e
h
ig
h
est
ac
cu
r
ac
y
an
d
F1
-
s
co
r
e,
f
o
llo
wed
b
y
B
i
-
L
STM
-
B
E
R
T
,
B
i
-
L
STM
-
Ar
wo
r
d
2
v
ec
,
B
E
R
T
C
AM
eL
-
d
a,
an
d
L
STM
-
B
E
R
T
.
‒
C
o
m
b
in
in
g
th
e
t
r
an
s
f
o
r
m
e
r
B
E
R
T
m
o
d
el
with
C
AM
eL
-
d
a
o
r
Ar
ab
b
er
t0
v
2
im
p
r
o
v
ed
th
e
L
STM
m
o
d
el’
s
o
u
tco
m
es
an
d
m
a
r
g
in
ally
im
p
r
o
v
e
d
B
i
-
L
STM
’
s
p
er
f
o
r
m
a
n
ce
to
9
7
%
a
n
d
ac
ce
le
r
ated
its
tr
ain
in
g
(
we
s
to
p
p
ed
th
e
h
y
b
r
i
d
B
i
-
L
STM
-
B
E
R
T
C
AM
eL
-
d
a
m
o
d
el’
s
tr
a
in
in
g
ea
r
ly
at
1
2
e
p
o
ch
)
.
‒
T
h
e
s
tu
d
y
'
s
d
em
o
n
s
tr
ated
wh
e
n
h
y
b
r
id
d
ee
p
lear
n
in
g
tr
an
s
f
o
r
m
er
s
an
d
Fas
tTe
x
t
-
em
b
e
d
d
in
g
d
ee
p
m
o
d
els
ca
n
b
e
ap
p
lied
th
ey
o
u
tp
e
r
f
o
r
m
o
n
b
aselin
e
m
ac
h
in
e
lear
n
in
g
m
o
d
els.
T
h
e
d
r
ew
lear
n
in
g
c
u
r
v
es o
f
B
i
-
L
STM
-
Fas
tTe
x
t in
Fig
u
r
e
1
3
an
d
th
e
h
y
b
r
id
tr
an
s
f
o
r
m
er
m
o
d
el
in
Fig
u
r
e
1
4
.
Fig
u
r
e
9
.
Fas
tTe
x
t
-
Bi
-
L
STM
’
s
class
if
y
r
ep
o
r
t
Fig
u
r
e
1
0
.
Fas
tTe
x
t
-
Bi
-
L
STM
’
s
Fig
u
r
e
1
1
.
B
i
-
L
STM
-
Fas
tTe
x
t
’
s
AUC
Fig
u
r
e
1
2
.
Hy
b
r
id
tr
an
s
f
o
r
m
er
L
STM
/B
i
-
L
STM
-
C
AM
eL
-
d
a’
s
AUC
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
E
n
h
a
n
ce
d
A
r
a
b
ic
-
l
a
n
g
u
a
g
e
cy
b
erb
u
llyin
g
d
etec
tio
n
:
d
ee
p
e
mb
ed
d
in
g
a
n
d
…
(
E
b
tesa
m
Ja
b
er A
l
jo
h
a
n
i
)
2267
Fig
u
r
e
1
3
.
B
i
-
L
STM
-
Fas
tTe
x
t
’
s
lear
n
in
g
cu
r
v
e
Fig
u
r
e
1
4
.
Hy
b
r
id
tr
an
s
f
o
r
m
er
B
i
-
L
STM
-
C
AM
eL
-
d
a’
s
lear
n
in
g
cu
r
v
e
4.
CO
NCLU
SI
O
N
T
h
i
s
s
t
u
d
y
f
o
c
u
s
es
o
n
t
h
e
d
et
e
c
t
i
o
n
o
f
c
y
b
e
r
b
u
l
l
y
i
n
g
o
n
s
o
c
i
a
l
m
e
d
i
a
,
p
a
r
t
i
c
u
l
a
r
l
y
o
n
th
e
A
r
a
b
ic
l
a
n
g
u
a
g
e
.
B
y
p
r
o
p
o
s
i
n
g
t
w
o
d
e
e
p
l
e
a
r
n
i
n
g
m
o
d
e
ls
,
t
h
e
f
i
r
s
t
m
o
d
e
l
is
a
B
i
-
L
S
T
M
-
F
as
tT
e
x
t
m
o
d
e
l
t
h
a
t
ca
n
r
e
c
o
g
n
i
z
e
A
r
a
b
i
c
-
l
a
n
g
u
a
g
e
b
u
l
l
y
i
n
g
p
o
s
ts
w
it
h
9
8
%
a
c
c
u
r
ac
y
.
W
e
c
o
m
p
a
r
e
d
t
h
e
e
x
p
e
r
im
e
n
t
a
l
r
e
s
u
l
t
s
w
i
t
h
d
i
f
f
e
r
e
n
t
b
a
s
el
i
n
e
m
o
d
e
ls
t
o
d
e
t
e
r
m
i
n
e
t
h
ei
r
a
c
c
u
r
a
c
y
r
at
e
s
.
T
h
e
s
e
c
o
n
d
m
o
d
e
l
i
s
a
c
o
m
b
in
a
t
i
o
n
o
f
B
i
-
L
S
T
M
-
B
E
R
T
a
n
d
L
S
T
M
-
B
E
R
T
m
o
d
e
l
s
,
b
o
t
h
a
c
h
i
e
v
e
d
s
a
t
is
f
a
ct
o
r
y
o
u
t
c
o
m
e
s
w
h
i
c
h
is
9
7
%
a
c
c
u
r
a
c
y
w
h
e
n
t
h
e
o
b
j
e
c
t
i
v
e
w
as
t
o
b
o
o
s
t
t
h
e
i
m
p
r
o
v
e
m
e
n
t
.
T
h
e
e
x
p
e
r
i
m
e
n
t
a
l
r
e
s
u
l
t
s
d
e
m
o
n
s
t
r
a
t
e
d
t
h
a
t
t
h
e
h
y
b
r
i
d
p
r
e
-
t
r
a
i
n
e
d
e
m
b
e
d
d
i
n
g
m
o
d
e
l
s
B
E
R
T
w
i
t
h
d
e
e
p
l
e
a
r
n
i
n
g
m
o
d
e
l
s
(
L
S
T
M
,
B
i
-
L
S
T
M
)
o
u
t
p
e
r
f
o
r
m
e
d
t
h
e
b
a
s
e
l
i
n
e
m
o
d
e
ls
.
I
n
a
d
d
i
t
i
o
n
,
t
h
e
n
o
v
e
l
d
at
a
s
e
t
is
i
n
t
r
o
d
u
c
e
d
w
h
i
c
h
c
o
n
s
i
s
ts
o
f
a
c
o
l
l
e
c
ti
o
n
w
h
i
c
h
wa
s
l
i
m
it
e
d
b
e
ca
u
s
e
it
o
n
l
y
a
p
p
l
i
ed
g
e
n
e
r
a
l
l
y
t
o
A
r
a
b
i
c
i
n
g
e
n
e
r
a
l
r
a
t
h
e
r
t
h
a
n
t
o
a
s
p
e
c
i
f
i
c
d
i
a
l
e
ct
.
F
u
t
u
r
e
r
e
s
e
a
r
c
h
c
o
u
l
d
u
s
e
d
a
t
ase
t
s
w
h
o
s
e
s
i
z
e
a
n
d
p
l
a
t
f
o
r
m
o
f
o
r
i
g
i
n
a
r
e
d
i
f
f
e
r
e
n
t f
r
o
m
o
u
r
s
t
o
as
s
ess
t
h
e
m
o
d
e
ls
’
p
e
r
f
o
r
m
a
n
c
e
o
n
o
t
h
e
r
A
r
a
b
i
c
N
L
P
ta
s
k
s
,
s
u
c
h
a
s
W
h
a
ts
A
p
p
o
r
o
t
h
e
r
s
o
c
i
a
l
m
e
d
i
a
n
e
t
w
o
r
k
s
.
F
u
r
t
h
e
r
m
o
r
e
,
r
e
s
ea
r
c
h
e
r
s
c
o
u
l
d
e
x
p
e
r
i
m
e
n
t
w
i
t
h
c
o
m
b
i
n
i
n
g
v
a
r
i
o
u
s
d
e
e
p
l
e
a
r
n
i
n
g
m
o
d
e
l
s
l
i
k
e
B
i
GR
U
wi
t
h
d
i
f
f
e
r
e
n
t
A
r
a
b
i
c
B
E
R
T
a
p
p
r
o
a
c
h
e
s
t
o
t
r
a
n
s
f
o
r
m
e
r
t
ec
h
n
i
q
u
e
s
a
n
d
a
s
s
ess
t
h
e
i
r
p
e
r
f
o
r
m
a
n
c
e
wi
t
h
o
u
r
d
a
t
a
s
et
a
n
d
t
h
e
n
i
n
t
e
g
r
a
t
e
t
h
o
s
e
c
o
m
b
i
n
a
ti
o
n
s
w
i
t
h
t
r
a
n
s
f
o
r
m
e
r
m
o
d
e
l
s
.
T
h
e
p
r
o
p
o
s
e
d
m
o
d
e
l
s
c
o
u
l
d
b
e
d
e
p
lo
y
e
d
a
s
t
o
o
l
s
o
n
s
o
c
i
a
l
m
e
d
i
a
to
d
e
t
e
c
t
A
r
a
b
ic
b
u
l
l
y
i
n
g
.
F
UNDING
I
NF
O
R
M
A
T
I
O
N
N
o
f
u
n
d
in
g
was in
v
o
lv
ed
.
AUTHO
R
CO
NT
RI
B
UT
I
O
NS ST
A
T
E
M
E
N
T
T
h
is
jo
u
r
n
al
u
s
es
th
e
C
o
n
tr
ib
u
to
r
R
o
les
T
ax
o
n
o
m
y
(
C
R
ed
iT)
to
r
ec
o
g
n
ize
in
d
iv
id
u
al
au
th
o
r
co
n
tr
ib
u
tio
n
s
,
r
ed
u
ce
au
th
o
r
s
h
ip
d
is
p
u
tes,
an
d
f
ac
ilit
ate
co
llab
o
r
atio
n
.
Evaluation Warning : The document was created with Spire.PDF for Python.