I
nte
rna
t
io
na
l J
o
urna
l o
f
E
lect
rica
l a
nd
Co
m
p
ute
r
E
ng
in
ee
ring
(
I
J
E
CE
)
Vo
l.
11
,
No
.
4
,
A
u
g
u
s
t
2021
,
p
p
.
3
5
1
9
~
3
5
2
8
I
SS
N:
2
0
8
8
-
8708
,
DOI
: 1
0
.
1
1
5
9
1
/
i
j
ec
e
.
v
1
1
i
4
.
pp
3
5
1
9
-
3
5
2
8
3519
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ec
e.
ia
esco
r
e.
co
m
Using
deep
l
ea
rni
ng
m
o
dels
f
o
r lea
rning
se
m
a
ntic
t
e
x
t
si
m
i
la
rity
o
f
Ara
bic
q
uestio
ns
M
a
h
m
o
ud
H
a
mm
a
d,
M
o
ha
mm
e
d Al
-
S
m
a
di
,
Q
a
nita
B
a
ni B
a
k
er
,
Sa
’
a
d A.
Al
-
Z
bo
o
n
Co
ll
e
g
e
o
f
Co
m
p
u
ter an
d
In
f
o
rm
a
ti
o
n
T
e
c
h
n
o
lo
g
y
,
Jo
rd
a
n
Un
iv
e
rsity
o
f
S
c
ien
c
e
a
n
d
T
e
c
h
n
o
lo
g
y
,
Irb
id
,
Jo
r
d
a
n
Art
icle
I
nfo
AB
ST
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
J
u
l 2
2
,
2
0
2
0
R
ev
i
s
ed
Dec
9
,
2
0
2
0
A
cc
ep
ted
J
an
13
,
2
0
2
1
Qu
e
stio
n
-
a
n
sw
e
rin
g
p
latf
o
rm
s se
r
v
e
m
il
li
o
n
s o
f
u
se
rs se
e
k
in
g
k
n
o
w
led
g
e
a
n
d
so
lu
ti
o
n
s
f
o
r
th
e
ir
d
a
il
y
li
f
e
p
ro
b
lem
s.
Ho
w
e
v
e
r,
m
a
n
y
k
n
o
w
led
g
e
se
e
k
e
r
s
a
re
f
a
c
in
g
th
e
c
h
a
ll
e
n
g
e
to
f
in
d
th
e
rig
h
t
a
n
sw
e
r
a
m
o
n
g
si
m
il
a
r
a
n
sw
e
re
d
q
u
e
stio
n
s
a
n
d
w
rit
e
r’s
re
sp
o
n
d
in
g
t
o
a
sk
e
d
q
u
e
stio
n
s
f
e
e
l
li
k
e
th
e
y
n
e
e
d
to
re
p
e
a
t
a
n
sw
e
rs
m
a
n
y
ti
m
e
s
f
o
r
sim
il
a
r
q
u
e
stio
n
s.
T
h
is
re
se
a
rc
h
a
im
s
a
t
tac
k
li
n
g
th
e
p
ro
b
lem
o
f
lea
rn
in
g
th
e
se
m
a
n
ti
c
tex
t
si
m
il
a
rit
y
a
m
o
n
g
d
iff
e
re
n
t
a
sk
e
d
q
u
e
stio
n
s
b
y
u
sin
g
d
e
e
p
lea
rn
in
g
.
T
h
re
e
m
o
d
e
l
s
a
r
e
i
m
p
l
e
m
e
n
t
e
d
t
o
a
d
d
r
e
s
s
t
h
e
a
f
o
r
e
m
e
n
t
i
o
n
e
d
p
r
o
b
l
e
m
:
i
)
a
s
u
p
e
r
v
i
s
e
d
-
m
a
c
h
i
n
e
l
e
a
r
n
i
n
g
m
o
d
e
l
u
s
i
n
g
XG
B
o
o
s
t
t
r
a
i
n
e
d
w
i
t
h
p
r
e
-
d
e
f
i
n
e
d
f
e
a
t
u
r
e
s
,
ii
)
a
n
a
d
a
p
t
e
d
S
i
a
m
e
s
e
-
b
a
s
e
d
d
e
e
p
l
e
a
r
n
i
n
g
r
e
c
u
r
r
e
n
t
a
r
c
h
i
t
e
c
t
u
r
e
t
r
a
i
n
e
d
w
i
t
h
p
r
e
-
d
e
f
i
n
e
d
f
e
a
tu
re
s
,
a
n
d
ii
i
)
a
p
re
-
train
e
d
d
e
e
p
b
id
irec
ti
o
n
a
l
tra
n
sf
o
rm
e
r
b
a
se
d
o
n
BERT
m
o
d
e
l.
P
r
o
p
o
se
d
m
o
d
e
ls
w
e
re
e
v
a
lu
a
ted
u
sin
g
a
re
fe
re
n
c
e
A
ra
b
ic
d
a
tas
e
t
f
ro
m
th
e
m
a
w
d
o
o
3
.
c
o
m
c
o
m
p
a
n
y
.
Ev
a
lu
a
ti
o
n
re
su
lt
s
sh
o
w
th
a
t
th
e
BERT
-
b
a
se
d
m
o
d
e
l
o
u
t
p
e
rf
o
r
m
s
th
e
o
th
e
r
tw
o
m
o
d
e
ls
w
it
h
a
n
F
1
=
9
2
.
9
9
%
,
w
h
e
r
e
a
s
t
h
e
S
i
a
m
e
s
e
-
b
a
s
e
d
m
o
d
e
l
c
o
m
e
s
i
n
t
h
e
s
e
c
o
n
d
p
l
a
c
e
w
i
t
h
F
1
=
8
9
.
0
4
8
%
,
a
n
d
f
i
n
a
l
l
y
,
t
h
e
XG
B
o
o
s
t
a
s
a
b
a
s
e
l
i
n
e
m
o
d
e
l
a
c
h
i
e
v
e
d
t
h
e
l
o
w
e
s
t
re
su
lt
o
f
F
1
=
8
6
.
0
8
6
%
.
K
ey
w
o
r
d
s
:
A
r
ab
ic
d
ataset
B
E
R
T
Dee
p
lear
n
in
g
Ma
ch
i
n
e
lear
n
i
n
g
Se
m
a
n
tic
te
x
t si
m
ilar
i
t
y
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
:
Ma
h
m
o
u
d
Ha
m
m
ad
C
o
lleg
e
o
f
C
o
m
p
u
ter
an
d
I
n
f
o
r
m
atio
n
T
ec
h
n
o
lo
g
y
J
o
r
d
an
Un
iv
er
s
it
y
o
f
Scie
n
ce
a
n
d
T
ec
h
n
o
lo
g
y
I
r
b
id
,
J
o
r
d
an
2
2
1
1
0
E
m
ail:
m
-
h
a
m
m
ad
@
j
u
s
t.e
d
u
.
j
o
1.
I
NT
RO
D
UCT
I
O
N
Se
m
a
n
tic
tex
t
s
i
m
ilar
it
y
(
ST
S)
is
a
m
ea
s
u
r
e
m
en
t
u
s
ed
to
d
eter
m
i
n
e
h
o
w
l
i
n
g
u
i
s
t
i
c
t
e
r
m
s
a
r
e
e
q
u
i
v
a
l
e
n
t
t
o
e
a
c
h
o
t
h
e
r
.
L
i
n
g
u
i
s
t
i
c
t
e
r
m
s
t
h
a
t
u
s
u
a
l
l
y
s
t
u
d
i
e
d
a
r
e
d
o
c
u
m
e
n
t
s
,
s
e
n
t
e
n
c
e
s
,
w
o
r
d
s
,
a
n
d
q
u
e
s
t
i
o
n
s
[
1
]
.
T
h
e
ST
S
im
p
r
o
v
es
t
h
e
u
n
d
er
s
tan
d
i
n
g
o
f
th
e
s
e
m
a
n
tic
s
i
m
i
lar
it
y
b
et
w
ee
n
li
n
g
u
is
tic
ter
m
s
an
d
in
cr
ea
s
e
s
th
e
ac
cu
r
ac
y
o
f
s
ev
er
al
k
n
o
w
led
g
e
-
b
ased
ap
p
licatio
n
s
.
T
h
is
u
n
d
er
s
tan
d
in
g
g
i
v
es
t
h
e
ST
S
a
g
r
ea
t
i
m
p
ac
t
o
n
m
a
n
y
ap
p
licatio
n
s
i
n
ar
tific
ial
i
n
t
ell
ig
en
ce
an
d
t
h
e
co
m
p
u
tatio
n
a
l
lin
g
u
is
tic
s
s
u
ch
as
i
n
f
o
r
m
ati
o
n
r
etr
iev
al,
w
o
r
d
s
en
s
e
d
is
a
m
b
i
g
u
atio
n
,
k
n
o
w
le
d
g
e
ac
q
u
is
it
io
n
,
an
d
n
a
tu
r
al
la
n
g
u
a
g
e
p
r
o
ce
s
s
in
g
(
N
L
P
)
[
2
,
3
]
.
R
eg
ar
d
in
g
to
th
e
N
L
P
f
ield
,
ST
S
ap
p
licatio
n
s
ca
n
b
e
e
x
ten
d
f
r
o
m
p
ar
ap
h
r
ase
id
e
n
ti
f
i
ca
tio
n
a
n
d
q
u
esti
o
n
s
i
m
i
lar
it
y
to
m
ac
h
in
e
tr
an
s
latio
n
[
4
]
.
T
h
e
r
esear
ch
i
n
t
h
e
ST
S
h
a
s
b
ee
n
g
r
ea
tl
y
i
n
cr
ea
s
ed
d
u
r
in
g
th
e
p
ast
f
e
w
y
ea
r
s
,
m
o
s
t
o
f
t
h
e
m
d
r
iv
en
b
y
th
e
an
n
u
al
Se
m
E
v
al
co
m
p
etitio
n
s
[
5
]
.
Se
m
E
v
al
is
an
in
ter
n
atio
n
al
w
o
r
k
s
h
o
p
f
o
r
s
e
m
an
tic
ev
al
u
a
tio
n
d
r
iv
en
b
y
th
e
SIG
L
E
X
[
1
]
.
W
ith
t
h
e
ad
v
e
n
t
o
f
W
eb
2
.
0
an
d
s
o
cial
co
m
p
u
ti
n
g
ad
v
a
n
ce
m
en
ts
,
p
lat
f
o
r
m
s
f
o
r
q
u
e
s
tio
n
an
s
w
er
i
n
g
h
as
b
ec
o
m
e
w
id
el
y
u
s
ed
.
Acc
o
r
d
in
g
to
t
h
e
Qu
o
r
a,
a
well
-
k
n
o
w
n
p
lat
f
o
r
m
f
o
r
q
u
est
io
n
a
n
s
w
er
i
n
g
an
d
k
n
o
w
led
g
e
s
h
ar
i
n
g
,
o
v
er
1
0
m
illi
o
n
u
s
e
r
s
v
is
it
Q
u
o
r
a
ev
e
r
y
m
o
n
t
h
.
W
it
h
s
u
c
h
a
lar
g
e
n
u
m
b
er
o
f
v
is
ito
r
s
,
s
i
m
ilar
q
u
esti
o
n
s
d
ef
i
n
itel
y
wo
u
ld
b
e
ask
ed
an
d
an
s
w
er
ed
b
y
s
e
v
er
al
u
s
er
s
,
h
en
ce
co
n
f
u
s
in
g
o
th
er
u
s
er
s
i
n
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8708
I
n
t J
E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
11
,
No
.
4
,
A
u
g
u
s
t 2
0
2
1
:
3
5
1
9
-
3528
3520
f
i
n
d
in
g
t
h
e
r
i
g
h
t
an
s
w
er
an
d
c
au
s
i
n
g
t
h
e
w
r
iter
s
to
f
ee
l
lik
e
th
e
y
h
a
v
e
to
w
r
ite
th
e
s
a
m
e
a
n
s
w
er
s
ev
er
al
ti
m
es
r
esp
o
n
d
in
g
to
s
i
m
i
lar
q
u
esti
o
n
s
.
T
w
o
q
u
est
io
n
s
a
s
k
i
n
g
ab
o
u
t
th
e
s
a
m
e
th
i
n
g
ca
n
b
e
f
o
r
m
ed
u
s
i
n
g
a
d
if
f
er
e
n
t
s
et
o
f
v
o
ca
b
u
lar
y
an
d
s
y
n
-
tactic
s
tr
u
ct
u
r
e.
T
h
is
m
a
k
es
d
etec
tin
g
t
h
e
s
e
m
an
t
ic
s
i
m
i
lar
it
y
b
et
w
ee
n
t
h
e
q
u
e
s
tio
n
s
i
s
a
ch
alle
n
g
i
n
g
tas
k
.
T
h
is
r
esear
ch
p
r
o
p
o
s
es
th
r
e
e
m
o
d
els
f
o
r
a
n
al
y
zi
n
g
t
h
e
s
e
m
an
tic
s
i
m
ilar
it
y
o
f
A
r
a
b
ic
q
u
esti
o
n
p
air
s
;
i
)
a
s
u
p
er
v
is
ed
-
m
ac
h
i
n
e
lear
n
in
g
m
o
d
el
u
s
i
n
g
XGB
o
o
s
t
[
6
]
tr
ain
ed
w
it
h
p
r
e
-
d
ef
i
n
ed
f
ea
t
u
r
es,
ii
)
an
ad
ap
ted
Sia
m
e
s
e
d
ee
p
lear
n
in
g
r
ec
u
r
r
en
t
ar
ch
itect
u
r
e
b
ased
o
n
th
e
w
o
r
k
o
f
[
7
]
,
an
d
iii
)
a
p
r
e
-
tr
ain
ed
d
ee
p
b
id
ir
ec
tio
n
al
tr
an
s
f
o
r
m
er
b
ase
d
o
n
B
E
R
T
m
o
d
el
[
8
]
.
T
h
is
p
ap
er
p
r
esen
ts
s
e
v
er
al
n
e
w
n
o
n
-
tr
i
v
ial
ex
te
n
s
io
n
s
to
o
u
r
p
r
elim
in
ar
y
w
o
r
k
d
escr
ib
ed
in
[
9
]
:
Ou
r
p
r
eli
m
in
ar
y
w
o
r
k
[
9
]
co
n
t
ain
s
o
n
l
y
tr
ad
itio
n
a
l
m
ac
h
in
e
l
ea
r
n
in
g
m
o
d
els
s
u
c
h
a
s
XGB
o
o
s
t,
SVM,
a
n
d
d
ec
is
io
n
tr
ee
.
I
n
t
h
is
m
a
n
u
s
cr
i
p
t,
w
e
h
a
v
e
d
esi
g
n
ed
an
d
i
m
p
le
m
e
n
ted
v
ar
io
u
s
d
ee
p
lear
n
in
g
m
o
d
els
u
s
in
g
tr
an
s
f
er
lear
n
i
n
g
tec
h
n
iq
u
e.
W
e
en
lar
g
ed
t
h
e
d
ataset
u
s
ed
f
o
r
t
r
a
i
n
i
n
g
a
n
d
t
e
s
t
i
n
g
.
I
n
[
9
]
,
w
e
t
r
a
i
n
e
d
o
u
r
m
o
d
e
l
s
o
n
9
,
5
6
8
p
a
i
r
s
o
f
q
u
e
s
t
i
o
n
s
w
h
e
r
e
a
s
i
n
t
h
i
s
p
a
p
e
r
,
w
e
t
r
a
i
n
e
d
o
u
r
m
o
d
e
l
s
o
n
1
5
,
7
1
2
p
a
i
r
s
o
f
q
u
e
s
t
i
o
n
s
,
i
.
e
.
,
3
1
,
4
2
4
d
i
s
t
i
n
c
t
q
u
e
s
tio
n
s
.
Si
m
i
lar
to
o
u
r
p
r
elim
in
ar
y
w
o
r
k
in
[
9
]
,
t
w
o
o
f
o
u
r
m
o
d
el
s
tr
ain
ed
u
s
in
g
p
r
e
-
e
n
g
i
n
ee
r
ed
f
e
atu
r
es
i
n
clu
d
i
n
g
ch
ar
ac
ter
-
le
v
el
f
ea
tu
r
e
s
,
w
o
r
d
-
lev
el
f
ea
t
u
r
es,
m
o
r
p
h
o
lo
g
i
ca
l
f
ea
t
u
r
es,
s
e
m
an
t
ic
f
ea
t
u
r
es,
an
d
w
o
r
d
e
m
b
ed
d
in
g
f
ea
tu
r
es.
Un
lik
e
o
u
r
p
r
eli
m
i
n
ar
y
w
o
r
k
,
o
u
r
b
est
-
ac
h
iev
ed
m
o
d
el,
t
h
e
B
E
R
T
-
b
a
s
ed
m
o
d
el,
w
a
s
ab
le
to
l
ea
r
n
t
h
e
s
e
m
a
n
tic
s
i
m
ilar
it
y
a
m
o
n
g
p
air
o
f
A
r
ab
ic
q
u
esti
o
n
s
w
it
h
o
u
t
t
h
e
n
ee
d
f
o
r
p
r
e
-
en
g
in
ee
r
ed
f
ea
t
u
r
es.
Hen
ce
,
i
n
cr
ea
s
i
n
g
t
h
e
g
en
er
alit
y
a
n
d
th
e
ap
p
licab
ilit
y
o
f
o
u
r
ap
p
r
o
ac
h
.
On
to
p
o
f
t
h
e
p
r
ev
io
u
s
tec
h
n
ic
al
co
n
tr
ib
u
tio
n
s
,
w
e
d
i
s
cu
s
s
ed
o
u
r
w
o
r
k
in
li
ght
o
f
o
t
h
e
r
r
e
l
a
t
e
d
r
e
s
e
a
r
c
h
e
f
f
o
r
t
i
n
t
h
e
a
r
e
a
o
f
A
r
a
b
i
c
t
e
x
t
s
i
m
i
l
a
r
i
t
y
d
e
t
e
c
t
i
o
n
u
s
i
n
g
d
e
e
p
l
e
a
r
n
i
n
g
.
M
o
r
e
o
v
e
r
,
t
h
e
p
a
p
e
r
p
r
o
v
i
d
e
s
d
e
t
a
i
l
e
d
d
e
s
c
r
i
p
t
i
o
n
o
f
t
h
e
m
o
d
e
l
s
a
l
o
n
g
w
i
t
h
t
h
e
u
s
e
d
p
a
r
a
m
e
t
e
r
s
f
o
r
t
r
a
i
n
i
n
g
o
u
r
m
o
d
e
l
s
t
o
g
i
v
e
t
h
e
b
est
r
esu
lts
.
T
h
e
r
est
o
f
th
is
p
ap
er
is
o
r
g
an
ized
as
f
o
llo
w
s
:
Sectio
n
2
.
p
r
esen
t
s
a
b
r
ief
s
u
r
v
e
y
o
f
t
h
e
lit
er
atu
r
e
f
o
r
ST
S
T
h
en
,
s
ec
tio
n
3
.
d
escr
ib
es
o
u
r
m
e
th
o
d
f
o
r
d
etec
tin
g
s
i
m
ilar
A
r
ab
ic
q
u
esti
o
n
s
.
Secti
o
n
4
.
p
r
esen
ts
th
e
r
esu
lt
s
o
f
o
u
r
in
te
n
s
i
v
e
ex
p
e
r
i
m
en
ts
.
I
n
s
ec
tio
n
5
,
r
esu
lts
ar
e
an
al
y
ze
d
an
d
d
is
cu
s
s
ed
.
Fin
all
y
,
s
ec
tio
n
6
.
co
n
clu
d
es t
h
e
p
ap
er
w
it
h
av
e
n
u
e
o
f
f
u
t
u
r
e
w
o
r
k
.
2.
RE
L
AT
E
D
WO
RK
Ma
n
y
r
esear
ch
er
s
f
r
o
m
v
ar
io
u
s
f
ield
s
u
ti
lized
s
e
m
a
n
ti
c
tex
t
s
i
m
ilar
it
y
(
ST
S)
o
n
d
if
f
er
en
t
ap
p
licatio
n
s
.
T
h
is
s
ec
tio
n
co
m
p
ar
es
an
d
co
n
tr
a
s
ts
o
u
r
r
esear
ch
co
n
tr
ib
u
tio
n
i
n
li
g
h
t
o
f
o
th
er
r
esear
ch
w
o
r
k
i
n
th
e
f
ield
.
O
u
r
w
o
r
k
i
s
r
elate
d
t
o
th
e
r
esear
ch
b
o
d
y
th
at
ap
p
lied
m
ac
h
i
n
e
lear
n
i
n
g
an
d
d
ee
p
l
ea
r
n
in
g
tec
h
n
iq
u
es
to
s
o
lv
e
ST
S
p
r
o
b
lem
s
i
n
clu
d
in
g
[
1
0
-
2
6
]
.
Ho
w
ev
er
,
all
o
f
th
e
p
r
ev
io
u
s
l
y
m
e
n
t
io
n
ed
ap
p
r
o
ac
h
es
d
esig
n
e
d
th
eir
ST
S
m
o
d
els
f
o
r
E
n
g
li
s
h
lan
g
u
ag
e
te
x
t
s
.
E
v
en
t
h
o
u
g
h
s
o
m
e
o
f
th
eir
m
o
d
els
ca
n
b
e
ap
p
lied
t
o
A
r
ab
ic
tex
ts
,
t
h
e
y
w
ill
n
o
t
p
r
o
d
u
ce
h
ig
h
ac
c
u
r
ac
y
s
in
ce
t
h
eir
m
o
d
els
ar
e
n
o
t
d
esig
n
ed
n
o
r
tr
ain
ed
o
n
A
r
ab
ic
tex
t.
T
h
er
ef
o
r
e,
th
ese
a
p
p
r
o
ac
h
es c
an
n
o
t
s
o
l
v
e
t
h
e
p
r
o
b
le
m
w
e
ar
e
tr
y
i
n
g
to
s
o
lv
e,
th
a
t i
s
ac
c
u
r
atel
y
a
n
d
e
f
f
icie
n
tl
y
d
etec
tin
g
s
i
m
ilar
A
r
ab
ic
q
u
est
io
n
s
.
A
lt
h
o
u
g
h
th
e
m
aj
o
r
it
y
o
f
t
h
e
r
esear
ch
er
s
in
t
h
e
ST
S
f
ield
d
ev
elo
p
ed
tech
n
iq
u
es
f
o
r
th
e
E
n
g
li
s
h
lan
g
u
a
g
es,
f
e
w
o
f
th
e
m
d
ev
elo
p
e
d
S
T
S
ap
p
r
o
ac
h
es
f
o
r
th
e
A
r
ab
ic
lan
g
u
ag
e.
Ne
x
t
we
d
is
cu
s
s
t
h
e
m
a
i
n
r
esear
ch
e
f
f
o
r
ts
f
o
r
d
etec
tin
g
th
e
s
e
m
an
tic
s
i
m
ilar
it
y
o
f
Ar
ab
ic
tex
ts
.
Mo
h
a
m
m
ad
et
a
l
.
[
2
7
]
p
r
o
p
o
s
ed
an
en
h
a
n
ce
d
ap
p
r
o
ac
h
f
o
r
p
ar
a
p
h
r
ase
id
en
tif
ica
tio
n
(
P
I
)
an
d
S
T
S
in
A
r
ab
ic
t
w
ee
t
s
.
Sag
h
e
r
et
a
l.
[
2
8
]
p
r
o
p
o
s
ed
a
C
NN
d
ee
p
lear
n
in
g
m
o
d
el
to
class
if
y
A
r
ab
ic
s
e
n
te
n
ce
s
in
to
th
r
ee
ca
te
g
o
r
ies.
[
2
9
]
u
s
ed
an
d
co
m
p
ar
ed
d
if
f
er
e
n
t
ST
S
m
et
h
o
d
s
to
m
e
asu
r
e
th
e
cr
o
s
s
-
la
n
g
u
a
g
e
s
e
m
an
tic
s
i
m
ilar
it
y
f
o
r
s
h
o
r
t
s
e
n
t
e
n
c
e
s
a
n
d
p
h
r
a
s
e
s
.
V
a
r
i
o
u
s
a
p
p
r
o
a
c
h
e
s
u
s
e
d
S
T
S
t
o
d
e
t
e
c
t
p
l
a
g
i
a
r
i
s
m
i
n
A
r
a
b
i
c
t
e
x
t
s
s
u
c
h
a
s
[
3
0
-
3
2
]
.
F
e
r
r
e
r
o
et
a
l.
[
3
3
]
p
r
o
p
o
s
ed
t
w
o
d
if
f
er
e
n
t
ap
p
r
o
ac
h
es
to
m
ea
s
u
r
e
t
h
e
ST
S
o
f
cr
o
s
s
-
la
n
g
u
ag
e
s
e
n
ten
ce
s
f
o
r
A
r
ab
ic
-
E
n
g
l
is
h
te
x
t.
Mo
r
eo
v
er
,
[
3
4
]
p
r
o
p
o
s
ed
a
q
u
er
y
-
b
ased
A
r
ab
ic
tex
t
s
u
m
m
ar
iza
tio
n
ap
p
r
o
ac
h
th
at
ac
ce
p
ts
A
r
ab
ic
d
o
cu
m
e
n
t
as
w
ell
as
u
s
er
q
u
er
ies.
Fin
a
ll
y
,
[
3
5
]
ad
o
p
ted
m
o
r
p
h
o
lo
g
ical
w
o
r
d
2
v
e
c
m
et
h
o
d
f
o
r
Neu
r
al
m
ac
h
i
n
e
tr
an
s
latio
n
f
o
r
lo
w
r
eso
u
r
ce
s
etti
n
g
s
3.
RE
S
E
ARCH
M
E
T
H
O
D
T
h
is
s
ec
tio
n
d
escr
ib
es
o
u
r
m
eth
o
d
in
d
ev
elo
p
in
g
a
m
ac
h
i
n
e
-
lear
n
i
n
g
ap
p
r
o
ac
h
f
o
r
ac
cu
r
atel
y
a
n
d
ef
f
icien
tl
y
d
etec
ti
n
g
if
t
w
o
A
r
ab
ic
q
u
esti
o
n
s
ar
e
s
i
m
i
lar
o
r
n
o
t.
3
.
1
.
Ara
bic
qu
estio
n
s
pa
ir’
s
da
t
a
s
et
I
n
o
r
d
er
to
ev
alu
a
te
o
u
r
m
o
d
els,
th
e
A
r
ab
ic
q
u
est
io
n
s
p
air
s
d
ataset
p
r
o
v
id
ed
b
y
ma
w
d
o
o
3
.
c
o
m
c
o
m
p
a
n
y
i
s
u
s
e
d
.
T
h
e
d
a
t
a
s
e
t
w
a
s
m
a
n
u
a
l
l
y
a
n
n
o
t
a
t
e
d
b
y
m
a
w
d
o
o
3
’
s
d
a
t
a
a
n
n
o
t
a
t
i
o
n
t
e
a
m
.
A
s
s
h
o
w
n
i
n
T
a
b
l
e
1
,
th
e
d
ataset
co
n
s
i
s
ts
o
f
ar
o
u
n
d
1
5
k
p
air
s
o
f
A
r
ab
ic
q
u
es
tio
n
s
an
n
o
tated
as
“
s
i
m
ilar
”
o
r
“
n
o
t
”
.
T
h
e
d
a
t
a
w
a
s
d
i
v
i
d
e
d
i
n
t
o
tw
o
f
i
l
e
s
,
t
r
a
i
n
i
n
g
d
a
t
a
w
i
t
h
1
1
.
9
9
7
p
a
i
r
s
o
f
q
u
e
s
t
i
o
n
s
a
n
d
t
e
s
t
i
n
g
f
i
l
e
w
i
t
h
3
.
7
1
5
p
a
i
r
s
o
f
q
u
est
io
n
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J
E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2
0
8
8
-
8708
Usi
n
g
d
ee
p
lea
r
n
in
g
mo
d
els fo
r
lea
r
n
in
g
s
ema
n
tic
text
s
imila
r
ity
o
f
A
r
a
b
ic
q
u
esti
o
n
s
(
Ma
h
mo
u
d
Ha
mma
d
)
3521
T
ab
le
2
s
h
o
w
s
a
n
ex
a
m
p
le
o
f
t
w
o
p
air
s
o
f
q
u
esti
o
n
s
s
elec
t
ed
f
r
o
m
t
h
e
d
ataset
an
d
r
ep
r
esen
ti
n
g
th
e
t
w
o
m
ai
n
ca
teg
o
r
ies:
s
i
m
ilar
s
h
o
w
n
as
“
Yes”
o
r
n
o
t
s
i
m
ila
r
s
h
o
w
n
a
s
“No
”.
I
n
th
e
f
ir
s
t
r
o
w
o
f
t
h
e
T
ab
le
2
,
Qu
est
io
n
1
as
k
s
ab
o
u
t
t
h
e
b
ir
th
cit
y
o
f
t
h
e
co
m
p
r
eh
e
n
s
iv
e
th
i
n
k
er
,
A
l
-
R
az
i,
a
n
d
Qu
e
s
tio
n
2
ask
s
ab
o
u
t
t
h
e
cit
y
o
f
A
l
-
R
az
i
m
u
s
e
u
m
.
C
lear
l
y
,
th
o
s
e
t
w
o
q
u
e
s
tio
n
s
ar
e
n
o
t
s
i
m
ilar
s
i
n
ce
th
e
y
ar
e
as
k
i
n
g
a
b
o
u
t
t
w
o
d
if
f
er
en
t
th
i
n
g
s
.
O
n
t
h
e
o
t
h
er
h
a
n
d
,
Q
u
esti
o
n
1
an
d
Qu
e
s
tio
n
2
i
n
t
h
e
s
ec
o
n
d
r
o
w
o
f
th
e
T
ab
le
2
ar
e
a
s
k
i
n
g
ab
o
u
t
t
h
e
f
ir
s
t
co
u
n
tr
y
w
h
er
e
co
m
m
u
n
i
s
m
p
o
liti
ca
l
id
eo
lo
g
y
w
as
s
t
ar
ted
.
T
h
o
s
e
t
w
o
q
u
e
s
tio
n
s
w
er
e
w
r
itte
n
i
n
t
w
o
d
if
f
er
e
n
t
w
a
y
s
b
u
t t
h
e
y
s
til
l h
a
v
e
a
s
i
m
ilar
m
ea
n
i
n
g
.
T
h
e
u
tili
z
ed
d
ataset
is
n
ea
r
l
y
b
alan
ce
d
w
it
h
5
5
.
0
1
%
lab
ele
d
as
“
n
o
t
s
i
m
ilar
”
an
d
th
e
r
est,
4
4
.
9
9
%,
lab
eled
as
“si
m
ilar
”
I
n
o
r
d
er
to
f
u
r
t
h
er
a
n
al
y
ze
th
e
d
ataset
,
w
e
co
m
p
u
ted
t
h
e
co
m
m
o
n
w
o
r
d
s
a
m
o
n
g
ea
c
h
q
u
esti
o
n
p
air
.
Fig
u
r
e
1
s
h
o
ws
th
a
t
t
h
e
n
u
m
b
er
o
f
o
v
er
lap
p
in
g
w
o
r
d
s
b
et
w
ee
n
p
air
o
f
“
s
i
m
ilar
”
an
d
“n
o
t
s
i
m
ilar
”
q
u
esti
o
n
s
i
s
ar
o
u
n
d
2
w
o
r
d
s
.
T
h
er
ef
o
r
e,
r
ely
i
n
g
o
n
th
e
o
v
er
lap
p
ed
w
o
r
d
s
b
et
w
ee
n
p
air
o
f
q
u
est
io
n
s
to
k
n
o
w
if
t
h
e
y
ar
e
s
i
m
ilar
w
il
l
n
o
t
g
i
v
e
g
o
o
d
r
esu
l
ts
.
T
h
u
s
,
Fi
g
u
r
e
1
s
h
o
w
s
t
h
at
o
u
r
p
r
o
b
lem
is
v
er
y
ch
alle
n
g
i
n
g
to
s
o
lv
e.
T
ab
le
1
.
Nu
m
b
er
o
f
in
s
ta
n
ce
s
in
th
e
d
ata
s
et
T
r
a
i
n
d
a
t
a
se
t
T
e
st
d
a
t
a
se
t
T
o
t
a
l
1
1
,
9
9
7
3
,
7
1
5
1
5
,
7
1
2
T
ab
le
2
.
A
n
ex
a
m
p
le
o
f
t
w
o
i
n
s
tan
ce
s
,
t
w
o
p
air
s
o
f
q
u
esti
o
n
s
,
f
r
o
m
th
e
t
w
o
cla
s
s
es
Q
u
e
st
i
o
n
1
Q
u
e
st
i
o
n
2
L
a
b
e
l
؟
ي
ز
ا
ر
لا
فح
ت
م
عق
ي
ةن
ي
د
م
ي
أ
ي
ف
ي
ز
ا
ر
لا
د
لو
ن
ي
أ
No
؟ةي
ع
وي
ش
ل
ا
ت
ر
ه
ظ
ةلود
ي
أ
ي
ف
؟
ةي
ع
وي
ش
لا
ت
أد
ب
ن
ي
أ
Y
e
s
Fig
u
r
e
1
.
T
h
e
d
is
tr
ib
u
tio
n
o
f
c
o
m
m
o
n
w
o
r
d
s
i
n
s
i
m
ilar
a
n
d
n
o
n
-
s
i
m
ilar
q
u
esti
o
n
s
f
o
r
th
e
tr
ain
i
n
g
d
atase
t
3
.
2
.
Da
t
a
pre
-
pro
ce
s
s
ing
I
n
o
r
d
er
to
p
r
ep
a
r
e
th
e
d
ataset
f
o
r
f
u
r
t
h
er
p
r
o
ce
s
s
in
g
an
d
to
en
h
a
n
ce
th
e
ac
c
u
r
ac
y
a
n
d
r
ed
u
ce
th
e
n
o
is
e
i
n
th
e
d
ata,
v
ar
io
u
s
A
r
ab
ic
p
r
e
-
p
r
o
ce
s
s
in
g
s
tep
s
w
er
e
a
p
p
lied
s
u
ch
as:
R
e
m
o
v
al
o
f
n
o
n
-
A
r
ab
ic
w
o
r
d
s
.
R
e
m
o
v
al
o
f
h
y
p
er
li
n
k
s
an
d
h
a
s
h
ta
g
s
i
n
all
p
o
s
ts
.
R
e
m
o
v
al
o
f
A
r
ab
ic
d
iacr
itics
s
u
ch
a
s
R
e
m
o
v
al
o
f
p
u
n
ct
u
atio
n
a
n
d
s
y
m
b
o
ls
s
u
c
h
as
“
?
,
(
,
)
,
,
’
!
@
$
#
—
”.
No
r
m
a
lizatio
n
,
w
h
ic
h
is
u
s
ed
to
r
em
o
v
e
“H
A
M
Z
A
/
”
f
r
o
m
t
h
e
“AL
E
F/
”
(
i.e
.
,
th
e
r
ep
lace
d
w
it
h
th
e
ab
s
tr
ac
t v
er
s
io
n
o
f
th
e
let
ter
(
(
(
(
)
)
)
.
R
e
m
o
v
al
o
f
A
r
ab
ic
s
to
p
w
o
r
d
s
.
T
h
e
NL
T
K
lib
r
ar
y
[
3
6
]
w
r
itte
n
i
n
p
y
t
h
o
n
w
as
u
s
ed
to
i
m
p
l
e
m
en
t
t
h
e
d
ata
p
r
e
-
p
r
o
ce
s
s
i
n
g
an
d
d
at
a
clea
n
i
n
g
p
h
a
s
e.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8708
I
n
t J
E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
11
,
No
.
4
,
A
u
g
u
s
t 2
0
2
1
:
3
5
1
9
-
3528
3522
3
.
3
.
F
ea
t
ures e
x
t
ra
ct
io
n
As
m
e
n
tio
n
ed
ea
r
lier
t
w
o
o
u
t
o
f
th
r
ee
m
o
d
els
d
ev
elo
p
ed
in
th
i
s
r
esear
ch
w
er
e
tr
ain
ed
w
it
h
p
r
e
-
en
g
i
n
ee
r
ed
f
ea
t
u
r
es.
Af
ter
th
e
d
ata
w
as c
lea
n
ed
,
th
e
f
o
llo
w
in
g
f
ea
t
u
r
es
w
er
e
ex
tr
ac
ted
:
C
h
ar
ac
ter
lev
el
f
ea
t
u
r
es:
T
h
is
s
et
o
f
f
ea
t
u
r
es
i
n
clu
d
e
s
:
th
e
to
tal
n
u
m
b
er
o
f
ch
ar
ac
ter
s
f
o
r
th
e
p
air
o
f
q
u
esti
o
n
s
,
th
e
n
u
m
b
er
o
f
d
if
f
e
r
en
t
ch
ar
ac
ter
s
a
m
o
n
g
th
e
q
u
e
s
tio
n
p
air
s
,
t
h
e
r
a
t
i
o
o
f
t
h
e
d
i
f
f
e
r
e
n
t
c
h
a
r
a
c
t
e
r
s
,
t
h
e
n
u
m
b
e
r
o
f
s
i
m
i
l
a
r
c
h
a
r
a
c
t
e
r
s
a
m
o
n
g
t
h
e
q
u
e
s
t
i
o
n
p
a
i
r
s
,
a
n
d
t
h
e
r
a
t
i
o
o
f
t
h
e
s
i
m
i
l
a
r
ch
ar
ac
te
r
s
.
W
o
r
d
lev
el
f
ea
t
u
r
es:
T
h
is
s
et
i
n
clu
d
e
s
:
th
e
to
tal
n
u
m
b
er
o
f
w
o
r
d
s
f
o
r
th
e
p
air
o
f
q
u
esti
o
n
s
,
th
e
n
u
m
b
er
o
f
d
if
f
er
e
n
t
w
o
r
d
s
a
m
o
n
g
th
e
q
u
esti
o
n
p
a
ir
s
,
an
d
t
h
e
r
atio
o
f
t
h
e
d
if
f
er
en
t
w
o
r
d
s
,
t
h
e
n
u
m
b
er
o
f
s
i
m
ila
r
w
o
r
d
s
a
m
o
n
g
t
h
e
q
u
e
s
tio
n
p
a
ir
s
,
an
d
th
e
r
atio
o
f
t
h
e
s
i
m
il
ar
w
o
r
d
s
.
Mo
r
eo
v
er
,
th
e
t
y
p
e
o
f
s
i
m
ilar
it
y
o
f
q
u
esti
o
n
p
air
s
is
u
s
ed
as
a
n
o
t
h
er
f
ea
tu
r
e.
T
h
is
f
ea
t
u
r
e
i
s
co
m
p
u
ted
a
s
a
b
i
n
ar
y
f
ea
t
u
r
e
an
d
d
ep
en
d
in
g
o
n
th
e
q
u
e
s
tio
n
in
ter
r
o
g
at
iv
e
p
a
r
ticles
(
i.e
.
s
i
m
ilar
it
y
o
f
th
e
f
ir
s
t
w
o
r
d
i
n
ea
ch
q
u
est
io
n
)
.
A
l
s
o
,
th
e
tex
t
o
v
er
lap
f
ea
tu
r
es
w
er
e
co
m
p
u
ted
o
n
th
e
w
o
r
d
lev
el
an
d
b
ased
o
n
o
u
r
p
r
ev
io
u
s
r
esear
ch
[
2
7
]
.
T
h
e
tex
t
o
v
er
lap
f
ea
tu
r
es
i
n
cl
u
d
e
th
e
n
u
m
b
er
o
f
o
v
er
lap
p
in
g
w
o
r
d
s
d
iv
id
ed
b
y
th
e
n
u
m
b
er
o
f
w
o
r
d
s
in
q
u
esti
o
n
1
,
th
e
n
u
m
b
er
o
f
o
v
er
lap
p
in
g
w
o
r
d
s
d
iv
id
ed
b
y
t
h
e
n
u
m
b
er
o
f
w
o
r
d
s
i
n
q
u
est
io
n
2
,
an
d
t
h
e
h
a
r
m
o
n
ic
m
ea
n
o
f
th
e
p
r
ev
io
u
s
t
w
o
f
ea
t
u
r
es.
Mo
r
p
h
o
lo
g
ical
f
ea
t
u
r
es:
Ste
m
m
in
g
w
a
s
u
s
ed
to
r
ep
r
esen
t
ea
ch
q
u
e
s
tio
n
.
T
h
e
A
r
ab
ic
lan
g
u
a
g
e
i
s
m
o
r
p
h
o
lo
g
icall
y
r
ic
h
[
3
7
]
,
th
er
ef
o
r
e,
r
ep
r
esen
tin
g
t
h
e
p
air
o
f
q
u
es
tio
n
s
u
s
i
n
g
t
h
eir
s
te
m
w
o
r
d
s
i
n
cr
ea
s
e
s
th
e
ch
a
n
ce
o
f
w
o
r
d
s
i
m
ilar
it
y
o
n
th
e
s
u
r
f
ac
e
le
v
el.
Se
m
a
n
tic
lev
el
f
ea
t
u
r
es:
T
h
is
f
ea
t
u
r
e
s
et
in
c
lu
d
es:
T
F
-
I
D
F,
J
a
cc
ar
d
,
an
d
C
o
s
in
e
s
i
m
il
ar
it
y
m
ea
s
u
r
es.
T
h
ese
f
ea
t
u
r
es
w
er
e
co
m
p
u
te
d
f
o
r
ea
ch
p
air
o
f
q
u
esti
o
n
s
.
T
h
e
lex
ical
s
i
m
ilar
it
y
m
ea
s
u
r
es
ar
e
co
m
p
u
te
d
o
n
b
o
th
th
e
o
r
ig
i
n
al
q
u
e
s
tio
n
s
an
d
th
eir
s
te
m
m
ed
f
ea
t
u
r
es.
W
o
r
d
E
m
b
ed
d
in
g
f
ea
t
u
r
es:
T
h
e
p
r
e
-
tr
ain
ed
m
o
d
el
f
o
r
A
r
ab
i
c
co
n
ten
t
A
r
aVe
c
3
.
0
[
3
8
]
is
u
s
ed
to
co
m
p
u
te
th
e
e
m
b
ed
d
in
g
s
f
ea
t
u
r
es
f
o
r
t
h
e
i
n
p
u
t
q
u
e
s
tio
n
s
.
T
h
e
T
w
itt
er
-
C
B
OW
w
it
h
e
m
b
ed
d
in
g
s
i
ze
o
f
1
0
0
is
u
s
ed
o
u
t o
f
th
e
A
r
aVe
c
a
v
ailab
le
p
r
e
-
tr
ain
ed
m
o
d
el
s
.
3
.
4
.
T
he
dev
elo
ped
m
o
dels
T
h
is
r
esear
ch
i
m
p
le
m
en
t
s
th
r
ee
m
o
d
e
ls
f
o
r
an
al
y
zi
n
g
th
e
s
e
m
an
tic
s
i
m
ilar
it
y
o
f
A
r
ab
ic
q
u
esti
o
n
s
’
p
air
s
:
i
)
a
s
u
p
er
v
i
s
ed
-
m
ac
h
in
e
lear
n
i
n
g
m
o
d
el
u
s
i
n
g
XGB
o
o
s
t
[
6
]
,
ii
)
an
ad
ap
ted
Sia
m
ese
d
ee
p
lear
n
i
n
g
r
ec
u
r
r
en
t
ar
ch
itect
u
r
e
b
ased
o
n
t
h
e
w
o
r
k
o
f
[
7
]
,
an
d
iii
)
a
p
r
e
-
tr
ain
ed
d
ee
p
b
id
ir
ec
tio
n
al
tr
an
s
f
o
r
m
er
b
ased
o
n
B
E
R
T
m
o
d
el
[
8
]
.
W
e
h
av
e
ex
tr
ac
ted
f
ea
t
u
r
es
f
r
o
m
th
e
d
ataset
to
tr
ai
n
t
h
e
f
ir
s
t
t
w
o
m
o
d
els,
t
h
e
XGB
o
o
s
t
a
n
d
t
h
e
Sia
m
e
s
e
n
eu
r
al
n
et
w
o
r
k
.
Ho
wev
er
,
th
e
B
E
R
T
m
o
d
el
is
ad
ap
ted
d
ir
ec
tly
w
it
h
o
u
t
an
y
f
ea
t
u
r
e
ex
tr
ac
tio
n
s
tep
.
W
e
h
av
e
ca
r
ef
u
ll
y
s
e
lecte
d
th
ese
t
h
r
ee
m
o
d
els
a
m
o
n
g
m
an
y
o
th
er
m
o
d
els
f
o
r
m
an
y
r
ea
s
o
n
s
.
T
h
e
XGB
o
o
s
t
w
a
s
t
h
e
b
est p
er
f
o
r
m
in
g
m
o
d
el
in
[
9
]
,
th
e
Sia
m
ese
d
ee
p
lear
n
in
g
m
o
d
el
w
o
r
k
s
w
ell
i
n
s
e
m
an
tic
te
x
t
s
i
m
ilar
it
y
[
7
,
3
9
]
,
an
d
th
e
Go
o
g
le
B
E
R
T
m
o
d
el
is
th
e
s
tate
-
of
-
t
h
e
-
ar
t
m
o
d
el
u
s
ed
f
o
r
s
e
v
er
al
n
at
u
r
al
-
la
n
g
u
a
g
e
p
r
o
ce
s
s
in
g
(
N
L
P
)
ap
p
licatio
n
s
.
3
.
4
.
1
.
Su
perv
is
ed
-
m
a
chi
ne
le
a
rning
m
o
de
l us
i
ng
XG
B
o
o
s
t
XGB
o
o
s
t
[
6
]
is
a
s
h
o
r
t
s
tan
d
in
g
f
o
r
eXtr
e
m
e
g
r
ad
ien
t
b
o
o
s
tin
g
.
XGB
o
o
s
t
is
a
s
ca
lab
le
m
ac
h
in
e
lear
n
-
i
n
g
s
y
s
te
m
f
o
r
tr
ee
b
o
o
s
tin
g
an
d
it
is
av
a
ilab
le
as
an
o
p
en
-
s
o
u
r
ce
p
ac
k
ag
e.
I
n
t
h
e
m
ac
h
i
n
e
lear
n
i
n
g
co
m
p
eti
tio
n
p
u
b
li
s
h
ed
b
y
Ka
g
g
le
in
2
0
1
5
,
am
o
n
g
th
e
2
9
w
i
n
n
i
n
g
s
o
lu
tio
n
s
,
1
7
s
o
lu
tio
n
s
ad
ap
ted
XGB
o
o
s
t.
Am
o
n
g
th
e
s
e
1
7
s
o
lu
tio
n
s
,
8
s
o
lu
tio
n
s
u
s
ed
XGB
o
o
s
t
to
tr
ai
n
th
e
m
o
d
el,
w
h
i
le
th
e
r
est
9
co
m
b
i
n
ed
XGB
o
o
s
t
w
it
h
t
h
e
ar
tific
ial
n
e
u
r
al
n
e
t
wo
r
k
as e
n
s
e
m
b
les.
XGB
o
o
s
t
ap
p
r
o
ac
h
p
r
o
v
id
es
a
p
ar
allel
tr
ee
b
o
o
s
tin
g
k
n
o
w
n
as
g
r
ad
ie
n
t
b
o
o
s
ted
r
eg
r
ess
io
n
tr
ee
(
GB
R
T
)
o
r
g
r
ad
ien
t
b
o
o
s
tin
g
m
ac
h
in
e
(
GB
M)
w
h
ich
is
a
s
ca
lab
le
an
d
e
f
f
icie
n
t
i
m
p
le
m
e
n
tatio
n
o
f
g
r
ad
ien
t
b
o
o
s
ti
n
g
f
r
a
m
e
w
o
r
k
p
r
o
p
o
s
ed
b
y
[
4
0
,
4
1
]
.
XGBo
o
s
t
alg
o
r
ith
m
co
m
b
i
n
es
w
ea
k
b
ase
lear
n
in
g
m
o
d
els
i
n
to
a
s
tr
o
n
g
er
lear
n
er
i
n
an
iter
ati
v
e
m
an
n
er
.
I
t
is
av
ailab
le
i
n
s
ev
er
al
lan
g
u
ag
e
s
s
u
ch
as
P
y
t
h
o
n
,
R
,
a
n
d
J
u
lia.
XGB
o
o
s
t
ca
n
b
e
in
teg
r
ated
w
ith
s
e
v
er
al
lan
g
u
a
g
e
d
ata
s
cie
n
ce
p
ip
elin
es
as
s
c
ik
i
tlear
n
.
T
h
e
XGB
o
o
s
t
m
o
d
el
is
tr
ain
ed
i
n
an
ad
d
iti
v
e
m
an
n
er
.
A
s
s
h
o
w
n
i
n
(
1
)
,
n
ee
d
s
to
b
e
ad
d
ed
t
o
m
i
n
i
m
ize
t
h
e
o
b
j
e
ctiv
e
(
)
.
W
h
er
e
−
1
is
th
e
p
r
ed
ictio
n
o
f
t
h
e
−
ℎ
in
s
tan
ce
at
th
e
−
ℎ
iter
atio
n
.
(
)
=
1
|
(
̂
)
−
1
+
(
)
)
+
Ω
(
)
(
1
)
I
n
th
i
s
w
o
r
k
,
w
e
u
s
ed
t
h
e
XG
B
o
o
s
t
P
y
th
o
n
p
ac
k
a
g
e
i
n
tr
o
d
u
ce
d
in
[
6
]
to
tr
ain
th
e
m
o
d
el
with
t
h
e
p
r
e
-
d
ef
in
ed
f
ea
t
u
r
es
i
n
o
r
d
er
to
e
n
h
a
n
ce
d
ap
p
r
o
ac
h
f
o
r
lear
n
in
g
s
e
m
an
t
ic
s
i
m
ilar
itie
s
in
A
r
a
b
ic
q
u
esti
o
n
s
.
T
h
e
XGB
o
o
s
t
class
i
f
ier
w
a
s
tr
ain
ed
u
s
i
n
g
th
e
e
x
tr
ac
ted
f
ea
tu
r
es
as
e
x
p
lain
ed
i
n
3
.
3
.
Th
e
X
G
B
o
o
s
t
m
o
d
e
l
w
a
s
t
r
a
i
n
e
d
w
i
t
h
a
m
a
x
i
m
u
m
t
r
e
e
d
e
p
t
h
o
f
6
a
n
d
a
l
e
a
r
n
i
n
g
r
a
t
e
(
e
t
a
)
=
0
.
0
6
,
0
.
0
4
,
a
n
d
0
.
0
2
f
o
r
6
.
0
0
0
e
p
o
c
h
s
o
n
ea
ch
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J
E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2
0
8
8
-
8708
Usi
n
g
d
ee
p
lea
r
n
in
g
mo
d
els fo
r
lea
r
n
in
g
s
ema
n
tic
text
s
imila
r
ity
o
f
A
r
a
b
ic
q
u
esti
o
n
s
(
Ma
h
mo
u
d
Ha
mma
d
)
3523
3
.
4
.
2
.
A
S
ia
m
ese
deep
lea
rni
ng
re
curr
ent
a
rc
hite
ct
ure
Sia
m
e
s
e
n
e
u
r
al
n
et
w
o
r
k
i
m
p
l
e
m
en
ts
t
w
o
s
y
m
m
etr
ic
n
eu
r
al
n
et
w
o
r
k
s
w
it
h
s
h
ar
ed
w
ig
h
t
s
to
lear
n
s
e
m
a
n
tic
s
i
m
ilar
itie
s
a
m
o
n
g
i
n
p
u
ts
.
Sia
m
e
s
e
n
eu
r
al
n
et
w
o
r
k
s
is
u
s
ed
i
n
m
a
n
y
s
e
m
an
t
ic
s
i
m
ilar
it
y
ap
p
licatio
n
s
s
u
c
h
as:
f
ac
e
v
er
if
icatio
n
u
s
in
g
s
y
m
m
e
tr
ic
co
n
v
o
lu
tio
n
al
n
e
t
w
o
r
k
s
[
4
2
]
,
s
p
ee
ch
u
n
d
er
s
tan
d
in
g
a
n
d
s
p
ea
k
er
-
s
p
ec
if
ic
i
n
f
o
r
m
ati
o
n
e
x
tr
ac
tio
n
[
4
3
]
,
an
d
in
s
e
m
a
n
tic
te
x
t si
m
ilar
it
y
[
7
,
39
].
I
n
th
i
s
r
esear
ch
,
w
e
u
til
ized
th
e
Sia
m
ese
n
e
u
r
al
n
et
w
o
r
k
ar
ch
itect
u
r
e
to
d
ev
elo
p
an
en
h
a
n
ce
d
ap
p
r
o
ac
h
f
o
r
lear
n
in
g
s
e
m
a
n
ti
c
s
i
m
ilar
it
ies
i
n
A
r
ab
ic
q
u
es
ti
o
n
s
.
T
h
e
m
o
d
el
co
n
s
i
s
ts
o
f
t
wo
s
y
m
m
etr
ic
la
y
er
s
ea
ch
co
n
ta
in
s
a
n
e
m
b
ed
d
in
g
l
a
y
er
f
o
llo
w
ed
b
y
a
b
i
-
d
ir
ec
ti
o
n
al
lo
n
g
s
h
o
r
t
ter
m
m
e
m
o
r
y
(
L
ST
M)
la
y
er
a
n
d
th
en
an
L
ST
M
la
y
er
.
B
ein
g
S
ia
m
ese
-
b
ased
t
h
e
w
e
ig
h
ts
u
s
e
d
to
tr
ain
b
o
th
b
i
-
L
ST
E
M
an
d
L
ST
M
la
y
er
s
f
o
r
b
o
th
q
u
es
t
io
n
s
ar
e
s
h
ar
ed
s
ee
Fig
u
r
e
2
.
T
h
e
o
u
tp
u
t
o
f
t
h
e
s
y
m
m
etr
ic
la
y
er
s
ar
e
th
e
n
co
n
ca
ten
a
ted
w
it
h
t
h
e
o
u
t
p
u
t
o
f
t
h
e
f
ea
tu
r
es
la
y
er
an
d
f
ed
to
f
u
ll
y
co
n
n
ec
t
d
en
s
e
l
a
y
er
s
w
it
h
b
atch
n
o
r
m
aliza
tio
n
[
4
4
]
an
d
Dr
o
p
o
u
t
[
4
5
]
lay
er
s
.
T
h
e
f
in
a
l
la
y
er
is
d
en
s
e
cla
s
s
i
f
icatio
n
la
y
er
w
it
h
ac
t
iv
at
io
n
o
f
’
s
i
g
m
o
id
’
to
g
et
b
i
n
ar
y
class
i
f
icatio
n
v
al
u
e
w
h
et
h
er
q
u
esti
o
n
s
ar
e
s
i
m
il
ar
o
r
n
o
t.
T
h
e
b
atch
n
o
r
m
aliza
tio
n
[
4
4
]
an
d
d
r
o
p
o
u
t
[
4
5
]
lay
er
s
ar
e
u
s
ed
to
r
eg
u
lar
ize
th
e
o
u
t
p
u
t
o
f
t
h
e
Sia
m
e
s
e
la
y
er
s
an
d
to
av
o
id
c
o
m
m
o
n
p
r
o
b
le
m
s
i
n
d
ee
p
lear
n
in
g
s
u
ch
as:
i)
I
n
ter
n
al
co
v
a
r
iate
s
h
i
f
t
(
i
.
e
.
“
t
h
e
c
h
a
n
g
e
i
n
t
h
e
d
i
s
t
r
i
b
u
t
i
o
n
o
f
n
e
t
w
o
r
k
a
c
t
i
v
a
t
i
o
n
s
d
u
e
t
o
t
h
e
c
h
a
n
g
e
i
n
n
e
t
w
o
r
k
p
a
r
a
m
e
t
e
r
s
d
u
r
i
n
g
t
r
a
i
n
i
n
g
.
”
[
4
4
]
)
an
d
ii)
Ov
er
f
itti
n
g
d
u
r
in
g
n
e
u
r
al
n
et
w
o
r
k
tr
ai
n
i
n
g
.
T
h
e
ter
m
“
d
r
o
p
o
u
t”
r
ef
er
s
to
“
r
a
n
d
o
m
l
y
d
r
o
p
p
in
g
o
u
t
u
n
i
ts
(
h
id
d
en
an
d
v
i
s
ib
le)
in
a
n
e
u
r
al
n
et
w
o
r
k
.
B
y
d
r
o
p
p
in
g
a
u
n
i
t
o
u
t,
w
e
m
ea
n
te
m
p
o
r
ar
il
y
r
em
o
v
i
n
g
it
f
r
o
m
t
h
e
n
et
w
o
r
k
,
alo
n
g
w
it
h
all
it
s
i
n
c
o
m
in
g
a
n
d
o
u
t
g
o
i
n
g
co
n
n
ec
tio
n
s
”
[
4
5
]
.
T
h
e
L
2
r
e
g
u
lar
izat
io
n
w
it
h
v
al
u
e=
0
.
0
0
1
w
a
s
u
s
ed
w
i
th
t
h
e
f
u
ll
y
co
n
n
e
ct
d
en
s
e
la
y
er
s
.
Fig
u
r
e
2
.
T
h
e
Siam
e
s
e
-
b
ased
m
o
d
el
ar
c
h
itect
u
r
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8708
I
n
t J
E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
11
,
No
.
4
,
A
u
g
u
s
t 2
0
2
1
:
3
5
1
9
-
3528
3524
3
.
4
.
3
.
A
P
re
-
t
r
a
ined deep
bi
direct
io
na
l t
ra
ns
f
o
r
m
er
ba
s
e
d o
n B
E
RT
m
o
del
I
n
o
u
r
w
o
r
k
,
w
e
al
s
o
u
s
ed
b
i
d
i
r
e
c
t
i
o
n
a
l
e
n
c
o
d
e
r
r
e
p
r
e
s
e
n
t
a
t
i
o
n
s
f
r
o
m
t
r
a
n
s
f
o
r
m
e
r
s
(
B
E
R
T
)
m
o
d
e
l
[
8
]
.
B
E
R
T
is
a
s
tate
-
of
-
th
e
-
ar
t
m
o
d
el
u
s
ed
f
o
r
s
e
v
er
al
n
atu
r
al
-
la
n
g
u
a
g
e
p
r
o
ce
s
s
in
g
(
N
L
P
)
ap
p
licati
o
n
s
.
B
E
R
T
is
a
lan
g
u
a
g
e
r
ep
r
esen
ta
tio
n
r
elea
s
ed
b
y
Go
o
g
le
i
n
Octo
b
er
2
0
1
8
u
tili
zin
g
t
h
e
en
co
d
er
-
d
ec
o
d
er
tr
an
s
f
o
r
m
er
ar
ch
itect
u
r
e
to
tr
ain
th
e
m
o
d
el
r
ep
r
esen
tatio
n
s
u
s
in
g
u
n
a
n
n
o
tated
d
ata.
T
h
e
B
E
R
T
m
o
d
el
r
ep
r
esen
tatio
n
s
ar
e
b
u
ilt o
v
er
co
n
te
x
t
u
al
r
ep
r
esen
t
at
io
n
s
li
k
e
Se
m
i
-
s
u
p
er
v
i
s
ed
L
ea
r
n
in
g
[
4
6
]
,
UL
MFit [
4
7
]
,
an
d
E
L
Mo
[
4
8
]
.
T
h
e
B
E
R
T
m
o
d
el
h
a
s
m
an
y
v
er
s
io
n
s
,
s
ee
T
ab
le
3
f
o
r
m
o
r
e
d
etails.
T
h
e
Un
ca
s
ed
v
er
s
io
n
s
o
f
B
E
R
T
m
o
d
el
m
ea
n
s
th
at
th
e
tex
t
h
a
s
b
ee
n
lo
w
er
-
ca
s
ed
b
ef
o
r
e
to
k
en
iza
tio
n
a
n
d
t
h
e
C
a
s
ed
v
er
s
io
n
s
m
ea
n
s
th
at
th
e
ca
s
e
o
f
th
e
tex
t
i
s
p
r
eser
v
ed
.
Fo
r
th
e
s
ak
e
o
f
th
is
r
esear
ch
,
th
e
B
E
R
T
-
B
ase,
Un
ca
s
ed
is
u
s
ed
.
T
h
e
B
E
R
T
-
B
ase,
Un
ca
s
ed
is
b
u
i
lt o
u
t o
f
1
2
lay
er
s
w
ith
7
6
8
h
id
d
en
la
y
er
s
,
w
i
th
1
2
h
ea
d
s
,
an
d
1
1
0
M
tr
ain
ed
p
ar
a
m
eter
s
.
T
ab
le
3
.
T
h
e
B
E
R
T
m
o
d
el
v
er
s
io
n
s
B
ER
T
v
e
r
si
o
n
#
L
a
y
e
r
s
#
H
i
d
d
e
n
#
H
e
a
d
s
#
P
a
r
a
me
t
e
r
s
B
ER
T
-
L
a
r
g
e
,
U
n
c
a
se
d
24
1
0
2
4
16
3
4
0
M
B
ER
T
-
L
a
r
g
e
,
C
a
se
d
24
1
0
2
4
16
3
4
0
M
B
ER
T
-
B
a
se
,
U
n
c
a
se
d
12
7
6
8
12
1
1
0
M
B
ER
T
-
B
a
se
,
C
a
se
d
12
7
6
8
12
1
1
0
M
As
s
h
o
w
n
i
n
Fig
u
r
e
3
,
in
t
h
e
B
E
R
T
m
o
d
el,
th
e
i
n
p
u
t
r
ep
r
es
en
ts
a
p
air
o
f
s
en
ten
ce
s
,
w
h
ic
h
is
h
er
e
a
p
air
o
f
q
u
est
io
n
s
,
in
o
n
e
to
k
en
s
eq
u
en
ce
.
As
s
h
o
w
n
,
t
w
o
q
u
esti
o
n
s
p
ac
k
ed
to
g
et
h
er
t
o
th
e
i
n
p
u
t
to
k
e
n
s
eq
u
en
ce
.
T
h
e
f
ir
s
t
to
k
en
o
f
th
e
s
eq
u
e
n
ce
i
s
a
s
p
ec
ial
cla
s
s
i
f
i
ca
tio
n
to
k
e
n
ca
lled
“
[
C
L
S]”
.
T
o
d
if
f
er
en
tiate
th
e
t
w
o
q
u
e
s
tio
n
s
in
t
h
e
to
k
e
n
s
eq
u
en
ce
,
a
s
p
ec
ial
to
k
e
n
ca
lled
“[
SEP
]
”
is
u
s
ed
to
s
ep
ar
ate
th
em
.
T
h
en
,
a
lear
n
ed
e
m
b
ed
d
in
g
i
s
ad
d
ed
to
ev
er
y
to
k
en
i
n
o
r
d
er
to
in
d
icate
w
h
eth
er
it
b
elo
n
g
s
to
Qu
est
io
n
1
o
r
Qu
esti
o
n
2
.
T
h
e
in
p
u
t
e
m
b
ed
d
in
g
i
s
d
en
o
ted
a
s
“
E
”.
T
h
e
f
i
n
al
h
id
d
en
v
ec
to
r
f
o
r
th
e
s
p
ec
ial
“CLS”
to
k
e
n
is
d
en
o
ted
as
“
C
”
an
d
th
e
f
i
n
al
h
id
d
en
v
ec
to
r
f
o
r
th
e
i
th
in
p
u
t
to
k
en
i
s
d
en
o
ted
as
T
i.
T
h
e
in
p
u
t
e
m
b
ed
d
in
g
s
ar
e
r
ep
r
esen
ted
as
th
e
s
u
m
m
a
tio
n
o
f
t
h
e
to
k
e
n
e
m
b
ed
d
i
n
g
s
,
th
e
s
eg
m
e
n
tatio
n
e
m
b
ed
d
in
g
s
,
an
d
th
e
p
o
s
itio
n
e
m
b
ed
d
in
g
s
.
Fig
u
r
e
3
.
T
h
e
B
E
R
T
m
o
d
el
em
b
ed
d
i
n
g
m
ec
h
an
i
s
m
As
p
r
esen
ted
i
n
Fi
g
u
r
e
4
,
th
e
B
E
R
T
-
b
ased
m
o
d
el
u
ti
l
izes
th
e
e
n
co
d
er
-
d
ec
o
d
er
tr
an
s
f
o
r
m
er
ar
ch
itect
u
r
e
to
lear
n
t
h
e
s
e
m
a
n
tic
s
i
m
ilar
it
y
o
f
t
h
e
i
n
p
u
t
q
u
esti
o
n
s
.
T
r
an
s
f
o
r
m
er
s
[
4
9
]
i
m
p
le
m
en
t
s
d
i
f
f
er
e
n
t
la
y
er
s
o
f
m
u
lti
-
h
ea
d
s
el
f
-
a
tt
en
tio
n
w
it
h
f
ee
d
-
f
o
r
w
ar
d
a
n
d
s
k
ip
p
in
g
m
ec
h
an
i
s
m
.
I
n
co
n
tr
ast
to
tr
ad
itio
n
al
atten
tio
n
m
ec
h
a
n
i
s
m
[
5
0
]
,
th
e
m
u
lti
-
h
ea
d
s
el
f
-
atte
n
tio
n
att
en
d
s
o
n
l
y
to
th
e
i
n
p
u
t
s
eq
u
e
n
ce
o
f
tex
t
a
n
d
th
e
m
u
lti
-
h
ea
d
f
u
n
ctio
n
a
lit
y
e
n
ab
l
es
ea
ch
la
y
er
to
atten
d
to
d
if
f
e
r
en
t
w
o
r
d
s
w
it
h
in
t
h
e
in
p
u
t
s
e
q
u
en
ce
o
f
tex
t.
T
h
e
p
o
s
itio
n
al
e
n
co
d
in
g
m
ec
h
a
n
is
m
r
ep
r
esen
t
s
t
h
e
i
n
p
u
t
s
eq
u
e
n
ce
o
r
d
er
,
w
o
r
d
s
p
o
s
itio
n
w
it
h
i
n
t
h
e
s
eq
u
en
ce
,
a
n
d
th
e
d
is
ta
n
ce
b
et
w
ee
n
w
o
r
d
s
as
a
v
ec
to
r
w
h
ich
i
s
th
e
n
ad
d
ed
to
th
e
em
b
ed
d
in
g
la
y
er
.
T
h
ese
v
ec
to
r
s
h
elp
i
n
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J
E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2
0
8
8
-
8708
Usi
n
g
d
ee
p
lea
r
n
in
g
mo
d
els fo
r
lea
r
n
in
g
s
ema
n
tic
text
s
imila
r
ity
o
f
A
r
a
b
ic
q
u
esti
o
n
s
(
Ma
h
mo
u
d
Ha
mma
d
)
3525
ca
p
tu
r
in
g
t
h
e
co
n
tex
t
u
al
in
f
o
r
m
ati
o
n
w
it
h
in
t
h
e
i
n
p
u
t
s
eq
u
en
ce
.
E
ac
h
s
el
f
-
at
ten
t
io
n
la
y
er
is
f
o
llo
w
ed
b
y
r
esid
u
al
co
n
n
ec
tio
n
r
ep
r
esen
te
d
b
y
a
n
o
r
m
al
izatio
n
la
y
er
t
h
a
t
ad
d
s
th
e
i
n
p
u
t
v
ec
to
r
o
f
th
e
s
elf
-
atte
n
tio
n
la
y
er
to
th
e
o
u
tp
u
t
v
ec
to
r
f
r
o
m
t
h
e
s
a
m
e
s
el
f
-
atte
n
tio
n
la
y
er
h
el
p
in
g
to
ca
r
r
y
f
o
r
g
o
tten
i
n
f
o
r
m
atio
n
to
th
e
n
ex
t
la
y
er
.
Fo
r
m
o
r
e
in
f
o
r
m
atio
n
t
h
e
r
ea
d
er
is
r
ed
ir
ec
te
d
to
[
4
9
]
.
Fig
u
r
e
4
.
A
tr
an
s
f
o
r
m
er
ar
ch
it
ec
tu
r
e
o
f
2
s
tack
ed
e
n
co
d
er
s
an
d
d
ec
o
d
er
s
4.
E
XP
E
R
I
M
E
NT
A
T
I
O
N
AN
D
RE
SU
L
T
S
I
n
th
i
s
s
tu
d
y
,
th
e
th
r
ee
d
ev
elo
p
ed
m
o
d
els
ar
e
e
v
alu
a
ted
u
s
in
g
t
h
e
A
r
ab
ic
q
u
est
io
n
s
d
ata
s
et
p
r
o
v
id
ed
b
y
ma
w
d
o
o
3
.
co
m
.
T
h
ese
th
r
e
e
m
o
d
els
ar
e
XGB
o
o
s
t
[
6
]
,
Si
a
m
ese
n
e
u
r
al
n
et
w
o
r
k
[
7
]
,
an
d
B
E
R
T
m
o
d
el
[
8
]
.
T
h
e
F1
m
ea
s
u
r
e
i
s
u
s
ed
to
e
v
alu
ate
t
h
e
p
er
f
o
r
m
an
ce
o
f
th
e
m
o
d
el
s
.
T
h
e
F1
m
ea
s
u
r
e
i
s
t
h
e
h
ar
m
o
n
ic
m
ea
n
o
f
p
r
ec
is
io
n
an
d
r
ec
all.
T
h
e
XGB
o
o
s
t
class
if
ier
w
as
tr
ain
ed
u
s
i
n
g
th
e
p
r
e
-
e
n
g
in
ee
r
ed
f
ea
tu
r
es
co
m
p
u
ted
o
n
th
e
tr
ain
in
g
d
ataset
w
it
h
m
a
x
tr
ee
d
ep
th
o
f
6
,
lear
n
in
g
r
ate
(
eta)
o
f
0
.
0
6
,
0
.
0
4
,
an
d
0
.
0
2
f
o
r
6
.
0
0
0
ep
o
ch
s
.
On
t
h
e
o
th
er
h
an
d
,
th
e
Sia
m
e
s
e
-
b
ased
m
o
d
el
w
as
tr
ain
ed
u
s
i
n
g
t
h
e
p
r
e
-
en
g
i
n
ee
r
ed
f
ea
tu
r
e
s
.
T
h
e
s
h
ar
ed
B
i
-
L
ST
M
an
d
L
ST
M
la
y
er
s
h
ad
1
0
0
h
id
d
en
la
y
er
s
a
n
d
an
in
p
u
t
s
ize
o
f
1
0
0
.
T
h
e
m
o
d
el
w
as
tr
ai
n
ed
f
o
r
1
0
0
ep
o
ch
s
w
it
h
ea
r
l
y
s
to
p
p
in
g
o
n
th
e
9
8
ep
o
ch
.
T
h
e
ea
r
l
y
s
to
p
p
in
g
i
s
u
s
ed
to
a
v
o
id
tr
ai
n
i
n
g
o
v
er
f
itti
n
g
is
b
a
s
ed
o
n
m
o
n
ito
r
i
n
g
t
h
e
v
alid
atio
n
lo
s
s
v
al
u
e.
On
l
y
th
e
m
o
d
el
w
it
h
th
e
b
est
w
ei
g
h
ts
w
as
s
a
v
e
d
an
d
th
en
u
s
ed
f
o
r
ev
alu
a
tin
g
t
h
e
te
s
t
d
ataset.
W
e
h
av
e
tr
ai
n
ed
t
h
e
m
o
d
el
w
it
h
t
h
e
f
o
llo
w
i
n
g
h
y
p
er
-
p
ar
a
m
ete
r
s
:
h
id
d
en
=1
0
0
,
e
m
b
ed
d
in
g
s
ize
=
1
0
0
,
b
atch
s
ize=
5
1
2
,
lear
n
in
g
r
ate=
0
.
0
0
1
,
a
n
d
n
u
m
b
er
o
f
ep
o
ch
s
=9
8
.
Fin
all
y
,
th
e
B
E
R
T
-
b
ased
m
o
d
el
w
as
tr
ai
n
ed
f
o
r
2
0
ep
o
ch
s
w
it
h
d
ata
e
m
b
ed
d
in
g
s
ize
o
f
1
0
0
,
b
atch
s
ize
(
B
S)=
16,
a
lear
n
i
n
g
r
ate
(
L
R
)
=(
2
e
-
5
-
1e
-
5
)
,
a
w
ar
m
-
u
p
p
r
o
p
o
r
tio
n
(
W
P)=
0
.
1
,
an
d
n
u
m
b
er
o
f
i
ter
atio
n
s
p
er
lo
o
p
(
I
P
L
)
=(
1
0
0
0
-
2
5
0
0
0
0
)
.
T
h
e
m
o
d
el
w
as
tr
ain
ed
o
n
t
h
e
s
te
m
m
ed
v
er
s
io
n
o
f
t
h
e
q
u
e
s
tio
n
s
’
p
air
s
w
it
h
o
u
t u
s
in
g
t
h
e
o
th
er
p
r
e
-
en
g
in
ee
r
ed
f
ea
tu
r
es
,
a
s
s
h
o
w
n
i
n
T
a
b
le
4.
T
ab
le
4
.
T
h
e
h
y
p
er
p
ar
am
eter
s
u
s
ed
to
tr
ain
t
h
e
B
E
R
T
-
b
ased
m
o
d
el
a
n
d
th
eir
r
es
u
lts
o
n
th
e
test
d
ata
BS
#
E
p
o
c
h
s
LR
WP
I
P
L
F1
-
S
c
o
r
e
16
2
2e
-
5
0
.
1
1
0
0
0
8
8
.
7
7
%
16
5
2e
-
5
0
.
1
1
0
0
0
0
9
0
.
1
2
%
16
10
1e
-
5
0
.
1
1
0
0
0
0
0
9
1
.
5
6
%
16
15
1e
-
5
0
.
1
1
5
0
0
0
0
9
1
.
2
0
%
16
20
1e
-
5
0
.
1
2
5
0
0
0
0
9
2
.
9
9
%
T
ab
le
5
s
h
o
w
s
t
h
at
t
h
e
B
E
R
T
-
b
ased
m
o
d
el
o
u
tp
er
f
o
r
m
s
th
e
o
th
er
t
w
o
m
o
d
els
w
i
th
a
n
F1
=9
2
.
9
9
%,
w
h
er
ea
s
t
h
e
Sia
m
ese
-
b
ased
m
o
d
el
co
m
e
s
i
n
t
h
e
s
ec
o
n
d
p
lace
w
ith
F1
=
8
9
.
0
4
8
%.
Fin
all
y
,
th
e
X
GB
o
o
s
t,
as
a
b
aselin
e
m
o
d
el,
ac
h
iev
ed
t
h
e
lo
w
est
r
e
s
u
lt
o
f
F1
=
8
6
.
0
8
6
%.
I
t
i
s
w
o
r
t
h
m
e
n
tio
n
i
n
g
t
h
at
t
h
e
r
esu
l
ts
w
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8708
I
n
t J
E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
11
,
No
.
4
,
A
u
g
u
s
t 2
0
2
1
:
3
5
1
9
-
3528
3526
o
b
tain
ed
in
th
i
s
r
esear
ch
is
th
e
b
est
r
esu
lts
o
n
t
h
is
d
ataset
i
n
clu
d
i
n
g
o
u
r
p
r
eli
m
in
ar
y
w
o
r
k
in
[
9
]
in
w
h
ic
h
th
e
b
est
m
o
d
el
in
[
9
]
ac
h
iev
ed
F
-
m
ea
s
u
r
e
o
f
8
2
.
6
1
%.
Fig
u
r
e
5
d
ep
icts
th
e
m
o
d
el
ac
cu
r
ac
y
o
n
b
o
th
t
h
e
tr
ain
in
g
a
n
d
v
alid
atio
n
d
atase
ts
d
u
r
i
n
g
t
h
e
tr
ain
i
n
g
p
h
ase.
Fi
g
u
r
e
6
s
h
o
w
s
t
h
e
lo
s
s
v
al
u
e
f
o
r
b
o
th
t
h
e
tr
ain
i
n
g
a
n
d
th
e
v
alid
atio
n
d
u
r
in
g
th
e
m
o
d
el
tr
ain
in
g
p
h
ase.
As d
ep
icted
in
b
o
th
f
i
g
u
r
e
s
n
o
m
o
d
el
o
v
er
f
itti
n
g
ca
n
b
e
s
ee
n
d
u
r
in
g
t
h
e
tr
ai
n
in
g
p
h
a
s
e.
T
ab
le
5
.
T
h
e
r
esu
lts
o
b
tain
ed
o
u
t o
f
th
e
t
h
r
ee
d
ev
elo
p
ed
m
o
d
els
M
o
d
e
l
X
G
B
o
o
st
S
i
a
me
se
-
b
a
se
d
B
ER
T
-
b
a
se
d
F1
8
6
.
0
8
6
%
8
9
.
0
4
8
%
9
2
.
9
9
%
Fig
u
r
e
5
.
T
h
e
co
m
p
u
ted
ac
cu
r
ac
y
o
f
t
h
e
Sia
m
ese
-
b
ased
m
o
d
el
o
n
t
h
e
tr
ain
i
n
g
a
n
d
v
alid
atio
n
d
ata
s
ets
Fig
u
r
e
6
.
T
h
e
lo
s
s
v
alu
e
s
o
n
t
h
e
tr
ain
i
n
g
an
d
v
alid
atio
n
o
n
ea
c
h
ep
o
ch
d
u
r
i
n
g
t
h
e
Sia
m
e
s
e
-
b
ased
tr
ain
i
n
g
5.
DIS
CU
SS
I
O
N
Hav
i
n
g
a
clo
s
er
lo
o
k
to
th
e
e
x
p
er
i
m
e
n
tatio
n
r
esu
lt
s
,
it
ca
n
b
e
s
ee
n
th
at
t
h
e
B
E
R
T
-
b
ased
m
o
d
el
is
o
u
tp
er
f
o
r
m
i
n
g
th
e
o
t
h
er
t
w
o
m
o
d
el
s
i
n
ter
m
s
o
f
F1
r
esu
lts
w
it
h
3
%
h
i
g
h
er
th
a
n
t
h
e
Sia
m
ese
-
b
ased
m
o
d
el
a
n
d
6
%
h
i
g
h
er
t
h
an
th
e
XGB
o
o
s
t
o
n
e.
T
h
e
r
es
u
lt
s
g
o
in
li
n
e
w
it
h
liter
at
u
r
e
as
th
e
T
r
an
s
f
o
r
m
er
s
b
ased
m
o
d
els
s
u
c
h
as
B
E
R
T
[
8
]
,
UL
MFit
[
4
7
]
,
an
d
E
L
MO
[
4
8
]
ar
e
r
ev
o
lu
tio
n
izi
n
g
t
h
e
N
L
P
r
esear
ch
.
T
h
e
s
e
t
e
c
h
n
i
q
u
e
s
a
r
e
l
e
a
d
i
n
g
t
h
e
d
e
v
e
l
o
p
e
d
m
o
d
e
l
s
i
n
m
a
n
y
N
L
P
t
a
s
k
s
s
u
c
h
a
s
t
e
x
t
c
l
a
s
s
i
f
i
c
a
t
i
o
n
a
n
d
s
e
q
u
e
n
c
e
-
to
-
s
e
q
u
e
n
c
e
lab
eli
n
g
.
I
n
co
n
tr
as
t
to
t
h
e
o
t
h
er
t
w
o
d
ev
elo
p
ed
m
o
d
els,
t
h
e
B
E
R
T
-
b
ased
m
o
d
el
w
a
s
ab
le
t
o
lear
n
t
h
e
s
e
m
a
n
tic
s
i
m
ilar
it
y
a
m
o
n
g
i
n
p
u
t
q
u
es
tio
n
s
’
p
air
s
w
it
h
o
u
t
t
h
e
n
e
ed
f
o
r
p
r
e
-
en
g
in
ee
r
ed
f
ea
t
u
r
es.
T
h
is
ex
p
lai
n
s
th
e
p
o
w
er
o
f
tr
a
n
s
f
o
r
m
er
s
i
n
h
an
d
li
n
g
N
L
P
task
s
m
o
r
e
ef
f
i
cien
tl
y
w
h
en
co
m
p
ar
ed
to
C
NN
an
d
R
NN
-
b
ased
m
o
d
el
s
.
C
o
m
p
u
ti
n
g
f
ea
t
u
r
es c
an
r
ed
u
ce
t
h
e
ap
p
licab
ilit
y
o
f
t
h
e
d
ev
elo
p
ed
m
o
d
el
f
o
r
p
r
o
d
u
c
t
i
o
n
s
e
r
v
i
c
e
s
.
U
s
e
r
s
m
a
y
g
e
t
n
e
g
a
t
i
v
e
e
x
p
e
r
i
e
n
c
e
w
a
i
t
i
n
g
f
o
r
t
h
e
m
o
d
e
l
t
o
c
o
m
p
u
t
e
t
h
e
f
e
a
t
u
r
e
s
a
n
d
t
h
e
n
c
l
a
s
s
i
f
y
th
e
in
p
u
t te
x
t.
Fo
cu
s
i
n
g
o
n
t
h
e
p
r
e
-
en
g
i
n
ee
r
ed
f
ea
tu
r
es,
s
elec
ted
f
ea
t
u
r
es
b
o
asted
th
e
r
esu
lts
o
f
th
e
d
ev
elo
p
ed
m
o
d
el
s
.
T
h
e
S
ia
m
e
s
e
-
b
a
s
ed
m
o
d
el
ac
h
iev
ed
o
n
l
y
a
n
F1
v
al
u
e
o
f
7
8
.
1
8
6
%
w
ith
o
u
t
th
e
p
r
e
-
e
n
g
i
n
ee
r
ed
f
ea
t
u
r
es
(
r
el
y
i
n
g
o
n
l
y
o
n
t
h
e
e
m
b
ed
d
in
g
f
ea
t
u
r
es).
T
h
is
in
d
icate
s
h
o
w
p
o
w
er
f
u
l
t
h
e
f
ea
tu
r
es
s
elec
ted
to
tr
ai
n
th
e
m
o
d
els
w
it
h
a
m
ar
g
i
n
o
f
r
esu
lts
’
en
h
a
n
ce
m
en
t
r
ea
ch
e
s
1
0
%
f
o
r
th
e
Sia
m
e
s
e
-
b
a
s
ed
m
o
d
el.
T
h
is
also
em
p
h
a
s
izes
h
o
w
p
o
w
er
f
u
l
o
u
r
B
E
R
T
-
b
ased
m
o
d
el
w
h
e
n
co
m
p
ar
ed
to
t
h
e
S
ia
m
e
s
e
-
b
ased
m
o
d
el
w
ith
o
u
t
f
ea
t
u
r
es
w
it
h
a
d
if
f
er
en
ce
o
f
a
r
o
u
n
d
1
5
% in
ter
m
s
o
f
ac
h
iev
ed
r
esu
lts
w
it
h
o
u
t f
ea
t
u
r
es.
6.
CO
NCLU
SI
O
N
AND
F
U
T
U
RE
WO
RK
T
h
is
r
esear
ch
p
r
o
p
o
s
es
th
r
ee
d
if
f
er
e
n
t
ap
p
r
o
ac
h
es
to
a
n
al
y
z
e
th
e
s
e
m
a
n
tic
s
i
m
ilar
it
y
b
et
wee
n
a
p
air
o
f
A
r
ab
ic
q
u
e
s
tio
n
s
.
T
h
e
f
ir
s
t
m
o
d
el
i
s
a
s
u
p
er
v
is
ed
-
m
ac
h
i
n
e
lear
n
in
g
m
o
d
el
u
s
i
n
g
XGB
o
o
s
t
tr
ain
ed
u
s
in
g
a
s
et
o
f
p
r
e
-
e
n
g
i
n
ee
r
ed
f
ea
t
u
r
es
,
th
e
s
ec
o
n
d
i
s
a
n
ad
ap
ted
Si
a
m
ese
-
b
ased
d
ee
p
lear
n
in
g
r
e
cu
r
r
en
t
ar
c
h
itect
u
r
e
also
tr
ain
ed
u
s
in
g
a
s
et
o
f
p
r
e
-
en
g
i
n
ee
r
ed
f
ea
t
u
r
es,
an
d
f
in
all
y
,
a
p
r
e
-
tr
ain
ed
d
ee
p
b
id
ir
e
ctio
n
al
tr
an
s
f
o
r
m
er
b
ased
o
n
B
E
R
T
m
o
d
el.
T
h
e
p
r
o
p
o
s
ed
ap
p
r
o
ac
h
es
w
er
e
ev
al
u
ated
u
s
i
n
g
a
d
ataset
co
llected
b
y
m
a
w
d
o
o
3
.
co
m
(
s
ee
s
ec
tio
n
3
.
1
.
)
.
T
h
e
ev
al
u
atio
n
r
es
u
lts
s
h
o
w
t
h
at
t
h
e
B
E
R
T
-
b
ased
m
o
d
el
o
u
tp
er
f
o
r
m
s
t
h
e
o
t
h
er
t
w
o
p
r
o
p
o
s
ed
m
o
d
els
w
it
h
6
% o
f
e
n
h
a
n
ce
m
en
t in
t
h
e
F1
-
s
co
r
e
(
s
ee
s
ec
tio
n
5
)
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J
E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2
0
8
8
-
8708
Usi
n
g
d
ee
p
lea
r
n
in
g
mo
d
els fo
r
lea
r
n
in
g
s
ema
n
tic
text
s
imila
r
ity
o
f
A
r
a
b
ic
q
u
esti
o
n
s
(
Ma
h
mo
u
d
Ha
mma
d
)
3527
I
n
th
is
r
esear
c
h
,
w
e
h
av
e
o
n
l
y
co
n
s
id
er
ed
d
etec
tin
g
i
f
t
w
o
q
u
esti
o
n
s
ar
e
s
i
m
ilar
o
r
n
o
t.
Dete
ctin
g
s
i
m
ilar
q
u
es
tio
n
s
to
a
g
i
v
en
q
u
esti
o
n
u
s
in
g
o
u
r
ap
p
r
o
ac
h
is
a
n
i
n
ter
esti
n
g
a
v
en
u
e
o
f
f
u
tu
r
e
w
o
r
k
.
B
es
id
es t
h
at,
w
e
p
la
n
to
e
n
h
a
n
ce
t
h
e
B
E
R
T
-
b
ased
m
o
d
el
ar
ch
itect
u
r
e
b
y
co
m
b
i
n
i
n
g
th
e
p
r
e
-
en
g
i
n
ee
r
ed
f
ea
tu
r
e
s
to
it,
a
n
d
in
v
e
s
ti
g
ate
t
h
eir
i
m
p
ac
t
o
n
t
h
e
m
o
d
el
r
esu
l
ts
.
Mo
r
eo
v
er
,
we
ar
e
p
lan
n
i
n
g
to
ex
tr
ac
t
f
ea
t
u
r
es
f
r
o
m
th
e
B
E
R
T
m
o
d
el
a
n
d
f
ee
d
t
h
e
m
to
o
th
er
m
ac
h
in
e
lear
n
i
n
g
ap
p
r
o
ac
h
es
u
tili
zi
n
g
t
h
e
f
lex
ib
le
ar
c
h
itec
tu
r
e
o
f
a
n
en
co
d
er
-
d
ec
o
d
er
a
r
ch
itectu
r
e
in
a
tr
a
n
s
f
er
lear
n
i
n
g
m
ec
h
an
is
m
.
ACK
NO
WL
E
D
G
E
M
E
NT
S
T
h
is
r
esear
ch
is
p
ar
tiall
y
f
u
n
d
ed
b
y
J
o
r
d
an
Un
i
v
er
s
it
y
o
f
S
cien
ce
an
d
T
ec
h
n
o
lo
g
y
,
R
ese
ar
ch
Gr
an
t
Nu
m
b
er
: 2
0
1
7
0
1
0
7
an
d
2
0
1
9
3
0
1
.
RE
F
E
R
E
NC
E
S
[1
]
J.
Ra
m
a
p
ra
b
h
a
,
S
.
Da
s,
a
n
d
P
.
M
u
k
e
rjee
,
“
S
u
rv
e
y
o
n
se
n
ten
c
e
sim
il
a
rit
y
e
v
a
lu
a
ti
o
n
u
sin
g
d
e
e
p
lea
r
n
in
g
,
”
J
o
u
rn
a
l
o
f
P
h
y
sic
s:
Co
n
fer
e
n
c
e
S
e
rie
s
,
v
o
l.
1
0
0
0
,
2
0
1
8
,
A
rt
.
n
o
.
0
1
2
0
7
0
.
[2
]
S
.
Zh
a
n
g
,
X
.
Zh
e
n
g
,
a
n
d
C.
Hu
,
“
A
su
rv
e
y
o
f
se
m
a
n
ti
c
si
m
il
a
rit
y
a
n
d
it
s
a
p
p
li
c
a
ti
o
n
to
s
o
c
ial
n
e
tw
o
rk
a
n
a
l
y
sis,
”
2
0
1
5
IE
EE
I
n
ter
n
a
t
io
n
a
l
C
o
n
fer
e
n
c
e
o
n
Bi
g
D
a
ta
(
Bi
g
Da
t
a
)
,
S
a
n
t
a
Clara
,
CA
,
US
A
,
2
0
1
5
,
p
p
.
2
3
6
2
-
2
3
6
7
.
[3
]
D.
S
a
´
n
c
h
e
z
,
M
.
Ba
tet,
D.
Ise
r
n
,
a
n
d
A
.
V
a
ll
s,
“
On
t
o
l
o
g
y
-
b
a
s
e
d
se
m
a
n
ti
c
sim
il
a
rit
y
:
A
n
e
w
f
e
a
tu
re
-
b
a
se
d
ap
p
ro
a
c
h
,
”
Exp
e
rt sy
ste
ms
wit
h
a
p
p
li
c
a
ti
o
n
s
,
v
o
l
.
39
,
n
o
.
9
,
p
p
.
7
7
1
8
-
7
7
2
8
,
2
0
1
2
.
[4
]
E.
Ag
irre,
M
.
Dia
b
,
D.
Ce
r,
a
n
d
A
.
G
o
n
z
a
lez
-
Ag
irre,
“
S
e
m
e
v
a
l
-
2
0
1
2
tas
k
6
:
A
p
il
o
t
o
n
se
m
a
n
ti
c
te
x
tu
a
l
si
m
il
a
rit
y
,
”
Pro
c
e
e
d
in
g
s
o
f
t
h
e
Fi
rs
t
J
o
in
t
Co
n
fer
e
n
c
e
o
n
L
e
x
ica
l
a
n
d
Co
m
p
u
t
a
ti
o
n
a
l
S
e
ma
n
ti
c
s
-
V
o
lu
me
1
:
Pro
c
e
e
d
in
g
s
o
f
th
e
ma
in
c
o
n
fer
e
n
c
e
a
n
d
t
h
e
sh
a
re
d
t
a
sk
,
a
n
d
Vo
lu
me
2
:
Pro
c
e
e
d
i
n
g
s
o
f
t
h
e
S
ixth
I
n
ter
n
a
ti
o
n
a
l
W
o
rk
sh
o
p
o
n
S
e
ma
n
ti
c
Eva
lu
a
ti
o
n
,
A
ss
o
c
iatio
n
f
o
r
C
o
m
p
u
tati
o
n
a
l
L
in
g
u
isti
c
s,
2
0
1
2
,
p
p
.
3
8
5
-
3
9
3
.
[5
]
S
e
m
E
v
a
l2
0
1
9
,
“
S
e
m
a
n
ti
c
e
v
a
lu
a
ti
o
n
2
0
1
9
,
”
2
0
1
9
.
[
On
li
n
e
]
.
A
v
a
il
a
b
le:
h
t
tp
:/
/alt
.
q
c
ri.
o
rg
/se
m
e
v
a
l2
0
1
9
/.
[6
]
T
.
Ch
e
n
,
a
n
d
C.
G
u
e
strin
,
“
X
g
b
o
o
st:
A
sc
a
lab
le
tre
e
b
o
o
stin
g
sy
ste
m
,
”
KDD
'
1
6
:
Pr
o
c
e
e
d
in
g
s
o
f
th
e
2
2
n
d
ACM
S
IGKD
D In
ter
n
a
ti
o
n
a
l
Co
n
fer
e
n
c
e
o
n
K
n
o
wle
d
g
e
Disc
o
v
e
ry
a
n
d
D
a
ta
M
i
n
in
g
,
2
0
1
6
,
p
p
.
7
8
5
-
7
9
4
.
[7
]
J.
M
u
e
ll
e
r,
a
n
d
A
.
T
h
y
a
g
a
r
a
jan
,
“
S
iam
e
se
re
c
u
rre
n
t
a
rc
h
it
e
c
tu
re
s
f
o
r
lea
rn
in
g
se
n
ten
c
e
si
m
il
a
r
it
y
,
”
AA
AI'
1
6
:
P
ro
c
e
e
d
in
g
s
o
f
t
h
e
T
h
irti
e
th
AA
A
I
Co
n
fer
e
n
c
e
o
n
Arti
fi
c
ia
l
I
n
telli
g
e
n
c
e
,
2
0
1
6
,
p
p
.
2
7
8
6
-
2
7
9
2
.
[8
]
J.
De
v
li
n
,
M
.
-
W
.
Ch
a
n
g
,
K.
L
e
e
,
K.
T
o
u
tan
o
v
a
,
“
Be
rt:
P
re
-
trai
n
in
g
o
f
d
e
e
p
b
id
irec
ti
o
n
a
l
tra
n
sf
o
rm
e
rs
f
o
r
lan
g
u
a
g
e
u
n
d
e
rsta
n
d
i
n
g
,
”
Co
mp
u
ta
t
io
n
a
n
d
L
a
n
g
u
a
g
e
,
2
0
1
9
.
[9
]
M
.
Ha
m
m
a
d
,
M
.
A
L
-
S
m
a
d
i,
Q.
Ba
n
i
Ba
k
e
r,
M
.
A
l
-
a
sa
’d
,
N.
A
l
-
k
h
d
o
u
r,
M
.
B.
Y
o
u
n
e
s,
E.
Kh
w
a
il
e
h
,
“
Qu
e
sti
o
n
t
o
q
u
e
stio
n
sim
il
a
rit
y
a
n
a
l
y
sis
u
sin
g
m
o
rp
h
o
l
o
g
ica
l,
s
y
n
tac
ti
c
,
se
m
a
n
ti
c
,
a
n
d
lex
ica
l
f
e
a
tu
re
s,
”
J
o
u
rn
a
l
o
f
Un
ive
rs
a
l
Co
m
p
u
ter
S
c
ien
c
e
,
v
o
l.
2
6
,
n
o
.
6
,
p
p
.
6
7
1
-
697
,
2
0
2
0
.
[1
0
]
F
.
S
a
ric,
G
.
G
la
v
a
s,
M
.
Ka
ra
n
,
J.
S
n
a
jd
e
r,
B.
D.
Ba
sic
,
“
T
a
k
e
lab
:
S
y
st
e
m
s
f
o
r
m
e
a
su
ri
n
g
se
m
a
n
ti
c
tex
t
si
m
il
a
rit
y
,
”
Fi
rs
t
J
o
in
t
Co
n
fer
e
n
c
e
o
n
L
e
x
ica
l
a
n
d
Co
m
p
u
t
a
ti
o
n
a
l
S
e
ma
n
t
ics
(
*
S
EM
),
2
0
1
2
,
p
p
.
4
4
1
-
4
4
8
.
[1
1
]
T
.
Zh
u
,
M
.
L
a
n
,
“
ECNU:
L
e
v
e
ra
g
in
g
o
n
e
n
se
m
b
le
o
f
h
e
tero
g
e
n
e
o
u
s
f
e
a
tu
re
s
a
n
d
in
f
o
rm
a
ti
o
n
e
n
rich
m
e
n
t
f
o
r
c
ro
ss
lev
e
l
se
m
a
n
ti
c
si
m
il
a
rit
y
e
sti
m
a
t
io
n
,
”
Pr
o
c
e
e
d
in
g
s
o
f
th
e
8
th
I
n
ter
n
a
ti
o
n
a
l
W
o
rk
sh
o
p
o
n
S
e
ma
n
ti
c
Eva
lu
a
ti
o
n
(
S
e
mEv
a
l
2
0
1
4
)
,
2
0
1
4
,
p
p
.
2
6
5
-
2
7
0
.
[1
2
]
N.
P
.
A
.
Vo
,
O.
P
o
p
e
sc
u
,
a
n
d
T
.
Ca
se
ll
i,
“
F
BK
-
TR
:
S
V
M
f
o
r
se
m
a
n
ti
c
re
late
d
n
e
ss
a
n
d
c
o
rp
u
s
p
a
tt
e
rn
s
f
o
r
RT
E
,
”
Pro
c
e
e
d
in
g
s
o
f
t
h
e
8
th
I
n
ter
n
a
ti
o
n
a
l
W
o
rk
sh
o
p
o
n
S
e
ma
n
ti
c
Eva
lu
a
ti
o
n
(
S
e
mEv
a
l
2
0
1
4
)
,
2
0
1
4
,
p
p
.
2
8
9
-
2
9
3
.
[1
3
]
J.
Zh
a
o
,
a
n
d
M
.
L
a
n
,
“
ECNU:
L
e
v
e
ra
g
in
g
w
o
rd
e
m
b
e
d
d
in
g
s
to
b
o
o
st
p
e
rf
o
rm
a
n
c
e
f
o
r
p
a
ra
p
h
ra
se
in
t
w
it
ter,
”
Pro
c
e
e
d
in
g
s
o
f
t
h
e
9
th
I
n
ter
n
a
ti
o
n
a
l
W
o
rk
sh
o
p
o
n
S
e
ma
n
ti
c
Eva
lu
a
ti
o
n
(
S
e
mEv
a
l
2
0
1
5
)
,
2
0
1
5
,
p
p
.
3
4
-
39.
[1
4
]
A
.
S
a
n
b
o
rn
,
a
n
d
J.
S
k
ry
z
a
li
n
,
“
De
e
p
lea
rn
in
g
f
o
r
se
m
a
n
ti
c
sim
il
a
rit
y
,
CS
2
2
4
d
:
De
e
p
L
e
a
rn
in
g
f
o
r
Na
tu
ra
l
L
a
n
g
u
a
g
e
P
r
o
c
e
ss
in
g
S
tan
f
o
rd
,
”
CA
,
USA
:
S
tan
f
o
rd
Un
iv
e
rsity
,
2
0
1
5
.
[1
5
]
S
.
A
ro
ra
,
Y.
L
ian
g
,
a
n
d
T
.
M
a
,
“
A
si
m
p
le
b
u
t
t
o
u
g
h
-
to
-
b
e
a
t
b
a
s
e
li
n
e
f
o
r
se
n
ten
c
e
e
m
b
e
d
d
in
g
s,”
Pu
b
li
s
h
e
d
a
s
a
c
o
n
fer
e
n
c
e
p
a
p
e
r a
t
IC
L
R
2
0
1
7
,
p
p
.
1
-
1
6
,
2
0
1
7
.
[1
6
]
S
.
W
a
n
,
Y.
L
a
n
,
J.
G
u
o
,
J.
X
u
,
L
.
P
a
n
g
,
X
.
C
h
e
n
g
,
“
A
d
e
e
p
a
r
c
h
it
e
c
tu
re
f
o
r
se
m
a
n
ti
c
m
a
tch
in
g
w
it
h
m
u
lt
ip
le
p
o
siti
o
n
a
l
se
n
ten
c
e
re
p
re
se
n
tatio
n
s,”
T
h
irti
e
th
AA
AI
Co
n
fer
e
n
c
e
o
n
Arti
fi
c
ia
l
I
n
telli
g
e
n
c
e
,
2
0
1
6
.
[1
7
]
J.
T
ian
,
Z.
Zh
o
u
,
M
.
L
a
n
,
Y.
W
u
,
“
ECNU
a
t
s
e
m
e
v
a
l
-
2
0
1
7
tas
k
1
:
L
e
v
e
ra
g
e
k
e
rn
e
l
-
b
a
se
d
trad
it
io
n
a
l
NL
P
f
e
a
tu
re
s
a
n
d
n
e
u
ra
l
n
e
tw
o
rk
s
to
b
u
il
d
a
u
n
iv
e
rsa
l
m
o
d
e
l
f
o
r
m
u
lt
il
in
g
u
a
l
a
n
d
c
ro
ss
-
li
n
g
u
a
l
se
m
a
n
ti
c
tex
t
u
a
l
sim
il
a
rit
y
,
”
Pro
c
e
e
d
in
g
s
o
f
t
h
e
1
1
t
h
In
ter
n
a
ti
o
n
a
l
W
o
rk
sh
o
p
o
n
S
e
ma
n
t
ic E
v
a
l
u
a
ti
o
n
(
S
e
mEv
a
l
-
2
0
1
7
)
,
2
0
1
7
,
p
p
.
1
9
1
-
1
9
7
.
[1
8
]
D.
P
rij
a
telj
,
J.
Ka
li
t
a
,
J.
V
e
n
tu
r
a
,
“
Ne
u
ra
l
n
e
t
w
o
rk
s
f
o
r
se
m
a
n
ti
c
tex
tu
a
l
si
m
il
a
rit
y
,
”
Pro
c
e
e
d
in
g
s
o
f
th
e
1
4
t
h
In
ter
n
a
t
io
n
a
l
C
o
n
fer
e
n
c
e
o
n
N
a
tu
ra
l
L
a
n
g
u
a
g
e
Pro
c
e
ss
in
g
(
ICON
-
2
0
1
7
)
,
2
0
1
7
,
p
p
.
4
5
6
-
4
6
5
.
[1
9
]
H.
He
,
a
n
d
J.
L
in
,
“
P
a
irw
ise
w
o
rd
in
tera
c
ti
o
n
m
o
d
e
li
n
g
w
it
h
d
e
e
p
n
e
u
ra
l
n
e
tw
o
rk
s
f
o
r
se
m
a
n
ti
c
sim
il
a
rit
y
m
e
a
su
re
m
e
n
t,
”
Pro
c
e
e
d
in
g
s o
f
N
AA
CL
-
HLT
2
0
1
6
,
2
0
1
6
,
p
p
.
9
3
7
-
9
4
8
.
[2
0
]
H.
He
,
K.
G
i
m
p
e
l,
J.
L
in
,
“
M
u
lt
i
-
p
e
rsp
e
c
ti
v
e
se
n
ten
c
e
si
m
il
a
rit
y
m
o
d
e
li
n
g
w
it
h
c
o
n
v
o
lu
ti
o
n
a
l
n
e
u
ra
l
n
e
t
w
o
rk
s,
”
Pro
c
e
e
d
in
g
s
o
f
t
h
e
2
0
1
5
C
o
n
fer
e
n
c
e
o
n
Emp
irica
l
M
e
th
o
d
s in
Na
t
u
ra
l
L
a
n
g
u
a
g
e
Pro
c
e
ss
in
g
,
2
0
1
5
,
p
p
.
1
5
7
6
-
1
5
8
6
.
[2
1
]
J.
M
u
e
ll
e
r,
a
n
d
A
.
T
h
y
a
g
a
r
a
jan
,
“
S
iam
e
se
r
e
c
u
rre
n
t
a
rc
h
it
e
c
tu
re
s
f
o
r
lea
rn
in
g
se
n
ten
c
e
si
m
il
a
rit
y
,
”
T
h
irti
e
th
AA
AI
Co
n
fer
e
n
c
e
o
n
Arti
fi
c
i
a
l
I
n
telli
g
e
n
c
e
,
2
0
1
6
,
p
p
.
2
7
6
6
-
2
7
9
2
.
[2
2
]
B.
Ag
a
r
w
a
l,
H.
Ra
m
a
m
p
iaro
,
H.
L
a
n
g
s
e
th
,
M
.
Ru
o
c
c
o
,
“
A
d
e
e
p
n
e
tw
o
rk
m
o
d
e
l
f
o
r
p
a
ra
p
h
ra
se
d
e
tec
ti
o
n
in
sh
o
rt
tex
t
m
e
ss
a
g
e
s,
”
In
fo
rm
a
ti
o
n
Pro
c
e
ss
in
g
&
M
a
n
a
g
e
me
n
t
,
v
o
l.
5
4
,
n
o
.
6
,
p
p
.
9
2
2
-
9
3
7
,
2
0
1
8
.
[2
3
]
Y.
L
e
,
Z.
J
.
W
a
n
g
,
Z
.
Q
u
a
n
,
J
.
H
e
,
B
.
Y
a
o
,
“
A
C
V
-
T
R
E
E
:
A
n
e
w
m
e
t
h
o
d
f
o
r
s
e
n
t
e
n
c
e
s
i
m
i
l
a
r
i
t
y
m
o
d
e
l
i
n
g
,
”
P
r
o
c
e
e
d
i
n
g
s
o
f
t
h
e
T
w
e
n
t
y
-
S
e
v
e
n
t
h
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
i
n
t
C
o
n
f
e
r
e
n
c
e
o
n
A
r
t
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
(
I
J
C
A
I
)
,
2
0
1
8
,
p
p
.
4
1
3
7
-
4
1
4
3
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8708
I
n
t J
E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
11
,
No
.
4
,
A
u
g
u
s
t 2
0
2
1
:
3
5
1
9
-
3528
3528
[2
4
]
J.
T
ian
,
M
.
L
a
n
,
Y.
W
u
,
J
.
W
a
n
g
,
L
.
Qiu
,
S
.
L
i,
L
.
Ju
n
,
L
.
S
i
,
“
A
n
a
d
v
e
rsa
rial
jo
in
t
lea
rn
i
n
g
m
o
d
e
l
fo
r
lo
w
-
re
so
u
rc
e
lan
g
u
a
g
e
se
m
a
n
ti
c
tex
tu
a
l
si
m
il
a
rit
y
,
”
Eu
ro
p
e
a
n
C
o
n
fer
e
n
c
e
o
n
I
n
fo
rm
a
t
io
n
Retrie
v
a
l
(
ECIR
2
0
1
8
)
,
v
o
l.
1
0
7
7
1
,
2
0
1
8
,
p
p
.
8
9
-
1
0
1
.
[2
5
]
X
.
T
a
n
g
,
S
.
C
h
e
n
,
L
.
Do
,
Z.
M
i
n
,
F
.
Ji,
H.
Y
u
e
t
a
l.
,
“
Im
p
ro
v
in
g
m
u
lt
il
in
g
u
a
l
se
m
a
n
ti
c
tex
tu
a
l
sim
il
a
rit
y
w
it
h
sh
a
re
d
se
n
ten
c
e
e
n
c
o
d
e
r
f
o
r
l
o
w
-
re
so
u
rc
e
lan
g
u
a
g
e
s,
”
G
ro
u
n
d
Al
,
2
0
1
8
.
[2
6
]
Y.
Ya
n
g
,
S
.
Yu
a
n
,
D.
Ce
r,
S
.
-
y
.
Ko
n
g
,
N.
Co
n
sta
n
t
,
P
.
P
il
a
r
e
t
a
l.
,
“
L
e
a
rn
in
g
se
m
a
n
ti
c
te
x
tu
a
l
sim
il
a
rit
y
f
ro
m
c
o
n
v
e
rsa
ti
o
n
s,
”
Pro
c
e
e
d
i
n
g
s
o
f
T
h
e
T
h
ir
d
W
o
rk
sh
o
p
o
n
Re
p
re
se
n
ta
ti
o
n
L
e
a
rn
i
n
g
f
o
r
NL
P
,
p
p
.
1
6
4
-
1
7
4
,
2
0
1
8
.
[2
7
]
A.
-
S
.
M
o
h
a
m
m
a
d
,
Z.
Ja
ra
d
a
t,
A
.
-
A
.
M
a
h
m
o
u
d
,
Y.
Ja
ra
rwe
h
,
“
P
a
ra
p
h
ra
se
id
e
n
t
if
ica
ti
o
n
a
n
d
se
m
a
n
ti
c
tex
t
sim
il
a
rit
y
a
n
a
l
y
sis in
A
ra
b
ic
n
e
w
s
tw
e
e
ts u
sin
g
lex
ica
l,
s
y
n
tac
ti
c
,
a
n
d
se
m
a
n
ti
c
f
e
a
tu
re
s,
”
In
fo
rm
a
ti
o
n
Pro
c
e
ss
in
g
&
M
a
n
a
g
e
me
n
t
,
v
o
l.
5
3
,
n
o
.
3
,
p
p
.
6
4
0
-
6
5
2
.
2
0
1
7
.
[2
8
]
D.
S
a
g
h
e
e
r,
a
n
d
F
.
S
u
k
k
a
r,
“
A
ra
b
ic
se
n
ten
c
e
s
c
las
si
f
ic
a
ti
o
n
v
ia
d
e
e
p
lea
rn
in
g
,
”
In
ter
n
a
ti
o
n
a
l
J
o
u
r
n
a
l
o
f
Co
mp
u
ter
Ap
p
li
c
a
ti
o
n
s
,
v
o
l
.
1
8
2
,
n
o
.
5
,
p
p
.
40
-
4
6
,
2
0
1
8
.
[2
9
]
S
.
A
lza
h
ra
n
i,
“
Cro
ss
-
lan
g
u
a
g
e
s
e
m
a
n
ti
c
si
m
il
a
rit
y
o
f
Ara
b
i
c
-
En
g
li
sh
sh
o
rt
p
h
ra
se
s
a
n
d
se
n
ten
c
e
s,”
J
o
u
rn
a
l
o
f
Co
mp
u
ter
S
c
ien
c
e
(
J
CS
)
,
v
o
l.
1
2
,
n
o
.
1
,
p
p
.
1
-
18
,
2
0
1
6
.
[3
0
]
D.
S
u
leim
a
n
,
A
.
Aw
a
jan
,
N.
A
l
-
M
a
d
i,
“
De
e
p
lea
rn
in
g
b
a
se
d
tec
h
n
iq
u
e
f
o
r
p
lag
iaris
m
d
e
t
e
c
t
i
o
n
i
n
a
r
a
b
i
c
t
e
x
t
s
,
”
2017
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
o
n
N
e
w
T
r
e
n
d
s
i
n
C
o
m
p
u
t
i
n
g
S
c
i
e
n
c
e
s
(
I
C
T
C
S
)
,
A
m
m
a
n
,
J
o
r
d
a
n
,
2
0
1
7
,
p
p
.
2
1
6
-
2
2
2
.
[3
1
]
M
.
A
l
-
S
u
h
a
i
q
i,
M
.
A
.
Ha
z
a
a
,
M
.
A
lb
a
re
d
,
“
A
r
a
b
ic
En
g
li
sh
c
r
o
ss
-
li
n
g
u
a
l
p
lag
iarism
d
e
tec
ti
o
n
b
a
se
d
o
n
k
e
y
p
h
ra
se
s
e
x
trac
ti
o
n
,
m
o
n
o
li
n
g
u
a
l
a
n
d
m
a
c
h
in
e
lea
rn
i
n
g
a
p
p
r
o
a
c
h
,
”
As
ia
n
J
o
u
rn
a
l
o
f
Res
e
a
rc
h
i
n
C
o
m
p
u
ter
S
c
ien
c
e
,
v
o
l.
2
,
n
o
.
3
,
p
p
.
1
-
12
,
2
0
1
8
.
[3
2
]
H.
Ch
e
rro
u
n
,
A
.
A
lsh
e
h
ri
e
t
a
l.
,
“
Disg
u
ise
d
p
lag
iarism
d
e
tec
ti
o
n
in
A
ra
b
ic
tex
t
d
o
c
u
m
e
n
ts,
”
2
0
1
8
2
n
d
In
ter
n
a
ti
o
n
a
l
Co
n
fer
e
n
c
e
o
n
Na
t
u
ra
l
L
a
n
g
u
a
g
e
a
n
d
S
p
e
e
c
h
Pro
c
e
ss
in
g
(
ICNL
S
P
)
,
A
lg
iers
,
A
lg
e
ria,
2
0
1
8
,
p
p
.
1
-
6.
[3
3
]
J.
F
e
rre
ro
,
D.
S
c
h
w
a
b
,
H.
Ch
e
rro
u
n
e
t
a
l
.
,
“
W
o
rd
e
m
b
e
d
d
in
g
-
b
a
se
d
a
p
p
r
o
a
c
h
e
s
f
o
r
m
e
a
su
rin
g
se
m
a
n
ti
c
sim
il
a
rit
y
o
f
A
r
a
b
ic
-
En
g
li
sh
se
n
ten
c
e
s,
”
In
t
e
rn
a
ti
o
n
a
l
Co
n
fer
e
n
c
e
o
n
Ara
b
ic
L
a
n
g
u
a
g
e
Pro
c
e
ss
in
g
(
ICAL
P
2
0
1
7
)
,
v
o
l.
7
8
2
,
2
0
1
7
,
p
p
.
1
9
-
3
3
.
[3
4
]
R.
M
.
Ba
d
ry
,
a
n
d
I
.
F
.
M
o
a
w
a
d
,
“
A
se
m
a
n
ti
c
tex
t
s
u
m
m
a
r
i
z
a
t
i
o
n
m
o
d
e
l
f
o
r
A
r
a
b
i
c
t
o
p
i
c
-
o
r
i
e
n
t
e
d
,
”
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
o
n
A
d
v
a
n
c
e
d
M
a
c
h
i
n
e
L
e
a
r
n
i
n
g
T
e
c
h
n
o
l
o
g
i
e
s
a
n
d
A
p
p
l
i
c
a
t
i
o
n
s
(
A
M
L
T
A
2
0
1
9
)
,
v
o
l
.
9
2
1
,
2
0
1
9
,
p
p
.
5
1
8
-
5
2
8
.
[3
5
]
P
.
S
h
a
p
iro
,
a
n
d
K.
Du
h
,
“
M
o
rp
h
o
lo
g
ica
l
w
o
rd
e
m
b
e
d
d
in
g
s
f
o
r
Ara
b
ic
n
e
u
ra
l
m
a
c
h
in
e
tran
sla
ti
o
n
in
lo
w
-
re
so
u
rc
e
se
tt
in
g
s,
”
Pro
c
e
e
d
in
g
s
o
f
t
h
e
S
e
c
o
n
d
W
o
rk
sh
o
p
o
n
S
u
b
w
o
rd
/Ch
a
ra
c
ter
L
Eve
l
M
o
d
e
ls
,
2
0
1
8
,
p
p
.
1
-
1
1
.
[3
6
]
E.
L
o
p
e
r,
a
n
d
S
.
Bird
,
“
NL
T
K
:
th
e
n
a
tu
ra
l
lan
g
u
a
g
e
to
o
lk
it
,
”
Pro
c
e
e
d
in
g
s
o
f
t
h
e
ACL
In
ter
a
c
ti
v
e
Po
ste
r
a
n
d
De
mo
n
stra
ti
o
n
S
e
ss
io
n
s
,
2
0
0
4
,
p
p
.
2
1
4
-
2
1
7
.
[3
7
]
M
.
A
b
d
u
l
-
M
a
g
e
e
d
,
M
.
T
.
Dia
b
,
M
.
Ko
ra
y
e
m
,
“
S
u
b
jec
ti
v
it
y
a
n
d
se
n
ti
m
e
n
t
a
n
a
l
y
sis
o
f
m
o
d
e
rn
sta
n
d
a
rd
A
ra
b
ic
,
”
Pro
c
e
e
d
in
g
s
o
f
t
h
e
4
9
t
h
A
n
n
u
a
l
M
e
e
ti
n
g
o
f
th
e
Asso
c
ia
ti
o
n
fo
r
Co
mp
u
ta
t
io
n
a
l
L
in
g
u
isti
c
s:
Hu
ma
n
L
a
n
g
u
a
g
e
T
e
c
h
n
o
l
o
g
ies
,
2
0
1
1
,
p
p
.
5
8
7
-
5
9
1
.
[3
8
]
A
.
B.
S
o
li
m
a
n
,
K.
Ei
ss
a
,
S
.
R.
El
-
Be
lt
a
g
y
,
“
A
R
A
V
EC:
A
se
t
o
f
Ara
b
ic
w
o
rd
e
m
b
e
d
d
in
g
m
o
d
e
ls
f
o
r
u
se
in
A
ra
b
ic
NL
P
,
”
Pro
c
e
d
i
a
Co
m
p
u
ter
S
c
ien
c
e
,
v
o
l.
1
1
7
,
p
p
.
2
5
6
-
2
6
5
,
2
0
1
7
.
[3
9
]
W
.
T
.
Yih
,
K.
T
o
u
tan
o
v
a
,
J.
C.
P
latt
,
C.
M
e
e
k
,
“
L
e
a
rn
in
g
d
isc
ri
m
in
a
ti
v
e
p
ro
jec
ti
o
n
s
f
o
r
tex
t
s
i
m
i
l
a
r
i
t
y
m
e
a
s
u
r
e
s
,
”
P
r
o
c
e
e
d
i
n
g
s
o
f
t
h
e
f
i
f
t
e
e
n
t
h
c
o
n
f
e
r
e
n
c
e
o
n
c
o
m
p
u
t
a
t
i
o
n
a
l
n
a
t
u
r
a
l
l
a
n
g
u
a
g
e
l
e
a
r
n
i
n
g
,
P
o
r
t
l
a
n
d
,
U
S
A
,
2
0
1
1
,
p
p
.
2
4
7
-
2
5
6
.
[4
0
]
J.
F
ried
m
a
n
,
T
.
Ha
stie,
R.
T
ib
sh
iran
i
e
t
a
l.
,
“
A
d
d
it
iv
e
lo
g
isti
c
re
g
re
ss
io
n
:
a
sta
ti
stica
l
v
ie
w
o
f
b
o
o
st
in
g
(w
it
h
d
isc
u
ss
io
n
a
n
d
a
re
jo
i
n
d
e
r
b
y
th
e
a
u
th
o
rs),
”
T
h
e
a
n
n
a
ls
o
f
st
a
ti
stics
,
v
o
l.
2
8
,
n
o
.
2
,
p
p
.
3
3
7
-
4
0
7
,
2
0
0
0
.
[4
1
]
J.
H.
F
rie
d
m
a
n
,
“
G
r
e
e
d
y
f
u
n
c
ti
o
n
a
p
p
r
o
x
im
a
ti
o
n
:
a
g
ra
d
ien
t
b
o
o
sti
n
g
m
a
c
h
in
e
,
”
An
n
a
ls
o
f
sta
ti
stics
,
v
o
l.
2
9
,
n
o
.
5
,
p
p
.
1
1
8
9
-
1
2
3
2
,
2
0
0
1
.
[4
2
]
S
.
Ch
o
p
ra
,
R.
Ha
d
se
ll
,
Y.
L
e
Cu
n
e
t
a
l.
,
“
L
e
a
rn
in
g
a
si
m
il
a
rit
y
m
e
tri
c
d
isc
ri
m
in
a
ti
v
e
l
y
,
w
it
h
a
p
p
li
c
a
ti
o
n
to
f
a
c
e
v
e
ri
f
ica
ti
o
n
,
”
CVP
R
,
v
o
l.
1
,
p
p
.
5
3
9
-
5
4
6
,
2
0
0
5
.
[4
3
]
K.
Ch
e
n
,
a
n
d
A
.
S
a
lma
n
,
“
E
x
trac
ti
n
g
sp
e
a
k
e
r
-
sp
e
c
if
i
c
in
f
o
r
m
a
ti
o
n
w
it
h
a
re
g
u
lariz
e
d
S
iam
e
se
d
e
e
p
n
e
tw
o
rk
,
”
Pro
c
e
e
d
in
g
s
o
f
t
h
e
2
4
t
h
In
ter
n
a
ti
o
n
a
l
Co
n
fer
e
n
c
e
o
n
Ne
u
ra
l
In
f
o
r
ma
ti
o
n
Pro
c
e
ss
in
g
S
y
ste
ms
,
2
0
1
1
,
p
p
.
2
9
8
-
3
0
6
.
[4
4
]
S
.
Io
f
f
e
,
a
n
d
C.
S
z
e
g
e
d
y
,
“
Ba
t
c
h
n
o
rm
a
li
z
a
ti
o
n
:
A
c
c
e
l
e
ra
ti
n
g
d
e
e
p
n
e
tw
o
rk
tr
a
in
in
g
b
y
re
d
u
c
in
g
in
tern
a
l
c
o
-
v
a
riate
sh
if
t,
”
Pro
c
e
e
d
in
g
s
o
f
th
e
3
2
n
d
In
ter
n
a
ti
o
n
a
l
Co
n
fer
e
n
c
e
o
n
I
n
ter
n
a
ti
o
n
a
l
Co
n
fer
e
n
c
e
o
n
M
a
c
h
in
e
L
e
a
rn
i
n
g
,
v
o
l.
3
7
,
2
0
1
5
,
p
p
.
4
4
8
-
4
5
6
.
[4
5
]
N.
S
riv
a
sta
v
a
,
G
.
Hin
to
n
,
A
.
Kriz
h
e
v
sk
y
,
I.
S
u
tsk
e
v
e
r,
R.
S
a
lak
h
u
t
d
in
o
v
,
“
Dro
p
o
u
t
:
a
sim
p
le
w
a
y
to
p
re
v
e
n
t
n
e
u
ra
l
n
e
tw
o
rk
s
f
ro
m
o
v
e
r
f
it
ti
n
g
,
”
T
h
e
j
o
u
rn
a
l
o
f
ma
c
h
in
e
le
a
rn
i
n
g
re
se
a
rc
h
,
v
o
l.
15
,
n
o
.
1
,
p
p
.
1
9
2
9
-
1
9
5
8
,
2
0
1
4
.
[4
6
]
A
.
M
.
Da
i,
a
n
d
Q.
V
.
L
e
,
“
S
e
m
i
-
su
p
e
rv
ise
d
s
e
q
u
e
n
c
e
lea
rn
in
g
,
”
Ad
v
a
n
c
e
s
in
n
e
u
r
a
l
in
fo
rm
a
t
i
o
n
p
r
o
c
e
ss
in
g
sy
ste
ms
,
2
0
1
5
,
p
p
.
3
0
7
9
-
3
0
8
7
.
[4
7
]
J.
Ho
w
a
rd
,
a
n
d
S
.
Ru
d
e
r,
“
Un
iv
e
rsa
l
lan
g
u
a
g
e
m
o
d
e
l
f
in
e
-
tu
n
in
g
f
o
r
tex
t
c
las
si
f
ica
ti
o
n
,
”
Pro
c
e
e
d
i
n
g
s
o
f
th
e
5
6
t
h
An
n
u
a
l
M
e
e
ti
n
g
o
f
t
h
e
Asso
c
ia
t
io
n
fo
r C
o
mp
u
ta
ti
o
n
a
l
L
i
n
g
u
isti
c
s,
M
e
lb
o
u
m
e
,
A
u
stra
li
a
,
v
o
l.
1
,
2
0
1
8
,
p
p
.
3
2
8
-
3
3
9
.
[4
8
]
M
.
E.
P
e
ters
,
M
.
Ne
u
m
a
n
n
,
M
.
Iy
y
e
r,
M
.
G
a
rd
n
e
r,
C.
Clark
,
K.
Lee
e
t
a
l.
,
“
De
e
p
c
o
n
tex
tu
a
li
z
e
d
w
o
rd
re
p
re
se
n
tatio
n
s,
”
Pr
o
c
.
o
f
NAA
CL
,
2
0
1
8
.
[4
9
]
A
.
V
a
s
w
a
n
i,
N.
S
h
a
z
e
e
r,
N.
P
a
r
m
a
r,
J.
Us
z
k
o
re
it
,
L
.
Jo
n
e
s,
A
.
N
.
G
o
m
e
z
,
Ka
is
e
r,
I.
P
o
l
o
su
k
h
i
n
,
“
A
tt
e
n
ti
o
n
is
a
ll
y
o
u
n
e
e
d
,
”
Ad
v
a
n
c
e
s in
n
e
u
ra
l
in
f
o
rm
a
ti
o
n
p
ro
c
e
ss
in
g
sy
ste
ms
,
2
0
1
7
,
p
p
.
5
9
9
8
-
6
0
0
8
.
[5
0
]
D.
Ba
h
d
a
n
a
u
,
K.
Ch
o
,
Y.
Be
n
g
io
,
“
Ne
u
ra
l
m
a
c
h
in
e
tran
sla
ti
o
n
b
y
jo
in
tl
y
lea
rn
in
g
to
a
li
g
n
a
n
d
tran
sla
te,
”
c
o
n
fer
e
n
c
e
p
a
p
e
r a
t
IC
L
R
,
2
0
1
5
,
p
p
.
1
-
15
.
Evaluation Warning : The document was created with Spire.PDF for Python.