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.
1
5
,
No
.
3
,
J
u
n
e
2
0
2
6
,
p
p
.
2849
~
2
8
6
2
I
SS
N:
2
2
5
2
-
8
9
3
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijai.v
15
.i
3
.
p
p
2
8
4
9
-
2
8
6
2
2849
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
i
.
ia
esco
r
e.
co
m
H
y
brid conv
o
luti
o
na
l neural net
wo
rk
-
bidir
ectiona
l l
o
ng
sho
rt
-
term
me
mo
ry
m
o
del f
o
r
Ara
bic sen
tence rea
da
bility
a
ss
ess
ment
M
o
ha
m
ed
Am
ine O
ua
s
s
il
1
,
M
o
ha
m
m
ed
J
ebba
ri
1
,
Ra
bia
Ra
chidi
2
,
M
o
ua
a
d E
rr
a
m
i
1
,
So
ufia
ne
H
a
m
ida
3,
5
,
B
o
ucha
i
b Cher
ra
di
1,
2,
3,
4
,
Abdelh
a
di
Ra
iha
ni
1,
3
1
EEI
S
La
b
o
r
a
t
o
r
y
,
EN
S
ET
o
f
M
o
h
a
m
med
i
a
,
H
a
ss
a
n
I
I
U
n
i
v
e
r
si
t
y
o
f
C
a
s
a
b
l
a
n
c
a
,
M
o
h
a
mm
e
d
i
a
,
M
o
r
o
c
c
o
2
La
R
O
S
E
R
I
La
b
o
r
a
t
o
r
y
,
F
a
c
u
l
t
y
o
f
S
c
i
e
n
c
e
,
C
h
o
u
a
i
b
D
o
u
k
k
a
l
i
U
n
i
v
e
r
si
t
y
,
El
J
a
d
i
d
a
,
M
o
r
o
c
c
o
3
2
I
A
C
S
La
b
o
r
a
t
o
r
y
,
EN
S
ET
o
f
M
o
h
a
mm
e
d
i
a
,
H
a
ss
a
n
I
I
U
n
i
v
e
r
si
t
y
o
f
C
a
s
a
b
l
a
n
c
a
,
M
o
h
a
mm
e
d
i
a
,
M
o
r
o
c
c
o
4
S
TI
E
Te
a
m,
C
R
M
EF C
a
sa
b
l
a
n
c
a
,
P
r
o
v
i
n
c
i
a
l
S
e
c
t
i
o
n
o
f
El
Ja
d
i
d
a
,
E
l
J
a
d
i
d
a
,
M
o
r
o
c
c
o
5
G
EN
I
U
S
La
b
o
r
a
t
o
r
y
,
S
u
p
M
TI
o
f
R
a
b
a
t
,
R
a
b
a
t
,
M
o
r
o
c
c
o
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Sep
2
5
,
2
0
2
4
R
ev
is
ed
Ma
r
6
,
2
0
2
6
Acc
ep
ted
Ap
r
2
1
,
2
0
2
6
In
th
e
c
u
r
re
n
t
e
d
u
c
a
ti
o
n
a
l
lan
d
s
c
a
p
e
,
a
larg
e
n
u
m
b
e
r
o
f
e
d
u
c
a
to
rs
p
re
fe
r
u
sin
g
g
e
n
e
ra
ti
v
e
a
rti
ficia
l
i
n
telli
g
e
n
c
e
tec
h
n
iq
u
e
s
t
o
p
r
o
d
u
c
e
tex
tu
a
l
c
o
n
ten
t
to
b
e
p
re
se
n
ted
fo
r
lea
rn
in
g
.
Ho
we
v
e
r,
th
e
se
g
e
n
e
ra
ted
tex
ts
m
a
y
n
o
t
m
e
e
t
th
e
sp
e
c
ifi
c
n
e
e
d
s
o
f
lea
rn
e
rs
o
r
a
li
g
n
wit
h
th
e
ir
a
b
i
li
ti
e
s.
M
a
n
y
t
ra
d
it
io
n
a
l
m
e
th
o
d
s
a
n
d
tec
h
n
i
q
u
e
s
c
a
n
b
e
e
m
p
lo
y
e
d
to
a
ss
e
ss
th
e
c
o
m
p
lex
it
y
o
f
a
tex
t,
su
c
h
a
s
t
ra
d
it
i
o
n
a
l
re
a
d
a
b
il
i
ty
f
o
rm
u
las
,
b
u
t
t
h
e
se
tec
h
n
i
q
u
e
s
a
re
ti
m
e
-
c
o
n
su
m
in
g
a
n
d
lab
o
r
-
i
n
ten
siv
e
.
I
n
th
is
p
a
p
e
r,
we
in
tr
o
d
u
c
e
a
d
e
e
p
lea
rn
in
g
a
p
p
ro
a
c
h
fo
r
a
u
t
o
m
a
ti
c
a
ll
y
e
v
a
l
u
a
ti
n
g
t
h
e
re
a
d
a
b
il
it
y
o
f
Ara
b
ic
tex
ts
b
y
a
n
a
ly
z
in
g
a
n
d
c
las
sify
i
n
g
se
n
ten
c
e
s
in
to
d
iffere
n
t
d
iffi
c
u
l
ty
lev
e
ls
with
i
n
e
d
u
c
a
ti
o
n
a
l
c
o
n
ten
t.
Th
e
i
n
it
i
a
l
sta
g
e
o
f
th
e
p
r
o
p
o
se
d
a
p
p
ro
a
c
h
is
p
re
p
ro
c
e
ss
in
g
tex
t
u
a
l
c
o
n
ten
t
a
n
d
lev
e
ra
g
i
n
g
n
a
tu
ra
l
lan
g
u
a
g
e
p
ro
c
e
ss
in
g
(NLP
)
m
e
th
o
d
o
lo
g
ies
fo
r
fe
a
tu
re
e
x
trac
ti
o
n
,
su
c
h
a
s
W
o
rd
2
Ve
c
.
Th
e
a
p
p
ro
a
c
h
th
e
n
c
o
n
c
e
n
trate
s
o
n
re
f
in
in
g
a
n
d
e
v
a
lu
a
ti
n
g
a
d
e
e
p
lea
rn
i
n
g
m
o
d
e
l
to
c
las
sify
tex
t
in
to
d
iffere
n
t
r
e
a
d
a
b
il
it
y
le
v
e
ls.
Th
is
p
a
p
e
r
in
t
ro
d
u
c
e
s
a
h
y
b
rid
c
las
sifica
ti
o
n
m
o
d
e
l
th
a
t
c
o
m
b
i
n
e
s
c
o
n
v
o
l
u
ti
o
n
a
l
n
e
u
ra
l
n
e
two
r
k
s
(CNN
s)
a
n
d
b
i
d
irec
ti
o
n
a
l
lo
n
g
sh
o
rt
-
term
m
e
m
o
ry
(BiL
S
TM
)
lay
e
rs,
a
tt
a
in
in
g
a
n
a
c
c
u
ra
c
y
o
f
9
6
.
6
8
%
.
Th
is
m
o
d
e
l
d
e
m
o
n
stra
tes
th
e
si
g
n
ifi
c
a
n
c
e
o
f
a
p
p
ly
in
g
h
y
b
r
id
d
e
e
p
lea
rn
in
g
m
o
d
e
ls
i
n
a
n
a
ly
z
in
g
e
d
u
c
a
ti
o
n
a
l
m
a
teria
ls
a
n
d
e
sta
b
li
sh
e
s
a
fo
u
n
d
a
ti
o
n
f
o
r
su
b
se
q
u
e
n
t
p
ro
g
re
ss
in
t
h
e
field
o
f
a
u
to
m
a
ted
Ara
b
ic rea
d
a
b
il
it
y
a
ss
e
ss
m
e
n
t.
K
ey
w
o
r
d
s
:
A
r
a
b
i
c
n
a
t
u
r
a
l
l
a
n
g
u
a
g
e
p
r
o
c
e
s
s
i
n
g
B
id
ir
ec
tio
n
al
lo
n
g
s
h
o
r
t
-
ter
m
m
em
o
r
y
C
o
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
R
ea
d
ab
ilit
y
ev
alu
atio
n
T
ex
t c
lass
if
icatio
n
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
:
B
o
u
ch
aib
C
h
er
r
ad
i
E
E
I
S L
ab
o
r
ato
r
y
,
E
NSET
o
f
Mo
h
am
m
ed
ia,
Hass
an
I
I
Un
iv
er
s
ity
o
f
C
asab
lan
ca
Mo
h
am
m
ed
ia
2
8
8
3
0
,
Mo
r
o
cc
o
E
m
ail:
b
o
u
ch
aib
.
ch
er
r
ad
i@
g
m
ail.
co
m
1.
I
NT
RO
D
UCT
I
O
N
I
n
m
o
d
er
n
ed
u
ca
tio
n
al
en
v
ir
o
n
m
en
ts
,
it
is
wid
ely
r
ec
o
g
n
ize
d
th
at
s
tu
d
en
ts
d
is
p
lay
d
iv
er
s
e
lev
els
o
f
lear
n
in
g
p
r
o
g
r
ess
.
W
ith
in
th
e
s
am
e
class
r
o
o
m
o
r
s
tu
d
y
g
r
o
u
p
,
t
h
e
k
n
o
wled
g
e
a
n
d
u
n
d
er
s
tan
d
in
g
o
f
o
n
e
s
tu
d
en
t
ca
n
s
ig
n
i
f
ican
tly
v
a
r
y
f
r
o
m
th
at
o
f
th
eir
p
ee
r
s
.
T
o
o
p
tim
ize
lear
n
in
g
o
u
tco
m
es
a
n
d
en
s
u
r
e
ef
f
ec
tiv
e
teac
h
in
g
,
ad
o
p
tin
g
in
n
o
v
ativ
e
lear
n
in
g
s
tr
ateg
ies
is
cr
u
cial
f
o
r
teac
h
er
s
[
1
]
.
On
e
s
o
lu
tio
n
to
ad
d
r
ess
in
g
th
is
ch
allen
g
e
is
th
e
teac
h
in
g
at
t
h
e
r
ig
h
t
lev
el
(
T
aRL)
ap
p
r
o
ac
h
,
a
p
er
s
o
n
alize
d
lear
n
in
g
m
eth
o
d
o
lo
g
y
th
at
h
el
p
s
lear
n
er
s
d
ev
elo
p
ess
en
tial
r
ea
d
in
g
an
d
m
at
h
s
k
ills
[
2
]
.
T
h
is
s
tr
ateg
y
r
ec
o
g
n
izes
th
e
in
d
iv
id
u
al
lear
n
i
n
g
d
if
f
er
en
ce
s
an
d
p
ac
e
o
f
ea
c
h
s
tu
d
en
t
[
3
]
.
I
t
is
p
iv
o
tal
in
ad
d
r
ess
in
g
th
e
v
ar
y
in
g
ac
ad
em
i
c
ab
ilit
ies
with
in
a
class
r
o
o
m
,
en
s
u
r
in
g
th
at
all
s
t
u
d
en
ts
,
r
eg
ar
d
less
o
f
th
eir
s
tar
tin
g
p
o
in
t,
an
d
h
a
v
e
th
e
o
p
p
o
r
tu
n
ity
to
ad
v
an
ce
in
th
eir
lear
n
in
g
jo
u
r
n
ey
[
4
]
.
I
t
ca
lls
f
o
r
ed
u
ca
to
r
s
to
tailo
r
th
e
ed
u
ca
tio
n
al
co
n
te
n
t
to
m
atch
ea
ch
s
tu
d
en
t'
s
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.
1
5
,
No
.
3
,
J
u
n
e
2
0
2
6
:
2
8
4
9
-
2
8
6
2
2850
u
n
iq
u
e
n
ee
d
s
[
5
]
,
[
6
]
.
E
d
u
ca
to
r
s
p
lay
an
im
p
o
r
tan
t
r
o
le
in
th
is
ad
ap
tiv
e
lear
n
in
g
m
o
d
el
[
7
]
.
B
y
m
ea
s
u
r
in
g
t
h
e
r
ea
d
ab
ilit
y
o
f
a
tex
t,
teac
h
er
s
ca
n
en
s
u
r
e
th
e
m
ater
ial
m
e
ets
th
e
s
p
ec
if
ic
n
ee
d
s
o
f
ea
c
h
lear
n
er
[
8
]
.
T
h
is
m
eth
o
d
o
l
o
g
y
allo
ws
lear
n
er
s
to
ad
v
an
ce
ac
c
o
r
d
in
g
to
th
eir
p
er
s
o
n
al
lear
n
in
g
r
h
y
th
m
,
av
o
id
in
g
an
y
s
en
s
e
o
f
b
ein
g
s
wam
p
ed
with
in
f
o
r
m
at
io
n
o
r
d
is
in
ter
est
d
u
e
to
lack
o
f
ch
allen
g
e.
R
ea
d
ab
ilit
y
is
in
f
lu
en
ce
d
b
y
a
r
an
g
e
o
f
f
ac
to
r
s
,
s
u
ch
as
v
o
ca
b
u
lar
y
co
m
p
lex
ity
,
s
en
ten
ce
an
d
p
ar
ag
r
ap
h
s
tr
u
ctu
r
e,
an
d
v
ar
i
o
u
s
t
ex
tu
al
f
ea
tu
r
es
[
9
]
.
T
h
ese
elem
en
ts
ar
e
cr
itical
f
o
r
ed
u
ca
to
r
s
to
co
n
s
id
er
wh
en
s
elec
tin
g
o
r
c
r
ea
tin
g
ed
u
ca
tio
n
al
m
ater
ials
,
allo
win
g
th
em
to
a
d
ju
s
t th
e
d
if
f
icu
lty
lev
el
to
m
atc
h
th
eir
s
tu
d
en
ts
'
ab
ilit
ie
s
.
Ho
wev
er
,
th
e
m
a
n
u
al
ev
al
u
atio
n
o
f
te
x
t
r
ea
d
a
b
ilit
y
is
b
o
th
tim
e
-
co
n
s
u
m
in
g
an
d
c
o
s
tly
,
lim
itin
g
its
p
r
ac
tical
ap
p
licatio
n
in
ed
u
ca
tio
n
al
s
ettin
g
s
[
1
0
]
.
Fo
r
th
e
Ar
ab
ic
lan
g
u
ag
e,
o
n
e
o
f
t
h
e
m
ain
ch
allen
g
es
th
at
f
ac
e
r
ea
d
ab
ilit
y
ass
ess
m
en
t
is
th
e
lan
g
u
a
g
e'
s
in
h
er
en
t
co
m
p
lex
ity
[
1
1
]
.
T
h
e
ass
ess
m
en
t
o
f
r
ea
d
ab
ilit
y
in
th
e
Ar
ab
ic
lan
g
u
a
g
e
p
o
s
es
ad
d
itio
n
al
ch
allen
g
es
d
u
e
to
th
e
lan
g
u
ag
e'
s
co
m
p
lex
m
o
r
p
h
o
lo
g
y
,
ex
ten
s
iv
e
lex
ico
n
,
an
d
d
iv
er
s
e
s
en
ten
ce
co
n
s
tr
u
ct
io
n
[
1
2
]
.
Ar
ab
ic'
s
h
ig
h
lev
el
o
f
in
f
lectio
n
,
wh
ich
allo
ws s
in
g
le
wo
r
d
s
to
co
n
v
ey
v
ar
io
u
s
g
r
am
m
atica
l n
u
an
ce
s
,
f
u
r
th
er
co
m
p
licates
th
is
task
.
T
h
ese
d
if
f
icu
lties
h
a
v
e
s
p
u
r
r
e
d
in
ter
est
in
s
ee
k
in
g
tech
n
o
lo
g
ical
i
n
n
o
v
atio
n
s
f
o
r
au
to
m
atin
g
t
h
e
p
r
o
ce
s
s
o
f
r
ea
d
ab
ilit
y
ass
ess
m
en
t.
Am
o
n
g
t
h
ese
s
o
lu
tio
n
s
,
d
ee
p
lear
n
in
g
an
d
n
atu
r
al
lan
g
u
ag
e
p
r
o
ce
s
s
in
g
(
NL
P)
h
av
e
em
er
g
ed
as
p
ar
ticu
lar
ly
p
r
o
m
is
in
g
ap
p
r
o
ac
h
es.
B
o
th
NL
P
an
d
d
ee
p
lear
n
in
g
ar
e
i
n
ter
d
is
cip
lin
ar
y
f
ield
s
th
at
m
er
g
e
c
o
m
p
u
ter
s
cien
ce
with
a
r
tific
ial
in
tellig
en
ce
[
1
3
]
,
[
1
4
]
,
eq
u
ip
p
in
g
m
ac
h
i
n
es
with
th
e
ab
ilit
y
to
p
r
o
ce
s
s
,
u
n
d
e
r
s
tan
d
,
an
d
in
ter
p
r
et
h
u
m
an
lan
g
u
ag
e
in
a
m
ea
n
in
g
f
u
l
an
d
co
n
tex
tu
ally
ap
p
r
o
p
r
iate
m
a
n
n
er
.
NL
P,
in
th
e
co
n
te
x
t
o
f
r
ea
d
ab
ilit
y
m
e
asu
r
em
en
t,
u
tili
ze
s
tex
t
class
if
icatio
n
tech
n
iq
u
es
to
ass
es
s
th
e
r
ea
d
ab
ilit
y
o
f
tex
ts
b
y
ca
teg
o
r
izin
g
th
em
in
t
o
v
ar
io
u
s
r
ea
d
ab
ilit
y
lev
els
o
r
ca
teg
o
r
ies
[
1
5
]
,
[
1
6
]
.
T
h
is
ap
p
r
o
ac
h
in
v
o
lv
es
tr
ai
n
in
g
m
ac
h
in
e
lear
n
i
n
g
m
o
d
el
s
to
an
aly
ze
tex
tu
al
f
ea
tu
r
es
an
d
m
ak
e
p
r
ed
ictio
n
s
ab
o
u
t
th
e
r
ea
d
ab
ilit
y
o
f
a
g
iv
en
tex
t
[
1
7
]
.
T
h
e
in
teg
r
atio
n
o
f
d
ee
p
lear
n
in
g
tech
n
iq
u
es
with
in
th
is
d
o
m
ain
o
f
f
er
s
th
e
p
o
ten
tial
f
o
r
ev
en
m
o
r
e
r
ef
in
ed
a
n
aly
s
is
.
D
ee
p
l
ea
r
n
in
g
'
s
ab
ilit
y
to
p
r
o
ce
s
s
an
d
lear
n
f
r
o
m
lar
g
e
d
atasets
ca
n
u
n
co
v
er
s
u
b
tle
p
atter
n
s
in
lan
g
u
ag
e
u
s
e
th
at
a
r
e
in
d
icativ
e
o
f
tex
t
co
m
p
lex
ity
,
t
h
er
eb
y
en
h
a
n
cin
g
th
e
p
r
ec
is
io
n
o
f
r
ea
d
ab
ilit
y
p
r
ed
ictio
n
s
[
1
8
]
.
T
h
e
f
ield
o
f
tex
t
r
ea
d
ab
ilit
y
a
s
s
es
s
m
en
t
is
an
ar
ea
o
f
in
cr
ea
s
in
g
in
ter
est
[
1
9
]
.
R
ev
iewin
g
p
r
ev
io
u
s
r
esear
ch
ef
f
o
r
ts
o
f
f
e
r
s
im
p
o
r
t
an
t
b
ac
k
g
r
o
u
n
d
f
o
r
th
e
cu
r
r
en
t
s
tu
d
y
,
em
p
h
asizin
g
th
eir
p
r
i
m
ar
y
co
n
tr
ib
u
tio
n
s
,
tech
n
iq
u
es,
an
d
r
esu
lts
,
with
a
p
ar
ticu
lar
f
o
cu
s
o
n
th
e
m
eth
o
d
o
lo
g
ies
em
p
lo
y
e
d
to
ad
d
r
es
s
th
e
is
s
u
es
o
f
tex
t
r
ea
d
ab
ilit
y
an
d
th
eir
ef
f
ec
tiv
en
ess
in
attain
in
g
h
ig
h
p
r
ed
i
ctio
n
ac
cu
r
ac
y
.
Fo
r
in
s
tan
ce
,
Nass
ir
i
et
a
l.
[
2
0
]
em
p
lo
y
ed
m
ac
h
in
e
lear
n
in
g
t
ec
h
n
iq
u
es
to
e
v
alu
ate
th
e
r
ea
d
ab
ilit
y
o
f
m
o
d
e
r
n
s
tan
d
a
r
d
Ar
ab
ic
(
MSA)
tex
ts
.
T
h
ey
m
ap
p
ed
th
e
r
esu
ltin
g
r
ea
d
ab
ilit
y
lev
els
to
th
e
in
ter
ag
e
n
cy
lan
g
u
a
g
e
r
o
u
n
d
ta
b
le
(
I
L
R
)
s
ca
le
f
o
r
ass
ess
in
g
s
ec
o
n
d
lan
g
u
a
g
e
p
r
o
f
icien
cy
.
Fro
m
an
in
itial
s
et
o
f
1
7
0
r
ea
d
ab
ilit
y
f
ea
tu
r
es,
th
e
p
r
o
p
o
s
ed
ap
p
r
o
ac
h
ac
h
iev
ed
an
ac
cu
r
ac
y
o
f
9
0
.
4
3
%
in
a
th
r
ee
-
class
clas
s
if
icatio
n
task
u
s
in
g
th
e
in
s
tan
ce
-
b
ased
k
-
n
ea
r
e
s
t
n
eig
h
b
o
r
s
(
I
B
K)
an
d
r
an
d
o
m
f
o
r
est
alg
o
r
ith
m
s
.
Ho
wev
er
,
th
is
s
tu
d
y
was
b
a
s
ed
o
n
tr
ad
itio
n
al
m
a
ch
in
e
le
ar
n
in
g
tech
n
iq
u
es,
wh
ich
h
av
e
lim
ited
ca
p
ac
ity
t
o
attain
h
ig
h
ac
cu
r
ac
y
.
Nass
ir
i
et
a
l.
[
2
1
]
ex
p
lo
r
e
th
e
u
s
e
o
f
s
tatis
tica
l
m
ac
h
in
e
lear
n
in
g
to
p
r
e
d
ict
th
e
r
ea
d
ab
i
lity
o
f
Ar
ab
ic
tex
ts
f
o
r
s
ec
o
n
d
-
lan
g
u
ag
e
lear
n
er
s
.
Usi
n
g
v
ar
i
o
u
s
f
ea
tu
r
e
s
ets an
d
alg
o
r
ith
m
s
,
th
e
s
tu
d
y
aim
s
t
o
m
ain
tain
o
r
in
cr
ea
s
e
p
r
e
d
ictio
n
ac
cu
r
ac
y
w
h
ile
s
im
p
lif
y
in
g
t
h
e
f
ea
t
u
r
e
s
et.
E
x
p
er
im
en
ts
s
h
o
wed
m
ax
im
u
m
ac
cu
r
ac
ies
o
f
u
p
t
o
8
6
.
1
5
%,
co
n
f
ir
m
in
g
th
at
a
r
ed
u
ce
d
f
ea
tu
r
e
s
et
co
u
l
d
ac
h
iev
e
co
m
p
ar
ab
le
r
esu
lts
.
Sad
d
ik
i
et
a
l.
[
2
2
]
p
r
esen
t
a
co
m
p
r
e
h
en
s
iv
e
co
m
p
u
tatio
n
al
s
tu
d
y
o
n
Ar
ab
ic
r
ea
d
ab
ilit
y
,
a
ch
allen
g
in
g
task
g
iv
en
Ar
a
b
ic'
s
m
o
r
p
h
o
l
o
g
ica
l
co
m
p
le
x
ity
a
n
d
lo
w
-
r
eso
u
r
c
e
s
tatu
s
.
T
h
e
s
tu
d
y
ev
alu
ates
b
o
th
f
ir
s
t
-
lan
g
u
ag
e
(
L
1
)
a
n
d
s
ec
o
n
d
-
la
n
g
u
a
g
e
(
L
2
)
Ar
ab
ic
r
ea
d
ab
ilit
y
,
u
n
i
q
u
ely
co
m
b
in
in
g
th
em
in
th
e
s
am
e
ex
p
er
im
en
tal
s
etti
n
g
.
B
y
in
co
r
p
o
r
atin
g
ad
v
an
c
ed
f
ea
tu
r
es
lik
e
lan
g
u
ag
e
m
o
d
elin
g
,
th
e
s
tu
d
y
ac
h
iev
es
a
9
4
.
8
%
ac
cu
r
ac
y
f
o
r
L
1
an
d
a
7
4
.
1
%
ac
cu
r
ac
y
f
o
r
L
2
,
m
ar
k
in
g
s
ig
n
if
ica
n
t
er
r
o
r
r
ed
u
ctio
n
s
co
m
p
ar
ed
to
tr
a
d
itio
n
al
m
et
h
o
d
s
.
T
h
e
p
a
p
er
s
u
g
g
ests
th
at
u
s
in
g
L
1
-
g
en
e
r
ated
m
o
d
el
s
ca
n
en
h
a
n
ce
L
2
r
ea
d
ab
ilit
y
ass
ess
m
en
t.
B
er
r
ich
i
et
a
l.
[
2
3
]
u
s
ed
v
ar
io
u
s
m
eth
o
d
s
f
o
r
ass
ess
in
g
th
e
co
m
p
le
x
ity
o
f
tex
t
f
o
r
n
a
tiv
e
(
L
1
)
s
p
ea
k
er
s
.
T
wo
p
r
im
ar
y
ap
p
r
o
ac
h
es
wer
e
ev
alu
ated
.
T
h
e
f
ir
s
t
ap
p
r
o
ac
h
in
v
o
lv
ed
a
n
aly
zin
g
s
p
ec
if
ic,
h
an
d
cr
a
f
ted
f
ea
tu
r
es
th
at
h
elp
d
eter
m
in
e
a
tex
t'
s
d
if
f
icu
lty
lev
el.
I
n
th
e
s
ec
o
n
d
a
p
p
r
o
ac
h
,
au
th
o
r
s
ex
p
l
o
r
ed
wo
r
d
em
b
e
d
d
in
g
s
(
in
clu
d
in
g
co
n
tin
u
o
u
s
b
ag
-
of
-
wo
r
d
s
m
o
d
el
(
C
B
O
W
)
,
s
k
ip
-
g
r
am
,
an
d
Ar
ab
ic
b
id
ir
ec
tio
n
al
en
co
d
er
r
ep
r
esen
tatio
n
s
f
r
o
m
tr
an
s
f
o
r
m
er
s
(
Ar
aBER
T
)
)
as
an
alter
n
ativ
e
to
th
e
h
an
d
cr
af
ted
f
ea
tu
r
es.
T
h
e
Ar
aBER
T
m
o
d
el
p
er
f
o
r
m
ed
b
est
am
o
n
g
th
e
tech
n
iq
u
es,
ac
h
iev
in
g
a
p
r
ed
i
ctio
n
ac
cu
r
ac
y
o
f
ap
p
r
o
x
im
ately
7
6
.
9
3
%.
B
ess
o
u
an
d
C
h
en
n
i
[
2
4
]
em
p
lo
y
m
a
ch
in
e
lear
n
in
g
to
p
r
ed
ict
te
x
t
co
m
p
lex
ity
,
aim
in
g
to
m
ak
e
Ar
ab
ic
tex
t
m
o
r
e
a
cc
ess
ib
le
to
lear
n
er
s
.
Var
io
u
s
class
if
ier
s
an
d
f
ea
tu
r
es
lik
e
c
o
u
n
t
a
n
d
ter
m
f
r
eq
u
e
n
cy
-
in
v
er
s
e
d
o
cu
m
e
n
t
f
r
eq
u
e
n
cy
(
T
F
-
I
DF)
wer
e
test
ed
,
r
ev
ea
lin
g
th
at
a
co
m
b
i
n
atio
n
o
f
n
-
g
r
a
m
f
ea
tu
r
es
an
d
s
p
ec
if
ic
d
ata
r
e
p
r
esen
tatio
n
s
s
ig
n
if
ican
tly
b
o
o
s
ted
p
er
f
o
r
m
a
n
ce
.
T
h
e
m
o
s
t
ac
cu
r
ate
r
esu
lts
wer
e
ac
h
iev
ed
u
s
in
g
T
F
-
I
DF
v
ec
to
r
s
an
d
s
u
p
p
o
r
t
v
ec
to
r
m
ac
h
in
e
(
SVM
)
,
r
ea
ch
in
g
a
n
o
v
er
all
a
cc
u
r
ac
y
o
f
8
7
.
1
4
%
ac
r
o
s
s
f
o
u
r
c
o
m
p
lex
ity
class
es.
Sad
d
ik
i
et
a
l.
[
2
5
]
ev
alu
ate
s
im
p
le
r
ea
d
ab
ilit
y
f
ea
tu
r
es
u
s
in
g
p
u
b
licly
ac
ce
s
s
ib
le
an
d
o
p
en
-
s
o
u
r
ce
to
o
ls
,
ac
h
iev
in
g
p
er
f
o
r
m
an
ce
co
m
p
ar
ab
le
t
o
ex
is
tin
g
Ar
ab
ic
r
ea
d
ab
ilit
y
s
tu
d
ies.
T
h
e
s
tu
d
y
f
o
u
n
d
th
at
co
s
t
-
ef
f
ec
tiv
e
f
ea
tu
r
es
lik
e
m
o
r
p
h
em
e
c
o
u
n
ts
,
ty
p
e
an
d
to
k
en
co
u
n
ts
,
an
d
p
ar
t
-
of
-
s
p
ee
c
h
tag
s
h
ad
a
g
o
o
d
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
Hyb
r
id
co
n
vo
lu
tio
n
a
l n
e
u
r
a
l n
etw
o
r
k
-
b
id
ir
ec
tio
n
a
l lo
n
g
s
h
o
r
t
-
term me
mo
r
y
…
(
Mo
h
a
med
A
min
e
Ou
a
s
s
il
)
2851
co
s
t/b
en
ef
it
r
atio
.
T
h
e
b
est
r
esu
lts
in
ter
m
s
o
f
ac
cu
r
ac
y
wer
e
7
3
.
7
4
%
ac
h
iev
ed
b
y
r
a
n
d
o
m
f
o
r
est
class
if
ier
.
B
aa
ze
em
et
a
l.
[
2
6
]
ex
p
l
o
r
ed
u
s
in
g
ey
e
-
t
r
ac
k
in
g
to
e
n
h
an
ce
Ar
ab
ic
tex
t
r
ea
d
a
b
ilit
y
p
r
e
d
ic
tio
n
,
r
e
v
ea
lin
g
t
h
at
ey
e
-
tr
ac
k
in
g
d
ata,
wh
en
co
m
b
in
ed
with
lin
g
u
is
tic
f
ea
tu
r
es,
ca
n
m
o
r
e
ac
cu
r
ately
ass
ess
r
ea
d
ab
ilit
y
.
T
h
e
ef
f
ec
tiv
en
ess
v
ar
ies
b
y
alg
o
r
it
h
m
,
s
u
g
g
esti
n
g
f
u
r
th
er
in
v
esti
g
atio
n
is
n
ee
d
ed
.
L
im
itatio
n
s
in
clu
d
e
in
s
u
f
f
icien
t
d
ata
an
d
th
e
u
s
e
o
f
b
asic
m
ac
h
in
e
lear
n
in
g
m
o
d
els.
Ou
ass
il
et
a
l.
[
2
7
]
p
r
esen
t
a
n
o
v
el
ap
p
r
o
ac
h
f
o
r
Ar
ab
ic
r
ea
d
ab
ilit
y
ass
ess
m
en
t
u
s
in
g
a
h
y
b
r
id
B
E
R
T
-
b
id
ir
ec
tio
n
al
lo
n
g
s
h
o
r
t
-
ter
m
m
em
o
r
y
(
B
iLST
M)
m
o
d
el.
T
h
is
m
eth
o
d
m
a
r
k
s
a
s
ig
n
if
ican
t
a
d
v
an
ce
m
e
n
t
in
tex
t
d
if
f
icu
lty
class
if
icatio
n
.
T
h
e
p
r
o
p
o
s
ed
m
o
d
el
ac
h
iev
e
d
an
ac
cu
r
ac
y
r
ate
o
f
8
9
.
5
3
%.
T
h
e
p
r
ec
ed
i
n
g
r
e
v
iew
r
ev
ea
ls
a
clea
r
o
p
p
o
r
tu
n
ity
t
o
im
p
r
o
v
e
u
p
o
n
t
h
e
p
er
f
o
r
m
a
n
ce
an
d
a
cc
u
r
ac
y
o
f
class
ical
m
o
d
els.
T
h
er
ef
o
r
e,
i
n
th
is
s
tu
d
y
a
n
o
v
el
ap
p
r
o
ac
h
b
ased
o
n
NL
P
an
d
h
y
b
r
id
d
ee
p
lear
n
in
g
m
o
d
el
is
p
r
esen
ted
to
ass
ess
Ar
ab
ic
te
x
t
r
ea
d
ab
ilit
y
b
y
class
if
y
in
g
s
en
ten
ce
s
in
to
d
if
f
er
en
t
lev
el
s
o
f
d
if
f
icu
lty
.
T
h
e
m
ajo
r
co
n
t
r
ib
u
tio
n
s
o
f
th
is
s
tu
d
y
in
clu
d
e:
‒
Dev
elo
p
in
g
a
n
ac
cu
r
ate
s
y
s
tem
to
au
t
o
m
atica
lly
class
if
y
th
e
d
if
f
ic
u
lty
o
f
Ar
ab
ic
s
en
ten
ce
s
u
s
in
g
a
co
m
b
in
atio
n
o
f
two
d
ee
p
lear
n
in
g
tech
n
iq
u
es:
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
(
C
NN)
an
d
B
iLST
M.
‒
An
aly
zin
g
th
e
ef
f
ec
t
o
f
co
n
v
o
lu
tio
n
al
d
ep
t
h
o
n
th
e
B
iLST
M
lay
er
b
y
s
tu
d
y
i
n
g
th
e
im
p
a
ct
o
f
ad
d
in
g
a
v
ar
y
in
g
n
u
m
b
er
o
f
co
n
v
o
lu
tio
n
al
lay
er
s
.
T
h
e
r
em
ain
d
e
r
o
f
t
h
is
p
ap
er
is
s
tr
u
ctu
r
ed
as f
o
llo
ws
.
Sectio
n
2
d
escr
ib
es th
e
ap
p
r
o
ac
h
,
m
at
er
ials
,
an
d
alg
o
r
ith
m
s
u
tili
ze
d
in
d
ev
elo
p
in
g
th
e
p
r
o
p
o
s
ed
s
y
s
tem
.
R
es
u
lts
ar
e
p
r
esen
ted
in
s
ec
tio
n
3
.
T
h
e
f
in
al
s
ec
tio
n
4
co
n
clu
d
es a
n
d
s
u
m
m
ar
izes th
e
p
ap
er
.
2.
M
E
T
H
O
D
T
h
is
wo
r
k
p
r
o
p
o
s
es
a
co
m
p
u
tatio
n
al
m
o
d
el
f
o
r
au
to
m
atica
lly
an
aly
zin
g
an
d
g
r
ad
in
g
th
e
lin
g
u
is
tic
co
m
p
lex
ity
le
v
els
o
f
Ar
ab
ic
tex
t,
class
if
y
in
g
it
as
ea
s
y
,
m
o
d
er
ately
d
if
f
ic
u
lt,
o
r
co
m
p
lex
to
r
ea
d
a
n
d
u
n
d
er
s
tan
d
.
A
d
ee
p
lear
n
in
g
m
o
d
el
was
em
p
lo
y
ed
to
ac
h
i
ev
e
th
is
,
an
d
it
was
tr
ain
ed
to
ac
cu
r
ately
class
if
y
s
en
ten
ce
s
b
ased
o
n
th
eir
lin
g
u
is
tic
an
d
s
em
an
tic
c
h
ar
ac
ter
i
s
tics
.
Fig
u
r
e
1
illu
s
tr
ates
th
e
p
r
o
p
o
s
ed
ap
p
r
o
ac
h
th
at
co
n
s
is
ts
o
f
f
o
u
r
m
ai
n
s
tag
es:
d
ata
r
ep
r
esen
tatio
n
,
p
r
ep
r
o
ce
s
s
in
g
,
wo
r
d
r
e
p
r
esen
tatio
n
,
an
d
class
if
icatio
n
.
I
n
th
e
d
ata
r
ep
r
esen
tatio
n
s
ta
g
e,
th
e
d
ataset
u
s
ed
in
t
h
e
e
x
p
er
im
en
ts
was
d
escr
ib
ed
,
i
n
clu
d
in
g
its
tex
tu
al
co
n
ten
t,
th
e
d
is
tr
ib
u
tio
n
o
f
s
en
ten
ce
s
b
y
co
m
p
lex
ity
,
k
ey
v
o
ca
b
u
la
r
y
s
tatis
tics
,
an
d
a
v
is
u
aliza
tio
n
o
f
th
e
d
ataset's
co
n
ten
t
u
s
in
g
a
wo
r
d
clo
u
d
tech
n
iq
u
e.
I
n
t
h
e
p
r
ep
r
o
ce
s
s
in
g
s
tag
e,
th
e
te
x
t
was
cl
ea
n
an
d
s
tan
d
ar
d
ize
to
p
r
ep
a
r
e
it
f
o
r
an
aly
s
is
.
T
h
is
co
n
s
is
ts
o
f
s
ev
er
al
k
ey
o
p
er
at
io
n
s
:
tex
t
clea
n
in
g
r
em
o
v
es
ir
r
elev
an
t
n
o
is
e
lik
e
p
u
n
ctu
atio
n
o
r
HT
ML
tag
s
,
n
o
r
m
aliza
tio
n
en
s
u
r
es
u
n
if
o
r
m
i
ty
b
y
s
tan
d
ar
d
izin
g
wo
r
d
f
o
r
m
s
,
an
d
to
k
e
n
izatio
n
b
r
ea
k
s
ea
ch
s
en
ten
ce
d
o
wn
in
t
o
its
in
d
iv
id
u
al
wo
r
d
s
ca
lled
t
o
k
en
s
.
I
n
th
e
w
o
r
d
r
e
p
r
esen
tat
io
n
s
tag
e,
th
e
wo
r
d
em
b
ed
d
in
g
tech
n
iq
u
e
W
o
r
d
2
Vec
was
em
p
lo
y
ed
to
co
n
v
er
t
th
e
clea
n
tex
t
in
to
a
n
u
m
er
ical
f
o
r
m
at
th
e
m
o
d
el
ca
n
p
r
o
ce
s
s
.
W
o
r
d
2
Vec
g
en
er
ates
a
u
n
iq
u
e
v
ec
t
o
r
f
o
r
ea
ch
wo
r
d
,
m
ath
em
atica
lly
ca
p
t
u
r
i
n
g
its
s
em
an
tic
an
d
s
y
n
tactic
co
n
tex
t.
L
astl
y
,
th
ese
co
n
tex
tu
al
wo
r
d
v
ec
to
r
s
as
in
p
u
t
wer
e
u
s
ed
f
o
r
th
e
p
r
o
p
o
s
ed
d
ee
p
lear
n
i
n
g
m
o
d
el.
T
h
e
m
o
d
el
was
f
in
et
u
n
ed
,
tr
ain
ed
,
an
d
test
ed
to
class
i
f
y
ea
ch
s
en
ten
ce
b
y
its
co
m
p
lex
ity
.
Fig
u
r
e
1
.
T
h
e
f
lo
wch
a
r
t o
f
t
h
e
p
r
o
p
o
s
ed
s
y
s
tem
2
.
1
.
Da
t
a
s
et
re
presenta
t
io
n
T
h
e
d
ataset
u
s
ed
in
th
is
r
esear
ch
was
co
llected
f
r
o
m
o
n
e
o
f
th
e
lar
g
est
Ar
ab
ic
lan
g
u
ag
e
lear
n
in
g
p
latf
o
r
m
s
(
Al
-
J
az
ee
r
a
L
ea
r
n
in
g
Ar
ab
ic)
,
wh
ich
is
an
o
p
en
-
s
o
u
r
ce
to
o
l
d
esig
n
e
d
to
s
im
p
lify
Ar
ab
ic
lan
g
u
ag
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.
1
5
,
No
.
3
,
J
u
n
e
2
0
2
6
:
2
8
4
9
-
2
8
6
2
2852
ac
q
u
is
itio
n
th
r
o
u
g
h
in
ter
ac
tiv
e
an
d
ea
s
y
-
to
-
n
av
ig
ate
co
n
ten
t
[
2
7
]
.
Fo
r
th
is
en
d
,
t
h
e
au
t
h
o
r
s
u
s
ed
a
s
cr
ap
in
g
to
o
l
to
c
o
llect
tex
ts
f
r
o
m
th
is
p
latf
o
r
m
t
h
at
wer
e
al
r
ea
d
y
lab
eled
an
d
class
if
ied
b
y
th
e
p
lat
f
o
r
m
'
s
ed
ito
r
s
in
to
th
r
ee
d
if
f
e
r
en
t
lear
n
in
g
ca
te
g
o
r
ies:
b
eg
in
n
e
r
,
in
ter
m
ed
iat
e,
an
d
ad
v
an
ce
d
.
T
h
is
d
ata
s
et
co
m
p
r
is
es
two
co
lu
m
n
s
:
th
e
'
T
ex
t'
co
lu
m
n
co
n
tain
s
s
en
ten
ce
s
as
tex
tu
al
co
n
ten
t,
an
d
th
e
'
L
ab
el'
co
lu
m
n
co
n
tain
s
n
u
m
b
er
s
f
r
o
m
1
to
3
,
r
ep
r
esen
tin
g
th
e
lev
el
o
f
s
en
ten
ce
d
if
f
icu
lty
.
E
ac
h
s
en
ten
ce
was
ca
teg
o
r
ized
in
to
o
n
e
o
f
th
r
ee
p
r
o
f
icien
c
y
lev
els:
b
e
g
in
n
in
g
,
in
ter
m
ed
iate,
o
r
ad
v
an
ce
d
.
B
eg
in
n
in
g
s
en
ten
ce
s
ar
e
s
im
p
le
in
s
tr
u
ct
u
r
e
a
n
d
v
o
ca
b
u
lar
y
f
o
r
b
eg
in
n
er
s
.
I
n
t
er
m
ed
iate
s
en
ten
ce
s
ar
e
m
o
d
er
ately
co
m
p
lex
,
s
u
itab
le
f
o
r
lear
n
er
s
with
b
asic
lan
g
u
ag
e
s
k
ills
.
Ad
v
an
ce
d
s
en
ten
ce
s
co
n
tain
co
m
p
lex
g
r
a
m
m
ar
an
d
e
x
ten
s
iv
e
v
o
ca
b
u
la
r
y
f
o
r
lea
r
n
er
s
with
h
ig
h
p
r
o
f
icien
c
y
.
T
ab
le
1
s
h
o
wca
s
es e
x
am
p
les f
r
o
m
th
e
d
at
aset.
T
ab
le
2
p
r
o
v
i
d
es
an
o
v
er
v
ie
w
o
f
s
en
ten
ce
d
is
tr
ib
u
tio
n
a
cr
o
s
s
th
r
ee
lan
g
u
ag
e
p
r
o
f
icie
n
cy
lev
els
with
in
th
e
d
ataset,
co
n
tain
in
g
4
,
5
1
9
s
en
ten
ce
s
,
8
1
,
7
4
3
to
k
en
s
,
an
d
2
3
,
1
8
8
u
n
iq
u
e
wo
r
d
s
.
I
t
illu
s
tr
ates
a
p
r
o
g
r
ess
io
n
f
r
o
m
th
e
b
e
g
in
n
e
r
class
,
with
1
,
4
1
8
s
en
ten
ce
s
an
d
8
,
4
4
4
to
k
e
n
s
,
to
th
e
ad
v
an
ce
d
class
,
with
1
,
3
5
6
s
en
ten
ce
s
an
d
4
1
,
9
0
4
to
k
en
s
,
in
d
icatin
g
in
cr
ea
s
i
n
g
s
en
ten
ce
co
m
p
lex
ity
a
n
d
v
o
ca
b
u
lar
y
s
ize.
Sen
ten
ce
len
g
th
v
ar
ies
f
r
o
m
2
to
4
0
to
k
e
n
s
,
an
d
th
e
n
u
m
b
er
o
f
letter
s
p
er
s
en
ten
ce
r
a
n
g
es
f
r
o
m
4
t
o
2
6
5
ac
r
o
s
s
th
e
class
es.
T
ab
le
1
.
L
a
b
elled
s
am
p
les
S
e
n
t
e
n
c
e
s
a
n
d
i
t
s
t
r
a
n
sl
a
t
i
o
n
s
C
l
a
s
ses
م
ا
ي
أ
ة
ر
ش
ع
ل
ر
م
ت
س
ي
و
ةن
س
لك
ن
م
ر
ب
م
فون
ي
ف
د
ا
ص
ح
لا
أد
ب
ي
Th
e
h
a
r
v
e
st
b
e
g
i
n
s
i
n
N
o
v
e
m
b
e
r
o
f
e
a
c
h
y
e
a
r
a
n
d
l
a
st
s fo
r
t
e
n
d
a
y
s.
B
e
g
i
n
n
i
n
g
ث
ي
د
ح
لا
ب
طلل
م
ك
م
و
ا
د
ع
ا
س
م
ن
ور
خ
آ
هر
ب
ت
ع
ي
و
ا
ي
لك
اد
ا
م
ت
ع
ا
هي
لع نود
م
ت
ع
ي
م
ه
ض
ع
ب
و
هب
ن
م
ؤي
سا
ن
ل
ا
ض
ع
ب
ف
ت
ا
ج
ع
لا
ن
م
ع
ون
لا
ا
ذ
ه نأ
ش
ب
ء
ار
ا
فل
ت
خ
ت
و
O
p
i
n
i
o
n
s
d
i
f
f
e
r
a
b
o
u
t
t
h
i
s
t
y
p
e
o
f
t
r
e
a
t
me
n
t
,
s
o
me
p
e
o
p
l
e
d
o
n
o
t
b
e
l
i
e
v
e
i
n
i
t
,
s
o
me
r
e
l
y
o
n
i
t
c
o
mp
l
e
t
e
l
y
,
a
n
d
o
t
h
e
r
s
c
o
n
si
d
e
r
i
t
a
n
a
ssi
s
t
a
n
t
a
n
d
c
o
m
p
l
e
me
n
t
t
o
m
o
d
e
r
n
me
d
i
c
i
n
e
.
I
n
t
e
r
med
i
a
t
e
لوان
ت
ي
م
ل
ف
ص
ن
و
م
اوعأ
ةي
ن
ا
م
ث
هت
لت
ج
ع
لا
ن
م
طق
ف
د
ح
او
م
ا
ع
د
ع
ب
سور
ي
ف
لا
ن
م
ا
م
ا
م
ت
هي
فا
ع
ت
ب
ء
ار
ب
خ
لا
ي
ف
ي
آ
شت
أ
سور
ي
ف
ب
د
لو
ا
ي
ق
ي
ر
ف
إ
ب
ون
ج
ن
م
لف
ط
أ
ج
ا
ف
ر
ي
قا
ق
ع
ي
أ
ا
ه
ل
خ
A
S
o
u
t
h
A
f
r
i
c
a
n
c
h
i
l
d
b
o
r
n
w
i
t
h
H
I
V
su
r
p
r
i
se
d
e
x
p
e
r
t
s
b
y
c
o
mp
l
e
t
e
l
y
r
e
c
o
v
e
r
i
n
g
f
r
o
m
t
h
e
v
i
r
u
s a
f
t
e
r
j
u
s
t
o
n
e
y
e
a
r
o
f
t
r
e
a
t
me
n
t
,
f
o
l
l
o
w
e
d
b
y
e
i
g
h
t
a
n
d
a
h
a
l
f
y
e
a
r
s w
i
t
h
o
u
t
t
a
k
i
n
g
a
n
y
me
d
i
c
a
t
i
o
n
.
A
d
v
a
n
c
e
d
T
ab
le
2
.
Nu
m
b
er
o
f
s
en
ten
ce
s
p
er
class
C
l
a
s
ses
N
u
mb
e
r
o
f
s
e
n
t
e
n
c
e
s
To
t
a
l
t
o
k
e
n
s
V
o
c
a
b
u
l
a
r
y
s
i
z
e
M
i
n
t
o
k
e
n
s
p
e
r
s
e
n
t
e
n
c
e
M
a
x
t
o
k
e
n
s
p
e
r
s
e
n
t
e
n
c
e
M
a
x
l
e
t
t
e
r
s
p
e
r
s
e
n
t
e
n
c
e
M
i
n
l
e
t
t
e
r
s
p
e
r
r
e
c
o
r
d
B
e
g
i
n
n
e
r
1
,
4
1
8
8
,
4
4
4
2
,
7
4
7
2
10
1
9
2
4
I
n
t
e
r
med
i
a
t
e
1
,
7
4
5
31
,
395
12
,
496
12
24
1
6
6
41
A
d
v
a
n
c
e
d
1
,
3
5
6
41
,
904
15
,
510
25
40
2
6
5
94
O
v
e
r
a
l
l
4
,
5
1
9
81
,
743
23
,
188
2
40
2
6
5
4
T
o
g
ai
n
a
n
in
itial
u
n
d
e
r
s
tan
d
i
n
g
o
f
th
e
te
x
tu
al
c
o
n
ten
t
with
in
a
d
ataset,
th
e
wo
r
d
-
clo
u
d
tech
n
iq
u
e
was
u
tili
ze
d
.
T
h
is
m
eth
o
d
s
er
v
es
as
a
to
o
l
f
o
r
th
e
v
is
u
al
r
ep
r
esen
tatio
n
o
f
w
o
r
d
f
r
eq
u
en
cy
,
th
er
eb
y
p
r
o
v
id
in
g
an
im
m
ed
iate
o
v
er
v
iew
o
f
t
h
e
m
o
s
t
p
r
o
m
in
en
t
te
r
m
s
an
d
k
e
y
ter
m
in
o
l
o
g
ies
o
f
ea
ch
class
with
in
th
e
d
ataset.
Fig
u
r
e
2
illu
s
tr
ate
wo
r
d
clo
u
d
p
r
esen
tatio
n
s
o
f
th
e
th
r
ee
class
es
with
in
th
e
d
atase
t.
Fig
u
r
e
2
(
a)
is
th
e
wo
r
d
clo
u
d
p
r
esen
tatio
n
o
f
b
eg
in
n
e
r
class
,
Fig
u
r
e
2
(
b
)
is
th
e
wo
r
d
clo
u
d
p
r
esen
tatio
n
o
f
in
te
r
m
ed
iate
class
,
an
d
Fig
u
r
e
2
(
c)
is
th
e
wo
r
d
cl
o
u
d
p
r
esen
tatio
n
o
f
ad
v
a
n
ce
d
class
.
(
a)
(
b
)
(
c)
Fig
u
r
e
2
.
W
o
r
d
clo
u
d
p
r
esen
ta
tio
n
s
f
o
r
th
e
t
h
r
ee
class
es with
in
th
e
d
ataset
of
(
a
)
b
e
g
in
n
er
,
(
b
)
in
ter
m
e
d
iate,
an
d
(
c)
ad
v
a
n
ce
d
2
.
2
.
P
re
pro
ce
s
s
ing
Pre
p
r
o
ce
s
s
in
g
is
an
ess
en
tial
s
tep
in
NL
P
,
aim
ed
at
en
s
u
r
in
g
d
ata
q
u
ality
an
d
co
n
s
is
ten
cy
[
2
8
]
.
I
n
th
is
r
esear
ch
,
th
r
ee
s
tep
s
ar
e
u
s
ed
.
T
h
e
f
ir
s
t
s
tep
is
id
e
n
tify
in
g
an
d
r
em
o
v
in
g
r
e
d
u
n
d
an
t
d
ata.
Sp
ec
ial
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
Hyb
r
id
co
n
vo
lu
tio
n
a
l n
e
u
r
a
l n
etw
o
r
k
-
b
id
ir
ec
tio
n
a
l lo
n
g
s
h
o
r
t
-
term me
mo
r
y
…
(
Mo
h
a
med
A
min
e
Ou
a
s
s
il
)
2853
ch
ar
ac
ter
s
,
HT
ML
tag
s
,
p
u
n
c
tu
atio
n
m
ar
k
s
,
an
d
u
n
n
ec
ess
ar
y
wh
ites
p
ac
e
ar
e
r
em
o
v
ed
.
T
h
e
s
ec
o
n
d
s
tep
is
r
em
o
v
in
g
s
to
p
wo
r
d
s
s
u
ch
as
co
m
m
o
n
ar
ticles
o
r
p
r
e
p
o
s
itio
n
s
;
th
ese
wo
r
d
s
co
n
tr
ib
u
te
l
ittl
e
to
th
e
o
v
er
all
m
ea
n
in
g
.
T
h
e
f
in
al
s
tep
is
b
r
ea
k
in
g
d
o
wn
th
e
tex
t
i
n
to
t
o
k
en
s
,
m
a
k
in
g
it
ea
s
ier
t
o
a
n
aly
ze
an
d
e
x
tr
ac
t
m
ea
n
in
g
f
u
l in
f
o
r
m
atio
n
.
2
.
3
.
Wo
rd
re
presenta
t
io
n
I
n
o
r
d
er
to
tr
an
s
f
o
r
m
tex
tu
al
p
r
ep
r
o
ce
s
s
ed
to
k
en
s
i
n
to
a
n
u
m
er
ical
r
ep
r
esen
tatio
n
th
at
ca
n
b
e
u
s
ed
as
in
p
u
t
f
o
r
d
ee
p
lear
n
in
g
alg
o
r
ith
m
s
,
th
e
C
B
O
W
-
W
o
r
d
2
Vec
tech
n
iq
u
e
is
p
r
o
p
o
s
ed
f
o
r
r
e
p
r
esen
tin
g
to
k
e
n
s
.
W
o
r
d
2
Vec
is
a
wo
r
d
em
b
ed
d
in
g
s
ap
p
r
o
ac
h
th
at
en
c
o
d
es
lin
g
u
is
tic
m
ea
n
in
g
b
y
p
r
o
je
ctin
g
wo
r
d
s
in
t
o
a
h
ig
h
-
d
im
e
n
s
io
n
al
s
p
ac
e
wh
er
e
th
eir
p
r
o
x
im
ity
r
e
f
lects
s
em
an
tic
s
im
ilar
ity
[
2
9
]
,
[
3
0
]
.
C
B
OW
is
a
v
ar
ian
t
o
f
W
o
r
d
2
Vec
th
at
p
r
ed
icts
a
wo
r
d
g
iv
e
n
its
co
n
tex
t
[
3
1
]
;
C
B
OW
m
o
d
els
ar
e
tr
ain
e
d
b
y
m
a
x
im
izin
g
th
e
a
v
er
ag
e
lo
g
p
r
o
b
ab
ilit
y
[
3
2
]
.
1
∑
(
|
−
,
…
,
−
1
,
+
1
,
…
,
+
)
=
1
(
1
)
W
h
er
e
{
−
,
…
,
−
1
,
+
1
,
…
,
+
}
is
a
s
et
o
f
wo
r
d
s
th
at
r
ep
r
esen
ts
th
e
co
n
tex
t
o
f
th
e
ter
m
.
I
n
th
is
s
tu
d
y
,
th
e
wo
r
d
em
b
e
d
d
in
g
lay
e
r
u
s
in
g
Ar
aVe
c
was
im
p
lem
en
ted
,
s
p
ec
if
ically
th
e
v
er
s
io
n
tr
ain
ed
o
n
th
e
Ar
ab
ic
W
ik
ip
ed
ia
co
r
p
u
s
u
s
in
g
t
h
e
C
B
O
W
ar
ch
itectu
r
e.
T
h
is
p
r
e
-
tr
ai
n
ed
m
o
d
el
p
r
o
v
id
es
300
-
d
im
en
s
io
n
al
d
en
s
e
v
ec
to
r
s
.
I
t
was
co
n
f
ig
u
r
ed
with
a
wi
n
d
o
w
s
ize
o
f
5
an
d
a
m
in
im
u
m
ter
m
f
r
eq
u
en
cy
o
f
2
0
d
u
r
in
g
its
tr
ain
in
g
p
h
ase
[
3
0
]
.
2
.
4
.
Cla
s
s
if
ica
t
io
n t
a
s
k
I
n
th
is
s
tep
,
a
h
y
b
r
id
C
NN
-
B
iLST
M
m
o
d
el
was
d
e
v
elo
p
ed
to
ad
d
r
ess
th
e
Ar
a
b
ic
r
ea
d
ab
ilit
y
ass
es
s
m
en
t.
T
h
e
C
NN
-
B
iLS
T
M
m
o
d
el
r
e
p
r
esen
ts
a
f
u
s
e
d
n
eu
r
al
ar
ch
itectu
r
e
th
at
in
t
eg
r
ates
th
e
f
ea
tu
r
e
ex
tr
ac
tio
n
ca
p
ab
ilit
ies
o
f
C
NNs
with
th
e
tem
p
o
r
al
m
o
d
elin
g
o
f
B
iLST
M
n
etwo
r
k
s
.
C
NN
s
ar
e
u
s
ed
f
o
r
lo
ca
l
f
ea
tu
r
e
ex
tr
ac
tio
n
,
o
f
ten
ca
p
tu
r
in
g
s
p
atial
h
ier
ar
c
h
ies
in
th
e
d
ata
[
3
3
]
.
B
iLST
Ms
ar
e
em
p
lo
y
ed
f
o
r
s
eq
u
e
n
ce
m
o
d
elin
g
,
tak
in
g
in
to
ac
co
u
n
t
b
o
th
p
ast
an
d
f
u
tu
r
e
co
n
tex
t
[
3
4
]
.
T
h
e
ar
c
h
itectu
r
e
is
o
f
ten
r
ep
r
esen
ted
b
y
a
s
eq
u
en
ce
o
f
eq
u
atio
n
s
,
b
u
t
a
s
im
p
lifie
d
f
o
r
m
f
o
r
th
e
B
iLST
M
ce
ll
co
u
ld
be
ℎ
=
(
ℎ
−
1
,
)
,
an
d
f
o
r
th
e
C
NN
lay
er
́
=
(
;
,
)
,
wh
er
e
an
d
ar
e
th
e
f
ilter
weig
h
ts
an
d
b
ias.
Fig
u
r
e
3
illu
s
tr
ates
th
e
ar
c
h
i
tectu
r
e
o
f
th
e
h
y
b
r
id
C
NN
-
B
iLST
M
m
o
d
el,
wh
ich
b
e
g
in
s
with
an
e
m
b
ed
d
in
g
lay
e
r
th
at
t
r
an
s
f
o
r
m
s
in
p
u
t
d
ata
in
to
d
en
s
e
v
ec
t
o
r
s
o
f
f
ix
ed
s
ize.
T
h
is
is
f
o
ll
o
wed
b
y
a
d
r
o
p
o
u
t
lay
er
to
p
r
ev
en
t
o
v
er
f
itti
n
g
b
y
r
an
d
o
m
ly
s
ettin
g
in
p
u
t
u
n
its
to
0
d
u
r
in
g
tr
ain
in
g
.
T
h
e
m
o
d
e
l
th
en
em
p
lo
y
s
two
C
o
n
v
1
D
lay
er
s
,
ea
ch
f
o
llo
we
d
b
y
a
Ma
x
Po
o
lin
g
lay
er
,
to
ex
tr
ac
t
an
d
d
o
w
n
s
am
p
le
f
ea
t
u
r
es
f
r
o
m
th
e
in
p
u
t
d
ata,
ca
p
tu
r
i
n
g
lo
ca
l
d
ep
e
n
d
e
n
cies.
Af
ter
war
d
,
a
B
iLST
M
lay
er
is
u
s
ed
to
lear
n
d
ep
e
n
d
en
cies
f
r
o
m
b
o
th
p
ast
an
d
f
u
t
u
r
e
co
n
tex
t
b
y
p
r
o
ce
s
s
in
g
d
ata
in
b
o
th
d
ir
ec
tio
n
s
.
T
o
en
h
an
ce
tr
ain
i
n
g
s
tab
ilit
y
an
d
s
p
ee
d
,
b
atc
h
n
o
r
m
aliza
tio
n
is
s
u
b
s
eq
u
en
tly
em
p
lo
y
e
d
to
n
o
r
m
ali
ze
th
e
in
p
u
t
d
is
tr
ib
u
tio
n
s
ac
r
o
s
s
ea
c
h
lay
er
.
Fin
ally
,
a
d
en
s
e
lay
er
f
o
llo
we
d
b
y
a
n
a
ctiv
atio
n
lay
er
is
u
s
ed
to
m
ak
e
p
r
ed
ictio
n
s
b
ased
o
n
th
e
lea
r
n
ed
f
ea
tu
r
es,
with
th
e
ac
tiv
atio
n
lay
er
d
eter
m
in
in
g
th
e
o
u
tp
u
t f
o
r
m
at:
b
e
g
in
n
er
,
in
ter
m
ed
iate
,
an
d
ad
v
a
n
ce
d
.
Fig
u
r
e
3
.
T
h
e
lay
er
s
o
f
th
e
h
y
b
r
id
C
NN
-
B
iLST
M
clas
s
if
ier
with
d
im
en
s
io
n
s
an
d
h
y
p
e
r
p
ar
am
eter
s
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.
1
5
,
No
.
3
,
J
u
n
e
2
0
2
6
:
2
8
4
9
-
2
8
6
2
2854
3.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
I
n
th
is
s
ec
tio
n
,
a
d
etailed
o
v
er
v
iew
o
f
t
h
e
ex
p
er
im
en
tal
s
ettin
g
s
an
d
th
e
co
r
r
esp
o
n
d
i
n
g
r
esu
lts
is
p
r
o
v
id
e
d
,
in
cl
u
d
in
g
d
escr
ip
ti
o
n
s
o
f
th
e
m
et
h
o
d
s
u
s
ed
an
d
a
s
u
m
m
ar
y
o
f
th
e
k
ey
f
in
d
i
n
g
s
.
T
o
ass
ess
th
e
p
er
f
o
r
m
an
ce
o
f
th
e
p
r
o
p
o
s
ed
m
o
d
el,
s
tan
d
ar
d
p
er
f
o
r
m
a
n
c
e
m
etr
ics
was
em
p
lo
y
ed
in
cl
u
d
in
g
class
if
icatio
n
ac
cu
r
ac
y
,
p
r
ec
is
io
n
,
r
ec
all,
F
1
-
s
co
r
e,
an
d
ar
ea
u
n
d
e
r
th
e
c
u
r
v
e
(
AUC
)
.
T
h
e
co
n
f
u
s
io
n
a
ls
o
p
r
o
v
id
e
d
.
T
h
ese
m
etr
ics
ar
e
wid
ely
ad
o
p
ted
i
n
th
e
f
ield
o
f
tex
t
class
if
icatio
n
.
Fu
r
th
er
m
o
r
e,
to
e
n
s
u
r
e
a
r
o
b
u
s
t
ev
alu
atio
n
,
a
f
iv
e
-
f
o
ld
c
r
o
s
s
-
v
alid
atio
n
a
p
p
r
o
ac
h
was
em
p
lo
y
e
d
f
o
r
all
ex
p
er
im
en
tal
p
r
o
ce
d
u
r
es.
T
ab
le
3
d
em
o
n
s
tr
ates
th
e
elem
en
ts
o
f
th
e
co
n
f
u
s
io
n
m
at
r
ix
.
T
ab
le
3
.
E
lem
e
n
ts
o
f
th
e
c
o
n
f
u
s
io
n
m
atr
ix
P
r
e
d
i
c
t
e
d
b
e
g
i
n
n
e
r
P
r
e
d
i
c
t
e
d
i
n
t
e
r
me
d
i
a
t
e
P
r
e
d
i
c
t
e
d
a
d
v
a
n
c
e
d
A
c
t
u
a
l
b
e
g
i
n
n
e
r
Tr
u
e
b
e
g
i
n
n
e
r
(
TB
)
M
i
s
c
l
a
ssi
f
i
e
d
a
s
i
n
t
e
r
me
d
i
a
t
e
M
i
s
c
l
a
ssi
f
i
e
d
a
s
a
d
v
a
n
c
e
d
A
c
t
u
a
l
i
n
t
e
r
me
d
i
a
t
e
M
i
s
c
l
a
ssi
f
i
e
d
a
s
b
e
g
i
n
n
e
r
Tr
u
e
i
n
t
e
r
me
d
i
a
t
e
(
TI
)
M
i
s
c
l
a
ssi
f
i
e
d
a
s
a
d
v
a
n
c
e
d
A
c
t
u
a
l
a
d
v
a
n
c
e
d
M
i
s
c
l
a
ssi
f
i
e
d
a
s
b
e
g
i
n
n
e
r
M
i
s
c
l
a
ssi
f
i
e
d
a
s
i
n
t
e
r
me
d
i
a
t
e
Tr
u
e
a
d
v
a
n
c
e
d
(
TA
)
T
h
e
p
r
esen
ted
ex
p
er
im
en
ts
w
er
e
co
n
d
u
cted
o
n
Go
o
g
le
C
o
ll
ab
o
r
ato
r
y
p
latf
o
r
m
,
o
n
a
v
ir
tu
al
m
ac
h
in
e
with
T
4
GPU
an
d
5
4
GB
o
f
R
AM
.
T
o
id
en
tify
o
p
tim
al
co
n
f
ig
u
r
atio
n
s
f
o
r
th
e
p
r
o
p
o
s
ed
m
o
d
el,
a
g
r
id
s
ea
r
ch
ap
p
r
o
ac
h
was
em
p
lo
y
ed
.
T
h
is
r
esear
ch
ap
p
r
o
ac
h
allo
wed
u
s
to
s
y
s
tem
atica
lly
ex
p
lo
r
e
a
r
an
g
e
o
f
p
ar
am
eter
v
alu
es
an
d
s
elec
t
th
e
s
et
th
at
r
esu
lted
in
t
h
e
b
est
p
er
f
o
r
m
a
n
ce
.
T
ab
le
4
p
r
o
v
id
es
a
d
etail
ed
o
v
er
v
iew
o
f
t
h
e
p
r
im
ar
y
p
ar
am
eter
s
an
d
th
eir
r
esp
ec
tiv
e
v
alu
e
r
an
g
es u
s
ed
in
th
is
g
r
id
s
ea
r
ch
.
T
ab
le
4
.
Set o
f
h
y
p
er
p
ar
a
m
eter
s
u
s
ed
b
y
th
e
s
ea
r
ch
g
r
i
d
H
y
p
e
r
p
a
r
a
me
t
e
r
s
V
a
l
u
e
s
W
o
r
d
e
m
b
e
d
d
e
d
d
i
m
e
n
s
i
o
n
[
1
0
0
,
3
0
0
]
N
u
mb
e
r
o
f
1
D
c
o
n
v
o
l
u
t
i
o
n
a
l
l
a
y
e
r
s
[
1
,
2
,
3
,
4
,
5
,
6
,
7
,
8
,
9
,
1
0
]
N
u
mb
e
r
o
f
e
p
o
c
h
s:
[
1
0
,
2
0
,
3
0
]
O
p
t
i
mi
z
e
r
:
[
A
d
a
m,
S
G
D
,
N
a
d
a
m]
D
r
o
p
o
u
t
r
a
t
e
[
0
.
1
,
0
.
2
,
0
.
3
]
Le
a
r
n
i
n
g
r
a
t
e
[
1
0
E
-
3]
B
ased
o
n
th
e
g
r
id
s
ea
r
ch
r
esu
lts
,
th
e
o
p
tim
al
h
y
p
er
p
ar
a
m
eter
co
n
f
ig
u
r
atio
n
f
o
r
th
e
m
o
d
el
was
d
eter
m
in
ed
.
T
h
is
s
etu
p
em
p
lo
y
s
a
wo
r
d
em
b
e
d
d
in
g
d
im
en
s
io
n
o
f
3
0
0
a
n
d
an
ar
c
h
itectu
r
e
th
at
in
clu
d
es
two
1D
-
co
n
v
o
lu
tio
n
al
lay
er
s
.
T
h
e
m
o
d
el
is
tr
ain
ed
u
s
in
g
th
e
Ad
a
m
o
p
tim
izer
f
o
r
2
0
ep
o
c
h
s
with
a
lear
n
in
g
r
ate
o
f
0
.
2
.
W
e
r
eser
v
ed
2
0
%
o
f
th
e
d
ataset
as
an
in
d
ep
en
d
en
t
h
eld
-
o
u
t
test
s
et.
T
h
e
r
em
ain
in
g
8
0
%
was
u
s
ed
f
o
r
m
o
d
el
tr
ain
in
g
an
d
v
alid
atio
n
u
s
in
g
5
-
f
o
ld
cr
o
s
s
-
v
alid
atio
n
.
3
.
1
.
T
ra
ini
ng
a
nd
v
a
lid
a
t
io
n
re
s
ults
I
n
th
is
s
u
b
s
ec
tio
n
,
t
h
e
p
er
f
o
r
m
an
ce
o
f
t
h
e
p
r
o
p
o
s
ed
h
y
b
r
id
m
o
d
el
was
e
x
am
in
ed
b
y
a
n
aly
zin
g
t
h
e
p
r
o
g
r
ess
io
n
o
f
lo
s
s
m
etr
ics
o
v
er
t
h
e
co
u
r
s
e
o
f
th
e
tr
ain
i
n
g
an
d
v
alid
atio
n
p
h
ases
.
Fig
u
r
e
4
s
h
o
ws
th
e
ev
o
lu
tio
n
o
f
v
alid
atio
n
a
n
d
tr
ain
in
g
lo
s
s
es
o
v
er
a
s
er
ies
o
f
2
0
ep
o
ch
s
.
I
n
itially
,
t
h
e
v
alid
atio
n
lo
s
s
s
tar
ts
at
a
v
alu
e
o
f
0
.
6
8
6
,
in
d
icatin
g
a
s
ig
n
if
ican
t
er
r
o
r
b
etwe
en
t
h
e
m
o
d
el'
s
p
r
ed
ictio
n
s
an
d
th
e
ac
tu
al
tar
g
ets.
Ho
wev
er
,
it
d
ec
r
ea
s
es
o
v
er
tim
e,
r
ea
c
h
i
n
g
a
lo
w
o
f
ap
p
r
o
x
im
ately
0
.
1
5
8
at
ep
o
c
h
8
,
b
e
f
o
r
e
ex
p
e
r
ien
cin
g
a
co
n
s
is
ten
t
in
cr
ea
s
e,
r
is
in
g
to
0
.
2
8
1
b
y
th
e
f
in
al
ep
o
c
h
.
T
h
e
tr
ai
n
in
g
lo
s
s
f
o
llo
ws
a
m
o
r
e
co
n
s
is
ten
t
d
o
wn
war
d
tr
ajec
to
r
y
,
s
tar
tin
g
f
r
o
m
0
.
4
2
9
an
d
s
ig
n
if
ican
tly
r
ed
u
cin
g
to
0
.
0
1
9
,
s
h
o
wca
s
in
g
th
e
m
o
d
el'
s
lear
n
in
g
an
d
ad
ap
tatio
n
to
th
e
tr
ain
in
g
d
ata.
T
h
ese
p
atter
n
s
in
d
icate
s
u
cc
ess
f
u
l
m
o
d
el
t
r
ain
in
g
r
e
g
ar
d
in
g
th
e
tr
ai
n
in
g
s
et,
h
ig
h
lig
h
ted
b
y
th
e
s
tead
y
d
ec
r
ea
s
e
in
lo
s
s
v
alu
es.
Ho
wev
er
,
th
e
s
u
b
s
eq
u
en
t
r
is
e
in
v
alid
atio
n
l
o
s
s
s
u
g
g
ests
d
i
s
tin
ct
o
v
er
f
itti
n
g
.
T
o
o
v
er
co
m
e
th
is
o
v
er
f
itti
n
g
p
r
o
b
lem
,
d
r
o
p
o
u
t
tech
n
iq
u
es
ca
n
b
e
i
n
tr
o
d
u
ce
d
,
ef
f
ec
tiv
el
y
im
p
r
o
v
in
g
th
e
m
o
d
el'
s
ab
ilit
y
to
g
en
e
r
alize
b
y
p
r
ev
en
tin
g
co
m
p
lex
c
o
-
ad
a
p
tatio
n
s
o
n
tr
ain
in
g
d
ata.
T
h
is
m
eth
o
d
r
a
n
d
o
m
l
y
d
is
ab
les
a
f
r
ac
tio
n
o
f
th
e
n
eu
r
o
n
s
d
u
r
in
g
th
e
tr
ain
in
g
p
r
o
ce
s
s
,
wh
ich
h
elp
s
in
m
ak
in
g
th
e
m
o
d
el
en
h
an
cin
g
g
en
er
aliza
tio
n
an
d
m
in
im
izin
g
th
e
r
is
k
o
f
o
v
er
f
itti
n
g
o
n
s
p
ec
if
ic
d
etails o
f
th
e
tr
ain
in
g
d
ata.
Fig
u
r
e
5
illu
s
tr
ates
th
e
p
r
o
g
r
e
s
s
o
f
tr
ain
in
g
an
d
v
alid
atio
n
a
cc
u
r
ac
ies
th
r
o
u
g
h
o
u
t
a
2
0
-
ep
o
ch
p
er
io
d
f
o
r
th
e
h
y
b
r
id
d
ee
p
lear
n
in
g
m
o
d
el.
T
h
e
tr
ai
n
in
g
ac
c
u
r
a
cy
co
m
m
en
ce
d
at
8
4
.
9
5
%,
wh
ich
co
n
s
is
ten
tly
im
p
r
o
v
e
d
,
c
u
lm
in
atin
g
i
n
9
9
.
2
9
%
b
y
th
e
e
n
d
o
f
t
h
e
tr
ain
in
g
.
T
h
is
s
tead
y
in
cr
ea
s
e
d
e
m
o
n
s
tr
ates
th
e
m
o
d
el'
s
ef
f
ec
tiv
e
lear
n
in
g
f
r
o
m
t
h
e
tr
ain
in
g
d
ata.
On
th
e
v
alid
atio
n
s
id
e,
ac
cu
r
ac
y
b
e
g
an
at
6
6
.
4
5
%,
a
f
ig
u
r
e
th
at
h
ig
h
lig
h
ts
th
e
in
itial
ch
allen
g
e
th
e
m
o
d
el
f
ac
ed
in
g
en
e
r
alizin
g
to
u
n
s
ee
n
d
ata.
Desp
ite
s
o
m
e
f
lu
ctu
atio
n
s
,
v
alid
atio
n
ac
cu
r
ac
y
s
aw
a
n
o
t
ab
le
r
is
e,
ac
h
iev
in
g
h
ig
h
s
s
u
c
h
as
9
6
.
6
0
%
an
d
co
n
clu
d
in
g
a
t
9
6
.
1
5
%
in
th
e
f
in
al
ep
o
ch
.
B
o
th
tr
ain
in
g
a
n
d
v
ali
d
atio
n
ac
cu
r
ac
ies
im
p
r
o
v
ed
p
r
o
g
r
ess
iv
ely
d
u
r
i
n
g
th
e
2
0
-
ep
o
ch
tr
ain
in
g
p
e
r
io
d
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
Hyb
r
id
co
n
vo
lu
tio
n
a
l n
e
u
r
a
l n
etw
o
r
k
-
b
id
ir
ec
tio
n
a
l lo
n
g
s
h
o
r
t
-
term me
mo
r
y
…
(
Mo
h
a
med
A
min
e
Ou
a
s
s
il
)
2855
h
ig
h
lig
h
ts
th
e
m
o
d
el'
s
s
u
cc
ess
f
u
l
ad
ap
tatio
n
an
d
o
p
tim
iz
atio
n
,
s
h
o
wca
s
in
g
its
ab
ilit
y
to
lear
n
f
r
o
m
t
h
e
tr
ain
in
g
d
ata
e
f
f
ec
tiv
ely
wh
ile
m
ain
tain
in
g
r
o
b
u
s
t p
er
f
o
r
m
a
n
ce
o
n
u
n
s
ee
n
d
ata.
Fig
u
r
e
4
.
Valid
atio
n
an
d
t
r
ain
i
n
g
lo
s
s
o
f
th
e
p
r
o
p
o
s
ed
m
o
d
el
Fig
u
r
e
5
.
T
r
ain
in
g
a
n
d
v
alid
atio
n
ac
cu
r
ac
y
o
f
th
e
p
r
o
p
o
s
ed
C
NN
-
B
iLST
M
m
o
d
el
3
.
2
.
E
v
a
lua
t
ing
t
he
im
pa
ct
o
f
inte
g
ra
t
ing
CNN
la
y
er
s
int
o
a
B
iL
ST
M
T
h
e
im
p
ac
t
o
f
in
teg
r
atin
g
C
NN
lay
er
s
in
to
a
B
iLST
M
was
ex
am
in
ed
b
y
s
tu
d
y
in
g
t
h
e
ef
f
ec
t
o
f
in
cr
em
en
tally
ad
d
in
g
v
ar
io
u
s
n
u
m
b
er
s
o
f
C
o
n
v
1
D
lay
er
s
.
T
h
is
ex
am
in
atio
n
aim
ed
to
d
eter
m
in
e
th
e
o
p
tim
al
co
n
v
o
l
u
tio
n
al
d
ep
th
th
at
en
h
an
ce
s
f
ea
tu
r
e
ex
tr
ac
tio
n
wh
il
e
m
ain
tain
in
g
th
e
tem
p
o
r
al
l
ea
r
n
in
g
s
tr
en
g
th
s
o
f
B
iLST
M.
T
h
e
r
esu
lts
illu
s
tr
ate
th
e
im
p
ac
t
o
f
C
o
n
v
1
D
lay
er
d
ep
th
e
v
alu
ated
ac
r
o
s
s
m
u
ltip
l
e
s
tan
d
ar
d
m
etr
ics,
s
p
ec
if
ically
ac
cu
r
ac
y
,
p
r
ec
is
io
n
,
r
ec
all
,
F1
-
s
co
r
e
,
an
d
AUC,
alo
n
g
s
id
e
test
in
g
ex
ec
u
tio
n
ti
m
e.
T
ab
le
5
an
d
Fig
u
r
e
6
d
em
o
n
s
tr
ate
th
at
au
g
m
en
tin
g
th
e
n
u
m
b
er
o
f
C
o
n
v
1
D
lay
er
s
to
2
en
h
an
ce
s
p
er
f
o
r
m
an
ce
ac
r
o
s
s
all
ev
alu
a
ted
m
etr
ics.
Sp
ec
if
ically
,
th
e
m
o
d
el
ac
h
iev
es
p
ea
k
v
alu
es
with
an
ac
cu
r
ac
y
o
f
9
6
.
6
8
%,
p
r
ec
is
io
n
o
f
9
6
.
7
0
%,
r
ec
all
o
f
9
6
.
6
8
%,
an
d
an
F1
-
s
co
r
e
o
f
9
6
.
6
8
%.
Ad
d
itio
n
ally
,
th
e
AUC
r
ea
ch
es
a
h
ig
h
o
f
9
9
.
0
4
%,
i
n
d
icatin
g
s
tr
o
n
g
d
is
cr
im
in
ativ
e
a
b
ilit
y
.
T
h
is
co
n
f
ig
u
r
atio
n
also
ac
h
iev
es
an
ef
f
icien
t
test
in
g
ex
ec
u
tio
n
tim
e
o
f
0
.
8
6
3
s
ec
o
n
d
s
,
as
s
h
o
wn
in
Fig
u
r
e
7
,
s
u
g
g
esti
n
g
an
im
p
r
o
v
em
en
t
i
n
f
ea
tu
r
e
ex
tr
ac
ti
o
n
with
o
u
t im
p
o
s
in
g
e
x
ce
s
s
iv
e
co
m
p
u
tatio
n
al
o
v
er
h
ea
d
.
T
ab
le
5
.
Acc
u
r
ac
y
,
p
r
ec
is
io
n
,
r
ec
all,
F1
-
s
co
r
e,
an
d
AUC m
etr
ics ac
r
o
s
s
d
if
f
er
en
t n
u
m
b
er
s
o
f
C
o
n
v
1
D
la
y
er
s
N
u
mb
e
r
o
f
a
d
d
e
d
C
o
n
v
1
D
l
a
y
e
r
s
A
c
c
u
r
a
c
y
(
%)
P
r
e
c
i
s
i
o
n
(
%)
R
e
c
a
l
l
(
%)
F1
-
s
c
o
r
e
(
%)
A
U
C
(
%)
1
9
5
.
5
8
±
0
.
8
5
9
5
.
6
6
±
0
.
7
8
9
5
.
5
8
±
0
.
8
5
9
5
.
5
8
±
0
.
8
6
9
9
.
0
9
±
0
.
1
4
2
9
6
.
6
8
±
0
.
2
5
9
6
.
7
0
±
0
.
2
6
9
6
.
6
8
±
0
.
2
5
9
6
.
6
8
±
0
.
2
5
9
9
.
0
4
±
0
.
1
5
3
9
6
.
1
3
±
0
.
4
9
9
6
.
1
9
±
0
.
4
4
9
6
.
1
3
±
0
.
4
9
9
6
.
1
2
±
0
.
4
9
9
8
.
8
2
±
0
.
2
2
4
9
6
.
0
2
±
0
.
4
4
9
6
.
0
5
±
0
.
4
5
9
6
.
0
2
±
0
.
4
4
9
6
.
0
2
±
0
.
4
4
9
8
.
7
6
±
0
.
1
1
5
9
6
.
2
8
±
0
.
1
7
9
6
.
3
2
±
0
.
1
8
9
6
.
2
8
±
0
.
1
7
9
6
.
2
8
±
0
.
1
7
9
8
.
9
1
±
0
.
1
8
6
9
6
.
0
2
±
0
.
5
2
9
6
.
0
7
±
0
.
5
0
9
6
.
0
2
±
0
.
5
2
9
6
.
0
2
±
0
.
5
1
9
8
.
9
9
±
0
.
1
1
7
9
5
.
6
6
±
0
.
3
5
9
5
.
7
0
±
0
.
3
4
9
5
.
6
6
±
0
.
3
5
9
5
.
6
6
±
0
.
3
4
9
9
.
1
0
±
0
.
1
2
8
9
5
.
7
5
±
0
.
7
8
9
5
.
7
8
±
0
.
7
6
9
5
.
7
5
±
0
.
7
8
9
5
.
7
5
±
0
.
7
8
9
9
.
0
6
±
0
.
1
7
9
9
6
.
1
3
±
0
.
2
0
9
6
.
1
6
±
0
.
2
1
9
6
.
1
3
±
0
.
2
0
9
6
.
1
3
±
0
.
2
0
9
9
.
2
2
±
0
.
0
8
10
9
4
.
9
8
±
0
.
5
5
9
5
.
0
6
±
0
.
5
2
9
4
.
9
8
±
0
.
5
5
9
4
.
9
8
±
0
.
5
6
9
8
.
9
6
±
0
.
2
6
Fig
u
r
e
6
.
Acc
u
r
ac
y
v
ar
iatio
n
a
cr
o
s
s
d
if
f
er
en
t n
u
m
b
e
r
s
o
f
C
o
n
v
1
D
lay
er
s
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.
1
5
,
No
.
3
,
J
u
n
e
2
0
2
6
:
2
8
4
9
-
2
8
6
2
2856
Fig
u
r
e
7
.
T
esti
n
g
tim
e
i
n
s
ec
o
n
d
s
ac
r
o
s
s
d
if
f
er
e
n
t n
u
m
b
er
s
o
f
C
o
n
v
1
D
l
ay
e
r
s
Ho
wev
er
,
b
ey
o
n
d
t
h
is
o
p
tim
al
p
o
in
t,
f
u
r
th
er
ad
d
itio
n
o
f
lay
er
s
lead
s
to
a
d
e
g
r
ad
atio
n
in
th
ese
co
m
p
r
eh
e
n
s
iv
e
m
etr
ics,
s
u
g
g
esti
n
g
o
v
er
f
itti
n
g
.
T
h
e
f
lu
ctu
atin
g
v
alu
es
with
lay
er
d
ep
th
s
b
ey
o
n
d
3
in
d
icat
e
d
im
in
is
h
in
g
r
etu
r
n
s
f
r
o
m
in
c
r
ea
s
ed
m
o
d
el
co
m
p
lex
ity
.
Fo
r
in
s
tan
ce
,
at
1
0
lay
e
r
s
,
th
e
ac
cu
r
ac
y
d
r
o
p
s
to
9
4
.
9
8
%,
with
co
r
r
esp
o
n
d
in
g
d
ec
lin
es
in
p
r
ec
is
io
n
9
5
.
0
6
%,
r
ec
all
9
4
.
9
8
%,
an
d
F1
-
s
co
r
e
9
4
.
9
8
%
.
Simu
ltan
eo
u
s
ly
,
th
e
co
m
p
u
t
atio
n
al
co
s
t
r
is
es,
in
d
icate
d
b
y
t
h
e
test
in
g
e
x
ec
u
tio
n
tim
e
p
ea
k
i
n
g
at
0
.
9
6
3
s
ec
o
n
d
s
,
wh
ich
f
u
r
th
e
r
s
u
g
g
ests
th
at
d
ee
p
er
m
o
d
els
m
ay
s
u
f
f
er
f
r
o
m
o
v
er
p
ar
am
eter
i
za
tio
n
an
d
a
f
ailu
r
e
to
g
en
er
alize
ef
f
ec
tiv
ely
.
T
h
e
f
in
d
in
g
s
r
ev
ea
led
th
at
th
e
in
teg
r
atio
n
o
f
two
C
o
n
v
1
D
lay
e
r
s
r
ep
r
esen
ts
th
e
o
p
tim
al
co
n
f
ig
u
r
atio
n
.
T
h
is
p
a
r
ticu
lar
s
etu
p
ac
h
iev
es
a
s
u
p
er
io
r
b
alan
ce
b
etwe
en
r
ep
r
esen
tatio
n
al
ca
p
ac
ity
an
d
g
en
er
aliza
tio
n
.
T
h
is
was
em
p
ir
ically
v
alid
ated
b
y
th
e
m
o
d
el
ac
h
iev
in
g
p
ea
k
s
co
r
es
a
cr
o
s
s
p
r
ec
is
io
n
,
r
ec
all,
an
d
F1
-
s
co
r
e
m
etr
ics wh
ile
m
ain
tain
in
g
co
m
p
u
tatio
n
al
e
f
f
ic
ien
t.
3
.
3
.
E
v
a
lua
t
ing
t
he
im
pa
ct
o
f
co
m
bin
ing
CNN
a
nd
B
iL
S
T
M
T
o
an
aly
ze
th
e
e
f
f
ec
t
o
f
th
e
co
m
b
in
atio
n
o
f
C
NN
an
d
B
i
L
STM
,
we
co
m
p
a
r
e
th
e
h
y
b
r
id
m
o
d
el'
s
p
er
f
o
r
m
an
ce
m
etr
ics
an
d
ex
e
cu
tio
n
tim
e
with
its
p
ar
ts
:
tr
ad
itio
n
al
C
NN
an
d
B
iLST
M
m
o
d
els.
Fig
u
r
e
8
d
em
o
n
s
tr
ates
th
e
s
u
p
er
io
r
ity
o
f
th
e
C
NN
-
B
iLST
M
h
y
b
r
id
m
o
d
el,
wh
ich
o
u
tp
e
r
f
o
r
m
s
th
e
in
d
iv
id
u
al
C
NN
an
d
B
iLST
M
m
o
d
els
ac
r
o
s
s
all
m
etr
ics.
I
t
lead
s
b
y
ap
p
r
o
x
im
ately
1
.
7
5
%
in
ac
c
u
r
ac
y
a
n
d
1
.
6
5
%
in
p
r
ec
is
io
n
co
m
p
ar
ed
to
B
iLST
M,
an
d
n
ea
r
ly
2
.
0
6
%
a
n
d
1
.
9
9
%
r
e
s
p
ec
tiv
ely
o
v
er
C
NN,
s
ig
n
if
y
in
g
a
s
ig
n
if
ican
t
im
p
r
o
v
em
e
n
t.
T
h
is
ad
v
a
n
tag
e
ex
ten
d
s
to
r
ec
all
an
d
F1
-
s
co
r
e
,
co
n
f
ir
m
i
n
g
th
at
th
e
h
y
b
r
i
d
m
o
d
el
ex
ce
ls
in
b
o
th
p
r
ec
is
io
n
an
d
co
m
p
r
eh
en
s
iv
e
d
ata
an
aly
s
is
,
m
ak
in
g
it a
m
o
r
e
ef
f
ec
tiv
e
to
o
l
f
o
r
c
o
m
p
lex
class
if
icatio
n
task
s
.
Fig
u
r
e
8
.
C
o
m
p
a
r
ativ
e
m
etr
ics
o
f
B
iLST
M
m
o
d
el
v
s
.
tr
a
d
itio
n
al
C
NN
an
d
B
iLST
M
m
o
d
els
Fig
u
r
e
9
illu
s
tr
ates
th
at
th
e
C
NN
-
B
iLST
M
m
o
d
el
m
ar
k
s
a
b
alan
ce
b
etwe
en
ef
f
ici
en
cy
an
d
co
m
p
u
tatio
n
al
tim
e.
Fig
u
r
e
9
(
a)
s
h
o
ws
th
at
th
e
tr
ain
in
g
tim
e
f
o
r
C
NN
-
B
iLST
M
is
1
6
2
.
9
8
s
,
wh
ich
is
n
o
tab
ly
less
th
an
B
iLST
M
'
s
2
4
4
.
9
5
s
,
s
u
g
g
esti
n
g
a
m
o
r
e
e
f
f
icien
t
tr
a
in
in
g
p
r
o
ce
s
s
wh
ile
m
ain
tain
i
n
g
p
e
r
f
o
r
m
an
ce
.
I
n
co
n
tr
ast,
C
NN
h
as
th
e
s
h
o
r
tes
t
tr
ain
in
g
tim
e
at
1
0
1
.
9
s
b
u
t
it
d
o
es
n
o
t
p
r
o
v
id
e
th
e
s
am
e
lev
el
o
f
ac
cu
r
ac
y
as
th
e
h
y
b
r
id
m
o
d
el.
Fig
u
r
e
9
(
b
)
s
h
o
ws
th
at
d
u
r
in
g
test
in
g
C
NN
-
B
iLST
M
an
d
C
NN
h
av
e
s
im
ilar
tim
es
0
.
8
6
s
an
d
0
.
9
s
,
b
u
t
B
iLST
M
is
s
ig
n
if
ican
tly
s
lo
wer
at
1
.
4
4
s
.
T
h
u
s
,
th
e
C
NN
-
B
iLST
M
h
y
b
r
i
d
o
f
f
er
s
an
o
p
tim
al
b
len
d
o
f
p
er
f
o
r
m
a
n
ce
an
d
tr
ai
n
in
g
ef
f
icien
cy
,
id
ea
l
f
o
r
s
ce
n
ar
io
s
wh
er
e
b
o
t
h
f
ac
to
r
s
ar
e
cr
itical.
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
Hyb
r
id
co
n
vo
lu
tio
n
a
l n
e
u
r
a
l n
etw
o
r
k
-
b
id
ir
ec
tio
n
a
l lo
n
g
s
h
o
r
t
-
term me
mo
r
y
…
(
Mo
h
a
med
A
min
e
Ou
a
s
s
il
)
2857
(
a)
(
b
)
Fig
u
r
e
9
.
E
x
ec
u
tio
n
tim
e
o
f
C
NN
-
B
iLST
M,
C
NN,
an
d
B
iLST
M
f
o
r
(
a)
t
r
ain
in
g
a
n
d
tim
e
(
b
)
test
in
g
tim
e
Fig
u
r
e
1
0
illu
s
tr
ates
th
e
co
n
f
u
s
io
n
m
atr
ices
f
o
r
th
r
ee
m
o
d
els
,
h
y
b
r
id
C
NN
-
B
iLST
M,
C
NN,
an
d
B
iLST
M
in
class
if
y
in
g
s
en
ten
ce
s
ac
r
o
s
s
th
r
ee
d
if
f
icu
lty
l
ev
els:
b
eg
in
n
e
r
,
in
ter
m
ed
iate,
an
d
ad
v
a
n
ce
d
.
I
n
Fig
u
r
e
1
0
(
a)
,
th
e
h
y
b
r
id
C
NN
-
B
iLST
M
m
o
d
el
d
em
o
n
s
tr
ated
n
o
tab
le
p
r
ec
is
io
n
,
co
r
r
ec
tly
id
en
tify
in
g
250
b
e
g
in
n
e
r
s
en
ten
ce
s
,
3
5
8
in
ter
m
ed
iate
s
en
ten
ce
s
,
a
n
d
2
6
5
ad
v
an
ce
d
s
en
te
n
ce
s
.
I
t
s
h
o
ws
m
in
im
al
m
is
class
if
icatio
n
s
,
with
5
in
ter
m
ed
iate
s
en
ten
ce
s
an
d
3
ad
v
a
n
ce
d
s
en
ten
ce
s
in
co
r
r
ec
tly
class
if
ied
as
b
eg
in
n
er
;
1
in
ter
m
ed
iate
s
en
ten
ce
m
is
class
if
ied
as
ad
v
an
ce
d
;
an
d
9
ad
v
an
ce
d
s
en
ten
ce
s
wr
o
n
g
ly
id
en
tifie
d
as
in
ter
m
ed
iate.
I
n
Fig
u
r
e
1
0
(
b
)
,
th
e
C
NN
m
o
d
el
ac
cu
r
atel
y
class
if
ied
2
5
0
b
e
g
in
n
er
s
en
ten
ce
s
b
u
t
f
ac
ed
ch
allen
g
es
with
h
ig
h
er
co
m
p
lex
ity
lev
els,
m
is
class
if
y
in
g
9
ad
v
an
ce
d
s
en
ten
ce
s
as
in
ter
m
ed
iate
an
d
16
in
ter
m
ed
iate
s
en
ten
ce
s
as
ad
v
an
ce
d
,
alo
n
g
s
id
e
s
m
aller
er
r
o
r
s
i
n
o
th
e
r
ca
teg
o
r
ies.
I
n
Fig
u
r
e
1
0
(
c
)
,
th
e
B
iLST
M
m
o
d
el
s
h
o
wed
co
n
s
is
ten
t
ac
cu
r
ac
y
,
co
r
r
ec
tly
class
if
y
in
g
2
4
4
b
e
g
in
n
er
,
3
5
0
in
t
er
m
ed
iate,
an
d
2
6
3
ad
v
an
ce
d
s
en
ten
ce
s
.
T
h
ese
o
u
tco
m
es
u
n
d
er
s
co
r
e
th
e
h
y
b
r
id
C
NN
-
B
iL
STM
m
o
d
el'
s
s
u
p
er
io
r
p
er
f
o
r
m
a
n
ce
i
n
ac
cu
r
ately
d
is
ce
r
n
in
g
a
wid
e
r
an
g
e
o
f
s
en
te
n
ce
c
o
m
p
lex
ities
with
h
i
g
h
p
r
ec
is
io
n
an
d
m
in
i
m
al
er
r
o
r
s
,
m
a
r
k
in
g
it a
s
p
ar
ticu
lar
ly
ef
f
ec
tiv
e
f
o
r
d
etailed
an
d
p
r
ec
is
e
class
if
icat
io
n
task
s
.
(
a)
(
b
)
(
c)
Fig
u
r
e
1
0
.
C
o
n
f
u
s
io
n
m
atr
ices o
f
th
e
th
r
ee
m
o
d
els:
(
a)
h
y
b
r
i
d
C
NN
-
B
iLST
M,
(
b
)
C
NN,
an
d
(
c)
B
iLST
M
3
.
4
.
Dis
cus
s
io
n
I
n
th
is
s
tu
d
y
,
a
n
o
v
el
s
y
s
tem
f
o
r
ass
ess
in
g
th
e
d
if
f
icu
lty
lev
els
o
f
Ar
a
b
ic
s
en
ten
ce
s
was
i
n
tr
o
d
u
ce
d
,
em
p
lo
y
in
g
a
h
y
b
r
id
d
ee
p
lear
n
in
g
m
o
d
el.
T
h
e
m
eth
o
d
in
v
o
lv
es
a
f
o
u
r
-
s
tep
p
r
o
ce
s
s
,
in
clu
d
in
g
d
ata
p
r
esen
tatio
n
,
p
r
ep
r
o
ce
s
s
in
g
,
s
em
an
tic
n
u
m
er
ical
co
n
v
er
s
io
n
u
s
in
g
th
e
W
o
r
d
2
Vec
tec
h
n
iq
u
e,
a
n
d
s
en
ten
ce
class
if
icatio
n
u
s
in
g
th
is
d
ee
p
lear
n
in
g
m
o
d
el.
T
h
e
s
tu
d
y
in
t
o
th
e
h
y
b
r
id
m
o
d
el
ar
ch
itectu
r
e
h
ig
h
lig
h
te
d
th
e
b
en
ef
its
o
f
co
m
b
in
in
g
C
NN
la
y
er
s
with
a
B
iLST
M
f
r
am
ewo
r
k
,
f
o
cu
s
in
g
o
n
f
in
d
in
g
th
e
o
p
tim
al
co
m
b
in
atio
n
f
o
r
ef
f
ec
tiv
e
f
ea
tu
r
e
ex
tr
ac
tio
n
.
T
h
e
in
teg
r
atio
n
o
f
two
C
o
n
v
1
D
lay
er
s
was
f
o
u
n
d
to
b
e
m
o
s
t
b
en
ef
icial,
ac
h
iev
in
g
an
im
p
r
ess
iv
e
ac
cu
r
ac
y
o
f
9
6
.
6
8
%,
wh
ich
s
h
o
wca
s
es
th
e
p
o
ten
tial
o
f
C
NN
an
d
B
iLST
M
lay
er
s
in
p
r
o
ce
s
s
in
g
co
m
p
lex
d
ata.
T
h
e
h
y
b
r
id
C
NN
-
B
iLST
M
m
o
d
el's
p
er
f
o
r
m
a
n
ce
was
co
m
p
ar
ed
with
tr
ad
itio
n
al
C
NN
an
d
B
iLST
M
m
o
d
els
a
n
d
n
in
e
tr
ad
itio
n
al
m
ac
h
in
e
lear
n
in
g
class
if
ier
s
.
Fig
u
r
e
1
1
illu
s
tr
ates
th
e
R
O
C
cu
r
v
es a
n
d
AUC s
co
r
es sh
o
win
g
th
e
p
er
f
o
r
m
an
ce
o
f
th
e
m
o
d
el
ac
r
o
s
s
all
tar
g
et
class
es.
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.
1
5
,
No
.
3
,
J
u
n
e
2
0
2
6
:
2
8
4
9
-
2
8
6
2
2858
Fig
u
r
e
1
1
.
R
OC
cu
r
v
es a
n
d
A
UC
s
co
r
es sh
o
win
g
th
e
p
er
f
o
r
m
an
ce
o
f
t
h
e
Ar
aVe
c
-
b
ased
m
o
d
el
ac
r
o
s
s
all
tar
g
et
class
es
T
h
is
r
esear
ch
u
n
d
e
r
s
co
r
es
th
e
ad
v
an
tag
es
o
f
h
y
b
r
id
d
ee
p
lear
n
in
g
m
o
d
els
in
lin
g
u
is
tic
an
aly
s
is
,
p
ar
ticu
lar
ly
f
o
r
lan
g
u
ag
es
with
co
m
p
lex
s
tr
u
ctu
r
es
lik
e
Ar
ab
ic.
T
h
e
o
u
ts
tan
d
in
g
p
er
f
o
r
m
an
ce
o
f
th
e
C
NN
-
B
iL
STM
m
o
d
el
s
ets
a
n
ew
b
en
c
h
m
ar
k
f
o
r
s
en
ten
ce
d
if
f
icu
lty
class
if
icatio
n
,
s
u
g
g
esti
n
g
a
p
r
o
m
is
in
g
d
ir
ec
tio
n
f
o
r
f
u
tu
r
e
ad
v
an
ce
m
en
ts
in
co
m
p
u
tatio
n
al
lin
g
u
is
tics
.
Fu
r
th
er
ex
p
lo
r
atio
n
co
u
ld
ex
ten
d
th
e
m
o
d
el'
s
ap
p
licatio
n
to
o
th
er
lin
g
u
is
tic
task
s
,
en
h
an
cin
g
its
u
tili
ty
an
d
im
p
ac
t in
th
e
f
ield
.
Fo
r
a
clo
s
er
l
o
o
k
at
th
e
er
r
o
r
c
lass
if
icatio
n
,
T
ab
le
6
s
h
o
ws
e
x
am
p
les
o
f
m
is
class
if
ied
tex
t.
I
n
th
e
f
ir
s
t
ex
am
p
le,
th
e
m
o
d
el
m
is
class
if
ied
an
i
n
ter
m
ed
iate
s
en
ten
ce
a
s
ad
v
an
ce
d
.
Sev
er
al
r
ea
s
o
n
s
c
o
u
ld
b
e
b
e
h
in
d
th
is
lik
e
v
o
ca
b
u
lar
y
an
d
th
e
f
r
e
q
u
e
n
t u
s
e
o
f
n
o
u
n
s
in
s
tead
o
f
v
e
r
b
s
(
n
o
m
in
aliza
tio
n
)
,
s
u
ch
as '
داص
ت
ق
ا
'
(
m
o
d
er
atio
n
)
an
d
'
ء
افتكا
'
(
co
n
ten
tm
en
t)
.
T
h
e
s
en
ten
ce
also
in
clu
d
es
m
ed
i
ca
l
ter
m
(
ca
lo
r
ies,
d
iab
etes)
a
n
d
f
o
r
m
al
wo
r
d
s
lik
e
'
مت
حي
'
(
n
ec
ess
itate
s
)
.
T
h
e
m
o
d
el
lik
ely
c
o
n
s
id
er
s
th
e
d
if
f
icu
lt
v
o
ca
b
u
lar
y
as
a
s
ig
n
o
f
th
e
ad
v
an
ce
d
class
.
I
n
th
e
s
ec
o
n
d
ex
am
p
le,
th
e
lev
el
o
f
an
ad
v
an
ce
d
s
en
ten
ce
was
co
n
s
id
er
ed
as
in
te
r
m
ed
iate
,
b
ec
au
s
e
th
e
o
f
th
e
s
im
p
licity
o
f
th
e
s
en
ten
ce
s
tr
u
ctu
r
e
a
n
d
t
h
e
u
s
e
o
f
th
e
ac
t
iv
e
p
r
esen
t
-
ten
s
e
v
er
b
s
lik
e
'
ر
ب
ت
ع
ت
'
an
d
'
مهست
'
a
n
d
co
m
m
o
n
c
o
n
n
ec
to
r
s
lik
e
'
نكل
',
r
ath
er
th
an
th
e
co
m
p
lex
n
o
u
n
p
h
r
ases
u
s
u
ally
f
o
u
n
d
in
ad
v
an
ce
d
Ar
ab
ic.
Alth
o
u
g
h
th
e
i
d
ea
s
ar
e
co
m
p
lex
(
co
g
n
itiv
e
f
ac
u
lties
an
d
d
ee
p
th
in
k
i
n
g
)
,
t
h
e
s
p
ec
if
ic
wo
r
d
s
ar
e
co
m
m
o
n
('
ة
ء
ارق
لا
'
f
o
r
r
ea
d
in
g
,
'
ع
ي
ر
س
'
f
o
r
f
ast).
T
h
e
m
o
d
el
ass
o
ciate
s
th
ese
co
m
m
o
n
wo
r
d
s
with
in
ter
m
ed
iate
to
p
ics
an
d
m
is
s
es
th
e
d
ee
p
er
ac
ad
em
ic
co
n
tex
t.
T
h
is
s
h
o
ws
th
at
th
e
m
o
d
el
r
elies
lar
g
ely
o
n
s
u
r
f
ac
e
-
lev
el
asp
ec
ts
;
er
r
o
r
s
o
cc
u
r
wh
e
n
s
im
p
le
co
n
ce
p
ts
u
tili
ze
f
o
r
m
al
lan
g
u
a
g
e,
o
r
wh
e
n
co
m
p
le
x
th
o
u
g
h
ts
ar
e
wr
itte
n
ex
p
licitly
.
T
ab
le
6
.
E
x
am
p
les o
f
m
is
class
if
ied
tex
t
Te
x
t
A
c
t
u
a
l
l
a
b
e
l
P
r
e
d
i
c
t
e
d
l
a
b
e
l
Ex
a
m
p
l
e
1
:
ةن
م
س
لا
ن
م
ي
ن
ا
ع
ي
ن
م
و
ي
ر
ك
س
لا
ض
ي
ر
م
ى
لع م
ت
ح
ي
ةي
ر
ا
ر
ح
لا
ت
ار
ع
س
لا
و
ر
ك
س
لا
ن
م
ي
لا
ع
لا
ا
هاوت
ح
م
ن
أ
ى
لإ
ها
ب
ت
ن
ا
ب
ج
ي
و
د
ا
ص
ت
ق
ا
.
ةد
ح
او
ةع
طق
ب
ء
ا
ف
ت
ك
او
،
ا
ه
لوان
ت
ي
ف
Tr
a
n
s
l
a
t
i
o
n
:
A
t
t
e
n
t
i
o
n
mu
s
t
b
e
p
a
i
d
t
o
i
t
s h
i
g
h
s
u
g
a
r
a
n
d
c
a
l
o
r
i
e
c
o
n
t
e
n
t
,
w
h
i
c
h
r
e
q
u
i
r
e
s
p
a
t
i
e
n
t
s
w
i
t
h
d
i
a
b
e
t
e
s
a
n
d
t
h
o
se
s
u
f
f
e
r
i
n
g
f
r
o
m
o
b
e
si
t
y
t
o
c
o
n
su
me
i
t
i
n
m
o
d
e
r
a
t
i
o
n
,
l
i
m
i
t
i
n
g
t
h
e
ms
e
l
v
e
s
t
o
j
u
st
o
n
e
p
i
e
c
e
.
I
n
t
e
r
med
i
a
t
e
A
d
v
a
n
c
e
d
Ex
a
m
p
l
e
2
:
ن
ك
ل
ي
ف
م
ه
س
ت
ي
ت
لا
ك
ار
د
او
أ
طب
ا
م
ه
ف
لا
ت
ا
ي
لم
ع
ر
وطت
د
ح
ي
ن
أ
ن
ك
م
ي
عي
ر
س
لا
ح
ف
ص
ت
لا
ن
م
ع
ون
ل
ا
اذ
ه نأ
ر
ب
ت
ع
ت
فل
وو
ن
ا
ي
ر
ا
م
قي
م
ع
ل
ا
ر
ي
ك
ف
ت
لا
و
ةء
ار
ق
لا
ت
ا
ك
لم
ن
ي
وك
ت
.
Tr
a
n
s
l
a
t
i
o
n
:
H
o
w
e
v
e
r
,
M
a
r
y
a
n
n
e
W
o
l
f
a
r
g
u
e
s
t
h
a
t
t
h
i
s
t
y
p
e
o
f
r
a
p
i
d
s
k
i
mm
i
n
g
c
a
n
l
i
mi
t
t
h
e
d
e
v
e
l
o
p
me
n
t
o
f
sl
o
w
e
r
c
o
mp
r
e
h
e
n
s
i
o
n
a
n
d
c
o
g
n
i
t
i
v
e
p
r
o
c
e
sse
s,
w
h
i
c
h
c
o
n
t
r
i
b
u
t
e
t
o
f
o
r
mi
n
g
t
h
e
f
a
c
u
l
t
i
e
s
o
f
r
e
a
d
i
n
g
a
n
d
d
e
e
p
t
h
i
n
k
i
n
g
.
A
d
v
a
n
c
e
d
I
n
t
e
r
med
i
a
t
e
Th
e
m
o
d
el
r
elies
o
n
f
o
r
m
al
s
tr
u
ctu
r
es,
lar
g
ely
b
ec
a
u
s
e
th
e
wo
r
d
em
b
e
d
d
in
g
s
wer
e
tr
ain
ed
o
n
W
ik
ip
ed
ia
d
ata.
T
h
is
r
estricts
th
e
m
o
d
el'
s
ab
ilit
y
to
g
en
er
alize
,
as
th
e
f
o
r
m
al
r
eg
is
ter
u
s
ed
in
W
ik
ip
ed
ia
d
o
es
n
o
t
r
ef
lect
th
e
d
ialec
ts
an
d
n
o
n
-
s
tan
d
ar
d
s
y
n
tax
o
f
ten
f
o
u
n
d
o
n
s
o
cial
p
latf
o
r
m
s
.
Ar
aV
ec
-
T
witter
co
u
ld
b
e
Evaluation Warning : The document was created with Spire.PDF for Python.