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.
15
,
No
.
4
,
A
u
g
u
s
t 2
0
2
6
,
p
p
.
3
6
9
2
~
3
7
0
2
I
SS
N:
2
2
5
2
-
8
9
3
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijai.v
15
.i
4
.
p
p
3
6
9
2
-
3
7
0
2
3692
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
i
.
ia
esco
r
e.
co
m
Text
emo
tion re
c
o
g
nition ba
sed o
n deep
l
ea
rning
an
d atten
tion
with
wha
le optimi
za
tion a
lg
o
rithm
T
a
ba
re
k
Alwa
n T
uib
1,
2
,
M
o
ha
m
m
a
d
-
Rez
a
F
eizi
-
Der
a
k
hs
hi
1
,
Ya
qd
ha
n M
a
hm
o
o
d H
us
s
ein
2
,3
,
F
a
ha
d T
a
ha
A
l
-
Dhief
4
1
C
o
m
p
u
t
e
r
i
z
e
d
I
n
t
e
l
l
i
g
e
n
c
e
S
y
s
t
e
ms
La
b
o
r
a
t
o
r
y
,
D
e
p
a
r
t
m
e
n
t
o
f
C
o
m
p
u
t
e
r
En
g
i
n
e
e
r
i
n
g
,
F
a
c
u
l
t
y
o
f
E
l
e
c
t
r
i
c
a
l
a
n
d
C
o
mp
u
t
e
r
E
n
g
i
n
e
e
r
i
n
g
,
U
n
i
v
e
r
si
t
y
o
f
Ta
b
r
i
z
,
Ta
b
r
i
z
,
I
r
a
n
2
C
o
l
l
e
g
e
o
f
E
n
g
i
n
e
e
r
i
n
g
El
e
c
t
r
o
n
i
c
s a
n
d
C
o
mm
u
n
i
c
a
t
i
o
n
En
g
i
n
e
e
r
i
n
g
,
A
l
-
M
u
t
h
a
n
n
a
U
n
i
v
e
r
si
t
y
,
S
a
maw
a
h
,
I
r
a
q
3
C
o
l
l
e
g
e
o
f
E
n
g
i
n
e
e
r
i
n
g
,
A
l
A
y
e
n
U
n
i
v
e
r
si
t
y
,
N
a
si
r
i
y
a
h
,
I
r
a
q
4
C
e
n
t
e
r
f
o
r
S
o
f
t
w
a
r
e
T
e
c
h
n
o
l
o
g
y
a
n
d
M
a
n
a
g
e
m
e
n
t
,
F
a
c
u
l
t
y
o
f
I
n
f
o
r
ma
t
i
o
n
S
c
i
e
n
c
e
a
n
d
T
e
c
h
n
o
l
o
g
y
,
U
n
i
v
e
r
s
i
t
i
K
e
b
a
n
g
sa
a
n
M
a
l
a
y
s
i
a
,
S
e
l
a
n
g
o
r
,
M
a
l
a
y
s
i
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Dec
2
7
,
2
0
2
4
R
ev
is
ed
Ap
r
2
6
,
2
0
2
6
Acc
ep
ted
J
u
l 9
,
2
0
2
6
Tex
t
e
m
o
ti
o
n
re
c
o
g
n
it
io
n
(T
ER)
r
e
p
re
se
n
ts
o
n
e
o
f
t
h
e
c
ru
c
ial
tas
k
s
in
n
a
tu
ra
l
lan
g
u
a
g
e
p
r
o
c
e
ss
in
g
(NL
P
),
a
n
d
i
t
is h
ig
h
ly
imp
o
rtan
t
fo
r
m
a
n
y
a
p
p
li
c
a
ti
o
n
s.
To
imp
ro
v
e
th
e
p
e
rfo
rm
a
n
c
e
o
f
TE
R,
a
d
v
a
n
c
e
d
d
e
e
p
lea
rn
in
g
(DL)
m
o
d
e
l
G
RU
-
Bi
G
R
U
-
CNN
-
Att
is
p
re
se
n
ted
in
t
h
is
re
se
a
rc
h
.
It
is
a
c
o
m
b
in
a
ti
o
n
o
f
g
a
ted
re
c
u
rre
n
t
u
n
it
s
(
G
RU),
b
id
irec
ti
o
n
a
l
g
a
ted
re
c
u
rre
n
t
u
n
it
s
(
BiG
RU
)
with
a
c
o
n
v
o
l
u
ti
o
n
a
l
n
e
u
ra
l
n
e
t
wo
rk
(CNN
),
a
n
d
a
n
a
tt
e
n
ti
o
n
m
e
c
h
a
n
ism
.
Us
in
g
a
G
RU,
BiG
RU
,
a
n
d
CNN
a
s
p
a
rt
o
f
a
d
e
e
p
fe
a
tu
re
e
x
trac
ti
o
n
m
o
d
e
l,
c
o
m
b
in
e
d
wit
h
a
n
a
tt
e
n
ti
o
n
m
e
c
h
a
n
ism
th
a
t
g
i
v
e
s
imp
o
rta
n
t
i
n
fo
rm
a
ti
o
n
d
iffere
n
t
we
ig
h
ts,
t
h
e
m
e
th
o
d
t
h
a
t
is
p
r
o
p
o
se
d
i
n
th
e
p
re
se
n
t
wo
r
k
e
n
h
a
n
c
e
s
th
e
q
u
a
li
ty
o
f
t
h
e
wo
rd
v
e
c
to
rs
a
n
d
lea
d
s
to
in
c
re
a
sin
g
se
n
ti
m
e
n
t
a
n
a
ly
sis
ju
d
g
m
e
n
t
a
c
c
u
ra
c
y
.
Th
e
wh
a
le
o
p
ti
m
iza
ti
o
n
a
lg
o
rit
h
m
(W
OA
)
,
wh
ich
se
lec
ts
th
e
m
o
st
in
f
o
rm
a
ti
v
e
fe
a
tu
re
s
fo
r
f
u
rth
e
r
e
n
h
a
n
c
i
n
g
t
h
e
m
o
d
e
l,
is
u
se
d
las
tl
y
f
o
r
th
e
o
p
ti
m
iza
ti
o
n
o
f
t
h
e
fe
a
tu
re
se
lec
ti
o
n
p
r
o
c
e
ss
.
Afte
r
su
c
h
o
p
ti
m
iza
ti
o
n
,
t
h
o
se
c
h
o
se
n
fe
a
t
u
r
e
s
a
re
train
e
d
o
n
a
m
u
lt
i
-
lay
e
r
p
e
rc
e
p
tro
n
(M
LP
),
s
u
c
c
e
ss
fu
ll
y
c
o
m
b
in
i
n
g
m
a
c
h
in
e
lea
rn
in
g
(M
L)
a
n
d
DL
a
p
p
ro
a
c
h
e
s
fo
r
imp
ro
v
i
n
g
TER
.
A
v
a
ried
c
o
r
p
u
s
o
f
twe
e
ts,
se
n
ten
c
e
s
,
a
n
d
d
ial
o
g
u
e
s
h
a
d
b
e
e
n
u
se
d
i
n
th
e
p
re
se
n
t
wo
rk
f
o
r
th
e
a
ss
e
ss
m
e
n
t
o
f
t
h
e
p
e
rfo
rm
a
n
c
e
o
f
th
e
su
g
g
e
ste
d
m
o
d
e
l.
T
h
e
p
r
o
p
o
se
d
m
e
th
o
d
a
c
h
iev
e
s
8
3
.
7
6
%
a
c
c
u
ra
c
y
in
e
m
o
ti
o
n
re
c
o
g
n
it
io
n
.
K
ey
w
o
r
d
s
:
B
i
d
i
r
e
ct
i
o
n
a
l
g
at
e
d
r
e
c
u
r
r
e
n
t
u
n
i
t
C
o
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
Gate
d
r
ec
u
r
r
e
n
t u
n
it
T
ex
t e
m
o
tio
n
r
ec
o
g
n
itio
n
W
h
ale
o
p
tim
izatio
n
alg
o
r
ith
m
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
:
Fah
ad
T
ah
a
AL
-
Dh
ie
f
C
en
ter
f
o
r
So
f
twar
e
T
ec
h
n
o
l
o
g
y
an
d
Ma
n
ag
em
e
n
t,
Facu
lty
o
f
I
n
f
o
r
m
atio
n
Scien
ce
an
d
T
e
ch
n
o
lo
g
y
Un
iv
er
s
iti Ke
b
an
g
s
aa
n
Ma
lay
s
ia
B
an
g
i 4
3
6
0
0
,
Selan
g
o
r
,
Ma
lay
s
ia
E
m
ail:
f
ah
ad
@
u
k
m
.
ed
u
.
m
y
1.
I
NT
RO
D
UCT
I
O
N
T
ex
t
em
o
tio
n
r
ec
o
g
n
itio
n
(
T
E
R
)
is
an
es
s
en
tial
an
d
r
ap
id
ly
d
ev
elo
p
in
g
ac
tiv
ity
in
th
e
s
p
h
er
e
o
f
n
atu
r
al
lan
g
u
ag
e
p
r
o
ce
s
s
in
g
(
NL
P)
.
I
ts
s
ig
n
if
ican
ce
ca
n
b
e
s
ee
n
in
its
wid
e
r
an
g
e
o
f
ap
p
l
icatio
n
s
in
d
if
f
er
e
n
t
ap
p
licatio
n
s
,
lik
e
s
o
cial
n
etw
o
r
k
a
n
aly
s
is
,
wh
er
e
it
ca
n
b
e
u
s
ed
t
o
g
a
u
g
e
p
u
b
lic
s
en
tim
en
t
f
r
o
m
e
x
ten
s
iv
e
o
n
lin
e
d
is
co
u
r
s
e;
m
en
tal
h
e
alth
ass
ess
m
en
t
,
d
eliv
er
in
g
i
n
s
ig
h
ts
in
to
em
o
tio
n
al
s
tates
b
ased
o
n
tex
tu
al
an
aly
s
is
;
an
d
p
u
b
lic
o
p
in
io
n
m
o
n
ito
r
in
g
,
wh
er
e
p
o
licy
m
a
k
er
s
an
d
b
u
s
in
ess
es
ca
n
b
e
g
i
v
en
th
e
n
ec
ess
ar
y
f
ee
d
b
ac
k
[
1
]
‒
[
3
]
.
R
eg
ar
d
less
o
f
its
im
p
o
r
ta
n
ce
,
t
h
e
in
ter
p
r
etatio
n
o
f
e
m
o
tio
n
s
i
n
a
t
ex
t
is
o
n
e
o
f
th
e
ch
allen
g
in
g
task
s
,
b
ec
a
u
s
e
o
f
th
e
ch
allen
g
in
g
n
atu
r
e
o
f
th
e
f
r
am
ewo
r
k
b
eh
i
n
d
h
u
m
an
lan
g
u
ag
e
an
d
ex
p
r
ess
io
n
.
T
h
e
r
ec
en
t
f
ast
an
d
u
n
ce
asin
g
d
ev
elo
p
m
en
t
o
f
d
ig
ital
co
m
m
u
n
icatio
n
tech
n
o
lo
g
y
h
as
led
to
th
e
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
Text
emo
tio
n
r
ec
o
g
n
itio
n
b
a
s
e
d
o
n
d
ee
p
lea
r
n
in
g
a
n
d
a
tten
ti
o
n
w
ith
w
h
a
le
…
(
Ta
b
a
r
ek
A
l
w
a
n
Tu
ib
)
3693
r
is
e
in
in
ter
est
in
th
e
T
E
R
s
u
b
ject.
T
h
e
T
E
R
will
b
e
ab
le
t
o
an
aly
ze
th
e
em
o
tio
n
al
co
n
te
n
t
o
f
th
e
d
o
c
u
m
en
t,
g
iv
in
g
it
an
ad
v
a
n
ce
d
i
n
s
ig
h
t
in
to
its
s
en
tim
en
t,
in
ten
t,
an
d
ex
p
r
ess
iv
e
s
u
b
tleties
.
T
h
e
t
ec
h
n
o
lo
g
y
h
as
b
ee
n
ap
p
lied
in
m
a
n
y
s
ec
to
r
s
s
u
ch
a
s
ed
u
ca
tio
n
,
co
m
m
u
n
icatio
n
s
,
an
d
b
u
s
in
ess
,
am
o
n
g
o
t
h
er
s
[
4
]
.
Hu
m
an
em
o
tio
n
s
ca
n
b
e
ca
teg
o
r
ized
[
5
]
t
h
e
em
o
tio
n
s
o
f
h
u
m
an
s
in
to
s
ix
n
am
ely
jo
y
,
s
u
r
p
r
is
e,
s
ad
n
ess
,
d
is
g
u
s
t,
f
ea
r
,
an
d
an
g
er
.
T
h
er
e
ar
e
m
an
y
co
m
p
licated
f
o
r
m
s
o
f
em
o
tio
n
s
s
u
ch
as
o
p
tim
is
m
an
d
lo
v
e.
T
h
e
d
im
en
s
io
n
al
ap
p
r
o
ac
h
was
in
s
p
ir
ed
b
y
th
e
th
e
o
r
y
d
escr
ib
ed
in
[
6
]
,
wh
er
e
f
ee
lin
g
s
(
n
o
t
to
b
e
co
n
f
u
s
e
d
with
em
o
tio
n
s
)
co
u
ld
b
e
d
e
s
cr
ib
ed
in
th
r
ee
d
im
en
s
io
n
s
:
ex
citem
en
t
-
in
h
ib
itio
n
,
ten
s
io
n
-
r
elax
atio
n
,
an
d
p
leasan
tn
ess
-
u
n
p
leasan
tn
ess
.
A
p
r
o
m
in
en
t
ex
am
p
le
is
Plu
t
ch
ik
’
s
wh
ee
l
o
f
em
o
tio
n
s
[
7
]
.
Plu
tch
ik
b
eliev
ed
th
at
eig
h
t
f
u
n
d
am
en
tal
em
o
ti
o
n
s
ar
e
p
r
esen
t
in
ev
e
r
y
o
n
e
a
n
d
m
o
d
eled
th
em
in
p
air
s
an
d
p
lace
d
th
em
o
n
a
wh
ee
l,
ea
ch
o
p
p
o
s
ite
to
th
e
o
th
er
:
an
ticip
atio
n
an
d
s
u
r
p
r
is
e,
an
g
er
an
d
f
ea
r
,
s
ad
n
ess
an
d
jo
y
,
an
d
d
is
g
u
s
t
an
d
tr
u
s
t.
T
h
e
eig
h
t
em
o
tio
n
s
ar
e
m
eth
o
d
ically
o
r
g
a
n
ized
d
ep
e
n
d
in
g
o
n
th
eir
p
h
y
s
io
lo
g
ical
f
u
n
ctio
n
s
.
An
ex
am
p
le
o
f
a
co
n
ce
p
tu
al
an
al
o
g
y
o
f
th
e
wh
ee
l
o
f
em
o
tio
n
s
b
y
Plu
tc
h
ik
is
a
co
n
e
th
at
u
n
f
o
ld
s
to
cr
ea
te
a
2
D
wh
ee
l.
Acc
o
r
d
in
g
to
Plu
tch
ik
'
s
th
eo
r
y
,
th
o
s
e
b
asic
em
o
tio
n
s
m
ay
b
ec
o
m
e
s
tr
o
n
g
er
,
wea
k
er
,
o
r
co
m
b
in
ed
to
f
o
r
m
a
b
r
o
ad
s
p
ec
tr
u
m
o
f
em
o
tio
n
s
.
T
h
e
s
ep
ar
atio
n
b
etwe
en
em
o
tio
n
s
,
s
p
ec
if
ically
th
e
s
ec
o
n
d
ar
y
,
p
r
im
ar
y
,
a
n
d
ter
tiar
y
d
y
a
d
s
in
d
icate
d
o
n
th
e
wh
ee
l o
f
em
o
tio
n
s
,
is
r
elativ
e
to
ea
ch
o
t
h
er
.
‒
Prim
ar
y
d
y
a
d
s
:
co
m
b
in
e
a
d
jac
en
t e
m
o
tio
n
s
(
e.
g
.
,
jo
y
an
d
tr
u
s
t)
.
‒
S
e
c
o
n
d
a
r
y
d
y
ad
s
:
co
m
b
i
n
e
e
m
o
t
io
n
s
w
i
th
o
n
e
in
t
e
r
v
e
n
in
g
e
m
o
t
i
o
n
(
e
.
g
.
jo
y
an
d
f
ea
r
,
w
i
th
tr
u
s
t
i
n
b
e
t
w
e
e
n
)
.
‒
T
er
tiar
y
d
y
ad
s
:
co
m
b
in
e
em
o
tio
n
s
s
ep
ar
ated
b
y
two
o
th
er
em
o
tio
n
s
(
e.
g
.
,
jo
y
an
d
s
u
r
p
r
i
s
e,
with
tr
u
s
t
an
d
f
ea
r
in
b
etwe
en
)
.
Su
ch
d
y
a
d
s
ex
p
lain
th
e
c
o
m
p
lex
in
ter
r
elatio
n
s
b
etwe
en
e
m
o
tio
n
s
,
an
d
th
is
g
i
v
es
an
o
r
g
an
ized
ap
p
r
o
ac
h
to
lear
n
in
g
th
e
s
y
s
t
em
atic
in
ter
co
n
n
ec
tio
n
s
b
etw
ee
n
th
em
o
n
th
e
wh
ee
l.
T
h
is
m
o
d
el
p
r
o
v
id
es
a
s
u
b
tle
ex
p
lan
atio
n
o
f
th
e
in
t
er
ac
tio
n
an
d
tr
a
n
s
f
o
r
m
atio
n
o
f
em
o
tio
n
s
in
th
e
h
u
m
an
e
m
o
tio
n
al
s
p
ec
tr
u
m
.
Alth
o
u
g
h
em
o
tio
n
s
ar
e
u
s
u
ally
co
n
v
ey
e
d
th
r
o
u
g
h
f
ac
ial
ex
p
r
ess
io
n
,
g
estu
r
es,
s
p
ee
ch
,
an
d
tex
t,
th
e
latter
d
o
es
n
o
t
p
o
s
s
ess
th
e
ex
p
r
ess
iv
e
elem
en
ts
in
h
er
en
t
in
s
p
o
k
en
la
n
g
u
ag
e
a
n
d
n
o
n
-
v
er
b
al
c
o
m
m
u
n
icatio
n
[
8
]
.
T
h
is
f
ac
t
co
m
p
licates
tex
t
-
b
ased
e
m
o
tio
n
r
ec
o
g
n
itio
n
an
d
m
ak
e
s
it
esp
ec
ially
d
if
f
icu
lt.
T
ex
t,
b
y
d
e
f
in
itio
n
,
d
o
es
n
o
t
h
av
e
an
y
f
ac
ial
o
r
to
n
al
in
d
icato
r
s
(
tast
eless
)
,
wh
ich
m
ak
es
th
e
s
itu
atio
n
o
f
id
en
tify
in
g
th
e
em
o
tio
n
a
l
s
tates
m
o
r
e
d
if
f
ic
u
lt.
B
esid
es
th
at,
th
e
co
m
p
lex
ity
an
d
a
m
b
ig
u
ity
o
f
th
e
i
n
h
er
e
n
t
tex
t
u
al
co
m
m
u
n
icatio
n
co
n
tr
ib
u
te
to
th
e
au
g
m
en
tatio
n
o
f
th
ese
is
s
u
es,
b
ec
au
s
e
t
h
e
wo
r
d
s
m
ig
h
t
h
av
e
a
wid
e
v
a
r
iety
o
f
m
ea
n
in
g
s
an
d
m
o
r
p
h
o
lo
g
ical
f
o
r
m
s
,
wh
ich
will
lead
to
co
m
p
licatin
g
th
e
p
r
o
ce
s
s
o
f
r
ec
o
g
n
izin
g
em
o
ti
o
n
s
in
wr
itten
tex
t.
T
h
e
T
E
R
r
elies
o
n
t
h
e
u
s
e
o
f
NL
P.
NL
P
is
a
d
y
n
am
ic
f
ield
o
f
s
tu
d
y
th
at
is
u
r
g
ed
b
y
th
e
in
cr
ea
s
in
g
am
o
u
n
t
o
f
o
n
lin
e
tex
tu
al
m
ater
ials
.
T
h
e
p
r
ep
ar
atio
n
o
f
th
e
tex
t
is
th
e
f
ir
s
t
s
tag
e
o
f
th
e
T
E
R
,
it
in
v
o
lv
es
wo
r
d
b
r
ea
k
in
g
,
p
ar
t
o
f
s
p
ee
ch
tag
g
in
g
,
m
ea
n
i
n
g
as
well
as
th
e
s
tr
u
ctu
r
e
o
f
th
e
s
en
ten
ce
s
.
T
h
e
p
r
ep
r
o
ce
s
s
in
g
will
b
e
f
o
llo
wed
b
y
th
e
f
ea
tu
r
e
ex
tr
ac
ti
o
n
s
tag
e
,
wh
er
eb
y
th
e
wo
r
d
v
ec
to
r
s
ar
e
q
u
an
tifie
d
,
f
o
r
m
atted
f
o
r
co
m
p
u
tatio
n
al
p
r
o
ce
s
s
in
g
,
a
n
d
f
ed
in
to
th
e
c
lass
if
icatio
n
s
y
s
tem
s
aim
in
g
a
t
id
en
tify
in
g
t
h
e
em
o
tio
n
al
s
tates
[
4
]
,
[
9
]
.
I
n
t
h
e
ca
s
e
o
f
f
ea
tu
r
e
ex
tr
ac
tio
n
,
s
o
m
e
co
m
m
o
n
al
g
o
r
ith
m
s
ar
e
a
p
p
lied
,
s
u
ch
as
o
n
e
-
h
o
t
en
co
d
in
g
wh
er
e
a
b
in
ar
y
r
ep
r
esen
tatio
n
o
f
th
e
ty
p
es
o
f
t
ex
ts
is
o
b
tain
e
d
[
4
]
,
[
9
]
,
a
n
d
t
h
e
ap
p
r
o
ac
h
o
f
te
r
m
f
r
eq
u
en
c
y
-
in
v
er
s
e
d
o
c
u
m
en
t
f
r
eq
u
e
n
cy
(
T
F
-
I
DF)
,
wh
e
r
e
t
h
e
wo
r
d
r
ele
v
an
ce
is
estab
lis
h
ed
b
ased
o
n
its
f
r
e
q
u
en
c
y
t
o
en
d
o
f
h
a
v
in
g
a
co
n
v
en
ie
n
t a
n
d
s
im
p
lis
tic
tex
tu
al
r
ep
r
esen
tatio
n
[
1
0
]
.
Desig
n
in
g
r
eliab
le
class
if
icat
io
n
alg
o
r
ith
m
s
an
d
ef
f
icien
tly
s
elec
tin
g
an
d
weig
h
tin
g
te
x
tu
al
d
at
a
p
r
esen
t
ch
allen
g
es
th
at
ex
ten
d
b
ey
o
n
d
j
u
s
t
th
e
tex
tu
al
f
ea
tu
r
e
ex
tr
ac
tio
n
.
T
h
e
p
r
ev
i
o
u
s
m
o
d
els
in
th
e
ar
ea
o
f
TER
h
ad
f
ac
e
d
a
wid
e
r
an
g
e
o
f
lim
itatio
n
s
.
Usu
ally
,
th
ey
lac
k
ed
a
c
o
m
p
r
eh
en
s
iv
e
ta
x
o
n
o
m
y
o
f
em
o
tio
n
s
an
d
h
av
e
b
ee
n
b
u
ilt o
n
lex
ico
n
s
with
in
s
u
f
f
icien
t k
in
d
s
o
f
v
o
ca
b
u
lar
y
.
Ma
n
y
d
is
r
eg
ar
d
ed
s
em
an
tic
co
n
tex
ts
,
f
ailed
to
ex
tr
ac
t
co
n
tex
tu
al
in
f
o
r
m
a
tio
n
ef
f
ec
tiv
ely
f
r
o
m
s
en
ten
c
es,
an
d
p
er
f
o
r
m
ed
p
o
o
r
ly
i
n
d
etec
tin
g
s
p
ec
if
ic
em
o
tio
n
s
.
Ad
d
itio
n
ally
,
th
ese
m
o
d
els
h
ad
tr
o
u
b
le
p
u
llin
g
o
u
t
im
p
o
r
tan
t
m
ea
n
in
g
f
r
o
m
th
e
tex
t,
wer
e
s
lo
w
to
p
r
o
ce
s
s
in
f
o
r
m
atio
n
,
m
is
s
ed
co
n
n
ec
tio
n
s
b
etwe
en
d
i
f
f
er
en
t
f
ea
tu
r
es,
h
ad
to
o
litt
le
d
ata
,
an
d
o
f
ten
m
ad
e
m
is
tak
es
in
clas
s
if
y
in
g
em
o
tio
n
s
.
Han
d
lin
g
o
f
co
m
m
o
n
l
y
ap
p
ea
r
in
g
em
o
jis
,
p
o
o
r
ex
tr
ac
tio
n
o
f
s
em
an
tic
in
f
o
r
m
atio
n
,
an
d
f
ailu
r
e
to
ad
ap
t
to
d
if
f
er
en
t
s
en
ten
ce
s
tr
u
c
tu
r
es
wer
e
also
o
th
er
m
ajo
r
co
n
ce
r
n
s
.
Alth
o
u
g
h
s
o
m
e
o
f
th
ese
g
ap
s
h
av
e
b
ee
n
ad
d
r
ess
ed
in
p
r
ev
io
u
s
s
tu
d
ies,
c
u
r
r
en
t
m
eth
o
d
o
lo
g
ies
o
f
ten
f
ac
e
ch
allen
g
es
in
co
llectin
g
co
m
p
lex
em
o
tio
n
a
l
p
atter
n
s
as
well
as
im
p
r
o
v
in
g
f
ea
tu
r
e
r
ep
r
esen
tatio
n
ef
f
ec
tiv
ely
an
d
lim
it
p
er
f
o
r
m
an
ce
.
I
n
th
is
p
ap
e
r
,
a
m
o
r
e
ad
v
a
n
ce
d
d
ee
p
lear
n
in
g
(
DL
)
ar
ch
itectu
r
e
,
GR
U
-
B
iGR
U
-
C
N
N
-
Att
is
p
r
o
p
o
s
ed
to
h
av
e
a
s
ig
n
if
ica
n
t
ef
f
ec
t
o
n
th
e
ac
cu
r
ac
y
an
d
s
tab
ilit
y
o
f
T
E
R
.
T
h
e
c
o
m
b
in
atio
n
o
f
g
ated
r
ec
u
r
r
en
t
u
n
its
(
GR
U)
,
b
id
ir
e
ctio
n
al
g
ated
r
ec
u
r
r
en
t
u
n
its
(
B
iG
R
U
)
,
a
co
n
v
o
lu
ti
o
n
al
n
eu
r
al
n
etwo
r
k
(
C
NN)
,
an
d
an
atten
tio
n
m
ec
h
a
n
is
m
(
AM
)
.
T
h
is
r
esear
ch
o
f
f
er
s
s
ev
er
al
k
ey
co
n
tr
ib
u
tio
n
s
.
First,
p
r
o
p
o
s
e
a
h
y
b
r
i
d
DL
ar
ch
itectu
r
e,
GR
U
-
B
i
GR
U
-
C
NN
-
Att,
wh
ic
h
in
teg
r
ates
GR
U
an
d
B
iGR
U
with
a
C
NN
to
ex
tr
ac
t
d
ee
p
f
ea
tu
r
es
f
r
o
m
tex
t
ef
f
ec
tiv
ely
.
An
atten
tio
n
m
ec
h
an
is
m
is
in
co
r
p
o
r
ated
,
allo
win
g
th
e
m
o
d
el
to
ass
ig
n
v
ar
y
in
g
weig
h
ts
to
cr
u
cial
tex
tu
al
in
f
o
r
m
atio
n
,
t
h
er
eb
y
e
n
h
an
cin
g
wo
r
d
v
ec
to
r
q
u
ality
an
d
s
en
tim
en
t
an
al
y
s
is
ac
cu
r
a
cy
.
Seco
n
d
,
th
e
u
s
e
o
f
wh
ale
o
p
tim
izatio
n
alg
o
r
ith
m
(
W
OA)
f
o
r
f
ea
tu
r
e
o
p
tim
iz
atio
n
m
ar
k
s
a
f
ir
s
t
in
th
is
co
n
tex
t,
id
en
tify
in
g
k
ey
f
ea
tu
r
es
to
im
p
r
o
v
e
m
o
d
el
p
r
ed
ictio
n
s
.
T
h
ir
d
,
t
h
e
s
tr
en
g
th
s
o
f
th
e
c
o
n
v
e
n
tio
n
al
m
ac
h
in
e
lear
n
in
g
(
ML
)
an
d
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.
15
,
No
.
4
,
Au
g
u
s
t 2
0
2
6
:
3
6
9
2
-
3
7
0
2
3694
DL
m
eth
o
d
s
ar
e
t
h
en
m
e
r
g
ed
th
r
o
u
g
h
tr
ain
i
n
g
o
p
tim
ized
f
ea
tu
r
es
with
(
ML
P)
to
ac
h
ie
v
e
b
etter
r
esu
lts
i
n
ter
m
s
o
f
em
o
tio
n
al
r
ec
o
g
n
iti
o
n
in
tex
t.
Fo
u
r
t
h
,
t
h
e
s
u
g
g
ested
m
o
d
el
wo
r
k
s
well
wh
en
it
is
a
p
p
lied
to
tex
t
-
b
ased
em
o
tio
n
r
ec
o
g
n
itio
n
,
th
o
u
g
h
th
e
r
e
is
n
o
v
o
ca
l
to
n
e
an
d
f
ac
e
e
x
p
r
ess
io
n
s
.
T
h
e
m
o
d
el
is
a
p
p
licab
le
to
u
s
e
o
n
d
if
f
er
en
t
p
latf
o
r
m
s
b
ec
au
s
e
th
r
ee
d
if
f
er
en
t
d
ata
s
ets
(
s
im
p
le
s
en
ten
ce
s
,
d
ialo
g
u
es
,
an
d
T
wee
ts
)
en
h
an
ce
r
o
b
u
s
tn
ess
an
d
g
en
e
r
aliza
b
ilit
y
o
v
er
o
th
e
r
p
r
ev
io
u
s
r
esear
ch
s
tu
d
ies
th
at
u
s
ed
a
s
in
g
le
d
ataset
o
n
ly
.
T
h
is
ad
ap
tab
ilit
y
is
u
s
ef
u
l
in
im
p
lem
en
tatio
n
lik
e
th
e
an
a
ly
s
is
o
f
cu
s
to
m
er
r
e
v
iews
an
d
en
h
a
n
cin
g
s
o
cia
l
m
ed
ia
u
s
er
s
ec
u
r
ity
.
T
h
e
r
est
o
f
th
is
r
esear
ch
p
a
p
er
is
s
tr
u
ctu
r
ed
as
f
o
llo
ws:
s
ec
tio
n
2
is
th
e
r
ev
iew
o
f
th
e
r
elate
d
liter
atu
r
e
ab
o
u
t
e
m
o
tio
n
d
etec
tio
n
.
Sectio
n
3
d
escr
ib
es
th
e
p
r
o
p
o
s
ed
m
eth
o
d
an
d
th
e
co
m
p
u
tatio
n
al
m
eth
o
d
s
to
o
v
er
c
o
m
e
th
e
p
r
ev
io
u
s
lim
itatio
n
s
.
Sectio
n
4
d
is
cu
s
s
es
an
d
p
r
esen
ts
th
e
r
esu
lts
o
f
th
e
ex
p
er
im
en
t
i
n
all
o
f
th
e
d
atasets
.
L
astl
y
,
s
ec
ti
o
n
5
also
p
r
o
v
id
es
th
e
co
n
c
lu
s
io
n
o
f
th
e
r
esear
ch
an
d
r
ec
o
m
m
en
d
s
f
u
tu
r
e
r
esear
ch
d
ir
ec
tio
n
s
.
2.
L
I
T
E
R
AT
U
RE
R
E
VI
E
W
T
E
R
h
as
u
tili
ze
d
d
if
f
er
en
t
m
eth
o
d
s
with
th
e
aim
o
f
im
p
r
o
v
in
g
th
e
em
o
tio
n
r
ec
o
g
n
itio
n
p
r
o
ce
s
s
o
f
tex
tu
al
in
f
o
r
m
atio
n
[
4
]
.
On
e
o
f
th
e
k
ey
g
o
als
o
f
s
u
ch
s
tu
d
ie
s
is
to
f
in
d
m
o
d
els
th
at
b
est
id
en
tify
em
o
tio
n
s
.
A
k
ey
wo
r
d
-
b
ased
ap
p
r
o
ac
h
was
ex
p
lo
r
e
d
in
[
1
1
]
u
s
in
g
th
e
I
SEAR
d
ataset
[
1
2
]
an
d
f
o
c
u
s
ed
o
n
th
e
p
h
r
asal
v
er
b
s
.
T
h
e
y
cr
ea
te
d
a
s
p
ec
if
i
c
d
atab
ase
f
o
llo
win
g
p
r
ep
r
o
c
ess
in
g
th
e
d
ata
an
d
f
i
n
d
in
g
m
an
y
p
h
r
asal
v
e
r
b
s
lin
k
ed
to
em
o
tio
n
s
th
at
h
a
d
n
o
t
b
ee
n
p
r
ev
io
u
s
ly
in
d
ex
e
d
.
T
h
ey
ac
h
iev
e
d
a
n
o
tewo
r
th
y
6
5
%
ac
cu
r
ac
y
r
ate
b
y
u
s
in
g
th
is
m
eth
o
d
to
class
if
y
p
h
r
asal
v
er
b
s
an
d
s
y
n
o
n
y
m
o
u
s
ter
m
s
b
y
e
m
o
tio
n
.
Nev
er
th
eless
,
th
eir
ap
p
r
o
ac
h
was
co
n
s
tr
ain
ed
by
a
d
ea
r
th
o
f
em
o
tio
n
-
s
p
ec
if
ic
k
ey
wo
r
d
s
an
d
a
d
is
r
eg
ar
d
f
o
r
th
e
s
em
an
tic
d
e
p
th
o
f
wo
r
d
m
ea
n
in
g
s
.
A
lear
n
in
g
-
b
ased
m
eth
o
d
was
d
ev
elo
p
ed
i
n
[
1
3
]
u
s
in
g
th
e
I
SEAR
d
ata
s
et
[
1
2
]
to
cr
ea
te
a
lear
n
in
g
-
b
ased
m
eth
o
d
.
Af
ter
co
m
p
ar
in
g
s
ev
er
al
class
if
ier
s
,
th
ey
co
n
clu
d
ed
th
at
lo
g
is
tic
r
eg
r
ess
io
n
p
er
f
o
r
m
e
d
b
etter
th
an
e
x
tr
em
e
g
r
ad
ien
t
b
o
o
s
tin
g
(
XGBo
o
s
t
)
,
k
-
n
ea
r
es
t
n
eig
h
b
o
r
(
KNN)
,
a
n
d
s
u
p
p
o
r
t
v
ec
to
r
m
ac
h
in
es
(
SVMs).
T
h
e
s
tu
d
y
also
s
u
g
g
e
s
ted
th
at
DL
m
eth
o
d
s
m
ay
b
e
u
s
ed
to
ac
h
iev
e
f
u
r
th
e
r
r
ef
i
n
e
m
en
ts
.
T
h
e
E
m
o
2
Vec
a
p
p
r
o
ac
h
,
wh
i
ch
en
co
d
es
em
o
tio
n
al
m
ea
n
in
g
s
in
to
v
ec
to
r
f
o
r
m
s
,
was
f
ir
s
t
p
r
esen
ted
in
[
1
4
]
.
T
h
ey
u
s
ed
b
o
t
h
lar
g
er
an
d
s
m
aller
d
atasets
,
s
u
ch
as
W
ASSA
an
d
I
SEAR,
f
o
r
tr
ain
in
g
E
m
o
2
Vec
in
s
id
e
a
m
u
ltit
ask
lear
n
in
g
f
r
a
m
ewo
r
k
.
T
h
ei
r
r
esu
lts
o
u
tp
er
f
o
r
m
ed
m
o
r
e
co
n
v
en
tio
n
al
m
o
d
els s
u
ch
as C
N
N
as
well
as
d
ee
p
m
o
ji
em
b
ed
d
in
g
s
,
an
d
th
eir
ap
p
licatio
n
s
r
an
g
e
d
f
r
o
m
s
tr
ess
d
etec
tio
n
to
s
ar
ca
s
m
class
if
icatio
n
.
E
m
o
2
Vec
p
r
o
d
u
ce
d
co
m
p
etitiv
e
r
esu
lts
wh
en
p
air
e
d
with
Glo
Ve
an
d
lo
g
is
tic
r
eg
r
ess
io
n
.
E
m
o
tio
n
d
etec
tio
n
f
r
o
m
tex
tu
al
co
n
v
er
s
atio
n
s
was
in
v
esti
g
ated
o
n
em
o
tio
n
d
etec
tio
n
f
r
o
m
tex
tu
al
co
n
v
er
s
atio
n
s
u
s
in
g
an
att
en
tio
n
-
b
ased
lear
n
in
g
m
o
d
e
l
[
1
5
]
.
T
h
eir
ar
c
h
itectu
r
e
u
s
ed
to
k
e
n
izatio
n
,
b
id
ir
ec
tio
n
al
lo
n
g
s
h
o
r
t
-
ter
m
m
em
o
r
y
(
B
i
-
L
STM
)
u
n
its
,
s
elf
-
atten
tio
n
,
an
d
d
e
n
s
e
class
if
icatio
n
lay
er
s
,
ac
h
iev
in
g
a
m
icr
o
-
F1
s
co
r
e
o
f
7
5
.
8
2
%.
A
two
-
s
tep
ap
p
r
o
ac
h
was
p
r
o
p
o
s
ed
a
two
-
s
tep
a
p
p
r
o
ac
h
f
o
r
e
m
o
tio
n
d
etec
tio
n
in
T
witter
d
ata
b
y
co
m
b
in
in
g
a
r
u
le
-
b
ased
m
eth
o
d
with
a
s
u
p
e
r
v
is
ed
n
aiv
e
B
ay
es
class
if
ier
.
T
h
e
r
u
le
-
b
ased
a
n
d
m
ac
h
in
e
-
lear
n
in
g
a
p
p
r
o
ac
h
es
ac
h
iev
e
d
ac
cu
r
ac
ies
o
f
a
p
p
r
o
x
im
ately
8
5
%
an
d
8
8
%,
r
esp
ec
tiv
ely
[
1
6
]
.
A
s
u
p
er
v
is
ed
lear
n
in
g
f
r
am
ewo
r
k
was
d
ev
elo
p
e
d
a
s
u
p
er
v
is
ed
lear
n
in
g
f
r
am
ewo
r
k
f
o
r
au
to
m
atic
em
o
tio
n
class
if
icati
o
n
in
tex
tu
al
s
tr
ea
m
s
.
T
h
e
a
p
p
r
o
ac
h
co
m
b
i
n
es
an
o
f
f
lin
e
tr
ai
n
in
g
s
tag
e
with
a
n
o
n
lin
e
class
if
icatio
n
s
tag
e
f
o
r
tr
ac
k
in
g
e
m
o
tio
n
s
in
r
ea
l
-
tim
e
T
witter
d
ata
[
1
7
]
.
E
m
o
ti
o
n
an
aly
s
is
tech
n
iq
u
es
h
av
e
also
b
ee
n
ap
p
lied
em
o
ti
o
n
a
n
aly
s
is
tech
n
iq
u
es
with
th
e
aim
o
f
id
e
n
tify
in
g
h
ate
s
p
ee
ch
o
n
s
o
cial
m
e
d
ia
,
with
a
f
o
cu
s
o
n
th
e
an
al
y
s
is
o
f
u
n
s
tr
u
ctu
r
e
d
d
ata
o
f
p
o
s
ts
th
at
wer
e
in
ten
d
ed
t
o
s
p
r
ea
d
h
atr
ed
[
1
8
]
.
Ma
ch
in
e
-
lear
n
in
g
an
d
d
ee
p
-
le
ar
n
in
g
a
p
p
r
o
ac
h
es
wer
e
r
ev
ie
wed
m
ac
h
in
e
-
lear
n
in
g
a
n
d
d
e
ep
lear
n
in
g
ap
p
r
o
ac
h
es
f
o
r
em
o
tio
n
a
n
aly
s
is
in
tex
tu
al
d
ata
an
d
d
is
cu
s
s
ed
th
e
m
ain
ch
allen
g
es
ass
o
ciate
d
with
tex
t
-
b
ased
em
o
tio
n
d
etec
tio
n
[
1
9
]
.
T
e
x
t
-
b
ased
em
o
tio
n
d
etec
tio
n
ap
p
r
o
ac
h
es
wer
e
s
u
r
v
ey
ed
t
ex
t
-
b
ased
em
o
tio
n
d
etec
tio
n
an
d
h
ig
h
lig
h
ted
m
ai
n
ap
p
r
o
ac
h
es
ad
o
p
ted
in
d
esi
g
n
o
f
tex
t
-
b
ased
em
o
tio
n
-
d
ete
ctio
n
s
y
s
tem
s
[
2
0
]
.
A
p
r
o
ce
s
s
f
o
r
cr
ea
tin
g
an
em
o
tio
n
lex
ico
n
was
d
escr
ib
ed
a
p
r
o
ce
s
s
f
o
r
cr
ea
tin
g
an
em
o
ti
o
n
lex
ico
n
en
r
ic
h
ed
with
wo
r
d
-
le
v
el
em
o
tio
n
al
in
t
en
s
ity
to
im
p
r
o
v
e
em
o
tio
n
a
n
a
ly
s
is
in
tex
ts
[
2
1
]
.
T
witter
d
at
a
wer
e
u
s
ed
T
witter
d
ata
to
i
d
en
tify
em
o
tio
n
s
an
d
s
en
tim
en
ts
an
d
em
p
lo
y
ed
th
e
r
esu
ltin
g
e
m
o
tio
n
an
d
s
en
tim
en
t
in
f
o
r
m
atio
n
to
g
en
er
ate
g
e
n
er
alize
d
an
d
p
e
r
s
o
n
alize
d
r
ec
o
m
m
en
d
atio
n
s
b
a
s
ed
o
n
u
s
er
s
’
T
witter
ac
tiv
ity
[
2
2
]
.
3.
M
E
T
H
O
D
T
h
is
s
ec
tio
n
d
escr
ib
es
a
h
y
b
r
id
ar
ch
itectu
r
e
f
o
r
im
p
r
o
v
i
n
g
T
E
R
,
illu
s
tr
ated
in
Fig
u
r
e
1
,
wh
ic
h
co
m
b
in
es
DL
an
d
ML
to
le
v
er
ag
e
th
e
s
tr
en
g
t
h
s
o
f
b
o
t
h
.
T
h
e
p
ip
elin
e
s
tar
ts
with
d
ata
co
llectio
n
an
d
p
r
e
-
p
r
o
ce
s
s
in
g
to
clea
n
an
d
p
r
ep
ar
e
r
aw
tex
t
d
ata
f
o
r
tr
ain
i
n
g
.
T
h
e
n
s
ev
er
al
DL
m
o
d
els
will
b
e
tr
ain
ed
an
d
test
ed
,
an
d
th
e
th
r
ee
b
est
m
o
d
els
ar
e
ch
o
s
en
ac
co
r
d
in
g
to
th
e
p
er
f
o
r
m
an
ce
m
etr
ics,
in
cl
u
d
in
g
ac
cu
r
ac
y
an
d
F1
-
s
co
r
e.
T
h
ese
b
est
DL
m
o
d
els
ar
e
th
en
co
m
b
in
ed
(
s
u
m
m
ed
)
to
cr
ea
te
a
s
in
g
le
laten
t
f
ea
tu
r
e
v
ec
to
r
,
wh
ich
ca
p
tu
r
es
r
ich
,
m
u
ltid
im
en
s
io
n
al
em
o
tio
n
al
in
f
o
r
m
atio
n
f
r
o
m
tex
t.
T
h
is
d
ee
p
laten
t
v
ec
to
r
i
s
th
en
p
ass
ed
o
v
er
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
Text
emo
tio
n
r
ec
o
g
n
itio
n
b
a
s
e
d
o
n
d
ee
p
lea
r
n
in
g
a
n
d
a
tten
ti
o
n
w
ith
w
h
a
le
…
(
Ta
b
a
r
ek
A
l
w
a
n
Tu
ib
)
3695
to
th
e
m
o
s
t
p
e
r
f
o
r
m
in
g
M
L
class
if
ier
,
p
air
in
g
s
tr
o
n
g
DL
f
ea
tu
r
e
e
x
tr
ac
tio
n
wit
h
th
e
s
tab
ilit
y
an
d
in
ter
p
r
etab
ilit
y
o
f
co
n
v
en
tio
n
al
class
if
icatio
n
m
eth
o
d
s
.
L
astl
y
,
th
e
p
er
f
o
r
m
an
ce
o
f
th
e
h
y
b
r
id
s
y
s
tem
is
co
m
p
ar
ed
to
th
e
p
er
f
o
r
m
an
c
e
o
f
in
d
iv
id
u
al
DL
an
d
ML
m
o
d
els
f
o
r
th
e
p
u
r
p
o
s
e
o
f
d
eter
m
in
in
g
wh
ich
co
n
f
ig
u
r
atio
n
ac
h
iev
es
th
e
h
i
g
h
est
o
v
er
all
ac
cu
r
ac
y
,
a
d
d
r
ess
in
g
th
e
wea
k
n
ess
es
o
f
th
e
s
tan
d
alo
n
e
m
eth
o
d
s
as
well
as
d
eliv
er
in
g
a
m
o
r
e
r
o
b
u
s
t T
E
R
s
y
s
tem
.
Fig
u
r
e
1
.
Su
g
g
ested
m
o
d
el
f
o
r
d
etec
tin
g
em
o
tio
n
s
f
r
o
m
tex
t
3
.
1
.
Da
t
a
s
et
d
escript
io
n
T
h
e
d
ataset
to
b
e
u
tili
ze
d
i
n
th
e
p
r
esen
t
s
tu
d
y
was
co
m
p
iled
o
n
t
h
r
ee
s
o
u
r
ce
s
W
ASSA
[
2
3
]
,
I
SEAR
[
1
2
]
,
an
d
E
m
o
tio
n
-
s
tim
u
lu
s
[
2
4
]
,
wh
e
r
e
ea
ch
o
n
e
o
f
f
er
ed
tex
t
an
d
em
o
tio
n
lab
els
in
d
if
f
er
en
t
f
o
r
m
ats
(
d
ialo
g
u
es,
twee
ts
,
an
d
s
en
ten
ce
s
)
.
I
SEAR
d
atab
ase
was
d
ev
elo
p
ed
f
o
llo
win
g
a
g
lo
b
al
ef
f
o
r
t
o
f
p
s
y
ch
o
lo
g
is
ts
in
th
e
1
9
9
0
s
an
d
co
n
tain
s
r
e
p
o
r
ts
o
f
ap
p
r
o
x
im
ately
3
0
0
0
p
ar
ticip
an
ts
in
3
7
n
atio
n
s
with
s
ev
en
f
u
n
d
am
en
tal
em
o
tio
n
s
(
a
n
g
er
,
jo
y
,
s
ad
n
ess
,
g
u
ilt,
f
ea
r
,
d
is
g
u
s
t,
an
d
s
h
am
e)
[
1
2
]
.
T
h
e
d
ata
was
o
r
ig
in
al
ly
ca
teg
o
r
ize
d
in
to
1
7
3
em
o
tio
n
s
an
d
later
co
n
s
o
lid
ated
in
to
s
ev
en
m
ain
ty
p
es:
s
ad
n
ess
,
f
ea
r
,
jo
y
,
an
g
e
r
,
s
u
r
p
r
is
e,
d
is
g
u
s
t,
an
d
s
h
am
e.
I
n
ad
d
itio
n
,
th
e
d
ataset
in
clu
d
es
E
m
o
tio
n
-
s
tim
u
lu
s
s
tatem
en
ts
,
wh
er
e
th
e
“c
au
s
e”
s
u
b
s
et
co
n
tain
s
8
2
0
s
en
ten
ce
s
an
n
o
tated
with
b
o
th
an
em
o
tio
n
lab
el
as
well
as
its
tr
ig
g
er
in
g
ca
u
s
e,
w
h
ile
th
e
“n
o
ca
u
s
e”
s
u
b
s
et
in
clu
d
es
1
5
9
4
s
en
ten
c
es
lab
eled
o
n
ly
with
em
o
tio
n
s
[
2
4
]
,
as
s
u
m
m
ar
ized
in
T
ab
le
1
.
T
h
is
s
tu
d
y
co
m
b
in
ed
t
h
r
ee
d
atasets
(
tex
t
s
en
ten
ce
s
,
d
ialo
g
u
es,
an
d
tw
ee
ts
)
in
to
a
s
in
g
le
co
r
p
u
s
o
f
o
v
er
1
4
,
5
0
0
E
n
g
lis
h
tex
t
s
am
p
les.
E
ac
h
s
am
p
le
was
lab
eled
with
o
n
e
o
f
s
ix
em
o
tio
n
s
—
d
is
g
u
s
t,
jo
y
,
f
ea
r
,
s
ad
n
ess
,
s
u
r
p
r
is
e,
an
d
an
g
er
—
b
ased
o
n
s
y
n
tactic
an
d
s
em
an
tic
p
o
lar
ity
[
5
]
,
[
2
3
]
.
As
s
h
o
wn
in
Fig
u
r
e
2
,
t
h
ese
em
o
tio
n
s
ar
e
f
o
r
m
ally
d
ef
in
ed
,
an
d
th
e
tex
t
m
ay
in
cl
u
d
e
ex
tr
a
p
u
n
ctu
atio
n
an
d
em
o
tico
n
s
.
T
h
e
co
m
b
in
e
d
co
r
p
u
s
ex
h
ib
its
an
u
n
ev
en
d
is
tr
ib
u
tio
n
ac
r
o
s
s
th
e
em
o
tio
n
class
es,
wh
ich
s
h
o
u
ld
b
e
c
o
n
s
id
er
ed
b
ec
au
s
e
class
im
b
a
lan
ce
ca
n
in
f
lu
e
n
ce
class
if
icatio
n
p
er
f
o
r
m
a
n
ce
[
2
5
]
.
T
h
e
f
in
al
d
ataset
co
n
tain
s
o
n
ly
th
e
s
en
ten
ce
s
an
d
th
eir
em
o
tio
n
lab
els,
a
n
d
it
h
as b
ee
n
s
p
lit to
tr
ain
in
g
an
d
test
in
g
s
ets u
s
in
g
a
r
atio
o
f
8
0
:
2
0
.
Fig
u
r
e
2
.
Six
ty
p
es o
f
em
o
tio
n
s
in
th
is
wo
r
k
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.
15
,
No
.
4
,
Au
g
u
s
t 2
0
2
6
:
3
6
9
2
-
3
7
0
2
3696
T
ab
le
1
.
Descr
ip
tio
n
o
f
ea
c
h
d
ataset
in
d
iv
id
u
ally
D
a
t
a
s
e
t
G
r
a
n
u
l
a
r
i
t
y
N
o
.
o
f
e
mo
t
i
o
n
s
S
i
z
e
D
e
scri
p
t
i
o
n
I
S
EA
R
S
e
n
t
e
n
c
e
s
7
e
mo
t
i
o
n
s
7
6
6
6
S
t
u
d
i
e
d
i
n
3
7
c
o
u
n
t
r
i
e
s
W
A
S
S
A
Tw
e
e
t
s
4
e
mo
t
i
o
n
s
4
3
3
4
Tw
e
e
t
s
Emo
t
i
o
n
-
s
t
i
m
u
l
u
s
D
i
a
l
o
g
s
7
e
mo
t
i
o
n
s
2
5
0
0
—
3
.
2
.
Da
t
a
p
re
-
pro
ce
s
s
ing
On
e
o
f
th
e
ess
en
tial
p
r
o
ce
s
s
es
in
d
ata
m
in
in
g
is
p
r
ep
r
o
ce
s
s
in
g
,
wh
ich
is
u
s
ed
to
elim
in
ate
u
n
n
ec
ess
ar
y
tex
ts
a
n
d
s
y
m
b
o
l
s
in
r
aw
d
ata
to
m
ak
e
th
e
m
c
lean
er
,
ea
s
ier
to
in
ter
p
r
et
,
an
d
m
o
r
e
ef
f
icien
t
f
o
r
an
aly
s
is
.
Du
e
to
th
e
f
ac
t
th
at
t
h
e
d
ata
is
co
llected
in
v
ar
io
u
s
s
o
u
r
ce
s
,
it
is
to
b
e
in
te
g
r
ated
a
n
d
f
o
r
m
atted
p
r
io
r
to
f
u
r
th
e
r
p
r
o
ce
s
s
in
g
to
ac
h
iev
e
ac
cu
r
ate
f
in
d
i
n
g
s
.
I
n
t
h
is
s
tu
d
y
,
p
r
ep
r
o
ce
s
s
in
g
was
tailo
r
ed
to
th
e
d
ataset
an
d
in
clu
d
ed
s
to
p
-
wo
r
d
r
em
o
v
al,
to
k
en
izatio
n
,
s
tem
m
in
g
,
em
o
ji
-
to
-
tex
t
co
n
v
er
s
io
n
,
an
d
lem
m
atiza
tio
n
u
s
in
g
NL
T
K
to
o
ls
.
As
p
ar
t
o
f
p
r
ep
r
o
ce
s
s
in
g
,
d
ata
clea
n
in
g
was
ap
p
lied
to
elim
in
ate
n
o
is
y
tex
t
(
e.
g
.
,
s
u
p
er
f
lu
o
u
s
s
y
m
b
o
ls
an
d
wo
r
d
s
)
th
at
d
o
e
s
n
o
t
co
n
tr
ib
u
te
to
m
o
d
el
lea
r
n
in
g
[
9
]
.
Af
ter
clea
n
in
g
,
th
e
r
esu
ltin
g
s
en
ten
ce
s
wer
e
to
k
en
ized
a
n
d
u
s
ed
as
m
o
d
el
in
p
u
ts
,
th
en
s
p
lit
to
tr
a
in
in
g
an
d
test
in
g
s
ets
an
d
ev
a
lu
ated
with
m
u
ltip
le
ML
an
d
DL
alg
o
r
ith
m
s
to
i
m
p
r
o
v
e
p
er
f
o
r
m
a
n
ce
,
esp
ec
i
ally
s
in
ce
th
e
d
atasets
wer
e
d
r
awn
f
r
o
m
th
r
e
e
r
ep
o
s
ito
r
ies (
W
ASSA,
I
SEA
R
,
an
d
E
m
o
tio
n
-
s
tim
u
lu
s
)
.
3
.
3
.
F
e
a
t
ure
e
x
t
r
a
ct
io
n
I
n
o
r
d
er
to
s
p
ee
d
u
p
p
r
o
ce
s
s
in
g
,
f
ea
tu
r
e
ex
tr
ac
tio
n
r
e
d
u
ce
s
d
im
en
s
io
n
ality
b
y
d
iv
id
in
g
lar
g
e
am
o
u
n
ts
o
f
r
aw
d
ata
in
to
s
m
aller
ca
teg
o
r
ies
[
1
0
]
.
T
h
is
p
r
o
ce
s
s
co
n
v
er
ts
th
e
p
r
ep
r
o
ce
s
s
ed
tex
tu
al
d
ata
in
to
n
u
m
er
ical
r
ep
r
esen
tatio
n
s
th
at
ca
n
b
e
ef
f
icien
tly
p
r
o
ce
s
s
ed
b
y
th
e
m
a
ch
in
e
lear
n
in
g
an
d
d
ee
p
lear
n
in
g
m
o
d
els.
I
n
th
is
s
tu
d
y
,
T
F
-
I
DF
an
d
W
o
r
d
2
v
ec
ar
e
u
s
ed
to
r
ep
r
esen
t
ter
m
im
p
o
r
tan
ce
an
d
s
em
an
tic
r
elatio
n
s
h
ip
s
,
r
esp
ec
tiv
ely
,
as e
x
p
lain
ed
in
th
e
f
o
llo
win
g
s
u
b
s
ec
tio
n
s
.
3
.
3
.
1
.
T
er
m
f
re
qu
ency
-
inv
er
s
e
do
cum
ent
f
re
qu
ency
B
y
c
o
m
b
i
n
in
g
t
er
m
f
r
eq
u
en
cy
w
i
t
h
in
v
er
s
e
d
o
c
u
m
e
n
t
f
r
eq
u
en
c
y
,
t
h
e
T
F
-
I
D
F
m
e
t
h
o
d
f
i
n
d
s
th
e
t
e
r
m
s
th
a
t
ar
e
u
t
i
l
i
z
ed
m
o
s
t
o
f
t
e
n
in
a
d
o
c
u
m
e
n
t
a
s
w
e
l
l
a
s
t
h
o
s
e
t
h
a
t
a
r
e
u
t
i
l
i
ze
d
in
f
r
e
q
u
e
n
t
ly
th
r
o
u
g
h
o
u
t
t
h
e
co
r
p
u
s
.
T
h
i
s
m
e
th
o
d
f
a
c
i
li
t
a
t
e
s
t
h
e
p
r
o
ce
s
s
o
f
ch
o
o
s
i
n
g
a
u
n
iq
u
e
t
e
r
m
v
e
c
to
r
t
o
a
c
t
as
a
f
ea
t
u
r
e
s
e
t
f
o
r
t
h
e
t
r
a
i
n
in
g
o
f
t
h
e
m
o
d
e
l
.
I
n
t
h
i
s
wo
r
k
,
te
x
t
i
s
c
o
n
v
er
te
d
t
o
f
ea
t
u
r
e
v
e
c
t
o
r
s
w
i
t
h
th
e
u
s
e
o
f
T
F
-
I
D
F
v
e
c
t
o
r
i
z
er
,
w
h
ic
h
a
l
lo
w
s
t
h
em
t
o
b
e
u
s
ed
a
s
i
n
p
u
t
f
o
r
m
o
d
e
l
e
s
t
im
a
t
o
r
s
.
A
d
i
c
t
io
n
a
r
y
-
l
i
k
e
v
o
ca
b
u
l
a
r
y
i
s
c
r
e
a
t
ed
,
w
i
th
e
ac
h
t
o
k
e
n
(
w
o
r
d
)
b
e
i
n
g
m
a
p
p
e
d
t
o
a
f
e
a
t
u
r
e
i
n
d
ex
a
cc
o
r
d
i
n
g
to
h
o
w
f
r
eq
u
e
n
t
l
y
i
t
ap
p
e
ar
s
i
n
t
h
e
m
a
t
r
i
x
.
A
s
a
r
e
s
u
l
t
,
a
f
ea
tu
r
e
in
d
e
x
i
s
g
iv
e
n
t
o
e
v
er
y
d
i
s
t
i
n
c
t
t
o
k
e
n
.
T
h
e
T
F
-
I
D
F
w
ei
g
h
t
c
a
l
c
u
la
t
i
o
n
i
s
e
x
p
r
e
s
s
e
d
in
(
1
)
.
W
(
d
,
t
)
=
TF
(
d
,
t
)
∗
l
og
(
N
df
(
t
)
)
(
1
)
I
n
wh
i
c
h
d
d
en
o
te
s
d
o
c
u
m
en
t
s
,
t
d
en
o
t
e
s
te
r
m
s
in
th
e
d
o
c
u
m
e
n
t,
an
d
N
r
ep
r
e
s
e
n
t
s
th
e
to
t
a
l
n
u
m
b
er
o
f
d
o
cu
m
en
t
s
.
3
.
3
.
2
.
Wo
rd
e
m
bedd
ing
s
Var
io
u
s
m
eth
o
d
s
o
f
w
o
r
d
em
b
ed
d
in
g
h
av
e
b
ec
o
m
e
c
o
m
m
o
n
;
th
ey
ar
e
Glo
Ve,
W
o
r
d
2
v
ec
,
f
astTex
t,
an
d
E
L
M
o
,
alth
o
u
g
h
th
is
s
tu
d
y
is
b
ased
o
n
W
o
r
d
2
v
ec
.
W
o
r
d
2
v
ec
is
tr
ain
ed
o
n
a
lar
g
e
tex
t
co
r
p
u
s
to
lear
n
wo
r
d
ass
o
ciatio
n
s
v
ia
a
n
eu
r
a
l
n
etwo
r
k
an
d
p
r
esen
ts
wo
r
d
s
(
an
d
s
en
ten
ce
s
)
as
n
u
m
er
ical
v
ec
to
r
s
,
wh
ich
ca
n
b
e
u
s
ed
to
ca
p
tu
r
e
s
em
an
tic
r
elatio
n
s
h
ip
s
[
2
6
]
.
D
u
e
to
th
e
v
ar
iatio
n
i
n
s
en
ten
ce
len
g
th
,
p
ad
d
in
g
is
u
s
ed
f
o
r
s
tan
d
ar
d
izin
g
in
p
u
t
s
ize,
allo
win
g
th
e
m
o
d
el
to
p
r
o
ce
s
s
tex
t
u
n
if
o
r
m
ly
,
wh
er
e
a
co
m
m
o
n
len
g
th
o
f
s
eq
u
e
n
ce
is
5
0
to
k
en
s
.
T
h
e
s
tu
d
y
cr
ea
tes
an
em
b
e
d
d
in
g
m
atr
ix
o
f
th
e
v
o
ca
b
u
lar
y
ter
m
s
(
s
ize
-
1
8
2
1
0
,
3
0
0
)
in
to
a
300
-
d
im
en
s
io
n
al
co
n
tin
u
o
u
s
v
ec
to
r
s
p
ac
e
with
a
p
r
e
-
tr
ain
e
d
em
b
ed
d
in
g
to
en
h
an
ce
s
em
an
tic
r
ep
r
esen
tatio
n
an
d
o
v
er
all
m
o
d
el
p
er
f
o
r
m
an
c
e.
3
.
4
.
Dee
p
lea
rning
mo
dels
Ar
tific
ial
in
tellig
en
ce
(
AI
)
s
y
s
tem
s
th
at
en
ab
le
ML
a
s
well
a
s
ev
o
lu
tio
n
with
o
u
t
th
e
n
ee
d
f
o
r
ex
p
licit
p
r
o
g
r
a
m
m
in
g
ar
e
k
n
o
wn
as
D
L
m
o
d
els [
2
7
]
.
Fo
r
tr
ain
in
g
th
e
m
o
d
el
as
well
as
p
r
e
d
ictin
g
em
o
tio
n
s
o
n
th
e
test
d
ataset,
th
e
p
r
ep
r
o
ce
s
s
ed
tr
ain
in
g
d
ataset
is
f
ed
in
to
a
n
u
m
b
er
o
f
co
m
p
o
n
en
ts
in
th
is
r
esear
ch
,
s
u
ch
as T
F
-
I
DF
tr
an
s
f
o
r
m
er
,
C
o
u
n
tVec
to
r
izer
,
an
d
ML
C
lass
if
ier
.
T
h
e
s
tu
d
y
m
ak
es
u
s
e
o
f
p
r
e
-
b
u
ilt
DL
a
n
d
ML
m
o
d
els.
An
ML
P is
u
s
ed
in
ML
f
o
r
p
r
ed
ict
in
g
em
o
tio
n
s
.
On
th
e
o
t
h
er
h
a
n
d
,
DL
u
s
es a
r
ch
itectu
r
es lik
e
B
iG
R
U,
C
NN
,
an
d
GR
U
f
o
r
em
o
tio
n
p
r
e
d
ictio
n
t
ask
s
[
2
7
]
,
[
2
8
]
.
T
h
e
th
r
ee
DL
co
m
p
o
n
en
ts
u
s
ed
in
th
e
p
r
o
p
o
s
ed
f
r
am
ewo
r
k
ar
e
d
escr
ib
ed
in
th
e
f
o
llo
win
g
s
u
b
s
ec
tio
n
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Text
emo
tio
n
r
ec
o
g
n
itio
n
b
a
s
e
d
o
n
d
ee
p
lea
r
n
in
g
a
n
d
a
tten
ti
o
n
w
ith
w
h
a
le
…
(
Ta
b
a
r
ek
A
l
w
a
n
Tu
ib
)
3697
3
.
4
.
1
.
G
a
t
ed
re
curr
ent
un
it
T
h
e
v
an
is
h
in
g
g
r
ad
ie
n
t
p
r
o
b
le
m
th
at
tr
ad
itio
n
al
r
ec
u
r
r
en
t
n
e
u
r
al
n
etwo
r
k
s
(
R
NNs)
f
ac
e
is
r
eso
lv
ed
b
y
th
e
GR
U
[
2
9
]
.
Sin
ce
th
e
y
f
u
n
ctio
n
s
im
ilar
ly
in
v
ar
io
u
s
s
itu
atio
n
s
an
d
s
h
ar
e
s
tr
u
ctu
r
al
s
im
ilar
ities
with
lo
n
g
s
h
o
r
t
-
te
r
m
m
em
o
r
y
(
L
S
T
M)
n
etwo
r
k
s
,
GR
Us
ar
e
f
r
eq
u
en
tly
r
e
g
ar
d
e
d
as
L
STM
v
ar
ian
ts
.
GR
U
is
im
p
lem
en
ted
as
a
s
in
g
le
lay
er
in
th
e
s
u
g
g
ested
m
o
d
el.
T
h
e
em
b
ed
d
in
g
lay
er
o
f
d
im
en
s
io
n
s
(
1
8
2
1
0
,
3
0
0
)
is
u
s
ed
as
in
p
u
t
f
o
r
GR
U
m
o
d
el
af
ter
f
ea
tu
r
e
e
x
tr
ac
tio
n
.
T
h
e
GR
U
m
o
d
el
th
e
n
u
s
es
th
e
tr
ai
n
in
g
v
ec
to
r
as
in
p
u
t
f
o
r
th
e
p
r
ed
ictio
n
o
f
e
m
o
tio
n
s
th
r
o
u
g
h
o
u
t t
h
e
d
ataset.
3
.
4
.
2
.
B
idi
re
ct
io
na
l
g
a
t
ed
re
c
urre
nt
un
it
A
s
eq
u
en
ce
p
r
o
ce
s
s
in
g
p
ar
ad
ig
m
k
n
o
wn
as
B
iGR
U
[
3
0
]
co
n
s
is
ts
o
f
two
GR
Us
r
u
n
n
in
g
s
im
u
ltan
eo
u
s
ly
,
o
n
e
b
ac
k
war
d
an
d
o
n
e
f
o
r
war
d
.
T
h
e
o
u
tp
u
t
an
d
in
p
u
t
g
ates
ar
e
th
e
o
n
ly
o
n
es
u
tili
ze
d
in
s
u
ch
a
b
id
ir
ec
tio
n
al
R
NN.
B
iGR
U
is
im
p
lem
en
ted
as
a
s
in
g
le
lay
er
in
o
u
r
s
u
g
g
ested
m
o
d
el.
T
h
e
em
b
e
d
d
in
g
lay
er
o
f
d
im
en
s
io
n
s
(
1
8
2
1
0
,
3
0
0
)
i
s
u
s
ed
as
in
p
u
t
f
o
r
B
iGR
U
m
o
d
el
af
ter
f
ea
tu
r
e
e
x
tr
ac
tio
n
.
B
iG
R
U
m
o
d
el
th
en
u
s
es th
e
tr
ain
in
g
v
ec
to
r
as in
p
u
t f
o
r
p
r
ed
ictin
g
em
o
tio
n
s
th
r
o
u
g
h
o
u
t th
e
d
ataset.
3
.
4
.
3
.
Co
nv
o
lutio
na
l
neura
l
net
wo
rk
I
n
th
e
co
n
tex
t
o
f
DL
,
th
e
C
NN
is
a
d
ee
p
n
eu
r
al
n
etwo
r
k
(
D
NN)
ar
ch
itectu
r
e
t
h
at
ca
n
also
b
e
ap
p
lied
to
tex
t
class
if
icatio
n
[
3
1
]
.
T
h
e
C
NN
w
ill
f
o
r
m
o
n
e
lay
er
in
th
e
p
r
o
p
o
s
ed
m
o
d
el.
T
h
e
C
NN
m
o
d
el
co
n
s
u
m
es
an
em
b
ed
d
i
n
g
lay
er
o
f
p
a
r
am
eter
s
(
1
8
2
1
0
,
3
0
0
)
o
n
ce
th
e
f
e
atu
r
es
ar
e
ex
tr
ac
ted
,
an
d
th
en
it
tak
es
th
e
tr
ain
in
g
v
ec
to
r
to
p
r
ed
ict
th
e
e
m
o
tio
n
s
h
o
r
izo
n
tally
i
n
a
d
ataset.
3
.
5
.
M
ulti
-
hea
d a
t
t
ent
io
n
Mu
lti
-
h
ea
d
atten
tio
n
r
ep
r
esen
ts
a
co
r
e
co
m
p
o
n
en
t
o
f
m
o
d
er
n
DL
—
esp
ec
ially
tr
an
s
f
o
r
m
er
-
b
ased
NL
P
m
o
d
els
—
b
ec
au
s
e
it
ca
n
f
o
cu
s
o
n
d
if
f
er
e
n
t
p
ar
ts
o
f
an
in
p
u
t
s
eq
u
en
ce
in
p
a
r
allel
an
d
in
teg
r
at
e
in
f
o
r
m
atio
n
ac
r
o
s
s
r
ep
r
esen
tatio
n
lay
er
s
.
T
h
is
m
ak
es
it
ex
tr
em
ely
ef
f
icien
t
in
th
e
ca
p
tu
r
e
o
f
th
e
co
m
p
lex
d
ep
en
d
e
n
cies
an
d
lo
n
g
-
r
an
g
e
r
elatio
n
s
h
ip
s
in
task
s
l
ik
e
t
ex
t
class
if
icatio
n
an
d
m
ac
h
in
e
tr
an
s
latio
n
[
3
2
]
.
Mu
lti
-
h
ea
d
atten
tio
n
is
ap
p
li
ed
in
th
is
wo
r
k
to
c
o
m
b
in
e
th
e
d
ee
p
f
ea
tu
r
es
e
x
tr
ac
ted
f
r
o
m
th
r
ee
n
eu
r
al
ar
ch
itectu
r
es,
n
am
ely
GR
U,
C
NN,
an
d
B
i
GR
U
s
,
th
at
ca
p
tu
r
es
th
e
v
ar
io
u
s
elem
en
ts
o
f
s
eq
u
en
tial
tex
t,
in
clu
d
in
g
lo
ca
l
p
atter
n
s
,
tem
p
o
r
al
d
e
p
en
d
e
n
cies,
an
d
b
i
d
ir
ec
tio
n
al
co
n
tex
t.
B
o
th
o
f
th
ese
co
m
p
lem
en
tar
y
r
ep
r
esen
tatio
n
s
g
i
v
e
th
e
m
o
d
e
l
a
d
ee
p
er
u
n
d
er
s
tan
d
in
g
o
f
th
e
in
p
u
t
an
d
en
h
a
n
ce
its
ca
p
ac
i
ty
to
id
en
tify
b
o
th
g
lo
b
al
an
d
l
o
ca
l
co
n
tex
t
u
al
cu
es,
r
esu
ltin
g
in
im
p
r
o
v
e
d
NL
P
p
er
f
o
r
m
an
ce
.
T
h
e
o
th
er
b
en
ef
it
is
its
co
m
p
u
tatio
n
al
ef
f
icien
c
y
:
m
u
lti
-
h
ea
d
atten
tio
n
is
ca
p
ab
le
o
f
m
o
d
elin
g
co
m
p
lex
p
atter
n
s
an
d
p
r
o
ce
s
s
in
g
m
u
ltip
le
in
p
u
t
co
m
p
o
n
e
n
ts
at
o
n
ce
,
h
elp
i
n
g
to
ad
d
r
ess
lo
n
g
-
r
an
g
e
d
e
p
en
d
e
n
cy
an
d
d
ec
r
ea
s
in
g
is
s
u
es
s
u
ch
as
v
an
is
h
in
g
g
r
ad
ie
n
ts
ass
o
ciate
d
with
d
e
ep
n
etwo
r
k
s
.
Giv
en
t
h
at
atten
tio
n
is
co
m
p
u
ted
in
p
ar
allel
with
s
ev
er
al
h
ea
d
s
,
tr
ain
in
g
a
n
d
in
f
e
r
en
ce
ar
e
m
o
r
e
s
ca
lab
le
an
d
f
aster
,
h
en
ce
m
u
lti
-
h
ea
d
atten
tio
n
is
ce
n
tr
al
at
th
e
co
r
e
o
f
s
tate
-
of
-
th
e
-
ar
t N
L
P sy
s
tem
s
[
3
2
]
.
3
.
6
.
F
e
a
t
ure
s
elec
t
io
n wit
h
wha
le
o
ptim
iza
t
io
n a
lg
o
rit
hm
T
h
e
W
OA
is
a
m
etah
eu
r
is
tic
o
p
tim
izatio
n
m
eth
o
d
,
w
h
ich
h
a
s
b
ee
n
b
ased
o
n
th
e
co
o
p
er
ati
v
e
h
u
n
tin
g
b
eh
av
io
r
o
f
th
e
h
u
m
p
b
ac
k
wh
ales.
I
t
s
ea
r
ch
es
f
o
r
o
p
tim
al
s
o
lu
tio
n
s
th
r
o
u
g
h
iter
ativ
ely
u
p
d
atin
g
a
p
o
p
u
latio
n
o
f
ca
n
d
id
ate
s
o
lu
tio
n
s
,
b
ala
n
c
in
g
ex
p
lo
r
ati
o
n
an
d
e
x
p
lo
itati
o
n
with
en
cir
clin
g
p
r
ey
an
d
b
u
b
b
le
-
n
et
f
ee
d
in
g
tech
n
iq
u
es
wh
er
e
p
r
o
m
is
in
g
r
eg
io
n
s
o
f
t
h
e
s
ea
r
ch
s
p
ac
e
[
3
3
]
.
Op
tim
izatio
n
-
b
ased
f
ea
tu
r
e
-
s
elec
tio
n
m
eth
o
d
s
h
av
e
b
ee
n
ap
p
lied
t
o
tex
tu
al
a
n
d
s
en
tim
en
t
-
an
al
y
s
is
task
s
to
r
etain
in
f
o
r
m
ativ
e
f
ea
tu
r
es
an
d
r
ed
u
ce
r
ed
u
n
d
an
t
attr
ib
u
tes
[
3
4
]
,
[
3
5
]
.
T
h
e
W
OA
m
o
d
el
is
u
s
ed
in
th
is
s
tu
d
y
to
f
in
d
th
e
m
o
s
t
u
s
ef
u
l
d
ee
p
f
ea
tu
r
es
p
r
o
p
o
s
ed
b
y
m
u
lti
-
h
ea
d
atten
tio
n
with
th
e
p
u
r
p
o
s
e
o
f
p
r
eser
v
in
g
in
f
o
r
m
ativ
e
f
ea
tu
r
es
an
d
lo
s
in
g
r
ed
u
n
d
an
t
o
r
ir
r
elev
an
t
f
ea
tu
r
es
[
3
4
]
,
[
3
5
]
.
T
h
is
r
ed
u
ce
s
th
e
f
ea
tu
r
e
-
s
p
ac
e
s
ize
an
d
im
p
r
o
v
es
th
e
e
f
f
icien
cy
,
in
ter
p
r
etab
ilit
y
,
a
n
d
p
er
f
o
r
m
an
ce
o
f
later
lear
n
in
g
task
s
.
T
h
r
o
u
g
h
t
h
e
co
m
b
in
atio
n
o
f
W
OA
with
m
u
lti
-
h
ea
d
att
en
tio
n
,
th
e
m
eth
o
d
en
ab
les
a
t
h
o
r
o
u
g
h
s
ea
r
ch
o
f
h
ig
h
-
d
im
e
n
s
io
n
al
f
ea
tu
r
e
s
p
ac
es
—
co
m
m
o
n
in
DL
—
wh
ile
r
em
ain
in
g
s
ca
lab
le
an
d
ad
a
p
tab
le,
h
elp
i
n
g
p
r
o
d
u
c
e
m
o
r
e
r
o
b
u
s
t a
n
d
g
en
e
r
aliza
b
le
f
ea
tu
r
e
r
ep
r
esen
tatio
n
s
[
3
3
]
.
3
.
7
.
H
y
brid
m
o
del
As
s
ee
n
in
Fig
u
r
e
3
,
th
e
s
u
g
g
ested
h
y
b
r
i
d
m
o
d
el
c
o
m
b
in
es
DL
alg
o
r
ith
m
s
f
o
r
p
r
e
d
ictin
g
em
o
tio
n
s
.
B
iG
R
U
an
d
C
N
N
ar
e
ex
am
p
les o
f
DL
co
m
p
o
n
en
ts
,
wh
er
ea
s
an
ML
P c
las
s
if
ier
is
u
s
ed
in
ML
[
2
8
]
.
W
o
r
d
2
v
ec
is
u
s
ed
to
tr
an
s
m
it
th
e
in
p
u
t
d
atasets
v
ia
th
e
wo
r
d
e
m
b
ed
d
in
g
lay
er
at
th
e
b
e
g
in
n
in
g
o
f
th
e
p
r
o
ce
d
u
r
e.
T
h
e
r
esu
ltan
t
em
b
ed
d
in
g
v
ec
to
r
is
th
en
s
en
t
in
to
C
NN
an
d
B
iG
R
U,
two
DL
a
lg
o
r
ith
m
s
.
T
h
e
B
iG
R
U
an
d
C
N
N
m
o
d
els o
n
ly
s
er
v
e
as e
n
co
d
er
s
b
ec
au
s
e
th
e
last
lay
er
was e
li
m
in
ated
[
2
8
]
.
Fu
r
th
er
m
o
r
e,
f
o
r
th
e
p
r
o
v
i
d
ed
in
p
u
t
em
b
ed
d
in
g
v
ec
to
r
,
b
o
t
h
o
f
s
u
ch
en
co
d
e
r
s
p
r
o
d
u
ce
a
laten
t
v
ec
to
r
.
T
h
ese
laten
t
v
ec
t
o
r
s
ar
e
th
en
s
en
t
in
to
th
e
ML
P
class
if
ier
af
ter
b
ein
g
co
n
ca
ten
ated
.
T
h
e
em
o
tio
n
r
eg
a
r
d
in
g
th
e
in
p
u
t
wo
r
d
s
is
af
ter
th
at
p
r
ed
icted
b
y
th
e
ML
P
clas
s
if
ier
.
B
ec
au
s
e
cla
s
s
if
icatio
n
m
o
d
els
f
r
o
m
b
o
th
th
e
DL
an
d
ML
d
o
m
ain
s
wer
e
ca
r
ef
u
lly
ch
o
s
en
,
th
e
s
u
g
g
ested
h
y
b
r
i
d
m
o
d
el
s
h
o
w
s
b
etter
r
esu
lts
in
ter
m
s
o
f
F1
-
s
co
r
e
an
d
ac
cu
r
ac
y
.
B
iGR
U
an
d
C
NN,
two
D
L
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
4
,
Au
g
u
s
t 2
0
2
6
:
3
6
9
2
-
3
7
0
2
3698
m
o
d
el
r
esu
lts
,
p
er
f
o
r
m
ed
wel
l
in
d
iv
id
u
ally
.
I
n
th
e
s
am
e
way
,
ML
P
s
to
o
d
o
u
t
as
a
s
tr
o
n
g
p
er
f
o
r
m
er
am
o
n
g
ML
alg
o
r
ith
m
s
[
3
6
]
.
T
o
im
p
r
o
v
e
o
v
er
all
p
er
f
o
r
m
an
ce
,
th
e
h
y
b
r
id
m
o
d
el
in
teg
r
ates
t
h
e
to
p
c
lass
if
ier
s
f
r
o
m
th
e
two
ca
teg
o
r
ies.
T
h
e
b
est
DL
m
o
d
els
ar
e
co
m
b
i
n
ed
to
cr
ea
t
e
th
e
h
y
b
r
i
d
m
o
d
el,
wh
ich
p
r
o
d
u
ce
d
t
h
e
h
ig
h
est
ac
cu
r
ac
y
as
well
as
F1
-
s
co
r
e.
T
h
e
to
p
-
p
e
r
f
o
r
m
in
g
ML
m
o
d
e
ls
,
wh
ich
ar
e
r
en
o
wn
ed
f
o
r
th
e
ir
h
ig
h
ac
cu
r
ac
y
in
em
o
tio
n
p
r
ed
ictio
n
task
s
,
wer
e
th
en
f
ed
th
e
laten
t
v
ec
t
o
r
th
at
was
ex
tr
ac
ted
f
r
o
m
s
u
ch
m
o
d
els
[
3
7
]
.
Fig
u
r
e
3
s
h
o
ws
h
o
w
th
e
DL
an
d
ML
m
o
d
els
ar
e
co
m
b
in
ed
to
p
r
o
d
u
c
e
b
etter
r
esu
lts
b
y
u
tili
zin
g
th
e
ad
v
an
tag
es
o
f
th
e
two
f
ield
s
.
Fig
u
r
e
3
.
C
NN
m
o
d
el
f
o
r
d
etec
tin
g
em
o
tio
n
s
f
r
o
m
tex
t
4.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
T
o
u
n
d
e
r
s
tan
d
th
e
b
est
p
er
f
o
r
m
an
ce
o
f
th
e
s
u
g
g
ested
h
y
b
r
id
m
o
d
el,
a
th
o
r
o
u
g
h
s
eq
u
en
ce
o
f
ex
p
er
im
en
ts
with
th
e
h
ig
h
est
n
u
m
b
er
o
f
DL
m
o
d
els
an
d
test
in
g
ap
p
r
o
ac
h
es
was
p
er
f
o
r
m
e
d
.
T
h
e
an
aly
s
is
was
b
ased
o
n
co
m
m
o
n
ly
u
s
ed
b
in
a
r
y
m
ea
s
u
r
es
in
class
if
icatio
n
:
p
r
ec
is
io
n
,
r
ec
all
,
F1
-
s
co
r
e,
a
n
d
ac
cu
r
ac
y
b
u
ilt
o
n
th
e
co
n
f
u
s
io
n
m
atr
ix
[
3
8
]
.
T
h
e
m
etr
ics
wer
e
ca
lcu
lated
b
y
ev
alu
atin
g
ea
c
h
o
f
t
h
e
class
es
o
f
em
o
tio
n
s
ep
ar
ately
an
d
th
e
n
g
ettin
g
th
e
m
ac
r
o
-
av
er
a
g
ed
a
cc
u
r
ac
y
o
f
o
v
e
r
all
m
o
d
el
p
e
r
f
o
r
m
an
ce
a
cr
o
s
s
all
class
es.
T
o
th
is
en
d
,
th
e
tr
u
e
p
o
s
itiv
e
(
T
P),
tr
u
e
n
eg
ativ
e
(
T
N)
,
f
alse
p
o
s
itiv
e
(
FP
)
,
an
d
f
alse
n
eg
ativ
e
(
FN)
wer
e
h
ar
v
ested
b
y
m
atch
in
g
m
o
d
el
-
g
en
er
ated
p
r
ed
ictio
n
s
ag
ai
n
s
t
th
e
ex
p
er
t
-
a
n
n
o
tated
g
r
o
u
n
d
t
r
u
th
o
n
ea
c
h
o
f
s
ix
d
is
tin
ct
em
o
tio
n
ca
teg
o
r
ies.
T
h
e
ca
lcu
latio
n
o
f
th
e
item
s
o
f
p
er
f
o
r
m
an
ce
was m
ad
e
b
y
(
2
)
to
(
5
)
.
=
=
+
(
2
)
=
=
+
(
3
)
1
−
=
2
×
×
+
=
2
×
2
×
(
+
+
)
(
4
)
=
=
+
+
+
+
(
5
)
T
h
e
test
in
g
was
d
o
n
e
o
n
a
m
u
lti
-
d
o
m
ain
te
x
tu
al
d
ataset
th
at
co
n
s
is
ts
o
f
a
m
ix
t
u
r
e
o
f
twee
ts
,
f
o
r
m
a
l
s
en
ten
ce
s
,
an
d
d
ialo
g
u
es.
T
h
i
s
d
ataset
in
clu
d
es
I
SEAR,
W
ASSA,
an
d
E
m
o
tio
n
-
Sti
m
u
lu
s
,
with
all
o
f
th
em
co
v
er
in
g
a
b
r
o
ad
ar
ea
o
f
lin
g
u
is
tic
ex
p
r
ess
io
n
s
an
d
em
o
tio
n
a
l
s
tim
u
li.
T
h
e
p
r
e
-
tr
ain
ed
wo
r
d
em
b
ed
d
in
g
s
h
av
e
b
ee
n
u
s
ed
f
o
r
in
itializin
g
in
p
u
t
f
ea
tu
r
es.
I
n
p
ar
ticu
lar
,
it
r
ef
er
r
e
d
to
th
e
em
b
ed
d
i
n
g
l
ay
er
th
at
t
o
o
k
in
a
p
ad
d
ed
v
er
s
io
n
o
f
th
e
in
p
u
t
v
ec
to
r
(
1
8
2
1
0
,
3
0
0
)
to
allo
w
co
n
s
is
ten
t
r
ep
r
esen
tatio
n
o
f
d
ata
o
n
te
x
t
o
f
a
d
if
f
er
en
t
len
g
t
h
.
As
ca
n
b
e
o
b
s
er
v
ed
in
T
ab
le
2
,
th
e
p
r
o
p
o
s
ed
h
y
b
r
i
d
m
o
d
el,
b
ased
o
n
G
R
U,
B
iGR
U,
C
NN
,
W
OA,
an
d
ML
P,
p
er
f
o
r
m
s
b
e
tter
th
an
th
e
b
aselin
e
m
o
d
els
in
all
o
f
th
e
ev
alu
atio
n
m
etr
ics.
T
h
e
h
y
b
r
id
m
o
d
el
h
as
a
p
r
ec
is
io
n
o
f
8
5
.
3
4
%,
a
r
ec
all
o
f
8
3
.
9
8
%,
an
d
an
F1
-
s
co
r
e
o
f
8
4
.
3
4
%
th
at
cu
lm
in
ate
in
o
v
er
all
ac
cu
r
ac
y
o
f
8
3
.
7
6
%.
T
h
e
b
ase
m
o
d
el
(
C
NN+
B
iG
R
U+
SVM)
ac
h
ie
v
ed
a
p
r
ec
is
io
n
o
f
8
2
.
3
9
%,
r
ec
all
o
f
8
0
.
4
0
%,
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
Text
emo
tio
n
r
ec
o
g
n
itio
n
b
a
s
e
d
o
n
d
ee
p
lea
r
n
in
g
a
n
d
a
tten
ti
o
n
w
ith
w
h
a
le
…
(
Ta
b
a
r
ek
A
l
w
a
n
Tu
ib
)
3699
F1
-
s
co
r
e
o
f
8
1
.
2
7
%,
an
d
ac
cu
r
ac
y
o
f
8
0
.
1
1
%
[
2
8
]
.
T
h
ese
i
n
cr
em
en
tal
im
p
r
o
v
em
e
n
t
ev
id
en
ce
in
d
icate
s
h
o
w
well
th
e
h
y
b
r
i
d
f
r
am
ew
o
r
k
ca
n
b
e
u
s
ed
in
th
e
d
escr
ip
tio
n
o
f
em
o
tiv
e
d
etails
in
a
s
u
p
er
io
r
way
as
co
m
p
ar
e
d
with
in
d
iv
id
u
al
o
r
s
im
p
ler
co
m
p
o
s
ite
m
o
d
elin
g
.
Pre
v
io
u
s
s
tu
d
ies
h
av
e
also
in
v
esti
g
ated
d
ee
p
-
lear
n
i
n
g
an
d
h
y
b
r
id
ar
ch
itectu
r
es
f
o
r
tex
t
-
b
ased
em
o
tio
n
r
ec
o
g
n
itio
n
,
in
clu
d
in
g
s
ev
er
al
d
ee
p
-
lea
r
n
in
g
a
n
d
h
y
b
r
id
ap
p
r
o
ac
h
es
r
ep
o
r
ted
in
[
2
8
]
,
[
3
6
]
,
[
3
7
]
,
[
3
9
]
.
Ad
d
itio
n
al
r
elate
d
wo
r
k
h
as
ex
am
in
e
d
e
m
o
tio
n
r
ec
o
g
n
itio
n
th
r
o
u
g
h
f
ac
ial
im
ag
es,
E
E
G
s
ig
n
als,
r
u
le
-
b
ased
im
p
licit
-
tex
t
an
aly
s
is
,
c
o
n
v
er
s
atio
n
al
tr
an
s
f
er
lear
n
in
g
,
m
u
ltimo
d
al
a
u
d
io
-
te
x
t
m
o
d
el
in
g
,
g
r
ap
h
-
b
ased
co
n
v
er
s
atio
n
m
o
d
elin
g
,
an
d
b
r
o
ad
r
ev
i
ews
o
f
tex
t
-
b
ased
em
o
tio
n
d
etec
tio
n
[
4
0
]
–
[
4
6
]
.
T
ab
le
2
.
E
v
alu
atio
n
r
esu
lts
o
f
d
if
f
er
en
t m
o
d
els
M
o
d
e
l
P
r
e
c
i
s
i
o
n
R
e
c
a
l
l
F1
-
sc
o
r
e
A
c
c
u
r
a
c
y
G
R
U
7
8
.
3
7
7
8
.
9
4
7
8
.
6
5
7
8
.
0
2
B
i
G
R
U
8
0
.
6
2
7
9
.
6
4
8
0
.
0
9
7
9
.
4
6
C
N
N
8
2
.
1
2
7
9
.
9
2
8
0
.
7
6
7
9
.
3
2
C
N
N
+
B
i
G
R
U
+
S
V
M
(
B
a
se
a
r
t
i
c
l
e
[
2
8
]
)
8
2
.
3
9
8
0
.
4
0
8
1
.
2
7
8
0
.
1
1
G
R
U
+
B
i
G
R
U
+
C
N
N
+
W
O
A
+
M
LP (
P
r
o
p
o
se
d
m
e
t
h
o
d
)
8
5
.
2
3
8
3
.
9
8
8
4
.
3
4
8
3
.
7
6
T
h
e
p
r
o
p
o
s
ed
m
o
d
el
wo
r
k
s
w
ell
as
it
is
a
co
m
b
in
atio
n
o
f
co
m
p
lem
en
tar
y
elem
en
ts
o
f
DL
,
GR
U,
an
d
B
iG
R
U
h
elp
to
ca
p
tu
r
e
th
e
s
eq
u
en
tial
an
d
co
n
tex
tu
al
d
ep
e
n
d
en
cies
in
a
tex
t,
an
d
C
NN
lay
er
s
ex
tr
ac
t
lo
ca
l
s
em
an
tic
p
atter
n
s
.
T
h
e
atten
ti
o
n
m
ec
h
an
is
m
em
p
h
asizes
h
ig
h
lig
h
ts
em
o
tio
n
-
r
elev
a
n
t
to
k
en
s
,
an
d
W
OA
is
u
s
ed
f
o
r
f
ea
tu
r
e
s
elec
tio
n
t
h
at
ar
e
m
o
s
t
in
f
o
r
m
ativ
e
a
n
d
d
ec
r
ea
s
e
r
ed
u
n
d
an
c
y
.
T
h
e
s
elec
ted
f
ea
tu
r
es
a
r
e
th
e
n
class
if
ied
th
r
o
u
g
h
an
ML
P to
g
iv
e
s
tab
le
em
o
tio
n
p
r
ed
ictio
n
s
.
T
h
e
co
m
p
ar
ativ
e
b
ar
ch
ar
ts
(
Fig
u
r
e
4
)
p
r
o
v
e
its
s
u
p
er
io
r
ity
b
ec
a
u
s
e
th
e
im
p
r
o
v
em
en
t
is
alwa
y
s
o
b
s
er
v
e
d
in
th
e
ev
alu
atio
n
m
etr
ics.
T
h
e
f
in
d
in
g
s
s
h
o
w
ex
ce
llen
t
em
p
ir
ical
f
u
n
ctio
n
a
lity
,
as
well
as
th
e
p
r
o
s
p
ec
ts
o
f
m
u
lti
-
c
o
m
p
o
n
en
t
d
esig
n
s
to
ad
v
a
n
ce
T
E
R
,
esp
ec
ially
in
r
ea
l
-
wo
r
ld
ap
p
licatio
n
s
th
at
wo
u
ld
b
en
ef
it
f
r
o
m
s
u
b
tle
k
n
o
wled
g
e
o
f
af
f
ec
tiv
e
lan
g
u
a
g
e,
in
clu
d
in
g
s
en
tim
en
t
-
s
en
s
itiv
e
ch
atb
o
ts
,
m
en
tal
h
ea
lth
an
aly
s
is
,
an
d
o
p
in
i
o
n
-
f
i
n
d
in
g
s
y
s
tem
s
.
Fig
u
r
e
4
.
E
v
alu
atio
n
r
esu
lts
o
f
d
if
f
er
en
t
m
o
d
els
5.
CO
NCLU
SI
O
N
T
h
i
s
p
a
p
e
r
i
s
a
h
y
b
r
i
d
f
r
a
m
e
wo
r
k
t
h
a
t
u
s
e
s
M
L
a
n
d
D
L
t
o
d
e
v
e
l
o
p
a
te
x
t
-
b
a
s
e
d
e
m
o
t
i
o
n
r
e
c
o
g
n
i
t
i
o
n
m
o
d
e
l
.
T
h
e
t
h
r
ee
d
a
t
as
et
s
,
w
h
ic
h
a
r
e
i
n
cl
u
d
e
d
i
n
t
h
e
p
r
o
p
o
s
ed
m
o
d
e
l
,
a
r
e
E
m
o
t
i
o
n
-
s
t
i
m
u
l
u
s
d
a
ta
s
et
,
W
AS
SA
,
a
n
d
I
S
E
AR
.
I
t
s
a
d
a
p
t
a
b
il
i
t
y
e
n
s
u
r
e
s
t
h
a
t t
h
e
m
o
d
el
is
e
f
f
e
c
t
i
v
e
i
n
p
r
o
c
es
s
i
n
g
a
w
i
d
e
r
a
n
g
e
o
f
t
e
x
t i
n
p
u
t
s
(
t
we
e
ts
,
d
i
a
l
o
g
u
e
s
,
k
e
y
w
o
r
d
s
,
s
e
n
t
e
n
c
es
,
a
n
d
l
e
x
i
c
o
n
w
o
r
d
s
)
t
h
a
t
a
r
e
a
s
s
o
ci
a
t
e
d
wi
t
h
e
m
o
ti
o
n
s
.
C
o
m
p
a
r
i
s
o
n
o
f
f
i
n
d
i
n
g
s
i
n
t
h
e
p
r
o
p
o
s
e
d
m
e
t
h
o
d
a
n
d
b
a
s
e
l
i
n
e
m
o
d
e
l
,
a
s
s
h
o
w
n
i
n
t
h
e
g
i
v
e
n
t
a
b
l
e
,
i
n
d
i
c
at
e
s
s
i
g
n
i
f
ic
an
t
d
i
f
f
e
r
e
n
c
e
s
i
n
t
h
e
p
e
r
f
o
r
m
a
n
c
e
m
e
a
s
u
r
e
s
.
I
n
p
a
r
t
i
c
u
l
a
r
,
t
h
e
g
i
v
e
n
m
o
d
e
l
h
a
s
b
e
t
t
e
r
a
c
c
u
r
a
c
y
,
r
e
c
a
l
l
,
a
n
d
p
r
e
c
i
s
i
o
n
as
w
e
l
l
as
F
-
m
e
a
s
u
r
e
c
o
m
p
a
r
e
d
t
o
t
h
e
b
as
e
l
i
n
e
b
a
s
e
a
r
ti
c
l
e
m
o
d
el
.
T
h
is
s
i
g
n
i
f
ic
a
n
t
a
d
v
a
n
c
e
m
e
n
t
h
i
g
h
li
g
h
t
s
t
h
e
s
u
i
t
a
b
il
i
t
y
o
f
t
h
e
d
e
v
e
l
o
p
e
d
m
e
t
h
o
d
i
n
t
e
x
t
-
b
a
s
e
d
e
m
o
t
i
o
n
r
e
c
o
g
n
i
t
i
o
n
.
T
h
e
a
m
a
l
g
a
m
a
t
i
o
n
o
f
s
e
v
e
r
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
a
r
c
h
i
t
e
c
t
u
r
es
a
n
d
m
a
c
h
i
n
e
l
e
a
r
n
i
n
g
m
e
t
h
o
d
s
i
n
t
h
e
s
u
g
g
e
s
te
d
m
o
d
e
l
s
e
e
m
s
t
o
p
r
o
v
i
d
e
s
y
n
e
r
g
i
s
ti
c
a
d
v
a
n
t
a
g
es
,
as
77
78
79
80
81
82
83
84
85
86
CN
N
+
BiG
RU
+
SVM
(
B
a
s
e
ar
ticle
[4
6
])
G
RU
+
BiG
RU
+
CN
N
+
WO
A
+
MLP
(P
ro
p
o
se
d
m
e
th
o
d
)
P
e
rc
e
n
tag
e
M
o
d
el
P
re
c
isio
n
Re
c
a
ll
F
1
-sc
o
re
Acc
u
ra
cy
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.
15
,
No
.
4
,
Au
g
u
s
t 2
0
2
6
:
3
6
9
2
-
3
7
0
2
3700
o
n
e
m
i
g
h
t
g
a
i
n
a
m
o
r
e
i
n
-
d
ep
t
h
i
n
s
i
g
h
t
a
n
d
c
la
s
s
i
f
i
ca
t
i
o
n
o
f
e
m
o
t
i
o
n
a
l
c
o
n
t
e
n
t
wi
t
h
i
n
te
x
t
u
a
l
d
a
t
a
.
M
o
v
i
n
g
f
o
r
w
a
r
d
,
w
e
w
il
l
c
o
n
s
i
d
e
r
t
h
e
u
s
e
o
f
o
t
h
e
r
c
l
ass
i
f
i
e
r
s
o
r
e
n
s
em
b
l
e
a
p
p
r
o
a
c
h
e
s
f
o
r
i
m
p
r
o
v
i
n
g
p
e
r
f
o
r
m
a
n
c
e
e
v
e
n
f
u
r
t
h
e
r
.
I
n
t
h
e
a
r
e
a
o
f
D
L
,
t
h
e
c
o
m
b
i
n
a
t
i
o
n
o
f
t
h
e
B
i
GR
U
,
C
N
N
,
a
n
d
L
S
T
M
a
r
c
h
i
t
e
ct
u
r
e
s
w
il
l
b
e
e
x
p
l
o
r
e
d
.
M
o
r
e
o
v
e
r
,
t
h
e
d
i
f
f
e
r
e
n
c
e
s
i
n
th
e
s
t
r
u
c
t
u
r
e
o
f
t
h
e
s
e
n
t
e
n
ce
w
i
l
l
b
e
e
x
p
l
o
r
e
d
,
a
n
d
t
h
e
a
n
a
ly
s
i
s
o
f
t
h
e
r
e
g
i
o
n
a
l
l
a
n
g
u
a
g
e
s
wi
l
l
b
e
i
n
c
l
u
d
e
d
.
G
i
v
e
n
t
h
e
p
r
e
v
a
l
e
n
t
u
s
e
o
f
te
x
t
-
b
a
s
e
d
c
o
m
m
u
n
i
c
at
i
o
n
c
h
an
n
e
l
s
s
u
c
h
a
s
t
e
x
t
m
e
s
s
a
g
i
n
g
,
s
o
c
i
al
m
e
d
i
a
,
a
n
d
o
n
l
i
n
e
r
e
v
i
e
w
s
,
t
h
e
r
e
e
x
is
t
s
a
n
o
p
p
o
r
t
u
n
i
t
y
t
o
d
e
v
e
l
o
p
a
r
e
a
l
-
t
i
m
e
e
m
o
ti
o
n
r
e
c
o
g
n
i
t
i
o
n
m
o
d
e
l
c
a
p
a
b
l
e
o
f
d
i
s
c
e
r
n
i
n
g
p
e
o
p
l
e'
s
e
m
o
ti
o
n
s
a
n
d
m
o
o
d
s
f
r
o
m
v
a
s
t
d
a
t
a
s
t
r
e
a
m
s
.
F
UNDING
I
NF
O
R
M
A
T
I
O
N
T
h
is
r
esear
ch
was
f
u
n
d
ed
b
y
Facu
lty
I
n
co
m
e
Gen
er
atio
n
G
r
an
t
(
FTM
1
)
o
f
Facu
lty
o
f
I
n
f
o
r
m
atio
n
Scien
ce
an
d
T
ec
h
n
o
lo
g
y
,
Un
i
v
er
s
iti Ke
b
an
g
s
aa
n
Ma
lay
s
ia.
AUTHO
R
CO
NT
RI
B
UT
I
O
NS ST
A
T
E
M
E
N
T
T
h
is
jo
u
r
n
al
u
s
es
th
e
C
o
n
tr
ib
u
to
r
R
o
les
T
ax
o
n
o
m
y
(
C
R
ed
iT)
to
r
ec
o
g
n
ize
in
d
iv
id
u
al
au
th
o
r
co
n
tr
ib
u
tio
n
s
,
r
ed
u
ce
au
th
o
r
s
h
ip
d
is
p
u
tes,
an
d
f
ac
ilit
ate
co
llab
o
r
atio
n
.
Na
m
e
o
f
Aut
ho
r
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
T
ab
ar
ek
Alwan
T
u
ib
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
Mo
h
am
m
ad
-
R
ez
a
Feizi
-
Der
ak
h
s
h
i
✓
✓
✓
✓
✓
✓
✓
Yaq
d
h
an
Ma
h
m
o
o
d
Hu
s
s
ein
✓
✓
✓
✓
✓
✓
Fah
ad
T
ah
a
A
l
-
Dh
ief
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
C
:
C
o
n
c
e
p
t
u
a
l
i
z
a
t
i
o
n
M
:
M
e
t
h
o
d
o
l
o
g
y
So
:
So
f
t
w
a
r
e
Va
:
Va
l
i
d
a
t
i
o
n
Fo
:
Fo
r
mal
a
n
a
l
y
s
i
s
I
:
I
n
v
e
s
t
i
g
a
t
i
o
n
R
:
R
e
so
u
r
c
e
s
D
:
D
a
t
a
C
u
r
a
t
i
o
n
O
:
W
r
i
t
i
n
g
-
O
r
i
g
i
n
a
l
D
r
a
f
t
E
:
W
r
i
t
i
n
g
-
R
e
v
i
e
w
&
E
d
i
t
i
n
g
Vi
:
Vi
su
a
l
i
z
a
t
i
o
n
Su
:
Su
p
e
r
v
i
s
i
o
n
P
:
P
r
o
j
e
c
t
a
d
mi
n
i
st
r
a
t
i
o
n
Fu
:
Fu
n
d
i
n
g
a
c
q
u
i
si
t
i
o
n
CO
NF
L
I
C
T
O
F
I
N
T
E
R
E
S
T
ST
A
T
E
M
E
NT
Au
th
o
r
s
s
tate
n
o
co
n
f
lict o
f
in
t
er
est.
DATA AV
AI
L
AB
I
L
I
T
Y
T
h
e
d
a
t
a
a
v
a
i
l
a
b
i
l
i
t
y
i
s
n
o
t
a
p
p
l
i
c
a
b
l
e
t
o
t
h
i
s
p
a
p
e
r
a
s
n
o
n
e
w
d
a
t
a
w
e
r
e
c
r
e
a
t
e
d
o
r
a
n
a
l
y
z
e
d
i
n
t
h
i
s
s
t
u
d
y
.
RE
F
E
R
E
NC
E
S
[
1
]
S
.
Zh
o
u
a
n
d
M
.
M
o
h
d
,
“
M
e
n
t
a
l
h
e
a
l
t
h
safe
t
y
a
n
d
d
e
p
r
e
ss
i
o
n
d
e
t
e
c
t
i
o
n
i
n
s
o
c
i
a
l
me
d
i
a
t
e
x
t
d
a
t
a
:
a
c
l
a
ssi
f
i
c
a
t
i
o
n
a
p
p
r
o
a
c
h
b
a
se
d
o
n
a
d
e
e
p
l
e
a
r
n
i
n
g
m
o
d
e
l
,
”
I
E
EE
A
c
c
e
ss
,
v
o
l
.
1
3
,
p
p
.
6
3
2
8
4
–
6
3
2
9
7
,
2
0
2
5
,
d
o
i
:
1
0
.
1
1
0
9
/
A
C
C
ESS
.
2
0
2
5
.
3
5
5
9
1
7
0
.
[
2
]
S
.
K
.
H
a
m
e
d
,
M
.
J.
A
.
A
z
i
z
,
a
n
d
M
.
R
.
Y
a
a
k
u
b
,
“
F
a
k
e
n
e
w
s
d
e
t
e
c
t
i
o
n
mo
d
e
l
o
n
s
o
c
i
a
l
m
e
d
i
a
b
y
l
e
v
e
r
a
g
i
n
g
s
e
n
t
i
m
e
n
t
a
n
a
l
y
si
s
o
f
n
e
w
s
c
o
n
t
e
n
t
a
n
d
e
m
o
t
i
o
n
a
n
a
l
y
si
s o
f
u
sers’
c
o
mm
e
n
t
s,
”
S
e
n
s
o
rs
,
v
o
l
.
2
3
,
n
o
.
4
,
F
e
b
.
2
0
2
3
,
d
o
i
:
1
0
.
3
3
9
0
/
s
2
3
0
4
1
7
4
8
.
[
3
]
S
.
Za
i
n
u
d
i
n
,
“
S
e
n
t
i
me
n
t
a
n
a
l
y
s
i
s
o
n
I
n
d
o
n
e
s
i
a
n
C
O
V
I
D
-
1
9
p
a
n
d
e
m
i
c
b
a
se
d
o
n
2
0
2
0
Tw
i
t
t
e
r
d
a
t
a
,
”
Asi
a
-
P
a
c
i
f
i
c
J
o
u
r
n
a
l
o
f
I
n
f
o
rm
a
t
i
o
n
T
e
c
h
n
o
l
o
g
y
a
n
d
M
u
l
t
i
m
e
d
i
a
,
v
o
l
.
1
4
,
n
o
.
0
2
,
p
p
.
4
3
8
–
4
5
2
,
D
e
c
.
2
0
2
5
,
d
o
i
:
1
0
.
1
7
5
7
6
/
a
p
j
i
t
m
-
2
0
2
5
-
1
4
0
2
-
0
6
.
[
4
]
C
.
R
.
C
h
o
p
a
d
e
,
“
T
e
x
t
-
b
a
se
d
e
mo
t
i
o
n
r
e
c
o
g
n
i
t
i
o
n
:
a
su
r
v
e
y
,
”
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
rn
a
l
o
f
S
c
i
e
n
c
e
a
n
d
R
e
se
a
rc
h
,
v
o
l
.
4
,
n
o
.
6
,
p
p
.
4
0
9
–
4
1
4
,
2
0
1
5
.
[
5
]
P
.
Ek
m
a
n
,
“
B
a
s
i
c
e
mo
t
i
o
n
s,”
i
n
H
a
n
d
b
o
o
k
o
f
C
o
g
n
i
t
i
o
n
a
n
d
Em
o
t
i
o
n
,
C
h
i
c
h
e
s
t
e
r
,
U
n
i
t
e
d
K
i
n
g
d
o
m:
W
i
l
e
y
,
1
9
9
9
,
p
p
.
4
5
–
6
0
,
d
o
i
:
1
0
.
1
0
0
2
/
0
4
7
0
0
1
3
4
9
4
.
c
h
3
.
[
6
]
W
.
W
u
n
d
t
,
O
u
t
l
i
n
e
s
o
f
Ps
y
c
h
o
l
o
g
y
,
v
o
l
.
1
,
n
o
.
5
4
4
5
.
L
e
i
p
z
i
g
,
G
e
r
ma
n
y
:
W
i
l
h
e
l
m
E
n
g
e
l
ma
n
n
,
1
9
0
7
,
d
o
i
:
1
0
.
1
1
3
6
/
b
mj
.
1
.
5
4
4
5
.
1
3
0
2
-
b.
[
7
]
R
.
P
l
u
t
c
h
i
k
,
“
T
h
e
n
a
t
u
re
o
f
e
m
o
t
i
o
n
s
,
”
A
meri
c
a
n
S
c
i
e
n
t
i
s
t
,
v
o
l
.
8
9
,
n
o
.
4
,
p
p
.
3
4
4
–
3
5
0
,
2
0
0
1
,
d
o
i
:
1
0
.
1
5
1
1
/
2
0
0
1
.
2
8
.
3
4
4
.
[
8
]
M
.
M
.
K
a
t
a
a
a
n
d
W
.
K
a
u
r
,
“
R
e
c
o
g
n
i
z
i
n
g
f
a
c
i
a
l
e
mo
t
i
o
n
i
n
r
e
a
l
-
t
i
me
u
si
n
g
M
u
W
N
e
t
a
n
o
v
e
l
d
e
e
p
l
e
a
r
n
i
n
g
n
e
t
w
o
r
k
,
”
Asi
a
-
P
a
c
i
f
i
c
J
o
u
rn
a
l
o
f
I
n
f
o
rm
a
t
i
o
n
T
e
c
h
n
o
l
o
g
y
a
n
d
M
u
l
t
i
m
e
d
i
a
,
v
o
l
.
1
3
,
n
o
.
0
1
,
p
p
.
0
1
–
2
0
,
J
u
n
.
2
0
2
4
,
d
o
i
:
1
0
.
1
7
5
7
6
/
a
p
j
i
t
m
-
2
0
2
4
-
1
3
0
1
-
0
1
.
[
9
]
A
.
A
.
N
a
f
e
a
e
t
a
l
.
,
“
A
b
r
i
e
f
r
e
v
i
e
w
o
n
p
r
e
p
r
o
c
e
ss
i
n
g
t
e
x
t
i
n
A
r
a
b
i
c
l
a
n
g
u
a
g
e
d
a
t
a
s
e
t
:
t
e
c
h
n
i
q
u
e
s
a
n
d
c
h
a
l
l
e
n
g
e
s,
”
B
a
b
y
l
o
n
i
a
n
J
o
u
rn
a
l
o
f
Art
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
,
v
o
l
.
2
0
2
4
,
p
p
.
4
6
–
5
3
,
M
a
y
2
0
2
4
,
d
o
i
:
1
0
.
5
8
4
9
6
/
B
JA
I
/
2
0
2
4
/
0
0
7
.
[
1
0
]
G
.
S
a
l
t
o
n
a
n
d
C
.
B
u
c
k
l
e
y
,
“
Te
r
m
-
w
e
i
g
h
t
i
n
g
a
p
p
r
o
a
c
h
e
s
i
n
a
u
t
o
mat
i
c
t
e
x
t
r
e
t
r
i
e
v
a
l
,
”
I
n
f
o
rm
a
t
i
o
n
Pr
o
c
e
ssi
n
g
&
Ma
n
a
g
e
m
e
n
t
,
v
o
l
.
2
4
,
n
o
.
5
,
p
p
.
5
1
3
–
5
2
3
,
J
a
n
.
1
9
8
8
,
d
o
i
:
1
0
.
1
0
1
6
/
0
3
0
6
-
4
5
7
3
(
8
8
)
9
0
0
2
1
-
0.
[
1
1
]
D
.
S
e
a
l
,
U
.
K
.
R
o
y
,
a
n
d
R
.
B
a
s
a
k
,
“
S
e
n
t
e
n
c
e
-
l
e
v
e
l
e
m
o
t
i
o
n
d
e
t
e
c
t
i
o
n
f
r
o
m
t
e
x
t
b
a
s
e
d
o
n
sem
a
n
t
i
c
r
u
l
e
s,
”
i
n
I
n
f
o
rm
a
t
i
o
n
a
n
d
C
o
m
m
u
n
i
c
a
t
i
o
n
T
e
c
h
n
o
l
o
g
y
f
o
r
S
u
s
t
a
i
n
a
b
l
e
D
e
v
e
l
o
p
m
e
n
t
,
S
i
n
g
a
p
o
r
e
:
S
p
r
i
n
g
e
r
N
a
t
u
r
e
,
2
0
2
0
,
p
p
.
4
2
3
–
430
,
d
o
i
:
1
0
.
1
0
0
7
/
9
7
8
-
9
8
1
-
13
-
7
1
6
6
-
0
_
4
2
.
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
Text
emo
tio
n
r
ec
o
g
n
itio
n
b
a
s
e
d
o
n
d
ee
p
lea
r
n
in
g
a
n
d
a
tten
ti
o
n
w
ith
w
h
a
le
…
(
Ta
b
a
r
ek
A
l
w
a
n
Tu
ib
)
3701
[
1
2
]
K
.
R
.
S
c
h
e
r
e
r
a
n
d
H
.
G
.
W
a
l
l
b
o
t
t
,
“
E
v
i
d
e
n
c
e
f
o
r
u
n
i
v
e
r
s
a
l
i
t
y
a
n
d
c
u
l
t
u
r
a
l
v
a
r
i
a
t
i
o
n
o
f
d
i
f
f
e
r
e
n
t
i
a
l
e
mo
t
i
o
n
r
e
s
p
o
n
s
e
p
a
t
t
e
r
n
i
n
g
,
”
J
o
u
rn
a
l
o
f
Pe
rs
o
n
a
l
i
t
y
a
n
d
S
o
c
i
a
l
Psy
c
h
o
l
o
g
y
,
v
o
l
.
6
6
,
n
o
.
2
,
p
p
.
3
1
0
–
3
2
8
,
1
9
9
4
,
d
o
i
:
1
0
.
1
0
3
7
/
0
0
2
2
-
3
5
1
4
.
6
6
.
2
.
3
1
0
.
[
1
3
]
F
.
M
.
A
l
o
t
a
i
b
i
,
“
C
l
a
ss
i
f
y
i
n
g
t
e
x
t
-
b
a
sed
e
m
o
t
i
o
n
s
u
si
n
g
l
o
g
i
st
i
c
r
e
g
r
e
ssi
o
n
,
”
VA
WK
U
M
T
r
a
n
sa
c
t
i
o
n
s
o
n
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
s
,
p
p
.
3
1
–
3
7
,
A
p
r
.
2
0
1
9
,
d
o
i
:
1
0
.
2
1
0
1
5
/
v
t
c
s.
v
1
6
i
2
.
5
5
1
.
[
1
4
]
P
.
X
u
,
A
.
M
a
d
o
t
t
o
,
C
.
S
.
W
u
,
J
.
H
.
P
a
r
k
,
a
n
d
P
.
F
u
n
g
,
“
Em
o
2
V
e
c
:
l
e
a
r
n
i
n
g
g
e
n
e
r
a
l
i
z
e
d
e
m
o
t
i
o
n
r
e
p
r
e
s
e
n
t
a
t
i
o
n
b
y
m
u
l
t
i
-
t
a
sk
t
r
a
i
n
i
n
g
,
”
i
n
9
t
h
W
o
r
k
sh
o
p
o
n
C
o
m
p
u
t
a
t
i
o
n
a
l
Ap
p
r
o
a
c
h
e
s
t
o
S
u
b
j
e
c
t
i
v
i
t
y
,
S
e
n
t
i
m
e
n
t
a
n
d
S
o
c
i
a
l
Me
d
i
a
An
a
l
y
s
i
s
,
2
0
1
8
,
p
p
.
2
9
2
–
298
,
d
o
i
:
1
0
.
1
8
6
5
3
/
v
1
/
W
1
8
-
6
2
4
3
.
[
1
5
]
W
.
R
a
g
h
e
b
,
J.
A
z
é
,
S
.
B
r
i
n
g
a
y
,
a
n
d
M
.
S
e
r
v
a
j
e
a
n
,
“
LI
R
M
M
-
a
d
v
a
n
se
a
t
S
e
mEv
a
l
-
2
0
1
9
Ta
s
k
3
:
a
t
t
e
n
t
i
v
e
c
o
n
v
e
r
sat
i
o
n
m
o
d
e
l
i
n
g
f
o
r
e
mo
t
i
o
n
d
e
t
e
c
t
i
o
n
a
n
d
c
l
a
s
si
f
i
c
a
t
i
o
n
,
”
i
n
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
1
3
t
h
I
n
t
e
rn
a
t
i
o
n
a
l
Wo
r
k
s
h
o
p
o
n
S
e
m
a
n
t
i
c
E
v
a
l
u
a
t
i
o
n
,
2
0
1
9
,
p
p
.
2
5
1
–
255
,
d
o
i
:
1
0
.
1
8
6
5
3
/
v
1
/
S
1
9
-
2
0
4
2
.
[
1
6
]
M
.
S
u
h
a
si
n
i
a
n
d
S
.
B
a
d
u
g
u
,
“
Tw
o
s
t
e
p
a
p
p
r
o
a
c
h
f
o
r
e
m
o
t
i
o
n
d
e
t
e
c
t
i
o
n
o
n
Tw
i
t
t
e
r
d
a
t
a
,
”
I
n
t
e
rn
a
t
i
o
n
a
l
J
o
u
r
n
a
l
o
f
C
o
m
p
u
t
e
r
Ap
p
l
i
c
a
t
i
o
n
s
,
v
o
l
.
1
7
9
,
n
o
.
5
3
,
p
p
.
1
2
–
1
9
,
J
u
n
.
2
0
1
8
,
d
o
i
:
1
0
.
5
1
2
0
/
i
j
c
a
2
0
1
8
9
1
7
3
5
0
.
[
1
7
]
M
.
H
a
sa
n
,
E
.
R
u
n
d
e
n
st
e
i
n
e
r
,
a
n
d
E
.
A
g
u
,
“
A
u
t
o
ma
t
i
c
e
m
o
t
i
o
n
d
e
t
e
c
t
i
o
n
i
n
t
e
x
t
s
t
r
e
a
ms
b
y
a
n
a
l
y
z
i
n
g
Tw
i
t
t
e
r
d
a
t
a
,
”
I
n
t
e
rn
a
t
i
o
n
a
l
J
o
u
rn
a
l
o
f
D
a
t
a
S
c
i
e
n
c
e
a
n
d
An
a
l
y
t
i
c
s
,
v
o
l
.
7
,
n
o
.
1
,
p
p
.
3
5
–
5
1
,
F
e
b
.
2
0
1
9
,
d
o
i
:
1
0
.
1
0
0
7
/
s
4
1
0
6
0
-
018
-
0
0
9
6
-
z.
[
1
8
]
A
.
R
o
d
r
i
g
u
e
z
,
Y
.
L.
C
h
e
n
,
a
n
d
C
.
A
r
g
u
e
t
a
,
“
F
r
a
mew
o
r
k
f
o
r
d
e
t
e
c
t
i
o
n
a
n
d
i
n
t
e
g
r
a
t
i
o
n
o
f
u
n
s
t
r
u
c
t
u
r
e
d
d
a
t
a
o
f
h
a
t
e
s
p
e
e
c
h
o
n
F
a
c
e
b
o
o
k
u
si
n
g
se
n
t
i
me
n
t
a
n
d
e
m
o
t
i
o
n
a
n
a
l
y
s
i
s,
”
I
EEE
Ac
c
e
ss
,
v
o
l
.
1
0
,
p
p
.
2
2
4
0
0
–
2
2
4
1
9
,
2
0
2
2
,
d
o
i
:
1
0
.
1
1
0
9
/
A
C
C
ESS
.
2
0
2
2
.
3
1
5
1
0
9
8
.
[
1
9
]
L.
C
a
o
,
S
.
P
e
n
g
,
P
.
Y
i
n
,
Y
.
Zh
o
u
,
A
.
Y
a
n
g
,
a
n
d
X
.
Li
,
“
A
su
r
v
e
y
o
f
e
mo
t
i
o
n
a
n
a
l
y
si
s
i
n
t
e
x
t
b
a
se
d
o
n
d
e
e
p
l
e
a
r
n
i
n
g
,
”
i
n
2
0
2
0
I
EE
E
8
t
h
I
n
t
e
rn
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
o
n
S
m
a
rt
C
i
t
y
a
n
d
I
n
f
o
rm
a
t
i
z
a
t
i
o
n
(
i
S
C
I
)
,
D
e
c
.
2
0
2
0
,
p
p
.
8
1
–
88
,
d
o
i
:
1
0
.
1
1
0
9
/
i
S
C
I
5
0
6
9
4
.
2
0
2
0
.
0
0
0
2
0
.
[
2
0
]
F
.
A
.
A
c
h
e
a
mp
o
n
g
,
C
.
W
e
n
y
u
,
a
n
d
H
.
N
.
M
e
n
sa
h
,
“
T
e
x
t
‐
b
a
se
d
e
m
o
t
i
o
n
d
e
t
e
c
t
i
o
n
:
a
d
v
a
n
c
e
s
,
c
h
a
l
l
e
n
g
e
s,
a
n
d
o
p
p
o
r
t
u
n
i
t
i
e
s,”
En
g
i
n
e
e
ri
n
g
Re
p
o
rt
s
,
v
o
l
.
2
,
n
o
.
7
,
J
u
l
.
2
0
2
0
,
d
o
i
:
1
0
.
1
0
0
2
/
e
n
g
2
.
1
2
1
8
9
.
[
2
1
]
A
.
S
.
N
a
v
a
r
r
e
t
e
,
C
.
M
.
A
r
a
n
e
d
a
,
C
.
V
.
C
a
st
r
o
,
a
n
d
C
.
R
.
M
a
n
z
a
n
o
,
“
A
n
o
v
e
l
a
p
p
r
o
a
c
h
t
o
t
h
e
c
r
e
a
t
i
o
n
o
f
a
l
a
b
e
l
l
i
n
g
l
e
x
i
c
o
n
f
o
r
i
mp
r
o
v
i
n
g
e
m
o
t
i
o
n
a
n
a
l
y
s
i
s i
n
t
e
x
t
,
”
T
h
e
El
e
c
t
r
o
n
i
c
L
i
b
r
a
ry
,
v
o
l
.
3
9
,
n
o
.
1
,
p
p
.
1
1
8
–
1
3
6
,
2
0
2
1
,
d
o
i
:
1
0
.
1
1
0
8
/
EL
-
04
-
2
0
2
0
-
0
1
1
0
.
[
2
2
]
K
.
S
a
i
l
u
n
a
z
a
n
d
R
.
A
l
h
a
j
j
,
“
Emo
t
i
o
n
a
n
d
se
n
t
i
me
n
t
a
n
a
l
y
si
s
f
r
o
m
Tw
i
t
t
e
r
t
e
x
t
,
”
J
o
u
r
n
a
l
o
f
C
o
m
p
u
t
a
t
i
o
n
a
l
S
c
i
e
n
c
e
,
v
o
l
.
3
6
,
S
e
p
.
2
0
1
9
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
j
o
c
s.
2
0
1
9
.
0
5
.
0
0
9
.
[
2
3
]
S
.
M
o
h
a
mm
a
d
a
n
d
F
.
B
.
M
a
r
q
u
e
z
,
“
W
A
S
S
A
-
2
0
1
7
sh
a
r
e
d
t
a
s
k
o
n
e
m
o
t
i
o
n
i
n
t
e
n
si
t
y
,
”
i
n
P
ro
c
e
e
d
i
n
g
s
o
f
t
h
e
8
t
h
W
o
rks
h
o
p
o
n
C
o
m
p
u
t
a
t
i
o
n
a
l
Ap
p
r
o
a
c
h
e
s
t
o
S
u
b
j
e
c
t
i
v
i
t
y
,
S
e
n
t
i
m
e
n
t
a
n
d
S
o
c
i
a
l
Me
d
i
a
An
a
l
y
si
s
,
2
0
1
7
,
p
p
.
3
4
–
49
,
d
o
i
:
1
0
.
1
8
6
5
3
/
v
1
/
W
1
7
-
5
2
0
5
.
[
2
4
]
D
.
G
h
a
z
i
,
D
.
I
n
k
p
e
n
,
a
n
d
S
.
S
z
p
a
k
o
w
i
c
z
,
“
D
e
t
e
c
t
i
n
g
e
mo
t
i
o
n
st
i
mu
l
i
i
n
e
mo
t
i
o
n
-
b
e
a
r
i
n
g
se
n
t
e
n
c
e
s,”
i
n
C
o
m
p
u
t
a
t
i
o
n
a
l
L
i
n
g
u
i
s
t
i
c
s
a
n
d
I
n
t
e
l
l
i
g
e
n
t
T
e
x
t
P
ro
c
e
ss
i
n
g
,
C
h
a
m,
S
w
i
t
z
e
r
l
a
n
d
:
S
p
r
i
n
g
e
r
,
2
0
1
5
,
p
p
.
1
5
2
–
1
6
5
,
d
o
i
:
1
0
.
1
0
0
7
/
9
7
8
-
3
-
3
1
9
-
1
8
1
1
7
-
2
_
1
2
.
[
2
5
]
A
.
Est
a
b
r
o
o
k
s
a
n
d
N
.
Ja
p
k
o
w
i
c
z
,
“
A
mi
x
t
u
r
e
-
of
-
e
x
p
e
r
t
s
f
r
a
mew
o
r
k
f
o
r
l
e
a
r
n
i
n
g
f
r
o
m
i
m
b
a
l
a
n
c
e
d
d
a
t
a
se
t
s,
”
i
n
A
d
v
a
n
c
e
s
i
n
I
n
t
e
l
l
i
g
e
n
t
D
a
t
a
An
a
l
y
si
s
,
B
e
r
l
i
n
,
G
e
r
man
y
:
S
p
r
i
n
g
e
r
,
2
0
0
1
,
p
p
.
3
4
–
43
,
d
o
i
:
1
0
.
1
0
0
7
/
3
-
540
-
4
4
8
1
6
-
0
_
4
.
[
2
6
]
T.
M
i
k
o
l
o
v
,
I
.
S
u
t
s
k
e
v
e
r
,
K
.
C
h
e
n
,
G
.
C
o
r
r
a
d
o
,
a
n
d
J.
D
e
a
n
,
“
D
i
s
t
r
i
b
u
t
e
d
r
e
p
r
e
s
e
n
t
a
t
i
o
n
s
o
f
w
o
r
d
s
a
n
d
p
h
r
a
ses
a
n
d
t
h
e
i
r
c
o
m
p
o
si
t
i
o
n
a
l
i
t
y
,
”
i
n
2
7
t
h
I
n
t
e
rn
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
o
n
N
e
u
ra
l
I
n
f
o
rm
a
t
i
o
n
Pro
c
e
ssi
n
g
S
y
s
t
e
m
s (
N
I
PS
’
1
3
)
,
2
0
1
3
,
p
p
.
3
1
1
1
–
3
1
1
9
.
[
2
7
]
S
.
K
.
B
h
a
r
t
i
e
t
a
l
.
,
“
T
e
x
t
-
b
a
se
d
e
mo
t
i
o
n
r
e
c
o
g
n
i
t
i
o
n
u
s
i
n
g
d
e
e
p
l
e
a
r
n
i
n
g
a
p
p
r
o
a
c
h
,
”
C
o
m
p
u
t
a
t
i
o
n
a
l
I
n
t
e
l
l
i
g
e
n
c
e
a
n
d
N
e
u
ro
s
c
i
e
n
c
e
,
p
p
.
1
–
8
,
A
u
g
.
2
0
2
2
,
d
o
i
:
1
0
.
1
1
5
5
/
2
0
2
2
/
2
6
4
5
3
8
1
.
[
2
8
]
N
.
A
.
M
o
h
a
m
e
d
,
M
.
A
.
Z
u
l
k
i
f
l
e
y
,
A
.
A
.
I
b
r
a
h
i
m,
a
n
d
M
.
A
o
u
a
c
h
e
,
“
O
p
t
i
ma
l
t
r
a
i
n
i
n
g
c
o
n
f
i
g
u
r
a
t
i
o
n
s
o
f
a
C
N
N
-
LSTM
-
b
a
se
d
t
r
a
c
k
e
r
f
o
r
a
f
a
l
l
f
r
a
m
e
d
e
t
e
c
t
i
o
n
s
y
st
e
m,”
S
e
n
s
o
rs
,
v
o
l
.
2
1
,
n
o
.
1
9
,
S
e
p
.
2
0
2
1
,
d
o
i
:
1
0
.
3
3
9
0
/
s2
1
1
9
6
4
8
5
.
[
2
9
]
K
.
C
h
o
e
t
a
l
.
,
“
L
e
a
r
n
i
n
g
p
h
r
a
se
r
e
p
r
e
se
n
t
a
t
i
o
n
s
u
si
n
g
R
N
N
e
n
c
o
d
e
r
-
d
e
c
o
d
e
r
f
o
r
s
t
a
t
i
st
i
c
a
l
ma
c
h
i
n
e
t
r
a
n
s
l
a
t
i
o
n
,
”
i
n
2
0
1
4
C
o
n
f
e
re
n
c
e
o
n
Em
p
i
ri
c
a
l
Me
t
h
o
d
s
i
n
N
a
t
u
r
a
l
L
a
n
g
u
a
g
e
Pro
c
e
ss
i
n
g
(
EM
N
L
P)
,
2
0
1
4
,
p
p
.
1
7
2
4
–
1
7
3
4
,
d
o
i
:
1
0
.
3
1
1
5
/
v
1
/
D
1
4
-
1
1
7
9
.
[
3
0
]
M
.
S
c
h
u
st
e
r
a
n
d
K
.
K
.
P
a
l
i
w
a
l
,
“
B
i
d
i
r
e
c
t
i
o
n
a
l
r
e
c
u
r
r
e
n
t
n
e
u
r
a
l
n
e
t
w
o
r
k
s
,
”
I
EEE
T
r
a
n
sa
c
t
i
o
n
s
o
n
S
i
g
n
a
l
Pr
o
c
e
ssi
n
g
,
v
o
l
.
4
5
,
n
o
.
1
1
,
p
p
.
2
6
7
3
–
2
6
8
1
,
1
9
9
7
,
d
o
i
:
1
0
.
1
1
0
9
/
7
8
.
6
5
0
0
9
3
.
[
3
1
]
Y
.
K
i
m,
“
C
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
s
f
o
r
s
e
n
t
e
n
c
e
c
l
a
ssi
f
i
c
a
t
i
o
n
,
”
i
n
2
0
1
4
C
o
n
f
e
r
e
n
c
e
o
n
Em
p
i
ri
c
a
l
M
e
t
h
o
d
s
i
n
N
a
t
u
r
a
l
L
a
n
g
u
a
g
e
Pr
o
c
e
ssi
n
g
(
EM
N
L
P)
,
2
0
1
4
,
p
p
.
1
7
4
6
–
1
7
5
1
,
d
o
i
:
1
0
.
3
1
1
5
/
v
1
/
D
1
4
-
1
1
8
1
.
[
3
2
]
A
.
V
a
sw
a
n
i
e
t
a
l
.
,
“
A
t
t
e
n
t
i
o
n
i
s
a
l
l
y
o
u
n
e
e
d
,
”
A
d
v
a
n
c
e
s
i
n
N
e
u
r
a
l
I
n
f
o
rm
a
t
i
o
n
Pr
o
c
e
ssi
n
g
S
y
s
t
e
m
s
,
p
p
.
5
9
9
9
–
6
0
0
9
,
2
0
1
7
,
d
o
i
:
1
0
.
1
2
0
1
/
9
7
8
1
0
0
3
5
6
1
4
6
0
-
19.
[
3
3
]
S
.
M
i
r
j
a
l
i
l
i
a
n
d
A
.
Le
w
i
s
,
“
Th
e
w
h
a
l
e
o
p
t
i
mi
z
a
t
i
o
n
a
l
g
o
r
i
t
h
m
,
”
Ad
v
a
n
c
e
s i
n
En
g
i
n
e
e
ri
n
g
S
o
f
t
w
a
re
,
v
o
l
.
9
5
,
p
p
.
5
1
–
6
7
,
M
a
y
2
0
1
6
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
a
d
v
e
n
g
so
f
t
.
2
0
1
6
.
0
1
.
0
0
8
.
[
3
4
]
S
.
R
.
A
h
m
a
d
,
A
.
A
.
B
a
k
a
r
,
a
n
d
M
.
R
.
Y
a
a
k
u
b
,
“
A
n
t
c
o
l
o
n
y
o
p
t
i
m
i
z
a
t
i
o
n
f
o
r
t
e
x
t
f
e
a
t
u
r
e
se
l
e
c
t
i
o
n
i
n
se
n
t
i
me
n
t
a
n
a
l
y
si
s
,
”
I
n
t
e
l
l
i
g
e
n
t
D
a
t
a
An
a
l
y
si
s
,
v
o
l
.
2
3
,
n
o
.
1
,
p
p
.
1
3
3
–
1
5
8
,
F
e
b
.
2
0
1
9
,
d
o
i
:
1
0
.
3
2
3
3
/
I
D
A
-
1
7
3
7
4
0
.
[
3
5
]
S
.
G
i
t
e
e
t
a
l
.
,
“
Te
x
t
u
a
l
f
e
a
t
u
r
e
e
x
t
r
a
c
t
i
o
n
u
si
n
g
a
n
t
c
o
l
o
n
y
o
p
t
i
m
i
z
a
t
i
o
n
f
o
r
h
a
t
e
s
p
e
e
c
h
c
l
a
ss
i
f
i
c
a
t
i
o
n
,
”
Bi
g
D
a
t
a
a
n
d
C
o
g
n
i
t
i
v
e
C
o
m
p
u
t
i
n
g
,
v
o
l
.
7
,
n
o
.
1
,
M
a
r
.
2
0
2
3
,
d
o
i
:
1
0
.
3
3
9
0
/
b
d
c
c
7
0
1
0
0
4
5
.
[
3
6
]
A
.
C
h
o
w
a
n
d
a
,
R
.
S
u
t
o
y
o
,
M
e
i
l
i
a
n
a
,
a
n
d
S
.
T
a
n
a
c
h
u
t
i
w
a
t
,
“
E
x
p
l
o
r
i
n
g
t
e
x
t
-
b
a
s
e
d
e
m
o
t
i
o
n
s
r
e
c
o
g
n
i
t
i
o
n
m
a
c
h
i
n
e
l
e
a
r
n
i
n
g
t
e
c
h
n
i
q
u
e
s
o
n
s
o
c
i
a
l
m
e
d
i
a
c
o
n
v
e
r
s
a
t
i
o
n
,
”
Pr
o
c
e
d
i
a
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
,
v
o
l
.
1
7
9
,
p
p
.
8
2
1
–
8
2
8
,
2
0
2
1
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
p
r
o
c
s
.
2
0
2
1
.
0
1
.
0
9
9
.
[
3
7
]
A
.
R
.
A
b
a
s,
I
.
El
h
e
n
a
w
y
,
M
.
Z
i
d
a
n
,
a
n
d
M
.
O
t
h
ma
n
,
“
B
ER
T
-
C
N
N
:
a
d
e
e
p
l
e
a
r
n
i
n
g
mo
d
e
l
f
o
r
d
e
t
e
c
t
i
n
g
e
mo
t
i
o
n
s
f
r
o
m
t
e
x
t
,
”
C
o
m
p
u
t
e
rs
,
M
a
t
e
r
i
a
l
s &
C
o
n
t
i
n
u
a
,
v
o
l
.
7
1
,
n
o
.
2
,
p
p
.
2
9
4
3
–
2
9
6
1
,
2
0
2
2
,
d
o
i
:
1
0
.
3
2
6
0
4
/
c
m
c
.
2
0
2
2
.
0
2
1
6
7
1
.
[
3
8
]
D
.
K
r
st
i
n
i
ć
,
M
.
B
r
a
o
v
i
ć
,
L.
Š
e
r
i
ć
,
a
n
d
D
.
B
.
Š
t
u
l
i
ć
,
“
M
u
l
t
i
-
l
a
b
e
l
c
l
a
ss
i
f
i
e
r
p
e
r
f
o
r
ma
n
c
e
e
v
a
l
u
a
t
i
o
n
w
i
t
h
c
o
n
f
u
si
o
n
ma
t
r
i
x
,
”
i
n
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
& In
f
o
rm
a
t
i
o
n
T
e
c
h
n
o
l
o
g
y
,
J
u
n
.
2
0
2
0
,
p
p
.
0
1
–
14
,
d
o
i
:
1
0
.
5
1
2
1
/
c
si
t
.
2
0
2
0
.
1
0
0
8
0
1
.
[
3
9
]
N
.
S
h
e
l
k
e
,
S
.
C
h
a
u
d
h
u
r
y
,
S
.
C
h
a
k
r
a
b
a
r
t
i
,
S
.
L
.
B
a
n
g
a
r
e
,
G
.
Y
o
g
a
p
r
i
y
a
,
a
n
d
P
.
P
a
n
d
e
y
,
“
A
n
e
f
f
i
c
i
e
n
t
w
a
y
o
f
t
e
x
t
-
b
a
se
d
e
m
o
t
i
o
n
a
n
a
l
y
s
i
s
f
r
o
m
s
o
c
i
a
l
m
e
d
i
a
u
s
i
n
g
L
R
A
-
D
N
N
,
”
N
e
u
r
o
s
c
i
e
n
c
e
I
n
f
o
rm
a
t
i
c
s
,
v
o
l
.
2
,
n
o
.
3
,
S
e
p
.
2
0
2
2
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
n
e
u
r
i
.
2
0
2
2
.
1
0
0
0
4
8
.
[
4
0
]
M
.
I
.
H
.
A
l
z
a
w
a
l
i
e
t
a
l
.
,
“
F
a
c
i
a
l
e
m
o
t
i
o
n
i
ma
g
e
s
r
e
c
o
g
n
i
t
i
o
n
b
a
s
e
d
o
n
b
i
n
a
r
i
z
e
d
g
e
n
e
t
i
c
a
l
g
o
r
i
t
h
m
-
r
a
n
d
o
m
f
o
r
e
st
,
”
Ba
g
h
d
a
d
S
c
i
e
n
c
e
J
o
u
rn
a
l
,
v
o
l
.
2
1
,
n
o
.
2
(
S
I
)
,
F
e
b
.
2
0
2
4
,
d
o
i
:
1
0
.
2
1
1
2
3
/
b
s
j
.
2
0
2
4
.
9
6
9
8
.
[
4
1
]
T.
A
.
T
u
i
b
,
B
.
H
.
S
a
o
u
d
i
,
Y
.
M
.
H
u
s
sei
n
,
T
.
H
.
M
a
n
d
e
e
l
,
a
n
d
F
.
T
.
A
l
-
D
h
i
e
f
,
“
C
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
w
i
t
h
b
i
n
a
r
y
m
o
t
h
f
l
a
me
o
p
t
i
mi
z
a
t
i
o
n
f
o
r
e
mo
t
i
o
n
d
e
t
e
c
t
i
o
n
i
n
e
l
e
c
t
r
o
e
n
c
e
p
h
a
l
o
g
r
a
m,
”
I
AE
S
I
n
t
e
rn
a
t
i
o
n
a
l
J
o
u
r
n
a
l
o
f
Ar
t
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
,
v
o
l
.
1
3
,
n
o
.
1
,
p
p
.
1
1
7
2
–
1
1
7
8
,
M
a
r
.
2
0
2
4
,
d
o
i
:
1
0
.
1
1
5
9
1
/
i
j
a
i
.
v
1
3
.
i
1
.
p
p
1
1
7
2
-
1
1
7
8
.
[
4
2
]
O
.
U
d
o
c
h
u
k
w
u
a
n
d
Y
.
H
e
,
“
A
r
u
l
e
-
b
a
se
d
a
p
p
r
o
a
c
h
t
o
i
m
p
l
i
c
i
t
e
m
o
t
i
o
n
d
e
t
e
c
t
i
o
n
i
n
t
e
x
t
,
”
i
n
N
a
t
u
r
a
l
L
a
n
g
u
a
g
e
Pro
c
e
ssi
n
g
a
n
d
I
n
f
o
rm
a
t
i
o
n
S
y
st
e
m
s
,
C
h
a
m,
S
w
i
t
z
e
r
l
a
n
d
:
S
p
r
i
n
g
e
r
,
2
0
1
5
,
p
p
.
1
9
7
–
203
,
d
o
i
:
1
0
.
1
0
0
7
/
9
7
8
-
3
-
3
1
9
-
1
9
5
8
1
-
0
_
1
7
.
[
4
3
]
D
.
H
a
z
a
r
i
k
a
,
S
.
P
o
r
i
a
,
R
.
Z
i
mm
e
r
ma
n
n
,
a
n
d
R
.
M
i
h
a
l
c
e
a
,
“
C
o
n
v
e
r
sat
i
o
n
a
l
t
r
a
n
sf
e
r
l
e
a
r
n
i
n
g
f
o
r
e
mo
t
i
o
n
r
e
c
o
g
n
i
t
i
o
n
,
”
I
n
f
o
rm
a
t
i
o
n
Fu
si
o
n
,
v
o
l
.
6
5
,
p
p
.
1
–
1
2
,
J
a
n
.
2
0
2
1
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
i
n
f
f
u
s.
2
0
2
0
.
0
6
.
0
0
5
.
[
4
4
]
X
.
Li
a
n
g
,
Y
.
Z
o
u
,
X
.
Zh
u
a
n
g
,
J
.
Y
a
n
g
,
T
.
N
i
u
,
a
n
d
R
.
X
u
,
“
M
M
A
T
ER
I
C
:
m
u
l
t
i
-
t
a
s
k
l
e
a
r
n
i
n
g
a
n
d
m
u
l
t
i
-
f
u
s
i
o
n
f
o
r
A
u
d
i
o
Te
x
t
e
mo
t
i
o
n
r
e
c
o
g
n
i
t
i
o
n
i
n
c
o
n
v
e
r
sa
t
i
o
n
,
”
El
e
c
t
r
o
n
i
c
s
,
v
o
l
.
1
2
,
n
o
.
7
,
M
a
r
.
2
0
2
3
,
d
o
i
:
1
0
.
3
3
9
0
/
e
l
e
c
t
r
o
n
i
c
s
1
2
0
7
1
5
3
4
.
[
4
5
]
W
.
N
i
e
,
R
.
C
h
a
n
g
,
M
.
R
e
n
,
Y
.
S
u
,
a
n
d
A
.
Li
u
,
“
I
-
G
C
N
:
i
n
c
r
e
m
e
n
t
a
l
g
r
a
p
h
c
o
n
v
o
l
u
t
i
o
n
n
e
t
w
o
r
k
f
o
r
c
o
n
v
e
r
sat
i
o
n
e
m
o
t
i
o
n
d
e
t
e
c
t
i
o
n
,
”
I
EEE
T
ra
n
s
a
c
t
i
o
n
s
o
n
M
u
l
t
i
m
e
d
i
a
,
v
o
l
.
2
4
,
p
p
.
4
4
7
1
–
4
4
8
1
,
2
0
2
2
,
d
o
i
:
1
0
.
1
1
0
9
/
T
M
M
.
2
0
2
1
.
3
1
1
8
8
8
1
.
[
4
6
]
S
.
K
u
sa
l
,
S
.
P
a
t
i
l
,
J.
C
h
o
u
d
r
i
e
,
K
.
K
o
t
e
c
h
a
,
D
.
V
o
r
a
,
a
n
d
I
.
P
a
p
p
a
s,
“
A
r
e
v
i
e
w
o
n
t
e
x
t
-
b
a
s
e
d
e
mo
t
i
o
n
d
e
t
e
c
t
i
o
n
:
t
e
c
h
n
i
q
u
e
s
,
a
p
p
l
i
c
a
t
i
o
n
s,
d
a
t
a
set
s
,
a
n
d
f
u
t
u
r
e
d
i
r
e
c
t
i
o
n
s
,
”
M
a
y
2
0
2
2
,
a
rXi
v
:
2
2
0
5
.
0
3
2
3
5
.
Evaluation Warning : The document was created with Spire.PDF for Python.