I
nte
rna
t
io
na
l J
o
urna
l o
f
E
lect
rica
l a
nd
Co
m
pu
t
er
E
ng
ineering
(
I
J
E
CE
)
Vo
l.
14
,
No
.
6
,
Dec
em
b
er
20
24
,
p
p
.
6
7
8
6
~
6
7
9
6
I
SS
N:
2088
-
8
7
0
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijece.
v
14
i
6
.
pp
6
7
8
6
-
6
7
9
6
6786
J
o
ur
na
l
ho
m
ep
a
g
e
:
h
ttp
:
//ij
ec
e.
ia
esco
r
e.
co
m
Deep learni
ng
-
ba
sed a
tt
en
tion mo
d
els for sa
rcas
m de
tect
io
n i
n
text
G
a
nes
h Cha
nd
ra
s
ek
a
ra
n
1
,
M
a
nd
a
la
pu
K
a
lpa
na
cho
wd
a
ry
2
,
J
y
o
t
hi Chin
na
B
a
bu
3
,
Aj
m
ee
ra
K
ir
a
n
4
,
K
o
t
t
hu
ru
Anil K
um
a
r
5
,
Seif
edine K
a
dry
6,
7
1
D
e
p
a
r
t
me
n
t
o
f
C
o
m
p
u
t
e
r
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
,
S
r
i
Es
h
w
a
r
C
o
l
l
e
g
e
o
f
En
g
i
n
e
e
r
i
n
g
,
C
o
i
m
b
a
t
o
r
e
,
I
n
d
i
a
2
D
e
p
a
r
t
me
n
t
o
f
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
a
n
d
E
n
g
i
n
e
e
r
i
n
g
,
M
LR
I
n
st
i
t
u
t
e
o
f
Te
c
h
n
o
l
o
g
y
,
H
y
d
e
r
a
b
a
d
,
I
n
d
i
a
3
D
e
p
a
r
t
me
n
t
o
f
El
e
c
t
r
o
n
i
c
s a
n
d
C
o
m
mu
n
i
c
a
t
i
o
n
En
g
i
n
e
e
r
i
n
g
,
A
n
n
a
ma
c
h
a
r
y
a
I
n
st
i
t
u
t
e
o
f
T
e
c
h
n
o
l
o
g
y
a
n
d
S
c
i
e
n
c
e
s,
R
a
j
a
m
p
e
t
,
I
n
d
i
a
4
D
e
p
a
r
t
me
n
t
o
f
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
a
n
d
E
n
g
i
n
e
e
r
i
n
g
,
M
LR
I
n
st
i
t
u
t
e
o
f
Te
c
h
n
o
l
o
g
y
,
H
y
d
e
r
a
b
a
d
,
I
n
d
i
a
5
D
e
p
a
r
t
me
n
t
o
f
El
e
c
t
r
o
n
i
c
s a
n
d
C
o
m
mu
n
i
c
a
t
i
o
n
En
g
i
n
e
e
r
i
n
g
,
M
a
l
l
a
R
e
d
d
y
En
g
i
n
e
e
r
i
n
g
C
o
l
l
e
g
e
,
H
y
d
e
r
a
b
a
d
,
I
n
d
i
a
6
M
EU
R
e
s
e
a
r
c
h
U
n
i
t
,
M
i
d
d
l
e
E
a
st
U
n
i
v
e
r
si
t
y
,
A
mm
a
n
,
Jo
r
d
a
n
7
A
p
p
l
i
e
d
S
c
i
e
n
c
e
R
e
se
a
r
c
h
C
e
n
t
e
r
,
A
p
p
l
i
e
d
S
c
i
e
n
c
e
P
r
i
v
a
t
e
U
n
i
v
e
r
s
i
t
y
,
A
mm
a
n
,
J
o
r
d
a
n
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Ma
r
2
1
,
2
0
2
4
R
ev
is
ed
J
u
l 2
3
,
2
0
2
4
Acc
ep
ted
Au
g
6
,
2
0
2
4
F
in
d
in
g
sa
rc
a
stic
sta
te
m
e
n
ts
h
a
s
re
c
e
n
tl
y
d
ra
wn
a
lo
t
o
f
c
u
ri
o
sity
in
so
c
ial
m
e
d
ia,
m
a
in
ly
b
e
c
a
u
se
sa
rc
a
stic
twe
e
ts
m
a
y
in
c
lu
d
e
fa
v
o
ra
b
le
p
h
ra
se
s
th
a
t
fil
l
i
n
u
n
a
tt
ra
c
ti
v
e
o
r
u
n
d
e
sira
b
le
a
tt
ri
b
u
tes
.
As
t
h
e
i
n
tern
e
t
b
e
c
o
m
e
s
in
c
re
a
sin
g
ly
in
g
ra
in
e
d
in
o
u
r
d
a
il
y
li
v
e
s,
m
a
n
y
m
u
lt
ime
d
ia
i
n
fo
r
m
a
ti
o
n
is
b
e
in
g
p
r
o
d
u
c
e
d
o
n
li
n
e
.
M
u
c
h
o
f
th
e
in
f
o
rm
a
ti
o
n
re
c
o
rd
e
d
m
o
st
ly
o
n
t
h
e
in
tern
e
t
is
tex
tu
a
l
d
a
ta.
It
is
c
r
u
c
ial
to
c
o
m
p
re
h
e
n
d
p
e
o
p
le'
s
se
n
ti
m
e
n
ts.
Ho
we
v
e
r,
sa
rc
a
stic
c
o
n
ten
t
wil
l
h
in
d
e
r
t
h
e
e
ffe
c
ti
v
e
n
e
ss
o
f
se
n
ti
m
e
n
t
a
n
a
ly
sis
sy
ste
m
s.
Co
rre
c
tl
y
i
d
e
n
ti
fy
i
n
g
sa
rc
a
sm
a
n
d
c
o
rre
c
tl
y
p
re
d
ictin
g
p
e
o
p
le'
s
m
o
ti
v
e
s
a
re
e
x
trem
e
ly
i
m
p
o
rtan
t.
S
a
rc
a
sm
is
p
a
rti
c
u
larl
y
h
a
r
d
t
o
re
c
o
g
n
ize
,
b
o
t
h
b
y
h
u
m
a
n
s
a
n
d
b
y
m
a
c
h
in
e
s.
We
e
m
p
lo
y
th
e
d
e
e
p
bi
-
d
irec
ti
o
n
a
l
l
o
n
g
-
sh
o
rt
m
e
m
o
ry
(Bi
-
LS
T
M
)
a
n
d
a
h
y
b
ri
d
a
rc
h
it
e
c
tu
re
o
f
th
e
c
o
n
v
o
lu
t
io
n
n
e
u
ra
l
n
e
two
rk
+
Bi
-
LS
TM
(CNN
+
Bi
-
LS
TM
)
wit
h
a
tt
e
n
t
io
n
n
e
two
rk
s
f
o
r
i
d
e
n
ti
f
y
in
g
sa
rc
a
stic
re
m
a
rk
s
in
a
c
o
rp
u
s.
Us
in
g
th
e
S
a
rc
a
sm
V2
d
a
tas
e
t,
we
tes
t
th
e
e
ffica
c
y
o
f
d
e
e
p
lea
rn
i
n
g
m
e
th
o
d
s
Bi
LS
TM
,
a
n
d
CNN
+
BiL
S
TM
with
a
tt
e
n
ti
o
n
n
e
two
rk
)
fo
r
th
e
tas
k
o
f
id
e
n
ti
fy
in
g
tex
t
sa
rc
a
s
m
.
Th
e
su
g
g
e
ste
d
a
p
p
ro
a
c
h
in
c
o
r
p
o
ra
ti
n
g
d
e
e
p
n
e
tw
o
rk
s
is
c
o
n
siste
n
t
with
v
a
ri
o
u
s
re
c
e
n
t
a
n
d
a
d
v
a
n
c
e
d
tec
h
n
iq
u
e
s
fo
r
sa
rc
a
sm
d
e
tec
ti
o
n
.
Wi
t
h
a
tt
e
n
ti
o
n
p
r
o
c
e
ss
e
s,
th
e
imp
r
o
v
e
d
CNN
+
Bi
-
LS
TM
m
o
d
e
l
a
c
h
iev
e
d
a
n
a
c
c
u
ra
c
y
ra
te o
f
9
1
.
7
6
%
,
wh
ich
is
a
n
o
ta
b
le i
n
c
re
a
se
o
v
e
r
e
a
rli
e
r
re
se
a
rc
h
.
K
ey
w
o
r
d
s
:
Atten
tio
n
n
etwo
r
k
s
C
NN+
B
i
-
L
STM
Dee
p
lear
n
in
g
Sar
ca
s
m
T
ex
t m
in
in
g
T
h
is i
s
a
n
o
p
e
n
a
c
c
e
ss
a
rticle
u
n
d
e
r th
e
CC B
Y
-
SA
li
c
e
n
se
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
J
y
o
th
i Ch
in
n
a
B
ab
u
Dep
ar
tm
en
t o
f
E
lectr
o
n
ics an
d
C
o
m
m
u
n
icatio
n
E
n
g
in
ee
r
i
n
g
,
An
n
am
ac
h
ar
y
a
I
n
s
titu
te
o
f
T
e
ch
n
o
lo
g
y
an
d
Scien
ce
s
R
ajam
p
et
5
1
6
1
2
6
,
AP,
I
n
d
ia
E
m
ail:
jch
in
n
ab
ab
u
@
g
m
ail.
co
m
1.
I
NT
RO
D
UCT
I
O
N
Sh
ar
in
g
in
f
o
r
m
ati
o
n
t
h
r
o
u
g
h
th
e
in
te
r
n
et
h
as
b
ec
o
m
e
i
n
cr
ea
s
in
g
ly
c
o
m
m
o
n
o
v
er
t
h
e
d
ec
a
d
es
b
ec
au
s
e
s
o
cial
m
ed
ia
is
th
e
s
o
le
m
eth
o
d
f
o
r
r
ea
ch
in
g
o
u
t
to
y
o
u
n
g
in
d
i
v
id
u
als
an
d
in
ter
p
r
eted
s
en
tim
en
tality
as
a
p
er
s
o
n
'
s
attitu
d
e
to
war
d
a
s
in
g
u
lar
o
b
jectiv
e.
T
h
e
ex
is
ten
ce
o
f
s
ar
ca
s
tic
co
n
ten
t
with
i
n
th
o
s
e
o
n
lin
e
d
ata
is
im
p
ed
in
g
s
en
tim
en
t
an
aly
s
is
,
wh
eth
er
th
r
o
u
g
h
a
co
m
p
u
ter
o
r
a
m
a
n
u
al
p
r
o
ce
s
s
.
I
d
en
tify
in
g
s
ar
ca
s
tic
in
f
o
r
m
atio
n
with
in
s
u
ch
a
lar
g
e
co
r
p
u
s
o
f
o
n
lin
e
in
f
o
r
m
ati
o
n
is
tim
e
-
co
n
s
u
m
in
g
an
d
d
if
f
icu
lt.
As
a
r
esu
lt,
th
er
e
is
a
s
tr
o
n
g
d
esire
to
cr
e
ate
a
h
u
m
an
-
c
o
m
p
u
ter
in
ter
ac
tio
n
s
y
s
tem
th
at
ca
n
ca
teg
o
r
iz
e
s
ar
ca
s
tic
tex
t
in
a
co
r
p
u
s
.
Giv
en
th
e
a
d
v
en
t
o
f
o
n
lin
e
p
latf
o
r
m
s
,
r
esear
ch
er
s
wo
r
k
in
g
in
a
b
r
o
ad
r
a
n
g
e
o
f
f
ield
s
,
with
m
ac
h
i
n
e
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:
2088
-
8
7
0
8
Dee
p
lea
r
n
in
g
-
b
a
s
ed
a
tten
tio
n
mo
d
els fo
r
s
a
r
ca
s
m
d
etec
tio
n
in
text
(
Ga
n
esh
C
h
a
n
d
r
a
s
ek
a
r
a
n
)
6787
v
is
io
n
,
n
atu
r
al
lin
g
u
is
tic
co
m
p
u
tatio
n
,
an
d
m
an
y
o
t
h
er
s
,
n
o
w
h
av
e
m
u
ch
ea
s
ier
ac
ce
s
s
to
ex
ten
s
iv
e
an
d
p
r
ec
is
e
d
ata
co
llectio
n
s
[
1
]
,
[
2
]
.
I
n
ter
p
r
etatio
n
o
f
th
e
a
cc
u
r
ate
s
en
tim
en
t
is
ex
tr
em
e
ly
im
p
o
r
tan
t
w
h
en
d
eter
m
in
in
g
p
eo
p
le'
s
o
p
in
io
n
s
an
d
f
ee
lin
g
s
r
eg
ar
d
in
g
a
n
y
it
em
o
r
b
r
an
d
b
ec
au
s
e
a
p
r
eju
d
iced
o
r
in
c
o
r
r
ec
tly
d
etec
ted
co
m
m
e
n
t
ca
n
c
o
n
f
u
s
e
b
o
th
p
eo
p
le
an
d
co
m
p
u
ter
s
t
h
em
s
elv
es
[
3
]
,
[
4
]
.
T
h
e
te
r
m
"
s
ar
ca
s
m
"
is
u
s
ed
to
ex
p
r
ess
an
in
d
iv
id
u
al’
s
n
eg
ati
v
e
f
ee
lin
g
s
o
n
a
p
o
s
itiv
e
n
o
te.
T
h
e
u
s
ag
e
o
f
s
ar
ca
s
tic
s
tatem
en
ts
in
tex
ts
s
h
o
u
ld
b
e
elim
in
ated
as
th
ese
s
tatem
en
ts
h
av
e
a
s
er
io
u
s
im
p
ac
t
o
n
s
en
tim
en
t
an
aly
s
is
[
5
]
.
Pas
t
r
esear
ch
r
elate
d
to
s
ar
ca
s
m
d
etec
tio
n
h
a
v
e
r
elied
o
n
r
u
le
-
b
ased
an
d
o
t
h
er
tec
h
n
iq
u
es
th
at
e
m
p
lo
y
i)
L
ex
ical
,
ii)
Pra
g
m
atic,
an
d
iii)
Po
lar
ity
v
ar
iatio
n
s
,
an
d
s
o
o
n
.
Sin
ce
s
ar
ca
s
m
'
s
o
r
ig
in
al
in
ten
t
is
f
r
eq
u
en
tly
th
e
ex
ac
t o
p
p
o
s
ite
o
f
wh
at
th
e
u
s
er
in
ten
d
s
,
m
an
y
p
eo
p
le
s
tr
iv
e
to
co
n
v
ey
th
eir
o
p
i
n
io
n
s
an
d
u
s
e
t
h
em
to
g
et
atten
tio
n
f
r
o
m
o
th
er
s
.
Af
f
ec
ti
v
e
co
m
p
u
tin
g
an
d
co
m
p
u
tatio
n
al
lin
g
u
is
tics
ca
n
s
tr
en
g
th
en
d
ec
is
io
n
s
u
p
p
o
r
t
s
y
s
tem
s
an
d
cu
s
to
m
e
r
s
er
v
i
ce
b
y
u
n
c
o
v
er
in
g
cu
s
to
m
er
s
'
/u
s
er
s
'
p
er
s
o
n
al
p
r
ef
er
en
ce
s
o
r
r
em
o
v
in
g
item
s
/p
r
o
d
u
ct
in
f
o
r
m
atio
n
th
at
g
o
t
n
eg
ativ
e
cu
s
to
m
er
r
ev
iews.
L
ac
k
o
f
f
am
iliar
ity
with
th
e
s
itu
atio
n
'
s
s
u
r
r
o
u
n
d
in
g
s
,
an
d
th
e
p
ar
ticu
lar
s
u
b
ject
c
o
u
ld
m
ak
e
it
h
ar
d
er
to
r
ec
o
g
n
ize
s
ar
ca
s
tic
r
em
ar
k
s
.
R
ec
o
g
n
izin
g
c
o
n
tex
t is am
o
n
g
th
e
m
o
s
t d
if
f
icu
lt p
ar
ts
o
f
co
n
ten
t m
o
d
e
r
atio
n
.
I
n
tellig
en
t
s
ar
ca
s
m
r
ec
o
g
n
itio
n
is
co
n
s
id
er
e
d
a
s
im
p
le
te
x
t
ca
teg
o
r
izatio
n
is
s
u
e
[
6
]
.
T
h
e
u
s
e
o
f
s
em
an
tics
,
m
ac
h
in
e
lear
n
in
g
,
an
d
r
u
le
-
b
ased
s
tr
ateg
ies
h
av
e
all
b
ee
n
s
u
g
g
ested
as
s
ar
ca
s
m
d
etec
tio
n
tech
n
iq
u
es
in
p
r
e
v
io
u
s
p
u
b
licatio
n
s
.
Ho
wev
er
,
th
ey
a
ll
h
av
e
d
r
awb
ac
k
s
.
Sem
an
tic
an
d
s
y
n
tactic
ch
ar
ac
ter
is
tics
ar
e
u
s
ed
b
y
Nag
wan
s
h
i
an
d
Ma
d
h
av
a
n
[
7
]
to
id
en
tify
s
ar
ca
s
m
in
liter
atu
r
e.
No
n
eth
eless
,
t
h
ey
ar
e
f
r
eq
u
e
n
tly
u
s
ed
tex
t
ca
teg
o
r
izatio
n
co
m
p
o
n
e
n
ts
th
at
ar
e
also
u
s
ef
u
l
f
o
r
s
ar
ca
s
m
d
etec
tio
n
.
Acc
o
r
d
in
g
to
Sh
ah
an
d
Sh
a
h
[
8
]
,
d
ee
p
lear
n
i
n
g
alg
o
r
ith
m
s
p
er
f
o
r
m
b
etter
a
t d
etec
tin
g
s
ar
ca
s
m
.
I
n
o
r
d
er
t
o
im
p
r
o
v
e
s
ar
ca
s
m
d
etec
tio
n
,
we
p
r
esen
t
in
t
h
is
wo
r
k
a
h
y
b
r
id
d
ee
p
lear
n
in
g
ar
ch
itectu
r
e
th
at
c
o
m
b
in
es
att
en
tio
n
m
ec
h
an
is
m
s
with
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
et
wo
r
k
s
(
C
NN)
an
d
bi
-
d
ir
ec
tio
n
al
lo
n
g
s
h
o
r
t
-
te
r
m
m
em
o
r
y
(
B
i
-
L
STM
)
n
etwo
r
k
s
.
W
h
er
ea
s
th
e
B
i
-
L
STM
co
m
p
o
n
en
t
ca
n
c
o
m
p
r
eh
en
d
lo
n
g
-
r
a
n
g
e
r
elatio
n
s
h
i
p
s
an
d
c
o
n
tex
t
u
al
in
f
o
r
m
atio
n
,
th
e
C
NN
co
m
p
o
n
en
t
is
m
ad
e
to
p
ick
u
p
o
n
lo
ca
l
ch
a
r
ac
ter
is
tics
an
d
p
atter
n
s
in
th
e
tex
t.
T
h
e
m
o
d
el
p
er
f
o
r
m
s
b
etter
o
v
er
all
as
th
e
atten
tio
n
m
ec
h
an
is
m
h
elp
s
it
f
o
cu
s
o
n
im
p
o
r
ta
n
t
tex
t
p
ass
ag
es.
T
h
e
atten
tio
n
n
etwo
r
k
p
er
f
o
r
m
s
b
etter
th
an
th
e
s
tate
-
of
-
th
e
-
a
r
t
tech
n
iq
u
es
r
e
g
ar
d
in
g
F1
-
s
co
r
e,
ac
c
u
r
ac
y
,
p
r
e
cisi
o
n
,
an
d
r
ec
all.
C
o
m
b
in
in
g
t
h
e
a
d
v
an
tag
es
o
f
C
NNs
with
B
i
-
L
STM
s
,
th
e
h
y
b
r
id
ar
ch
itectu
r
e
o
f
f
er
s
a
m
o
r
e
t
h
o
r
o
u
g
h
co
m
p
r
eh
e
n
s
io
n
o
f
th
e
tex
t.
B
y
p
r
o
v
id
in
g
a
u
n
iq
u
e
m
eth
o
d
t
h
at
in
co
r
p
o
r
ates
tech
n
iq
u
es
f
r
o
m
d
ee
p
lear
n
in
g
to
en
h
an
ce
th
e
d
ete
r
m
in
atio
n
o
f
s
ar
ca
s
m
with
in
tex
tu
al
d
ata,
th
is
wo
r
k
co
n
tr
ib
u
tes
to
th
e
ar
ea
.
T
h
e
f
in
d
i
n
g
s
s
h
o
w
h
o
w
th
e
s
u
g
g
ested
ap
p
r
o
ac
h
m
ay
im
p
r
o
v
e
s
en
tim
en
t
an
aly
s
is
to
o
ls
,
en
ab
lin
g
m
o
r
e
p
r
ec
is
e
in
ter
p
r
etatio
n
s
o
f
u
s
er
-
g
en
er
ated
m
ater
ial
o
n
s
o
cial
m
ed
ia.
I
n
c
o
n
clu
s
io
n
,
th
is
wo
r
k
an
aly
s
es
p
er
tin
en
t
liter
atu
r
e,
tack
les
th
e
cr
u
cial
is
s
u
e
o
f
r
ec
o
g
n
itio
n
o
f
s
ar
ca
s
m
in
s
en
tim
en
t
an
aly
s
is
,
an
d
p
r
esen
ts
a
b
r
an
d
-
n
ew
co
m
p
o
s
ite
d
ee
p
lea
r
n
in
g
m
o
d
e
l.
2.
RE
L
AT
E
D
WO
RK
S
I
n
p
r
i
o
r
s
tu
d
ies,
v
ar
i
o
u
s
r
esear
ch
er
s
h
av
e
co
m
m
o
n
ly
e
m
p
lo
y
ed
th
r
ee
s
tr
ateg
ies
to
u
n
d
er
s
tan
d
s
ar
ca
s
m
in
tex
t.
T
h
e
f
ir
s
t
is
a
lex
ico
n
-
b
ased
s
tr
ateg
y
th
at
co
m
b
in
es
d
ictio
n
ar
ies
with
f
ea
t
u
r
es
lik
e
u
n
ig
r
am
s
,
b
ig
r
am
s
,
n
-
g
r
am
s
,
an
d
p
ar
t
o
f
s
p
ee
ch
(
POS)
ch
ar
ac
ter
is
tic
s
.
T
h
e
s
ec
o
n
d
m
eth
o
d
to
id
e
n
tify
s
ar
ca
s
m
u
s
es
m
ac
h
in
e
o
r
d
ee
p
lear
n
in
g
tec
h
n
iq
u
es.
T
h
e
t
h
ir
d
ap
p
r
o
ac
h
is
th
e
c
o
n
tex
t
-
b
ased
ap
p
r
o
ac
h
th
at
in
v
o
lv
es
th
e
Stu
d
y
o
f
R
elatio
n
s
h
ip
s
am
o
n
g
wo
r
d
s
in
a
co
r
p
u
s
to
d
etec
t
t
h
e
p
r
esen
ce
o
f
s
ar
ca
s
m
.
T
h
e
v
ar
io
u
s
m
eth
o
d
s
f
o
r
s
ar
ca
s
m
d
etec
tio
n
em
p
lo
y
ed
in
ex
is
tin
g
liter
atu
r
e
is
s
h
o
w
n
in
th
e
Fig
u
r
e
1
.
I
n
th
eir
wo
r
k
,
B
o
u
az
izi
an
d
Oh
ts
u
k
i
[
9
]
u
s
ed
th
e
POS
f
ea
tu
r
e
with
o
th
e
r
p
er
tin
e
n
t
v
ar
iab
les
to
an
aly
ze
twee
ts
f
o
r
s
ar
ca
s
m
u
s
in
g
a
p
atter
n
-
b
ased
ap
p
r
o
ac
h
.
T
h
ey
m
ay
d
e
m
o
n
s
tr
ate
th
at
ad
d
in
g
v
ar
i
o
u
s
elem
en
ts
will e
n
h
an
ce
th
e
p
er
f
o
r
m
an
ce
in
d
icato
r
s
f
o
r
s
ar
ca
s
m
d
etec
tio
n
.
T
h
ey
u
s
ed
d
if
f
er
en
t
class
if
icatio
n
tec
h
n
iq
u
es
f
o
r
s
ar
ca
s
m
id
en
tific
a
tio
n
;
am
o
n
g
th
em
,
r
an
d
o
m
f
o
r
est
(
R
F)
p
er
f
o
r
m
e
d
th
e
b
est.
Nig
am
an
d
Yad
a
v
[
1
0
]
u
s
ed
a
lex
ico
n
-
b
ased
ap
p
r
o
ac
h
i
n
v
o
lv
i
n
g
an
ex
ter
n
al
d
ictio
n
a
r
y
in
t
h
eir
w
o
r
k
f
o
r
s
ar
ca
s
m
d
etec
tio
n
.
T
h
e
lim
itatio
n
o
f
th
is
s
tr
ateg
y
is
th
at
it
p
r
ev
en
ts
th
e
d
ictio
n
ar
y
f
r
o
m
in
clu
d
i
n
g
wo
r
d
s
f
r
o
m
h
y
b
r
id
lan
g
u
ag
es.
C
lew
s
an
d
Ku
zm
a
[
1
1
]
u
s
ed
a
lex
ico
n
-
b
ase
d
ap
p
r
o
ac
h
to
h
an
d
le
s
ar
ca
s
tic
in
f
o
r
m
atio
n
b
y
m
atch
i
n
g
a
s
tr
in
g
with
p
o
s
itiv
e
an
d
in
ter
jectio
n
lex
ico
n
s
.
T
h
e
y
co
u
ld
m
in
im
ize
th
e
tim
e
n
ee
d
ed
to
co
llect
an
d
an
aly
ze
th
e
d
ata
f
o
r
s
ar
ca
s
tic
id
en
tific
atio
n
.
Z
h
an
g
et
a
l.
[
1
2
]
h
av
e
u
s
ed
a
n
eu
r
al
n
etwo
r
k
-
b
ased
m
o
d
el
to
d
etec
t
s
ar
ca
s
m
f
r
o
m
T
wee
ts
.
T
h
eir
m
o
d
e
l
co
u
ld
av
o
id
th
e
d
r
awb
ac
k
s
ass
o
ciate
d
with
m
an
u
ally
e
x
tr
ac
ted
f
ea
tu
r
es
lik
e
POS
tag
s
an
d
s
en
tim
en
t
lex
ico
n
s
.
W
ith
th
e
ass
is
tan
ce
o
f
a
n
ap
p
r
o
p
r
iate
wo
r
d
em
b
ed
d
i
n
g
a
p
p
r
o
ac
h
,
t
h
eir
ap
p
r
o
ac
h
r
esu
lted
i
n
im
p
r
o
v
e
d
o
u
tco
m
es.
Sev
er
al
class
if
ier
s
wer
e
em
p
lo
y
ed
b
y
Pra
s
ad
et
a
l.
[
1
3
]
to
d
e
ter
m
in
e
th
e
s
ar
ca
s
tic
co
n
ten
t
p
r
esen
t
in
twee
ts
b
y
u
tili
zin
g
em
o
ji a
n
d
s
lan
g
d
icti
o
n
ar
ies.
T
h
e
b
est p
er
f
o
r
m
a
n
ce
is
o
b
tain
ed
with
th
e
Gr
ad
ien
t
B
o
o
s
tin
g
alg
o
r
ith
m
am
o
n
g
s
ev
er
al
class
if
ier
s
em
p
lo
y
ed
f
o
r
s
ar
ca
s
m
d
etec
tio
n
.
T
h
e
d
ata
p
r
e
-
p
r
o
ce
s
s
in
g
was
d
o
n
e
with
tech
n
iq
u
es
lik
e
s
tem
m
in
g
an
d
lem
m
atiza
t
io
n
with
th
e
n
atu
r
al
lan
g
u
a
g
e
p
r
o
ce
s
s
in
g
to
o
l
k
it
(
NL
T
K)
to
o
l
k
it.
T
h
e
r
elatio
n
b
etwe
en
two
s
p
ea
k
er
s
wh
o
s
h
ar
e
th
eir
v
iews
in
T
wee
ts
is
u
s
ed
b
y
B
ali
an
d
Sin
g
h
[
1
4
]
f
o
r
s
ar
ca
s
m
d
etec
tio
n
with
lo
g
is
tic
r
eg
r
ess
io
n
(
L
R
)
.
I
t
in
v
o
lv
es
u
n
d
er
s
tan
d
i
n
g
co
n
te
x
tu
al
in
f
o
r
m
atio
n
/r
e
latio
n
b
etwe
en
th
e
p
ar
ticip
an
ts
o
r
s
p
ea
k
er
s
in
id
en
tify
in
g
s
ar
ca
s
m
.
A
f
u
zz
y
-
b
ased
ap
p
r
o
ac
h
was
u
s
ed
b
y
Me
r
iem
et
a
l.
[
1
5
]
in
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
14
,
No
.
6
,
Dec
em
b
e
r
20
24
:
6
7
8
6
-
6
7
9
6
6788
th
eir
wo
r
k
o
n
s
ar
ca
s
m
d
etec
tio
n
to
id
en
tif
y
s
ar
ca
s
tic
twee
ts
in
v
o
l
v
in
g
d
if
f
e
r
en
t
d
ata
p
r
e
-
p
r
o
ce
s
s
in
g
m
eth
o
d
s
.
T
h
ey
co
u
ld
u
s
e
th
e
f
u
zz
y
-
b
as
ed
ap
p
r
o
ac
h
as
m
o
s
t
o
f
th
e
t
wee
ts
ar
e
f
u
zz
y
an
d
s
u
itab
le
t
o
b
e
an
al
y
ze
d
u
s
in
g
f
u
zz
y
m
o
d
els.
Ah
u
ja
et
a
l.
[
1
6
]
h
av
e
tr
ied
a
s
ar
ca
s
m
d
etec
tio
n
ap
p
r
o
ac
h
u
s
in
g
Ad
aBo
o
s
t,
g
r
ad
ien
t
b
o
o
s
t
an
d
n
aïv
e
B
ay
es
tech
n
iq
u
es
o
n
f
o
u
r
d
atasets
in
v
o
lv
in
g
s
ar
ca
s
tic
co
n
ten
t.
B
ased
o
n
th
eir
r
esu
lts
,
it
was
co
n
clu
d
ed
th
at
th
e
ML
a
p
p
r
o
ac
h
will
y
ield
b
etter
r
esu
lts
co
m
p
a
r
ed
to
co
n
t
ex
t
-
b
ased
m
et
h
o
d
s
an
d
th
at
p
er
f
o
r
m
a
n
ce
ca
n
s
till
b
e
im
p
r
o
v
ed
u
s
in
g
d
ee
p
lea
r
n
in
g
m
et
h
o
d
s
.
Go
v
in
d
a
n
an
d
B
alak
r
is
h
n
an
[
1
7
]
c
o
n
s
id
er
e
d
h
y
p
er
b
o
lic
twee
ts
wh
en
u
s
in
g
1
2
d
if
f
er
e
n
t
m
ac
h
in
e
-
lear
n
in
g
ap
p
r
o
ac
h
es
th
at
a
r
e
lik
ewise
f
ea
tu
r
e
-
b
ased
t
o
id
en
tify
s
ar
ca
s
m
.
B
y
u
tili
zin
g
s
ev
er
al
d
ata
p
r
e
-
p
r
o
ce
s
s
in
g
tech
n
iq
u
es,
th
e
p
er
f
o
r
m
an
ce
o
f
s
ar
ca
s
m
d
etec
ti
o
n
is
en
h
an
ce
d
b
y
ex
tr
ac
tin
g
th
e
h
y
p
er
b
o
le
-
r
elat
ed
f
ea
tu
r
es
u
s
in
g
th
e
Py
th
o
n
lib
r
ar
y
.
B
h
ar
ti
et
a
l.
[
1
8
]
u
s
e
d
h
y
p
er
b
o
le
twee
ts
,
wh
ich
ar
e
id
en
tifie
d
b
y
r
e
co
g
n
izin
g
ad
v
e
r
b
s
.
Sev
er
al
r
u
les
wer
e
f
r
am
ed
b
y
Ma
y
n
ar
d
a
n
d
Gr
ee
n
wo
o
d
[
1
9
]
to
d
eter
m
in
e
th
e
s
ar
ca
s
tic
co
n
ten
t
f
r
o
m
th
e
tex
t
d
ata
in
th
eir
wo
r
k
.
T
h
e
y
co
u
ld
d
ete
r
m
in
e
h
o
w
s
ar
ca
s
m
af
f
ec
ts
s
en
tim
en
t
p
o
lar
ity
an
d
d
em
o
n
s
tr
ate
h
o
w
s
ar
ca
s
m
d
etec
tio
n
h
elp
s
th
e
s
en
tim
en
t
ass
ess
m
en
t
m
o
d
el
p
e
r
f
o
r
m
b
etter
.
T
h
ey
co
u
ld
r
ec
o
g
n
ize
th
e
s
ar
ca
s
tic
co
n
ten
t
in
twe
ets
b
y
ass
ess
in
g
th
e
s
en
tim
en
t
lin
k
ed
with
t
h
e
h
ash
tag
s
,
wh
ich
was u
tili
ze
d
to
id
en
tify
s
ar
ca
s
m
in
twee
ts
.
Felb
o
et
a
l.
[
2
0
]
d
e
v
elo
p
e
d
a
p
r
e
-
tr
ain
e
d
alg
o
r
ith
m
th
at
r
elied
o
n
an
ar
r
ay
o
f
em
o
jis
an
d
co
u
ld
r
ec
o
g
n
ize
twee
ts
th
at
co
n
tai
n
e
d
s
ar
ca
s
m
.
T
o
r
e
d
u
ce
th
e
o
v
e
r
-
f
itti
n
g
o
f
th
e
m
o
d
els,
th
ey
h
av
e
u
s
ed
wo
r
d
an
d
p
h
r
ase
co
v
e
r
ag
e
-
b
ased
p
r
e
-
tr
ain
in
g
tech
n
i
q
u
es.
T
h
eir
wo
r
k
is
b
u
ilt
o
n
L
STM
ar
c
h
itectu
r
e,
an
d
t
o
attain
cu
ttin
g
-
ed
g
e
p
e
r
f
o
r
m
an
ce
,
s
o
m
e
lay
er
s
o
f
th
eir
m
o
d
el
ar
e
f
r
o
ze
n
an
d
u
n
f
r
o
ze
n
.
Dav
id
o
v
et
a
l.
[
2
1
]
u
s
ed
a
q
u
asi
-
tech
n
iq
u
e
with
two
p
h
ases
to
d
eter
m
in
e
s
ar
ca
s
m
in
r
ev
iews
r
elate
d
to
s
o
m
e
p
r
o
d
u
cts.
T
h
e
y
h
av
e
em
p
h
asized
th
e
d
ata
ac
q
u
is
itio
n
p
r
o
ce
s
s
b
y
em
p
l
o
y
in
g
v
ar
y
in
g
tec
h
n
iq
u
es
b
e
f
o
r
e
p
r
o
ce
ed
in
g
with
th
e
class
if
icatio
n
task
.
Haz
ar
ik
a
et
a
l.
[
2
2
]
h
av
e
r
etr
i
ev
ed
co
n
tex
tu
al
clu
es
f
r
o
m
s
o
cial
m
ed
ia
m
ater
ial
to
id
e
n
tif
y
s
ar
ca
s
tic
twee
ts
to
g
et
b
ey
o
n
d
th
e
r
estrictio
n
s
o
f
u
s
in
g
m
a
n
u
al
ch
ar
ac
ter
is
tics
f
o
r
s
ar
ca
s
m
id
en
tific
atio
n
.
Fo
r
class
if
icatio
n
,
th
ey
h
av
e
u
s
ed
C
NN
an
d
with
th
e
h
elp
o
f
co
n
tex
tu
al
f
ea
tu
r
es,
th
eir
m
o
d
e
l
co
u
ld
o
u
tp
e
r
f
o
r
m
o
th
er
s
in
d
etec
tin
g
s
ar
ca
s
tic
c
o
n
ten
t.
T
o
d
etec
t
s
ar
ca
s
tic
co
n
ten
t,
th
e
co
n
ten
t
an
d
c
o
n
tex
t
-
b
ased
f
ea
tu
r
es
ar
e
em
p
lo
y
ed
b
y
Re
n
et
a
l.
[
2
3
]
i
n
th
eir
wo
r
k
with
C
NN.
T
h
e
c
o
n
tex
tu
al
m
o
d
elin
g
o
f
te
x
t
is
d
o
n
e
f
ir
s
t
with
h
elp
o
f
th
e
d
o
cu
m
e
n
t
m
o
d
elin
g
tec
h
n
iq
u
e,
an
d
t
h
e
s
y
n
tactica
l
f
ea
tu
r
es
ar
e
ex
tr
ac
ted
,
f
ed
to
th
e
C
NN
th
at
id
en
tifie
s
s
ar
ca
s
m
.
Fig
u
r
e
1
.
Stra
teg
ies f
o
r
s
ar
ca
s
m
r
ec
o
g
n
itio
n
3.
P
RO
P
O
SE
D
M
E
T
H
O
D
T
h
is
s
ec
tio
n
co
n
tain
s
a
b
lo
ck
d
iag
r
am
o
f
th
e
s
u
g
g
ested
tex
t
s
ar
ca
s
m
d
etec
tio
n
s
ch
e
m
e,
wh
ich
co
m
b
in
es
d
ee
p
lear
n
in
g
an
d
a
lear
n
in
g
alg
o
r
ith
m
.
Fig
u
r
e
2
d
ep
icts
th
e
a
n
ticip
ated
a
n
aly
s
is
o
n
s
ar
ca
s
m
d
etec
tio
n
u
s
in
g
tex
tu
al
in
f
o
r
m
atio
n
.
T
h
e
p
o
in
ts
th
at
f
o
llo
w
em
p
h
asize
th
e
v
ar
io
u
s
p
r
o
ce
s
s
es
in
v
o
lv
ed
in
u
n
d
er
s
tan
d
i
n
g
s
ar
ca
s
m
.
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:
2088
-
8
7
0
8
Dee
p
lea
r
n
in
g
-
b
a
s
ed
a
tten
tio
n
mo
d
els fo
r
s
a
r
ca
s
m
d
etec
tio
n
in
text
(
Ga
n
esh
C
h
a
n
d
r
a
s
ek
a
r
a
n
)
6789
Fig
u
r
e
2
.
Pro
ce
s
s
f
lo
w
d
iag
r
a
m
o
f
th
e
s
u
g
g
ested
m
eth
o
d
f
o
r
d
etec
tin
g
s
ar
ca
s
m
3
.
1
.
Da
t
a
a
cquis
it
io
n
Ma
n
y
ap
o
s
tr
o
p
h
es,
s
y
m
b
o
ls
,
s
to
p
wo
r
d
s
,
an
d
p
o
s
s
ib
ly
o
t
h
er
r
ed
u
n
d
an
t
elem
en
ts
in
th
e
d
a
taset
'
s
r
aw
tex
t
h
in
d
e
r
s
ar
ca
s
m
r
ec
o
g
n
it
io
n
.
B
ef
o
r
e
ex
tr
ac
tin
g
wo
r
d
v
ec
to
r
s
,
r
em
o
v
e
s
u
ch
d
ata
f
r
o
m
th
e
d
ataset.
T
o
k
en
izin
g
tex
t
with
th
e
NL
T
K
p
ac
k
ag
e
h
elp
s
f
in
d
an
d
r
e
m
o
v
e
d
u
p
licate
co
n
te
n
t.
T
h
e
d
at
a
ca
p
tu
r
e
p
h
ase
o
f
th
e
s
ar
ca
s
m
class
if
icatio
n
s
tar
ts
with
g
ath
er
in
g
tex
t.
T
h
e
r
e
ar
e
a
t
o
tal
o
f
9
,
3
8
6
in
s
tan
ce
s
o
f
s
ar
ca
s
m
in
th
is
d
ataset,
o
f
wh
ich
6
,
5
2
0
ar
e
g
en
er
alize
d
(
GE
N)
,
1
,
7
0
2
co
n
tain
r
h
eto
r
ical
in
q
u
i
r
ies
(
R
Q)
,
an
d
1
,
1
6
4
ar
e
h
y
p
er
b
o
lic
(
HYP)
.
3
.
2
.
P
re
-
pro
ce
s
s
ing
o
f
t
ex
t
u
a
l da
t
a
T
h
e
d
ataset'
s
r
aw
tex
t
in
clu
d
es
m
an
y
ap
o
s
tr
o
p
h
es,
s
y
m
b
o
ls
,
s
to
p
wo
r
d
s
,
an
d
p
er
h
a
p
s
o
th
er
r
ed
u
n
d
an
t
asp
ec
ts
th
at
d
o
n
o
t
h
elp
with
s
ar
ca
s
m
r
ec
o
g
n
itio
n
.
C
o
n
s
eq
u
e
n
tly
,
b
ef
o
r
e
e
x
tr
ac
tin
g
wo
r
d
v
ec
to
r
s
,
it
is
cr
itical
to
elim
in
ate
th
at
k
in
d
o
f
in
f
o
r
m
atio
n
f
r
o
m
th
e
d
ataset.
T
h
e
NL
T
K
p
ac
k
ag
e
aid
s
in
t
o
k
en
izin
g
tex
t,
wh
ich
d
iv
id
es
lar
g
e
am
o
u
n
ts
o
f
tex
t
in
to
tin
y
to
k
en
s
th
at
ass
is
t
in
d
etec
tin
g
an
d
d
eletin
g
d
u
p
l
icativ
e
co
n
ten
t.
T
h
e
p
r
e
-
p
r
o
ce
s
s
in
g
s
tep
in
cl
u
d
es
r
em
o
v
in
g
co
m
m
o
n
ly
en
co
u
n
ter
ed
p
u
n
ctu
atio
n
m
ar
k
s
in
ad
d
itio
n
to
s
p
ec
ial
ch
ar
ac
ter
s
(
@
,
#
,
an
d
s
o
o
n
)
.
I
t
is
ca
r
r
ied
o
u
t
with
th
e
aid
o
f
th
e
Py
th
o
n
"r
e
g
u
lar
ex
p
r
ess
io
n
"
p
ac
k
a
g
e,
wh
ic
h
in
clu
d
es
a
v
ar
iety
o
f
m
eth
o
d
s
f
o
r
s
ca
n
n
i
n
g
f
o
r
an
d
elim
in
at
in
g
u
n
wan
ted
r
ep
r
esen
tatio
n
s
.
T
h
e
NL
T
K
to
o
lk
it
is
th
en
u
s
ed
t
o
o
p
tim
ize
th
e
p
r
e
-
p
r
o
ce
s
s
in
g
o
u
tco
m
es
u
s
i
n
g
tex
t
n
o
r
m
aliza
tio
n
s
tr
ateg
ies,
wh
ich
in
clu
d
e
s
tem
m
in
g
,
to
k
e
n
izatio
n
,
an
d
p
ar
ts
o
f
s
p
ee
c
h
m
a
r
k
in
g
.
T
o
p
r
ep
ar
e
th
e
m
o
d
el,
th
e
p
r
e
-
p
r
o
ce
s
s
ed
in
f
o
r
m
atio
n
m
u
s
t b
e
tr
an
s
f
o
r
m
ed
in
to
th
e
i
n
p
u
t v
ec
t
o
r
s
in
th
e
f
o
llo
win
g
s
tep
.
3
.
3
.
F
e
a
t
ure
v
ec
t
o
rs g
ener
a
t
io
n wit
h
g
lo
ba
l v
ec
t
o
rs f
o
r
wo
rd
g
ener
a
t
io
n
(
G
lo
Ve)
m
o
d
el
I
n
c
o
n
tr
ast
to
o
th
er
p
r
o
ce
d
u
r
e
s
f
o
r
tex
tu
al
f
ea
tu
r
es
th
at
o
f
f
e
r
ce
r
tain
lim
itatio
n
s
,
th
e
g
lo
b
al
v
ec
to
r
s
f
o
r
wo
r
d
g
e
n
er
atio
n
(
Glo
Ve
)
wo
r
d
v
ec
to
r
s
p
r
o
ce
d
u
r
e
em
p
lo
y
ed
in
th
e
s
u
g
g
ested
m
o
d
el
o
f
f
er
s
m
an
y
m
er
its
.
A
Glo
Ve
is
a
cu
ttin
g
-
ed
g
e
ap
p
r
o
ac
h
th
at
g
en
er
ates
a
lo
w
-
d
im
e
n
s
io
n
al
ten
s
o
r
p
r
o
to
t
y
p
e
with
o
u
t
s
ac
r
if
icin
g
wo
r
d
co
n
tex
tu
al
r
esem
b
lan
ce
.
T
h
e
f
ir
s
t
s
tep
en
tails
cr
ea
tin
g
a
wo
r
d
d
ictio
n
ar
y
f
r
o
m
th
e
wo
r
d
s
f
o
u
n
d
i
n
th
e
d
ataset.
T
h
is
p
r
o
ce
s
s
g
en
er
ates
a
s
u
b
s
p
ac
e
o
f
s
tatem
en
ts
,
ea
ch
with
a
s
ep
ar
ate
id
en
tity
.
I
n
th
e
f
o
llo
win
g
p
h
ase,
t
h
e
s
tatis
t
ical
d
etail
o
f
d
ataset
w
o
r
d
s
is
cr
af
ted
an
d
co
n
ca
ten
ated
to
th
e
f
ea
tu
r
e
s
p
ac
e.
I
n
lig
h
t
o
f
p
r
ev
io
u
s
in
v
esti
g
atio
n
s
o
n
wo
r
d
em
b
e
d
d
in
g
m
o
d
els,
we
co
u
ld
ap
p
r
aise
th
e
ad
v
an
tag
e
o
f
in
co
r
p
o
r
atin
g
th
e
Glo
Ve
m
o
d
el
with
th
e
B
i
-
L
STM
a
p
p
r
o
ac
h
f
o
r
b
etter
o
v
er
all
m
o
d
el
p
er
f
o
r
m
an
ce
.
T
h
e
d
at
a
co
llectio
n
f
o
r
th
e
Sar
ca
s
m
C
o
r
p
u
s
V2
was
d
iv
id
ed
in
to
tr
ain
in
g
an
d
test
in
g
s
ets,
an
d
it
co
n
tain
ed
1
,
1
6
4
s
am
p
les
o
f
h
y
p
er
b
o
lic
s
ar
ca
s
m
m
ess
ag
es
in
ad
d
itio
n
to
6
,
5
2
0
in
s
tan
ce
s
o
f
g
e
n
er
ic
s
ar
ca
s
m
an
d
1
,
7
0
2
s
am
p
les
o
f
r
h
eto
r
ical
d
ialo
g
u
e
s
ar
ca
s
m
.
T
h
e
tr
ain
in
g
s
et
in
clu
d
es 7
,
5
0
8
s
am
p
les,
wh
ile
th
e
t
esti
n
g
s
et
in
clu
d
es 1
,
8
7
8
s
am
p
les.
3
.
4
.
T
ra
ini
ng
t
he
a
t
t
ent
i
o
n m
o
dels
f
o
r
s
a
rc
a
s
m
det
ec
t
io
n
T
h
e
m
ain
g
o
al
o
f
ad
o
p
tin
g
d
e
ep
lear
n
in
g
d
esig
n
s
ee
m
s
to
b
e
to
av
o
id
th
e
lim
itatio
n
s
o
r
o
b
s
tr
u
ctio
n
s
th
at
tr
ad
itio
n
al
m
ac
h
in
e
lear
n
in
g
d
esig
n
s
h
av
e
wh
e
n
p
r
o
v
id
i
n
g
th
e
b
est s
o
lu
tio
n
.
Alg
o
r
ith
m
s
f
o
r
d
ee
p
lear
n
in
g
ca
n
au
to
m
atica
lly
ex
tr
ac
t
f
r
o
m
a
p
ar
ticu
lar
d
ataset
with
o
u
t
h
u
m
an
in
ter
v
e
n
tio
n
an
d
ar
e
ex
tr
em
ely
ef
f
icien
t
at
ca
p
tu
r
in
g
ad
d
itio
n
al
co
n
tex
t.
W
ith
m
ac
h
in
e
lear
n
in
g
an
d
p
e
r
h
ap
s
o
th
er
alg
o
r
ith
m
s
,
it
is
i
m
p
o
s
s
ib
le
to
ca
p
tu
r
e
co
n
tex
tu
al
in
f
o
r
m
atio
n
.
T
h
e
at
ten
tio
n
ap
p
r
o
ac
h
is
e
m
p
lo
y
e
d
to
em
p
h
asize
s
p
ec
if
ic
wo
r
d
s
t
h
at
aid
in
d
etec
tin
g
s
ar
ca
s
m
an
d
m
u
ch
less
atten
ti
o
n
to
in
s
ig
n
i
f
ican
t te
r
m
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
14
,
No
.
6
,
Dec
em
b
e
r
20
24
:
6
7
8
6
-
6
7
9
6
6790
3
.
5
.
Sa
rc
a
s
m
det
ec
t
io
n wit
h
Bi
-
L
S
T
M
wit
h
a
t
t
ent
io
n m
o
del
T
h
e
B
i
-
L
STM
f
r
a
m
ewo
r
k
in
c
o
r
p
o
r
ates
an
atten
ti
o
n
s
y
s
tem
th
at
en
h
a
n
ce
s
th
e
s
ar
ca
s
m
id
e
n
tific
atio
n
m
o
d
el'
s
ef
f
ec
tiv
en
ess
.
T
h
is
m
eth
o
d
s
co
r
es
ea
ch
k
ey
in
th
e
d
atab
ase
ag
ain
s
t
th
e
v
ec
t
o
r
s
p
ac
e
ass
o
ciate
d
with
th
e
p
ar
ticu
lar
wo
r
d
in
t
h
e
in
p
u
t
wh
en
a
s
et
o
f
w
o
r
d
s
is
p
r
esen
ted
.
T
o
cr
ea
te
atten
tio
n
o
u
tp
u
t
f
o
r
th
e
wo
r
d
b
ein
g
co
n
s
id
er
e
d
,
th
e
v
al
u
es a
r
e
s
ca
led
in
ac
co
r
d
an
ce
with
t
h
e
atten
tio
n
weig
h
ts
an
d
co
n
c
en
tr
ate
o
n
th
e
wo
r
d
s
th
at
ar
e
p
er
tin
en
t
to
th
e
r
eq
u
e
s
t.
A
wo
r
d
's
atten
tio
n
weig
h
t
is
ca
lcu
lated
b
y
co
n
s
id
er
in
g
t
h
e
to
tal
am
o
u
n
t
o
f
atten
tio
n
it
h
as
r
ec
eiv
ed
th
r
o
u
g
h
o
u
t
lay
er
s
an
d
af
ter
war
d
av
er
ag
in
g
t
h
e
weig
h
ts
am
o
n
g
r
esp
ec
tiv
e
h
ea
d
s
in
s
id
e
th
is
lay
er
.
T
h
e
weig
h
ts
ass
ig
n
ed
to
ea
ch
wo
r
d
ar
e
t
h
en
av
er
a
g
ed
o
v
e
r
th
e
en
tire
p
h
r
ase.
T
h
e
m
o
d
el
g
iv
es th
e
wo
r
d
a
h
ig
h
er
atten
tio
n
weig
h
t w
h
en
class
if
y
in
g
a
s
en
ten
ce
if
it h
as a
s
tr
o
n
g
er
h
ig
h
lig
h
t/imp
o
r
tan
ce
.
T
h
e
b
lo
ck
d
iag
r
a
m
in
Fig
u
r
e
3
s
h
o
ws h
o
w
th
e
B
i
-
L
STM
+A
tten
tio
n
m
o
d
el
is
u
s
ed
to
id
e
n
tify
s
ar
ca
s
m
.
Fig
u
r
e
3
.
B
i
-
L
STM
with
a
tten
tio
n
tex
t sar
ca
s
m
d
etec
to
r
b
lo
c
k
d
iag
r
am
3.
5
.
1
.
Arc
hite
ct
ure
o
f
t
he
B
i
-
L
S
T
M
-
ba
s
e
d a
t
t
ent
io
n m
o
de
l
T
h
e
f
o
llo
win
g
T
ab
le
1
d
etail
s
th
e
d
if
f
er
en
t
lay
er
s
in
clu
d
ed
in
th
e
B
i
-
L
STM
m
o
d
el,
f
e
atu
r
in
g
an
atten
tiv
e
lay
er
an
d
s
ev
er
al
ad
d
itio
n
al
lay
er
s
u
s
ed
f
o
r
th
e
p
la
n
n
ed
s
tu
d
y
o
n
s
ar
ca
s
m
d
etec
tio
n
.
T
h
e
s
u
m
m
ar
y
o
f
th
e
B
i
-
L
STM
ap
p
r
o
ac
h
u
tili
ze
d
to
u
n
d
e
r
s
tan
d
s
ar
ca
s
m
is
d
es
cr
ib
ed
in
T
ab
le
1
.
T
h
e
Ker
as
lib
r
ar
y
was
u
s
ed
to
b
u
ild
th
e
m
o
d
el
s
u
m
m
ar
y
ab
o
v
e,
h
ig
h
lig
h
tin
g
th
e
s
ev
e
r
al
lay
er
s
o
f
f
er
ed
th
r
o
u
g
h
o
u
t
th
e
B
i
-
L
STM
s
tr
u
ctu
r
e
co
m
p
r
is
in
g
th
e
atten
ti
o
n
lay
er
.
Ad
d
itio
n
al
lay
er
s
,
esp
ec
ially
d
r
o
p
o
u
t
an
d
th
ick
lay
e
r
s
h
o
ld
i
n
g
ce
r
tain
u
n
its
ar
e
ad
d
ed
to
in
cr
ea
s
e
class
if
icatio
n
p
er
f
o
r
m
an
ce
.
T
h
e
wo
r
d
em
b
ed
d
in
g
v
ec
to
r
o
b
tain
ed
b
y
th
e
"Glo
Ve"
em
b
ed
d
in
g
ap
p
r
o
ac
h
is
f
ed
in
to
th
e
d
ee
p
lear
n
in
g
n
etwo
r
k
.
T
h
e
id
ea
o
f
atten
tio
n
is
ac
co
m
p
lis
h
ed
b
y
u
s
in
g
lay
er
s
s
u
ch
as p
er
m
u
te,
m
u
ltip
ly
,
an
d
lam
b
d
a.
T
ab
le
1
.
B
i
-
L
STM
m
o
d
el
with
atten
tio
n
–
Ker
as su
m
m
a
r
y
La
y
e
r
(
t
y
p
e
)
O
u
t
p
u
t
S
h
a
p
e
P
a
r
a
m
#
C
o
n
n
e
c
t
e
d
t
o
i
n
p
u
t
4
(
I
n
p
u
t
La
y
e
r
)
[
(
N
o
n
e
,
5
0
)
]
0
[
]
e
mb
e
d
d
i
n
g
3
(
Em
b
e
d
d
i
n
g
)
(
N
o
n
e
,
5
0
,
3
0
0
)
7
1
9
2
5
0
0
[
'i
n
p
u
t
_
4
[
0
]
[
0
]
']
B
i
d
i
r
e
c
t
i
o
n
a
l
2
(
B
i
d
i
r
e
c
t
i
o
n
a
l
)
(
N
o
n
e
,
5
0
,
1
2
8
)
4
3
9
2
9
6
[
'emb
e
d
d
i
n
g
_
3
[
0
]
[
0
]
']
d
e
n
se
_
1
0
(
D
e
n
s
e
)
(
N
o
n
e
,
5
0
,
1
)
1
2
9
[
'b
i
d
i
r
e
c
t
i
o
n
a
l
_
2
[
0
]
[
0
]
']
f
l
a
t
t
e
n
_
1
(
F
l
a
t
t
e
n
)
(
N
o
n
e
,
5
0
)
0
[
'd
e
n
se
_
1
0
[
0
]
[
0
]
']
a
c
t
i
v
a
t
i
o
n
1
(
A
c
t
i
v
a
t
i
o
n
)
(
N
o
n
e
,
5
0
)
0
[
'f
l
a
t
t
e
n
_
1
[
0
]
[
0
]
']
r
e
p
e
a
t
_
v
e
c
t
o
r
_
1
(
R
e
p
e
a
t
V
e
c
t
o
r
)
(
N
o
n
e
,
1
2
8
,
5
0
)
0
[
'ac
t
i
v
a
t
i
o
n
1
[
0
]
[
0
]
]
p
e
r
m
u
t
e
_
1
(
P
e
r
m
u
t
e
)
(
N
o
n
e
,
5
0
,
1
2
8
0
[
'r
e
p
e
a
t
_
v
e
c
t
o
r
_
1
[
0
]
[
0
]
']
mu
l
t
i
p
l
y
1
(
M
u
l
t
i
p
l
y
)
(
N
o
n
e
,
5
0
,
1
2
8
)
0
[
'b
i
d
i
r
e
c
t
i
o
n
a
l
_
2
[
0
]
[
0
]
',
'
p
e
r
m
u
t
e
_
1
[
0
]
[
0
]
']
l
a
m
b
d
a
_
1
(
L
a
m
b
d
a
)
(
N
o
n
e
,
1
2
8
)
0
[
'mu
l
t
i
p
l
y
_
1
[
0
]
[
0
]
']
d
r
o
p
o
u
t
6
(
D
r
o
p
o
u
t
)
(
N
o
n
e
,
1
2
8
)
0
[
'l
a
mb
d
a
_
1
[
0
]
[
0
]
']
d
e
n
se
1
1
(
D
e
n
se)
(
N
o
n
e
,
2
5
6
)
3
3
0
2
4
[
'd
r
o
p
o
u
t
_
6
[
0
]
[
0
]
']
d
e
n
se
_
1
2
(
D
e
n
s
e
)
(
N
o
n
e
,
1
2
8
)
3
2
8
9
6
[
'd
e
n
se
_
1
1
[
0
]
[
0
]
']
d
e
n
se
_
1
3
(
D
e
n
s
e
)
(
N
o
n
e
,
1
)
1
2
9
[
'd
e
n
se
1
2
[
0
]
[
0
]
']
To
t
a
l
p
a
r
a
ms
:
7
,
6
9
7
,
9
7
4
Tr
a
i
n
a
b
l
e
p
a
r
a
ms
:
5
0
5
,
4
7
4
N
o
n
-
t
r
a
i
n
a
b
l
e
p
a
r
a
ms:
7
,
1
9
2
,
5
0
0
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:
2088
-
8
7
0
8
Dee
p
lea
r
n
in
g
-
b
a
s
ed
a
tten
tio
n
mo
d
els fo
r
s
a
r
ca
s
m
d
etec
tio
n
in
text
(
Ga
n
esh
C
h
a
n
d
r
a
s
ek
a
r
a
n
)
6791
3
.
6
.
Rec
o
g
nizin
g
s
a
rc
a
s
m
co
m
bin
ing
CNN,
B
i
-
L
S
T
M
,
a
n
d a
t
t
ent
io
n m
o
del
T
o
en
h
an
ce
t
h
e
m
o
d
el'
s
ef
f
icac
y
,
th
e
p
o
te
n
tial
b
en
e
f
its
o
f
two
m
ajo
r
m
o
d
els,
i.e
.
,
C
NN
an
d
Bi
-
L
STM
,
ar
e
m
er
g
ed
in
th
is
s
ch
em
e
f
o
r
s
ar
ca
s
m
d
etec
tio
n
.
T
h
e
c
o
n
v
o
lu
tio
n
ca
n
r
etr
iev
e
th
e
m
o
s
t
ch
ar
ac
ter
is
tics
f
r
o
m
th
e
in
p
u
t
tex
tu
al
in
f
o
r
m
atio
n
,
wh
er
ea
s
th
e
B
i
-
L
STM
m
o
d
el
ca
n
co
llect
d
ata
b
o
th
way
s
.
T
h
e
B
i
-
L
STM
m
o
d
el'
s
u
s
ag
e
o
f
v
ar
io
u
s
g
ates
aid
s
in
d
is
r
eg
ar
d
in
g
u
n
d
esire
d
wo
r
d
s
/p
h
r
ases
.
T
h
is
m
eth
o
d
wo
r
k
s
b
est
wh
en
d
ea
lin
g
wit
h
a
b
r
ief
an
d
co
n
tin
u
o
u
s
f
lo
w
o
f
in
f
o
r
m
atio
n
.
T
h
e
co
n
v
o
lu
tio
n
lay
er
ex
tr
ac
ts
r
elev
an
t
attr
ib
u
tes
f
r
o
m
te
x
t
in
f
o
r
m
atio
n
,
wh
ile
th
e
B
i
-
L
STM
m
o
d
el
ex
tr
ac
ts
co
n
tex
t
u
a
l
in
f
o
r
m
atio
n
.
T
h
e
b
lo
ck
d
ia
g
r
am
o
f
th
e
e
n
v
is
io
n
ed
C
NN+
B
i
-
L
STM
ap
p
r
o
ac
h
f
o
r
s
ar
ca
s
m
r
ec
o
g
n
itio
n
is
s
h
o
wn
in
Fig
u
r
e
4
.
A
s
in
g
le
C
NN
m
o
d
el
ig
n
o
r
es
th
e
s
em
an
tic
an
d
g
r
am
m
atica
l
co
n
ten
t
o
f
wo
r
d
s
in
th
e
co
n
te
x
t,
an
d
th
e
ch
allen
g
e
o
f
v
a
n
is
h
in
g
g
r
ad
ie
n
ts
in
class
ic
r
ec
u
r
r
en
t
n
eu
r
al
n
etwo
r
k
s
(
R
NN)
is
ea
s
ily
h
a
n
d
led
u
s
in
g
L
STM
.
Af
ter
p
r
ep
r
o
ce
s
s
in
g
th
e
co
r
p
u
s
,
th
e
tex
t
is
en
co
d
ed
li
k
e
a
t
wo
-
d
im
en
s
io
n
al
w
o
r
d
v
ec
to
r
m
atr
ix
,
an
d
th
e
lo
ca
l
in
f
o
r
m
atio
n
f
ea
tu
r
es
ar
e
r
ec
o
v
er
ed
u
s
in
g
C
NN.
T
h
e
in
f
o
r
m
atio
n
is
th
en
s
en
t
to
B
iLST
M,
wh
ich
lear
n
s
th
e
lin
k
b
etwe
en
wo
r
d
s
an
d
p
h
r
ases
.
Fo
llo
win
g
th
e
B
i
-
L
STM
p
ar
ad
ig
m
,
th
e
atten
tio
n
lay
e
r
ca
p
tu
r
es
v
ital
tex
tu
al
in
f
o
r
m
atio
n
wh
ile
ig
n
o
r
in
g
t
h
e
u
n
in
ter
esti
n
g
ter
m
s
.
Fig
u
r
e
4
.
T
e
x
t sar
ca
s
m
r
ec
o
g
n
itio
n
em
p
lo
y
i
n
g
C
NN+
B
iLST
M
with
a
tten
tio
n
3.
6
.
1
.
CNN+B
i
-
L
ST
M
m
o
del w
it
h a
t
t
ent
io
n la
y
er
T
h
e
T
ab
le
2
p
r
o
v
id
es
th
e
d
eta
ils
o
f
th
e
s
ev
e
r
al
lay
er
s
in
th
e
p
r
o
p
o
s
ed
n
etwo
r
k
.
T
ab
le
2
s
u
m
m
ar
izes
th
e
Ker
as
d
escr
ip
tio
n
o
f
th
e
C
NN+
B
i
-
L
STM
n
etwo
r
k
,
wit
h
atten
tio
n
b
ein
g
u
s
ed
f
o
r
s
ar
ca
s
m
d
etec
tio
n
.
T
h
e
Ker
as
lib
r
ar
y
was
u
s
ed
to
co
n
s
tr
u
ct
th
e
m
o
d
el
s
u
m
m
ar
y
a
b
o
v
e,
wh
ic
h
h
ig
h
lig
h
ts
th
e
v
ar
io
u
s
lev
els
m
ad
e
ac
ce
s
s
ib
le
th
r
o
u
g
h
o
u
t
th
e
C
NN
an
d
B
i
-
L
STM
f
r
am
ewo
r
k
th
at
co
m
p
r
is
e
th
e
atten
tio
n
lay
er
.
T
o
im
p
r
o
v
e
class
if
icatio
n
p
er
f
o
r
m
an
ce
,
f
u
r
th
er
s
tag
es
a
r
e
a
d
d
ed
,
esp
ec
i
ally
lay
er
s
o
f
d
r
o
p
o
u
ts
an
d
th
ick
lay
er
s
h
o
ld
i
n
g
p
ar
ticu
lar
u
n
its
.
T
h
e
"Glo
Ve"
em
b
ed
d
in
g
ap
p
r
o
ac
h
y
ield
s
a
wo
r
d
em
b
e
d
d
in
g
v
ec
to
r
,
w
h
ich
is
in
jecte
d
in
to
th
e
d
ee
p
lear
n
in
g
n
etwo
r
k
.
T
h
e
co
n
ce
p
t
o
f
atten
tio
n
is
ac
h
iev
ed
u
s
in
g
lay
er
s
lik
e
p
er
m
u
te,
m
u
ltip
ly
,
an
d
lam
b
d
a.
T
ab
le
2
.
Ker
as d
escr
ip
tio
n
o
f
C
NN+
B
i
-
L
STM
n
etwo
r
k
f
o
r
s
ar
ca
s
m
d
etec
tio
n
with
atten
tio
n
La
y
e
r
(
t
y
p
e
)
O
u
t
p
u
t
S
h
a
p
e
P
a
r
a
m
#
C
o
n
n
e
c
t
e
d
t
o
i
n
p
u
t
4
(
I
n
p
u
t
La
y
e
r
)
[
(
N
o
n
e
,
5
0
)
]
0
[
]
e
mb
e
d
d
i
n
g
3
(
Em
b
e
d
d
i
n
g
)
(
N
o
n
e
,
5
0
,
3
0
0
)
7
1
9
2
5
0
0
[
'i
n
p
u
t
_
4
[
0
]
[
0
]
']
d
r
o
p
o
u
t
7
(
D
r
o
p
o
u
t
)
(
N
o
n
e
,
1
2
8
)
0
[
'emb
e
d
d
i
n
g
_
3
[
0
]
[
0
]
']
C
o
n
v
1
d
(
C
o
n
v
1
D
)
(
N
o
n
e
,
4
6
,
3
2
)
4
8
0
3
2
[
'd
r
o
p
o
u
t
_
6
[
0
]
[
0
]
']
b
i
d
i
r
e
c
t
i
o
n
a
l
2
(
B
i
d
i
r
e
c
t
i
o
n
a
l
)
(
N
o
n
e
,
5
0
,
1
2
8
)
4
3
9
2
9
6
[
'emb
e
d
d
i
n
g
_
3
[
0
]
[
0
]
']
d
e
n
se
_
1
0
(
D
e
n
s
e
)
(
N
o
n
e
,
5
0
,
1
)
1
2
9
[
'b
i
d
i
r
e
c
t
i
o
n
a
l
_
2
[
0
]
[
0
]
']
f
l
a
t
t
e
n
_
1
(
F
l
a
t
t
e
n
)
(
N
o
n
e
,
5
0
)
0
[
'd
e
n
se
_
1
0
[
0
]
[
0
]
']
a
c
t
i
v
a
t
i
o
n
1
(
A
c
t
i
v
a
t
i
o
n
)
(
N
o
n
e
,
5
0
)
0
[
'f
l
a
t
t
e
n
_
1
[
0
]
[
0
]
']
r
e
p
e
a
t
_
v
e
c
t
o
r
_
1
(
R
e
p
e
a
t
V
e
c
t
o
r
)
(
N
o
n
e
,
1
2
8
,
5
0
)
0
[
'ac
t
i
v
a
t
i
o
n
1
[
0
]
[
0
]
]
p
e
r
m
u
t
e
_
1
(
P
e
r
m
u
t
e
)
(
N
o
n
e
,
5
0
,
1
2
8
0
[
'r
e
p
e
a
t
_
v
e
c
t
o
r
_
1
[
0
]
[
0
]
']
mu
l
t
i
p
l
y
1
(
M
u
l
t
i
p
l
y
)
(
N
o
n
e
,
5
0
,
1
2
8
)
0
[
'b
i
d
i
r
e
c
t
i
o
n
a
l
_
2
[
0
]
[
0
]
',
'
p
e
r
m
u
t
e
_
1
[
0
]
[
0
]
']
l
a
m
b
d
a
_
1
(
L
a
m
b
d
a
)
(
N
o
n
e
,
1
2
8
)
0
[
'mu
l
t
i
p
l
y
_
1
[
0
]
[
0
]
']
d
r
o
p
o
u
t
7
(
D
r
o
p
o
u
t
)
(
N
o
n
e
,
1
2
8
0
[
'l
a
mb
d
a
_
1
[
0
]
[
0
]
']
d
e
n
se
1
1
(
D
e
n
se)
(
N
o
n
e
,
2
5
6
)
3
3
0
2
4
[
'd
r
o
p
o
u
t
_
7
[
0
]
[
0
]
']
d
e
n
se
_
1
2
(
D
e
n
s
e
)
(
N
o
n
e
,
1
2
8
)
3
2
8
9
6
[
'd
e
n
se
_
1
1
[
0
]
[
0
]
']
d
e
n
se
_
1
3
(
D
e
n
s
e
)
(
N
o
n
e
,
1
)
1
2
9
[
'd
e
n
se
1
2
[
0
]
[
0
]
']
To
t
a
l
p
a
r
a
ms
:
7
,
4
7
1
,
5
7
4
Tr
a
i
n
a
b
l
e
p
a
r
a
ms
:
2
7
9
,
0
7
4
N
o
n
-
t
r
a
i
n
a
b
l
e
p
a
r
a
ms:
7
,
1
9
2
,
5
0
0
4.
E
XP
E
R
I
M
E
N
T
A
L
RE
SUL
T
S AN
D
D
I
SC
USS
I
O
NS
4.
1
.
Resul
t
s
T
h
e
s
o
u
r
ce
co
d
in
g
f
o
r
th
is
tex
t
s
ar
ca
s
m
p
r
o
jectio
n
ex
p
e
r
im
e
n
t
was
d
o
n
e
in
Py
th
o
n
3
.
5
wi
th
th
e
aid
o
f
Go
o
g
le
C
o
llab
o
r
ato
r
y
.
I
t
u
s
es
an
id
ea
l
T
esla
GPU
f
ea
tu
r
in
g
a
2
.
3
0
GHz
clo
ck
s
p
ee
d
an
d
1
6
GB
o
f
R
AM
.
T
h
e
Ker
as
Py
th
o
n
p
ac
k
ag
e
is
u
s
ed
to
cr
ea
te
th
e
r
ef
in
e
d
tr
an
s
f
er
lear
n
in
g
m
o
d
el.
T
h
e
r
esu
lts
wer
e
p
lo
tted
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
14
,
No
.
6
,
Dec
em
b
e
r
20
24
:
6
7
8
6
-
6
7
9
6
6792
u
s
in
g
th
e
Ma
tPl
o
tLib
p
ac
k
ag
e,
an
d
th
e
co
n
f
u
s
io
n
m
atr
ix
was
cr
ea
ted
u
s
in
g
th
e
Scik
it
-
L
ea
r
n
lib
r
ar
y
.
T
h
e
Glo
v
e
(
g
lo
ve
.
6
B
.
3
0
0
d
.
txt
)
wo
r
d
em
b
e
d
d
in
g
is
em
p
lo
y
ed
t
o
g
en
er
ate
w
o
r
d
v
ec
to
r
s
b
ef
o
r
e
b
u
ild
in
g
u
p
o
f
th
e
d
ee
p
lear
n
in
g
m
o
d
el.
4
.
1
.
1
.
Sa
rc
a
s
m
V2
d
a
t
a
s
et
Sar
ca
s
tic
ar
ticles
ar
e
in
clu
d
ed
in
th
is
d
ataset
(
h
ttp
s
:
//n
ld
s
.
s
o
e.
u
csc.
ed
u
/s
a
r
ca
s
m2
)
.
T
h
e
Gen
er
ic
(
Gen
)
ca
teg
o
r
y
in
clu
d
es
g
e
n
e
r
al
s
ar
ca
s
tic
co
m
m
en
t
s
eq
u
e
n
ce
s
.
Hy
p
er
b
o
le
co
n
tain
s
in
f
lated
,
ev
en
ab
s
u
r
d
,
d
ec
ep
tio
n
s
th
at
s
h
o
u
ld
n
o
t
b
e
tak
en
liter
ally
.
A
r
h
eto
r
ical
in
q
u
ir
y
co
n
s
titu
tes
a
q
u
er
y
th
at
is
ask
ed
s
o
lely
f
o
r
ef
f
ec
t
an
d
with
n
o
r
eg
ar
d
f
o
r
th
e
o
u
tco
m
e.
T
h
e
f
ea
tu
r
e
s
in
clu
d
ed
with
in
th
e
Sar
ca
s
m
C
o
r
p
u
s
V2
ar
e
r
ep
r
esen
ted
in
T
ab
le
3.
T
h
er
e
ar
e
9
,
3
8
6
i
n
s
tan
ce
s
o
f
s
ar
ca
s
m
in
th
is
d
ataset,
o
f
wh
ich
6
,
5
2
0
ar
e
g
e
n
er
al
(
G
E
N)
,
1
,
7
0
2
co
n
s
titu
te
q
u
esti
o
n
s
ab
o
u
t
r
h
eto
r
ic
(
R
Q)
,
an
d
1
,
1
6
4
ar
e
h
y
p
er
b
o
le
(
HYP)
.
T
h
e
in
p
u
t
tex
t
d
ata
(
Gen
er
ic
:
6
,
5
2
0
s
am
p
les,
r
h
eto
r
ical
q
u
esti
o
n
s
:
1
,
7
0
2
s
am
p
les,
h
y
p
e
r
b
o
le
:
1
,
1
6
4
s
am
p
les)
a
r
e
d
i
v
id
ed
in
t
o
th
e
tr
ain
(
7
,
5
0
8
s
am
p
les)
an
d
test
(
1
,
8
7
8
s
am
p
les)
with
a
v
o
ca
b
u
lar
y
s
ize
o
f
2
,
3
9
7
5
w
o
r
d
s
to
c
o
m
p
lete
th
e
s
ar
ca
s
m
d
etec
tio
n
task
.
T
h
e
wo
r
d
em
b
ed
d
in
g
s
wer
e
in
itially
s
to
r
e
d
i
n
a
d
ictio
n
ar
y
with
ter
m
s
th
at
s
h
ar
ed
th
e
s
am
e
k
ey
an
d
v
alu
e
.
T
h
e
p
r
o
p
o
s
ed
n
etwo
r
k
s
wer
e
tr
ain
e
d
o
v
er
5
0
iter
atio
n
s
u
s
in
g
th
e
'
B
in
ar
y
cr
o
s
s
-
en
tr
o
p
y
'
ap
p
r
o
ac
h
.
T
h
e
co
n
f
u
s
io
n
m
atr
i
x
d
escr
ib
i
n
g
th
e
r
ec
o
m
m
en
d
ed
s
tr
ateg
y
is
s
h
o
wn
in
Fig
u
r
e
5
.
T
ab
les
4
an
d
5
d
is
p
lay
th
e
p
er
f
o
r
m
a
n
ce
in
d
icato
r
s
o
f
t
h
e
B
i
-
d
ir
ec
tio
n
al
L
STM
an
d
C
N
N
with
B
i
-
d
ir
ec
tio
n
al
L
STM
m
o
d
els
e
m
p
lo
y
in
g
atten
tio
n
m
ec
h
a
n
is
m
.
Acc
o
r
d
i
n
g
t
o
th
e
p
er
f
o
r
m
an
ce
m
etr
ics,
th
e
C
NN+
B
i
-
L
STM
clas
s
if
ier
wit
h
atten
tio
n
lay
er
p
r
o
d
u
ce
s
th
e
b
est
ac
cu
r
ac
y
(
9
1
.
7
6
%)
f
o
r
th
e
s
p
ec
if
ied
in
p
u
t,
tex
t
d
ata,
co
m
p
a
r
ed
to
t
h
e
B
i
-
L
STM
with
atten
tio
n
(
8
6
.
2
5
%)
alg
o
r
ith
m
f
o
r
tex
t
s
ar
c
asm
d
etec
tio
n
.
T
h
e
m
o
d
el'
s
ca
p
ac
ity
to
co
n
ce
n
tr
a
te
o
n
im
p
o
r
tan
t
s
ec
tio
n
s
o
f
t
h
e
tex
t
was
m
u
ch
en
h
a
n
ce
d
b
y
ad
d
in
g
atten
tio
n
m
ec
h
an
is
m
s
,
wh
ich
in
cr
ea
s
e
d
p
er
f
o
r
m
an
ce
m
etr
ics
f
o
r
al
l
tak
en
in
to
ac
co
u
n
t.
C
o
m
p
a
r
in
g
o
u
r
r
esu
lts
to
cu
r
r
en
t sar
ca
s
m
d
etec
tio
n
m
et
h
o
d
s
p
u
b
lis
h
ed
in
th
e
liter
atu
r
e,
we
f
in
d
a
s
ig
n
i
f
ican
t im
p
r
o
v
em
en
t.
T
ab
le
3
.
Sar
ca
s
m
C
o
r
p
u
s
V2
d
etails
A
t
t
r
i
b
u
t
e
/
F
e
a
t
u
r
e
N
a
me
D
e
scri
p
t
i
o
n
C
o
r
p
u
s
Th
e
c
o
r
p
u
s
t
y
p
e
:
g
e
n
e
r
i
c
s
a
r
c
a
sm
(
G
E
N
)
,
h
y
p
e
r
b
o
l
e
(
HYP
)
,
o
r
r
h
e
t
o
r
i
c
a
l
q
u
e
st
i
o
n
s
(
RQ
).
La
b
e
l
Th
e
r
e
s
p
o
n
se
u
t
t
e
r
a
n
c
e
's
c
l
a
ss
n
a
me
-
o
n
e
o
f
"
sa
r
c
"
(
sarc
a
st
i
c
)
o
r
"
n
o
t
sa
rc
"
(
n
o
t
-
sarca
st
i
c
)
ID
A
n
i
d
e
n
t
i
f
i
c
a
t
i
o
n
n
u
m
b
e
r
f
o
r
t
h
e
p
o
s
t
.
P
a
i
r
s
c
o
n
t
a
i
n
i
n
g
i
d
e
n
t
i
c
a
l
I
D
n
u
m
b
e
r
s i
n
v
a
r
i
o
u
s
d
a
t
a
se
t
s
a
r
e
u
n
r
e
l
a
t
e
d
.
R
e
s
p
o
n
se
t
e
x
t
Th
e
p
o
s
t
's t
e
x
t
i
s m
a
r
k
e
d
w
i
t
h
s
a
r
c
a
s
m (i.
e
.
t
h
i
s
u
t
t
e
r
a
n
c
e
i
s a
sso
c
i
a
t
e
d
w
i
t
h
t
h
e
sar
c
a
sm
l
a
b
e
l
)
.
Fig
u
r
e
5
.
Dee
p
lear
n
in
g
m
o
d
els with
atten
tio
n
–
co
n
f
u
s
io
n
m
a
tr
ix
T
ab
le
4
.
Bi
-
L
STM
s
y
s
tem
p
er
f
o
r
m
an
ce
m
ea
s
u
r
es with
em
p
h
asis
o
n
s
ar
ca
s
m
id
en
tific
atio
n
P
e
r
f
o
r
ma
n
c
e
me
t
r
i
c
s
V
a
l
u
e
s
o
b
t
a
i
n
e
d
S
e
n
s
i
t
i
v
i
t
y
8
0
.
7
2
S
p
e
c
i
f
i
c
i
t
y
9
1
.
2
7
P
r
e
c
i
s
i
o
n
8
8
.
5
4
F1
-
S
c
o
r
e
8
4
.
6
7
A
c
c
u
r
a
c
y
8
6
.
2
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:
2088
-
8
7
0
8
Dee
p
lea
r
n
in
g
-
b
a
s
ed
a
tten
tio
n
mo
d
els fo
r
s
a
r
ca
s
m
d
etec
tio
n
in
text
(
Ga
n
esh
C
h
a
n
d
r
a
s
ek
a
r
a
n
)
6793
T
ab
le
5
.
Per
f
o
r
m
an
ce
m
etr
ics o
f
C
NN+
B
i
-
L
STM
s
y
s
tem
wit
h
atten
tio
n
f
o
r
r
ec
o
g
n
izin
g
s
ar
ca
s
m
d
etec
tio
n
P
e
r
f
o
r
ma
n
c
e
me
t
r
i
c
s
V
a
l
u
e
s
o
b
t
a
i
n
e
d
S
e
n
s
i
t
i
v
i
t
y
9
0
.
2
7
S
p
e
c
i
f
i
c
i
t
y
9
0
.
3
8
P
r
e
c
i
s
i
o
n
8
7
.
5
6
F1
-
S
c
o
r
e
8
9
.
2
5
A
c
c
u
r
a
c
y
9
1
.
7
6
4
.
2
.
Dis
cus
s
io
n
T
h
e
m
ain
co
n
cl
u
s
io
n
f
r
o
m
o
u
r
f
in
d
in
g
s
is
th
e
ef
f
icac
y
o
f
th
e
s
u
g
g
ested
h
y
b
r
id
m
o
d
el
(
C
NN+
B
i
-
L
STM
+
Atten
tio
n
)
in
s
ar
ca
s
m
d
etec
tio
n
in
tex
tu
al
d
ata.
Sp
ec
if
ically
,
th
e
atten
tio
n
p
r
o
ce
s
s
im
p
r
o
v
es
th
e
m
o
d
el'
s
ca
p
ac
it
y
to
ass
ess
th
e
r
elativ
e
r
elev
a
n
ce
o
f
v
ar
io
u
s
wo
r
d
s
a
n
d
p
h
r
ases
,
wh
ich
r
aises
d
etec
tio
n
ac
cu
r
ac
y
.
T
ab
le
6
d
em
o
n
s
tr
ates
co
n
clu
s
iv
ely
h
o
w
m
u
ch
b
etter
o
u
r
m
o
d
el
p
e
r
f
o
r
m
s
th
an
c
u
r
r
en
t
m
o
d
els.
Ou
r
ap
p
r
o
ac
h
clea
r
ly
o
u
tp
er
f
o
r
m
s
alter
n
ativ
e
ap
p
r
o
ac
h
es wh
en
co
m
p
ar
e
d
to
p
r
ev
io
u
s
s
tu
d
y
f
in
d
in
g
s
.
T
h
is
en
h
an
ce
m
en
t
h
ig
h
lig
h
ts
th
e
b
en
ef
it
o
f
ap
p
ly
i
n
g
atten
tio
n
m
ec
h
an
is
m
s
in
d
ee
p
le
ar
n
in
g
s
y
s
tem
s
f
o
r
s
ar
ca
s
m
d
etec
tio
n
.
Sev
er
al
a
r
tific
ial
in
tellig
en
ce
a
n
d
d
ee
p
l
ea
r
n
in
g
m
eth
o
d
o
lo
g
ies
h
av
e
b
ee
n
u
s
ed
in
a
lar
g
e
n
u
m
b
er
o
f
p
u
b
lis
h
ed
r
esear
ch
p
r
o
jects
to
ev
alu
ate
tex
t
s
ar
ca
s
m
.
W
e
h
av
e
co
n
tr
asted
o
u
r
r
esear
ch
with
o
th
er
r
elev
an
t
s
tu
d
ies
co
n
d
u
cte
d
u
s
in
g
th
e
Sar
ca
s
m
C
o
r
p
u
s
V2
co
llectio
n
.
T
h
e
f
o
llo
win
g
t
ab
le
co
m
p
ar
es
o
u
r
m
eth
o
d
f
o
r
tex
t sar
ca
s
m
d
etec
t
io
n
to
v
a
r
io
u
s
ex
is
tin
g
s
tu
d
ies
in
T
ab
le
6
.
Ou
r
s
tu
d
y
'
s
th
o
r
o
u
g
h
ass
ess
m
en
t
o
f
v
ar
io
u
s
m
o
d
el
d
esig
n
s
,
wh
ich
em
p
h
asizes
th
e
ef
f
icien
cy
o
f
atten
tio
n
p
r
o
ce
s
s
es,
is
o
n
e
o
f
its
k
ey
s
tr
en
g
th
s
.
B
u
t
th
er
e
ar
e
s
ev
er
al
r
estricti
o
n
s
to
tak
e
i
n
to
ac
co
u
n
t:
C
o
m
p
u
tin
g
co
m
p
lex
ity
:
n
o
t
all
ex
p
er
ts
m
ay
b
e
ab
le
to
tr
ain
th
e
s
u
g
g
ested
m
o
d
el
d
u
e
to
its
s
ig
n
if
ican
t
co
m
p
u
tin
g
r
eq
u
ir
em
en
ts
.
Data
d
ep
e
n
d
en
cy
:
t
h
e
ca
lib
er
an
d
v
o
lu
m
e
o
f
t
h
e
s
am
p
le
d
ataset
s
ig
n
if
ican
tly
im
p
ac
t
th
e
m
o
d
e
l's
p
er
f
o
r
m
an
ce
.
T
h
e
m
o
d
el'
s
g
en
er
aliza
b
ilit
y
m
ay
b
e
im
p
a
cted
b
y
in
c
o
m
p
lete
o
r
s
k
ewe
d
d
ata.
T
h
e
r
esu
lts
o
f
th
e
cu
r
r
e
n
t
in
v
esti
g
atio
n
h
av
e
s
ig
n
if
ican
t
r
am
if
icatio
n
s
f
o
r
th
e
cr
ea
tio
n
o
f
s
en
tim
en
t
an
aly
s
is
alg
o
r
ith
m
s
th
at
ar
e
m
o
r
e
p
r
ec
is
e.
T
h
es
e
alg
o
r
ith
m
s
'
ab
ilit
y
to
r
ec
o
g
n
ize
s
ar
ca
s
m
well
allo
ws
th
em
to
p
r
o
v
id
e
s
en
t
im
en
t
r
atin
g
s
th
at
ar
e
m
o
r
e
ac
cu
r
ate,
wh
ich
im
p
r
o
v
es
ap
p
licatio
n
s
s
u
ch
as
cu
s
to
m
er
f
ee
d
b
ac
k
an
al
y
s
is
an
d
s
o
cial
m
ed
ia
m
o
n
ito
r
in
g
.
Po
ten
tial
ar
ea
s
o
f
f
u
tu
r
e
r
esear
ch
m
ig
h
t
i
n
clu
d
e:
i)
C
r
ea
tin
g
s
tr
ateg
ies
to
lo
we
r
th
e
m
o
d
el'
s
co
m
p
u
tin
g
d
em
an
d
s
with
o
u
t
co
m
p
r
o
m
is
in
g
its
f
u
n
ctio
n
ality
an
d
ii)
T
o
test
th
e
m
o
d
el's
g
en
er
aliza
b
ilit
y
f
u
r
th
er
,
co
n
s
id
er
ev
al
u
atin
g
it
o
n
a
wid
er
v
ar
iety
o
f
d
atasets
,
s
u
ch
a
s
th
o
s
e
f
r
o
m
m
u
ltip
le
s
o
cial
m
e
d
ia
s
ites
an
d
lan
g
u
ag
es.
T
ab
le
6
.
Per
f
o
r
m
an
ce
m
etr
ics o
f
ex
is
tin
g
wo
r
k
s
v
er
s
u
s
th
e
s
u
g
g
ested
f
o
r
tex
t sar
ca
s
m
d
ete
ctio
n
A
u
t
h
o
r
D
a
t
a
s
e
t
M
e
t
h
o
d
o
l
o
g
y
u
se
d
O
u
t
c
o
m
e
s
G
o
n
z
á
l
e
z
-
I
b
á
ñ
e
z
e
t
a
l
.
[
2
4
]
Tw
i
t
t
e
r
F
e
a
t
u
r
e
e
x
t
r
a
c
t
i
o
n
u
si
n
g
u
n
i
g
r
a
m
a
n
d
d
i
c
t
i
o
n
a
r
y
m
e
t
h
o
d
t
o
o
b
t
a
i
n
Le
x
i
c
a
l
a
n
d
p
r
a
g
ma
t
i
c
f
e
a
t
u
r
e
.
S
V
M
c
l
a
ssi
f
i
e
r
A
c
c
u
r
a
c
y
:
7
5
.
8
9
%
(
p
o
l
a
r
i
t
y
-
b
a
s
e
d
c
l
a
ss
i
f
i
c
a
t
i
o
n
)
Lu
n
a
n
d
o
e
t
a
l
.
[
2
5
]
S
o
c
i
a
l
me
d
i
a
d
a
t
a
o
f
I
n
d
o
n
e
si
a
N
e
g
a
t
i
v
i
t
y
f
r
o
m
t
h
e
st
a
t
e
m
e
n
t
s w
i
t
h
t
h
e
i
r
c
o
r
r
e
l
a
t
i
o
n
w
i
t
h
w
o
r
d
s
.
Tr
a
d
i
t
i
o
n
a
l
mac
h
i
n
e
l
e
a
r
n
i
n
g
met
h
o
d
s
a
r
e
u
s
e
d
.
A
c
c
u
r
a
c
y
:
5
3
.
1
%
(
N
a
ï
v
e
B
a
y
e
s)
A
c
c
u
r
a
c
y
:
5
3
.
8
%
(
M
a
x
i
mu
m E
n
t
r
o
p
y
)
A
c
c
u
r
a
c
y
:
5
4
.
1
%
(
S
V
M
)
B
a
r
b
i
e
r
i
e
t
a
l
.
[
2
6
]
Tw
i
t
t
e
r
F
e
a
t
u
r
e
s
l
i
k
e
w
o
r
d
s fr
e
q
u
e
n
c
y
,
w
r
i
t
i
n
g
a
n
d
s
p
e
a
k
i
n
g
s
t
y
l
e
,
I
mp
a
c
t
o
f
w
o
r
d
s
,
S
t
r
u
c
t
u
r
a
l
f
o
r
ma
t
i
o
n
,
c
o
n
t
e
x
t
u
a
l
sen
t
i
m
e
n
t
s
a
n
d
s
y
n
o
n
y
ms.
D
e
c
i
si
o
n
t
r
e
e
a
s c
l
a
ss
i
f
i
e
r
P
r
e
c
i
s
i
o
n
:
9
0
%
,
R
e
c
a
l
l
:
9
0
%
(
P
o
l
i
t
i
c
s
)
P
r
e
c
i
s
i
o
n
:
9
9
%
,
R
e
c
a
l
l
:
9
6
%
(
N
e
w
sp
a
p
e
r
)
P
r
e
c
i
s
i
o
n
:
8
8
%
,
R
e
c
a
l
l
:
8
7
%
(
H
u
m
o
r
)
P
r
e
c
i
s
i
o
n
:
8
7
%
,
R
e
c
a
l
l
:
9
0
%
(
E
d
u
c
a
t
i
o
n
)
Tu
n
g
t
h
a
mt
h
i
t
i
e
t
a
l
.
[
2
7
]
Tw
i
t
t
e
r
n
-
g
r
a
m
f
e
a
t
u
r
e
s,
p
a
t
t
e
r
n
f
e
a
t
u
r
e
s a
n
d
P
u
n
c
t
u
a
t
i
o
n
f
e
a
t
u
r
e
s
a
r
e
e
x
t
r
a
c
t
e
d
.
S
e
mi
S
u
p
e
r
v
i
se
d
c
l
a
ss
i
f
i
c
a
t
i
o
n
me
t
h
o
d
.
R
e
c
a
l
l
:
7
9
%
P
r
e
c
i
si
o
n
:
7
8
%
F
-
mea
su
r
e
:
7
9
%
A
c
c
u
r
a
c
y
:
7
9
.
5
%.
B
o
u
a
z
i
z
i
e
t
a
l
.
[
9
]
Tw
i
t
t
e
r
o
n
l
i
n
e
d
a
t
a
P
a
t
t
e
r
n
f
e
a
t
u
r
e
s
a
n
d
C
N
N
A
c
c
u
r
a
c
y
:
8
3
.
1
0
%
P
r
e
c
i
s
i
o
n
:
9
1
.
1
0
%
R
a
j
a
d
e
si
n
g
a
n
e
t
a
l
.
[
2
8
]
Tw
i
t
t
e
r
S
C
U
B
A
a
p
p
r
o
a
c
h
t
o
p
r
e
d
i
c
t
t
h
e
p
r
e
v
i
o
u
s
t
w
e
e
t
s
o
f
u
sers
b
a
s
e
d
o
n
b
e
h
a
v
i
o
r
a
l
m
o
d
e
l
.
A
c
c
u
r
a
c
y
:
7
1
.
5
2
%
(
Emo
t
i
o
n
)
A
c
c
u
r
a
c
y
:
7
6
.
7
%
(
Te
x
t
)
Lu
k
i
n
a
n
d
W
a
l
k
e
r
[
2
9
]
D
i
a
l
o
g
u
e
a
v
a
i
l
a
b
l
e
i
n
o
n
l
i
n
e
P
a
t
t
e
r
n
b
a
se
d
f
e
a
t
u
r
e
e
x
t
r
a
c
t
i
o
n
a
n
d
B
o
o
t
s
t
r
a
p
p
i
n
g
a
p
p
r
o
a
c
h
P
r
e
c
i
s
i
o
n
:
7
5
%
R
e
c
a
l
l
:
6
2
%
G
u
p
t
a
e
t
.
a
l
.
[
3
0
]
Tw
i
t
t
e
r
d
a
t
a
s
e
t
W
o
r
d
e
m
b
e
d
d
i
n
g
i
s
a
p
p
r
o
a
c
h
.
S
t
a
t
i
st
i
c
a
l
met
h
o
d
f
o
r
c
l
a
ssi
f
i
c
a
t
i
o
n
.
A
c
c
u
r
a
c
y
:
7
4
.
5
5
%
J
e
n
a
e
t
.
a
l
.
[
3
1
]
R
e
d
d
i
t
a
n
d
Tw
i
t
t
e
r
d
a
t
a
s
e
t
C
o
n
t
e
x
t
u
a
l
n
e
t
w
o
r
k
F1
-
sc
o
r
e
:
7
5
%
(
Tw
i
t
t
e
r
)
F1
-
sc
o
r
e
:
6
6
.
3
1
%
(
R
e
d
d
i
t
)
T
h
e
p
r
o
p
o
s
e
d
w
o
r
k
(
C
N
N
+
B
i
-
L
S
T
M
+
a
t
t
e
n
t
i
o
n
)
S
a
r
c
a
sm
C
o
r
p
u
s V
2
d
a
t
a
se
t
C
N
N
,
LST
M
a
n
d
B
i
-
LSTM
w
i
t
h
G
l
o
v
e
w
o
r
d
e
m
b
e
d
d
i
n
g
A
c
c
u
r
a
c
y
:
9
1
.
7
6
%
P
r
e
c
i
s
i
o
n
:
8
7
.
5
6
%
F1
-
S
c
o
r
e
–
8
9
.
2
5
%
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
14
,
No
.
6
,
Dec
em
b
e
r
20
24
:
6
7
8
6
-
6
7
9
6
6794
5.
CO
NCLU
SI
O
N
I
n
th
is
r
esear
ch
,
we
h
a
v
e
s
h
o
wn
th
at
s
ev
er
al
d
ee
p
lear
n
i
n
g
m
o
d
els
-
Glo
v
e
wo
r
d
em
b
e
d
d
in
g
s
i
n
p
ar
ticu
lar
-
ef
f
ec
ti
v
ely
d
etec
t
s
ar
ca
s
tic
m
ater
ial
with
in
th
e
Sar
ca
s
m
V2
d
ataset.
Ou
r
m
o
d
els
ar
e
b
ased
o
n
p
r
ep
r
o
ce
s
s
in
g
tex
t
with
Py
th
o
n
m
o
d
u
les
an
d
co
n
v
er
tin
g
p
h
r
ases
in
to
g
e
o
m
etr
ic
v
ec
to
r
s
u
s
in
g
Glo
Ve
em
b
ed
d
in
g
s
.
T
h
e
C
NN
co
u
p
led
with
a
B
i
-
L
STM
m
o
d
els
p
er
f
o
r
m
e
d
b
etter
th
an
th
e
o
th
e
r
ex
am
in
ed
d
esig
n
s
.
Acc
o
r
d
in
g
to
o
u
r
r
esear
c
h
,
th
e
C
NN+
B
i
-
L
STM
m
o
d
el
in
co
r
p
o
r
atin
g
atten
tio
n
m
ec
h
a
n
is
m
s
g
r
ea
tly
in
cr
ea
s
ed
ac
cu
r
ac
y
to
9
1
.
7
6
%,
b
u
t
th
e
B
i
-
L
STM
m
o
d
el
o
n
ly
r
ea
ch
e
d
8
6
.
2
5
%
ac
cu
r
ac
y
.
T
h
is
s
ig
n
if
ican
t
im
p
r
o
v
em
e
n
t
em
p
h
asizes
h
o
w
cr
u
cial
it
is
to
in
clu
d
e
atten
tio
n
p
r
o
ce
s
s
es
to
im
p
r
o
v
e
th
e
m
o
d
el'
s
ca
p
ac
ity
to
co
n
ce
n
tr
ate
o
n
th
e
m
o
s
t p
er
tin
en
t
p
ass
ag
es o
f
th
e
tex
t,
wh
ich
will im
p
r
o
v
e
t
h
e
d
etec
tio
n
o
f
s
ar
ca
s
m
.
T
h
e
r
em
a
r
k
ab
le
ef
f
icac
y
o
f
th
e
s
u
g
g
ested
a
p
p
r
o
ac
h
in
i
d
en
t
if
y
in
g
s
ar
ca
s
m
with
in
tex
t
u
al
d
ata
b
ea
r
s
s
ig
n
if
ican
t
im
p
licatio
n
s
f
o
r
th
e
s
en
tim
en
t
an
aly
s
is
d
o
m
ai
n
.
Sen
tim
en
t
m
o
d
elin
g
a
p
p
r
o
ac
h
es,
wh
ich
ar
e
em
p
lo
y
ed
in
m
an
y
a
p
p
licatio
n
s
,
in
clu
d
i
n
g
au
to
m
ated
co
n
t
en
t
m
o
d
er
atio
n
an
d
c
u
s
to
m
er
f
ee
d
b
ac
k
an
aly
s
is
m
ay
b
e
m
a
d
e
m
u
ch
m
o
r
e
r
el
iab
le
b
y
ac
cu
r
ately
d
etec
tin
g
s
ar
ca
s
m
.
Ou
r
m
o
d
el
ad
v
a
n
ce
s
th
e
m
o
r
e
g
en
er
al
o
b
jectiv
e
o
f
c
r
ea
tin
g
n
atu
r
al
lan
g
u
a
g
e
p
r
o
ce
s
s
in
g
(
NL
P)
s
y
s
tem
s
th
at
ar
e
m
o
r
e
ac
c
u
r
ate
b
y
tack
lin
g
th
e
p
r
o
b
lem
o
f
s
ar
ca
s
m
d
etec
tio
n
.
T
h
is
wo
r
k
d
em
o
n
s
tr
ates
h
o
w
cu
ttin
g
-
ed
g
e
d
ee
p
lear
n
in
g
ar
ch
itectu
r
es
m
ay
h
elp
o
v
er
c
o
m
e
th
e
d
r
aw
b
ac
k
s
o
f
co
n
v
e
n
tio
n
al
s
en
tim
en
t
an
aly
s
is
tech
n
iq
u
es.
T
h
e
s
u
p
p
o
r
tin
g
in
f
o
r
m
atio
n
o
f
th
is
m
an
u
s
cr
ip
t is
av
ailab
le
in
th
e
f
o
llo
win
g
UR
L
:
h
ttp
s
:
//g
it
h
u
b
.
co
m/Ga
n
esh
C
8
6
/S
a
r
ca
s
m
-
id
en
tifi
ca
tio
n
.
RE
F
E
R
E
NC
E
S
[
1
]
R
.
K
.
B
e
h
e
r
a
,
D
.
N
a
i
k
,
S
.
K
.
R
a
t
h
,
a
n
d
R
.
D
h
a
r
a
v
a
t
h
,
“
G
e
n
e
t
i
c
a
l
g
o
r
i
t
h
m
-
b
a
s
e
d
c
o
m
mu
n
i
t
y
d
e
t
e
c
t
i
o
n
i
n
l
a
r
g
e
-
sc
a
l
e
s
o
c
i
a
l
n
e
t
w
o
r
k
s
,
”
N
e
u
ra
l
C
o
m
p
u
t
i
n
g
a
n
d
Ap
p
l
i
c
a
t
i
o
n
s
,
v
o
l
.
3
2
,
n
o
.
1
3
,
p
p
.
9
6
4
9
–
9
6
6
5
,
J
u
l
.
2
0
2
0
,
d
o
i
:
1
0
.
1
0
0
7
/
s0
0
5
2
1
-
019
-
0
4
4
8
7
-
0.
[
2
]
A
.
A
l
s
a
y
a
t
,
“
I
mp
r
o
v
i
n
g
se
n
t
i
m
e
n
t
a
n
a
l
y
s
i
s
f
o
r
so
c
i
a
l
me
d
i
a
a
p
p
l
i
c
a
t
i
o
n
s
u
s
i
n
g
a
n
e
n
sem
b
l
e
d
e
e
p
l
e
a
r
n
i
n
g
l
a
n
g
u
a
g
e
mo
d
e
l
,
”
Ara
b
i
a
n
J
o
u
rn
a
l
f
o
r
S
c
i
e
n
c
e
a
n
d
En
g
i
n
e
e
ri
n
g
,
v
o
l
.
4
7
,
n
o
.
2
,
p
p
.
2
4
9
9
–
2
5
1
1
,
F
e
b
.
2
0
2
2
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
3
3
6
9
-
0
2
1
-
0
6
2
2
7
-
w.
[
3
]
H
.
T
h
a
k
k
a
r
a
n
d
D
.
P
a
t
e
l
,
“
A
p
p
r
o
a
c
h
e
s
f
o
r
se
n
t
i
me
n
t
a
n
a
l
y
si
s
o
n
t
w
i
t
t
e
r
:
a
st
a
t
e
-
of
-
a
r
t
st
u
d
y
,
”
a
r
Xi
v
p
re
p
ri
n
t
a
rXi
v
:
1
5
1
2
.
0
1
0
4
3
,
2
0
1
5
.
[
4
]
G
.
C
h
a
n
d
r
a
s
e
k
a
r
a
n
a
n
d
J
.
H
e
ma
n
t
h
,
“
D
e
e
p
l
e
a
r
n
i
n
g
a
n
d
Te
x
t
B
l
o
b
b
a
se
d
sen
t
i
m
e
n
t
a
n
a
l
y
s
i
s
f
o
r
c
o
r
o
n
a
v
i
r
u
s
(
C
O
V
I
D
-
1
9
)
u
si
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
r
n
a
l
o
n
Ar
t
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
T
o
o
l
s
,
v
o
l
.
3
1
,
n
o
.
1
,
2
0
2
2
,
d
o
i
:
1
0
.
1
1
4
2
/
S
0
2
1
8
2
1
3
0
2
2
5
0
0
1
1
7
.
[
5
]
M
.
S
y
k
o
r
a
,
S
.
El
a
y
a
n
,
a
n
d
T.
W
.
Jac
k
so
n
,
“
A
q
u
a
l
i
t
a
t
i
v
e
a
n
a
l
y
s
i
s o
f
s
a
r
c
a
s
m,
i
r
o
n
y
a
n
d
r
e
l
a
t
e
d
#
h
a
s
h
t
a
g
s
o
n
Tw
i
t
t
e
r
,
”
Bi
g
D
a
t
a
&
S
o
c
i
e
t
y
,
v
o
l
.
7
,
n
o
.
2
,
J
u
l
.
2
0
2
0
,
d
o
i
:
1
0
.
1
1
7
7
/
2
0
5
3
9
5
1
7
2
0
9
7
2
7
3
5
.
[
6
]
A
.
A
sh
w
i
t
h
a
,
G
.
S
h
r
u
t
h
i
,
H
.
R
.
S
h
r
u
t
h
i
,
M
.
U
p
a
d
h
y
a
y
a
,
A
.
P
.
R
a
y
,
a
n
d
T.
C
.
M
a
n
j
u
n
a
t
h
,
“
S
a
r
c
a
sm
d
e
t
e
c
t
i
o
n
i
n
n
a
t
u
r
a
l
l
a
n
g
u
a
g
e
p
r
o
c
e
ss
i
n
g
,
”
M
a
t
e
ri
a
l
s T
o
d
a
y
:
Pro
c
e
e
d
i
n
g
s
,
v
o
l
.
3
7
,
n
o
.
P
a
r
t
2
,
p
p
.
3
3
2
4
–
3
3
3
1
,
2
0
2
0
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
ma
t
p
r
.
2
0
2
0
.
0
9
.
1
2
4
.
[
7
]
P
.
N
a
g
w
a
n
s
h
i
a
n
d
C
.
E
.
V
.
M
a
d
h
a
v
a
n
,
“
S
a
r
c
a
s
m
d
e
t
e
c
t
i
o
n
u
s
i
n
g
s
e
n
t
i
m
e
n
t
a
n
d
s
e
m
a
n
t
i
c
f
e
a
t
u
r
e
s
,
”
i
n
P
r
o
c
e
e
d
i
n
g
s
o
f
t
h
e
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
K
n
o
w
l
e
d
g
e
D
i
s
c
o
v
e
r
y
a
n
d
I
n
f
o
r
m
a
t
i
o
n
R
e
t
r
i
e
v
a
l
,
2
0
1
4
,
p
p
.
4
1
8
–
4
2
4
,
d
o
i
:
1
0
.
5
2
2
0
/
0
0
0
5
1
5
3
5
0
4
1
8
0
4
2
4
.
[
8
]
B
.
S
h
a
h
a
n
d
M
.
S
h
a
h
,
“
A
s
u
r
v
e
y
o
n
mac
h
i
n
e
l
e
a
r
n
i
n
g
a
n
d
d
e
e
p
l
e
a
r
n
i
n
g
b
a
s
e
d
a
p
p
r
o
a
c
h
e
s
f
o
r
s
a
r
c
a
sm
i
d
e
n
t
i
f
i
c
a
t
i
o
n
i
n
so
c
i
a
l
med
i
a
,
”
L
e
c
t
u
r
e
N
o
t
e
s
o
n
D
a
t
a
E
n
g
i
n
e
e
ri
n
g
a
n
d
C
o
m
m
u
n
i
c
a
t
i
o
n
s
T
e
c
h
n
o
l
o
g
i
e
s
,
v
o
l
.
5
2
,
p
p
.
2
4
7
–
2
5
9
,
2
0
2
1
,
d
o
i
:
1
0
.
1
0
0
7
/
9
7
8
-
9
8
1
-
15
-
4
4
7
4
-
3
_
2
9
.
[
9
]
M
.
B
o
u
a
z
i
z
i
a
n
d
T.
O
t
su
k
i
,
“
A
p
a
t
t
e
r
n
-
b
a
s
e
d
a
p
p
r
o
a
c
h
f
o
r
s
a
r
c
a
sm
d
e
t
e
c
t
i
o
n
o
n
Tw
i
t
t
e
r
,
”
I
EEE
A
c
c
e
s
s
,
v
o
l
.
4
,
p
p
.
5
4
7
7
–
5
4
8
8
,
2
0
1
6
,
d
o
i
:
1
0
.
1
1
0
9
/
A
C
C
ESS
.
2
0
1
6
.
2
5
9
4
1
9
4
.
[
1
0
]
N
.
N
i
g
a
m
a
n
d
D
.
Y
a
d
a
v
,
“
Le
x
i
c
o
n
-
b
a
s
e
d
a
p
p
r
o
a
c
h
t
o
s
e
n
t
i
me
n
t
a
n
a
l
y
s
i
s
o
f
t
w
e
e
t
s
u
s
i
n
g
R
l
a
n
g
u
a
g
e
,
”
C
o
m
m
u
n
i
c
a
t
i
o
n
s
i
n
C
o
m
p
u
t
e
r
a
n
d
I
n
f
o
rm
a
t
i
o
n
S
c
i
e
n
c
e
,
v
o
l
.
9
0
5
,
p
p
.
1
5
4
–
1
6
4
,
2
0
1
8
,
d
o
i
:
1
0
.
1
0
0
7
/
9
7
8
-
981
-
13
-
1
8
1
0
-
8
_
1
6
.
[
1
1
]
P
.
C
l
e
w
s
a
n
d
J.
K
u
z
m
a
,
“
R
u
d
i
m
e
n
t
a
r
y
l
e
x
i
c
o
n
b
a
se
d
m
e
t
h
o
d
f
o
r
sarc
a
sm d
e
t
e
c
t
i
o
n
,
”
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
r
n
a
l
o
f
Ac
a
d
e
m
i
c
Re
s
e
a
rc
h
a
n
d
Re
f
l
e
c
t
i
o
n
,
v
o
l
.
5
,
n
o
.
4
,
2
0
1
7
.
[
1
2
]
M
.
Z
h
a
n
g
,
Y
.
Z
h
a
n
g
,
a
n
d
G
.
F
u
,
“
Tw
e
e
t
sar
c
a
sm
d
e
t
e
c
t
i
o
n
u
si
n
g
d
e
e
p
n
e
u
r
a
l
n
e
t
w
o
r
k
,
”
C
O
L
I
N
G
2
0
1
6
-
2
6
t
h
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
C
o
m
p
u
t
a
t
i
o
n
a
l
L
i
n
g
u
i
st
i
c
s
,
Pr
o
c
e
e
d
i
n
g
s
o
f
C
O
L
I
N
G
2
0
1
6
:
T
e
c
h
n
i
c
a
l
Pa
p
e
r
s
,
p
p
.
2
4
4
9
–
2
4
6
0
,
2
0
1
6
.
[
1
3
]
A
.
G
.
P
r
a
sad
,
S
.
S
a
n
j
a
n
a
,
S
.
M
.
B
h
a
t
,
a
n
d
B
.
S
.
H
a
r
i
s
h
,
“
S
e
n
t
i
m
e
n
t
a
n
a
l
y
s
i
s
f
o
r
sarca
sm
d
e
t
e
c
t
i
o
n
o
n
st
r
e
a
m
i
n
g
s
h
o
r
t
t
e
x
t
d
a
t
a
,
”
i
n
2
0
1
7
2
n
d
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
K
n
o
w
l
e
d
g
e
En
g
i
n
e
e
r
i
n
g
a
n
d
A
p
p
l
i
c
a
t
i
o
n
s,
I
C
K
EA
2
0
1
7
,
2
0
1
7
,
v
o
l
.
2
0
1
7
-
Jan
u
a
,
p
p
.
1
–
5
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
K
EA
.
2
0
1
7
.
8
1
6
9
8
9
2
.
[
1
4
]
T.
B
a
l
i
a
n
d
N
.
S
i
n
g
h
,
“
S
a
r
c
a
sm
d
e
t
e
c
t
i
o
n
:
B
u
i
l
d
i
n
g
a
c
o
n
t
e
x
t
u
a
l
h
i
e
r
a
r
c
h
y
,
”
i
n
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
Wo
r
k
sh
o
p
o
n
C
o
m
p
u
t
a
t
i
o
n
a
l
Mo
d
e
l
i
n
g
o
f
Pe
o
p
l
e
’
s O
p
i
n
i
o
n
s,
Pe
rs
o
n
a
l
i
t
y
,
a
n
d
Em
o
t
i
o
n
s
i
n
S
o
c
i
a
l
Me
d
i
a
(
PEO
PL
ES
)
,
2
0
1
6
,
p
p
.
1
1
9
–
127
.
[
1
5
]
A
.
B
e
n
M
e
r
i
e
m,
L.
H
l
a
o
u
a
,
a
n
d
L.
B
e
n
R
o
m
d
h
a
n
e
,
“
A
f
u
z
z
y
a
p
p
r
o
a
c
h
f
o
r
sarca
sm
d
e
t
e
c
t
i
o
n
i
n
so
c
i
a
l
n
e
t
w
o
r
k
s
,
”
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
9
2
,
p
p
.
6
0
2
–
6
1
1
,
2
0
2
1
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
p
r
o
c
s.
2
0
2
1
.
0
8
.
0
6
2
.
[
1
6
]
R
.
A
h
u
j
a
,
S
.
B
a
n
s
a
l
,
S
.
P
r
a
k
a
sh
,
K
.
V
e
n
k
a
t
a
r
a
m
a
n
,
a
n
d
A
.
B
a
n
g
a
,
“
C
o
m
p
a
r
a
t
i
v
e
s
t
u
d
y
o
f
d
i
f
f
e
r
e
n
t
sarc
a
sm
d
e
t
e
c
t
i
o
n
a
l
g
o
r
i
t
h
m
s
b
a
s
e
d
o
n
b
e
h
a
v
i
o
r
a
l
a
p
p
r
o
a
c
h
,
”
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
4
3
,
p
p
.
4
1
1
–
4
1
8
,
2
0
1
8
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
p
r
o
c
s.
2
0
1
8
.
1
0
.
4
1
2
.
[
1
7
]
V
.
G
o
v
i
n
d
a
n
a
n
d
V
.
B
a
l
a
k
r
i
s
h
n
a
n
,
“
A
mac
h
i
n
e
l
e
a
r
n
i
n
g
a
p
p
r
o
a
c
h
i
n
a
n
a
l
y
si
n
g
t
h
e
e
f
f
e
c
t
o
f
h
y
p
e
r
b
o
l
e
s
u
s
i
n
g
n
e
g
a
t
i
v
e
se
n
t
i
m
e
n
t
t
w
e
e
t
s
f
o
r
s
a
r
c
a
sm
d
e
t
e
c
t
i
o
n
,
”
J
o
u
rn
a
l
o
f
K
i
n
g
S
a
u
d
U
n
i
v
e
rsi
t
y
-
C
o
m
p
u
t
e
r
a
n
d
I
n
f
o
rm
a
t
i
o
n
S
c
i
e
n
c
e
s
,
v
o
l
.
3
4
,
n
o
.
8
,
p
p
.
5
1
1
0
–
5
1
2
0
,
2
0
2
2
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
j
k
su
c
i
.
2
0
2
2
.
0
1
.
0
0
8
.
[
1
8
]
S
.
K
.
B
h
a
r
t
i
,
R
.
N
a
i
d
u
,
a
n
d
K
.
S
.
B
a
b
u
,
“
H
y
p
e
r
b
o
l
i
c
f
e
a
t
u
r
e
-
b
a
se
d
sarc
a
sm
d
e
t
e
c
t
i
o
n
i
n
t
w
e
e
t
s:
a
m
a
c
h
i
n
e
l
e
a
r
n
i
n
g
a
p
p
r
o
a
c
h
,
”
i
n
2
0
1
7
1
4
t
h
I
EEE
I
n
d
i
a
C
o
u
n
c
i
l
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
,
I
N
D
I
C
O
N
2
0
1
7
,
2
0
1
8
,
p
p
.
1
–
6
,
d
o
i
:
1
0
.
1
1
0
9
/
I
N
D
I
C
O
N
.
2
0
1
7
.
8
4
8
7
7
1
2
.
[
1
9
]
D
.
M
a
y
n
a
r
d
a
n
d
M
.
A
.
G
r
e
e
n
w
o
o
d
,
“
W
h
o
c
a
r
e
s
a
b
o
u
t
sarc
a
st
i
c
t
w
e
e
t
s
?
I
n
v
e
st
i
g
a
t
i
n
g
t
h
e
i
m
p
a
c
t
o
f
sar
c
a
s
m
o
n
se
n
t
i
me
n
t
a
n
a
l
y
si
s
,
”
i
n
Pro
c
e
e
d
i
n
g
s
o
f
t
h
e
9
t
h
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
L
a
n
g
u
a
g
e
R
e
s
o
u
r
c
e
s
a
n
d
Ev
a
l
u
a
t
i
o
n
,
L
REC
2
0
1
4
,
2
0
1
4
,
p
p
.
4
2
3
8
–
4
2
4
3
.
[
2
0
]
B
.
F
e
l
b
o
,
A
.
M
i
s
l
o
v
e
,
A
.
S
ø
g
a
a
r
d
,
I
.
R
a
h
w
a
n
,
a
n
d
S
.
L
e
h
ma
n
n
,
“
U
si
n
g
mi
l
l
i
o
n
s
o
f
e
mo
j
i
o
c
c
u
r
r
e
n
c
e
s
t
o
l
e
a
r
n
a
n
y
-
d
o
ma
i
n
r
e
p
r
e
s
e
n
t
a
t
i
o
n
s
f
o
r
d
e
t
e
c
t
i
n
g
s
e
n
t
i
me
n
t
,
e
mo
t
i
o
n
a
n
d
sar
c
a
sm
,
”
a
rX
i
v
p
r
e
p
r
i
n
t
a
rX
i
v
:
1
7
0
8
.
0
0
5
2
4
,
2
0
1
7
.
[
2
1
]
O
.
Ts
u
r
,
D
.
D
mi
t
r
y
,
a
n
d
R
.
A
r
i
,
“
S
e
mi
-
su
p
e
r
v
i
s
e
d
r
e
c
o
g
n
i
t
i
o
n
o
f
sarc
a
st
i
c
se
n
t
e
n
c
e
s
i
n
t
w
i
t
t
e
r
a
n
d
A
m
a
z
o
n
,
”
i
n
C
o
N
L
L
2
0
1
0
-
Fo
u
rt
e
e
n
t
h
C
o
n
f
e
re
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
,
Pro
c
e
e
d
i
n
g
s
o
f
t
h
e
C
o
n
f
e
r
e
n
c
e
,
2
0
1
0
,
p
p
.
1
0
7
–
1
1
6
.
[
2
2
]
D
.
H
a
z
a
r
i
k
a
,
S
.
P
o
r
i
a
,
S
.
G
o
r
a
n
t
l
a
,
E.
C
a
m
b
r
i
a
,
R
.
Z
i
mm
e
r
m
a
n
n
,
a
n
d
R
.
M
i
h
a
l
c
e
a
,
“
C
A
S
C
A
D
E
:
c
o
n
t
e
x
t
u
a
l
sarc
a
sm
d
e
t
e
c
t
i
o
n
i
n
o
n
l
i
n
e
d
i
s
c
u
ss
i
o
n
f
o
r
u
ms,
”
i
n
C
O
L
I
N
G
2
0
1
8
-
2
7
t
h
I
n
t
e
rn
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
C
o
m
p
u
t
a
t
i
o
n
a
l
L
i
n
g
u
i
st
i
c
s,
Pr
o
c
e
e
d
i
n
g
s
,
2
0
1
8
,
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:
2088
-
8
7
0
8
Dee
p
lea
r
n
in
g
-
b
a
s
ed
a
tten
tio
n
mo
d
els fo
r
s
a
r
ca
s
m
d
etec
tio
n
in
text
(
Ga
n
esh
C
h
a
n
d
r
a
s
ek
a
r
a
n
)
6795
p
p
.
1
8
3
7
–
1
8
4
8
.
[
2
3
]
Y
.
R
e
n
,
D
.
J
i
,
a
n
d
H
.
R
e
n
,
“
C
o
n
t
e
x
t
-
a
u
g
me
n
t
e
d
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
t
w
i
t
t
e
r
sarc
a
sm
d
e
t
e
c
t
i
o
n
,
”
N
e
u
ro
c
o
m
p
u
t
i
n
g
,
v
o
l
.
3
0
8
,
p
p
.
1
–
7
,
2
0
1
8
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
n
e
u
c
o
m.
2
0
1
8
.
0
3
.
0
4
7
.
[
2
4
]
R
.
G
o
n
z
á
l
e
z
-
I
b
á
ñ
e
z
,
S
.
M
u
r
e
s
a
n
,
a
n
d
N
.
W
a
c
h
o
l
d
e
r
,
“
I
d
e
n
t
i
f
y
i
n
g
sarc
a
s
m
i
n
Tw
i
t
t
e
r
:
A
c
l
o
ser
l
o
o
k
,
”
i
n
A
C
L
-
H
L
T
2
0
1
1
-
Pro
c
e
e
d
i
n
g
s o
f
t
h
e
4
9
t
h
A
n
n
u
a
l
M
e
e
t
i
n
g
o
f
t
h
e
Ass
o
c
i
a
t
i
o
n
f
o
r C
o
m
p
u
t
a
t
i
o
n
a
l
L
i
n
g
u
i
st
i
c
s:
H
u
m
a
n
L
a
n
g
u
a
g
e
T
e
c
h
n
o
l
o
g
i
e
s
,
2
0
1
1
,
v
o
l
.
2
,
p
p
.
5
8
1
–
5
8
6
.
[
2
5
]
E.
L
u
n
a
n
d
o
a
n
d
A
.
P
u
r
w
a
r
i
a
n
t
i
,
“
I
n
d
o
n
e
s
i
a
n
so
c
i
a
l
me
d
i
a
s
e
n
t
i
me
n
t
a
n
a
l
y
s
i
s
w
i
t
h
s
a
r
c
a
sm
d
e
t
e
c
t
i
o
n
,
”
i
n
2
0
1
3
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
A
d
v
a
n
c
e
d
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
a
n
d
I
n
f
o
rm
a
t
i
o
n
S
y
s
t
e
m
s
(
I
C
AC
S
I
S
)
,
S
e
p
.
2
0
1
3
,
p
p
.
1
9
5
–
1
9
8
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
A
C
S
I
S
.
2
0
1
3
.
6
7
6
1
5
7
5
.
[
2
6
]
F
.
B
a
r
b
i
e
r
i
,
H
.
S
a
g
g
i
o
n
,
a
n
d
F
.
R
o
n
z
a
n
o
,
“
M
o
d
e
l
l
i
n
g
sa
r
c
a
s
m
i
n
T
w
i
t
t
e
r
,
a
n
o
v
e
l
a
p
p
r
o
a
c
h
,
”
i
n
P
ro
c
e
e
d
i
n
g
s
o
f
t
h
e
5
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
ro
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
A
n
a
l
y
s
i
s
,
2
0
1
4
,
p
p
.
5
0
–
5
8
,
d
o
i
:
1
0
.
3
1
1
5
/
v
1
/
W
1
4
-
2
6
0
9
.
[
2
7
]
P
.
T
u
n
g
t
h
a
mt
h
i
t
i
,
K
.
S
h
i
r
a
i
,
a
n
d
M
.
M
o
h
d
,
“
R
e
c
o
g
n
i
t
i
o
n
o
f
sarc
a
sm
i
n
m
i
c
r
o
b
l
o
g
g
i
n
g
b
a
s
e
d
o
n
se
n
t
i
me
n
t
a
n
a
l
y
si
s
a
n
d
c
o
h
e
r
e
n
c
e
i
d
e
n
t
i
f
i
c
a
t
i
o
n
,
”
J
o
u
r
n
a
l
o
f
N
a
t
u
r
a
l
L
a
n
g
u
a
g
e
Pr
o
c
e
ssi
n
g
,
v
o
l
.
2
3
,
n
o
.
5
,
p
p
.
3
8
3
–
4
0
5
,
2
0
1
6
,
d
o
i
:
1
0
.
5
7
1
5
/
j
n
l
p
.
2
3
.
3
8
3
.
[
2
8
]
A
.
R
a
j
a
d
e
s
i
n
g
a
n
,
R
.
Z
a
f
a
r
a
n
i
,
a
n
d
H
.
Li
u
,
“
S
a
r
c
a
sm
d
e
t
e
c
t
i
o
n
o
n
Tw
i
t
t
e
r
,
”
i
n
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
E
i
g
h
t
h
A
C
M
I
n
t
e
rn
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
W
e
b
S
e
a
r
c
h
a
n
d
D
a
t
a
Mi
n
i
n
g
,
F
e
b
.
2
0
1
5
,
p
p
.
9
7
–
1
0
6
,
d
o
i
:
1
0
.
1
1
4
5
/
2
6
8
4
8
2
2
.
2
6
8
5
3
1
6
.
[
2
9
]
S
.
L
u
k
i
n
a
n
d
M
.
W
a
l
k
e
r
,
“
R
e
a
l
l
y
?
w
e
l
l
.
A
p
p
a
r
e
n
t
l
y
b
o
o
t
st
r
a
p
p
i
n
g
i
mp
r
o
v
e
s
t
h
e
p
e
r
f
o
r
ma
n
c
e
o
f
sar
c
a
sm
a
n
d
n
a
st
i
n
e
ss
c
l
a
ssi
f
i
e
r
s
f
o
r
o
n
l
i
n
e
d
i
a
l
o
g
u
e
,
”
a
r
Xi
v
p
re
p
ri
n
t
a
rXi
v
:
1
7
0
8
.
0
8
5
7
2
,
2
0
1
7
.
[
3
0
]
R
.
G
u
p
t
a
,
J
.
K
u
m
a
r
,
H
.
A
g
r
a
w
a
l
,
a
n
d
K
u
n
a
l
,
“
A
st
a
t
i
st
i
c
a
l
a
p
p
r
o
a
c
h
f
o
r
sarca
sm
d
e
t
e
c
t
i
o
n
u
s
i
n
g
Tw
i
t
t
e
r
d
a
t
a
,
”
i
n
2
0
2
0
4
t
h
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
I
n
t
e
l
l
i
g
e
n
t
C
o
m
p
u
t
i
n
g
a
n
d
C
o
n
t
r
o
l
S
y
st
e
m
s
(
I
C
I
C
C
S
)
,
M
a
y
2
0
2
0
,
p
p
.
6
3
3
–
6
3
8
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
I
C
C
S
4
8
2
6
5
.
2
0
2
0
.
9
1
2
0
9
1
7
.
[
3
1
]
A
.
K
u
m
a
r
Je
n
a
,
A
.
S
i
n
h
a
,
a
n
d
R
.
A
g
a
r
w
a
l
,
“
C
-
N
e
t
:
c
o
n
t
e
x
t
u
a
l
n
e
t
w
o
r
k
f
o
r
sarc
a
sm
d
e
t
e
c
t
i
o
n
,
”
i
n
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
S
e
c
o
n
d
Wo
r
k
sh
o
p
o
n
Fi
g
u
ra
t
i
v
e
L
a
n
g
u
a
g
e
Pr
o
c
e
ssi
n
g
,
2
0
2
0
,
p
p
.
6
1
–
6
6
,
d
o
i
:
1
0
.
1
8
6
5
3
/
v
1
/
2
0
2
0
.
f
i
g
l
a
n
g
-
1
.
8
.
B
I
O
G
RAP
H
I
E
S O
F
AUTH
O
RS
G
a
n
e
sh
Cha
n
d
r
a
se
k
a
r
a
n
h
a
il
s
fro
m
S
a
lem
d
istri
c
t
in
th
e
sta
t
e
o
f
Tam
il
Na
d
u
.
He
re
c
e
iv
e
d
h
is
B.
E
d
e
g
re
e
in
e
le
c
tro
n
ics
a
n
d
i
n
stru
m
e
n
tati
o
n
e
n
g
i
n
e
e
rin
g
fr
o
m
M
u
t
h
a
y
a
m
m
a
l
En
g
i
n
e
e
rin
g
C
o
ll
e
g
e
a
ffil
iate
d
to
An
n
a
Un
i
v
e
rsity
,
C
h
e
n
n
a
i
i
n
th
e
y
e
a
r
2
0
0
8
,
h
is
m
a
ste
rs
in
a
p
p
li
e
d
e
lec
tro
n
ics
fr
o
m
P
a
a
v
a
i
E
n
g
i
n
e
e
rin
g
Co
ll
e
g
e
a
ffil
iate
d
to
A
n
n
a
Un
iv
e
rsity
,
C
h
e
n
n
a
i
in
th
e
y
e
a
r
2
0
1
3
.
He
h
a
s
1
0
y
e
a
rs
o
f
tea
c
h
in
g
e
x
p
e
rien
c
e
a
n
d
2
y
e
a
rs
o
f
i
n
d
u
strial
e
x
p
e
rien
c
e
.
His
re
se
a
rc
h
in
tere
sts
in
c
lu
d
e
m
a
c
h
i
n
e
lea
rn
i
n
g
,
d
e
e
p
lea
rn
i
n
g
,
ima
g
e
p
ro
c
e
ss
in
g
a
n
d
se
n
t
ime
n
t
a
n
a
ly
sis.
He
h
a
s
a
u
th
o
re
d
se
v
e
ra
l
re
se
a
rc
h
p
a
p
e
rs
a
n
d
p
re
se
n
ted
p
a
p
e
rs
in
re
p
u
te
d
in
tern
a
ti
o
n
a
l
c
o
n
fe
re
n
c
e
s.
His
re
se
a
r
c
h
a
rti
c
les
a
re
c
it
e
d
b
y
m
a
n
y
in
tern
a
ti
o
n
a
l
jo
u
r
n
a
ls
o
v
e
r
th
e
y
e
a
rs
.
He
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
c
sg
a
n
e
sh
8
6
@
g
m
a
il
.
c
o
m
.
Ma
n
d
a
la
p
u
K
a
lp
a
n
a
Cho
wd
a
r
y
wo
rk
i
n
g
a
s
a
ss
istan
t
p
ro
fe
ss
o
r,
Co
m
p
u
ter
S
c
ien
c
e
a
n
d
En
g
in
e
e
rin
g
De
p
a
rt
m
e
n
t
a
t
M
LR
In
stit
u
te
o
f
Tec
h
n
o
lo
g
y
,
Hy
d
e
ra
b
a
d
,
Tela
n
g
a
n
a
,
In
d
ia.
S
h
e
re
c
e
iv
e
d
th
e
B
.
Tec
h
.
d
e
g
re
e
in
e
lec
tro
n
ics
a
n
d
c
o
m
m
u
n
ica
ti
o
n
e
n
g
i
n
e
e
rin
g
fro
m
Ch
irala
En
g
in
e
e
rin
g
C
o
ll
e
g
e
,
C
h
i
ra
la
a
ffil
iate
d
to
JN
TU,
Ka
k
in
a
d
a
in
2
0
1
1
.
In
2
0
1
3
,
re
c
e
iv
e
d
th
e
M
Tec
h
d
e
g
re
e
in
e
lec
tro
n
ics
a
n
d
c
o
m
m
u
n
ica
ti
o
n
e
n
g
i
n
e
e
rin
g
with
VL
S
I
S
p
e
c
ializa
ti
o
n
fro
m
K
L
Un
i
v
e
rsity
,
Vija
y
a
wa
d
a
.
In
2
0
2
2
,
re
c
e
iv
e
d
th
e
P
h
.
D.
d
e
g
re
e
in
e
lec
tro
n
ics
a
n
d
c
o
m
m
u
n
ica
ti
o
n
e
n
g
in
e
e
ri
n
g
fro
m
Ka
ru
n
y
a
Un
i
v
e
rsity
,
C
o
imb
a
to
re
,
I
n
d
ia.
S
h
e
h
a
s
7
-
y
e
a
r
e
x
p
e
rien
c
e
in
tea
c
h
i
n
g
a
n
d
3
.
5
y
e
a
rs
in
re
se
a
rc
h
,
sh
e
p
u
b
l
ish
e
d
a
ro
u
n
d
2
0
p
u
b
li
c
a
ti
o
n
s
i
n
in
tern
a
ti
o
n
a
l/
n
a
ti
o
n
a
l
jo
u
rn
a
ls
a
n
d
c
o
n
fe
re
n
c
e
s
in
c
l
u
d
e
s
5
S
CI
p
a
p
e
rs.
2
P
a
ten
ts
in
m
a
c
h
in
e
lea
rn
in
g
a
n
d
d
e
e
p
lea
rn
in
g
.
S
h
e
i
s
a
li
fe
m
e
m
b
e
r
o
f
p
r
o
fe
ss
io
n
a
l
b
o
d
ies
l
ik
e
I
n
d
ian
S
o
c
iet
y
f
o
r
Tec
h
n
ica
l
Ed
u
c
a
ti
o
n
(IS
TE
),
I
AENG
.
Re
se
a
r
c
h
in
tere
st
in
c
lu
d
e
s
m
a
c
h
in
e
lea
rn
in
g
,
d
e
e
p
lea
rn
in
g
,
a
rti
ficia
l
in
telli
g
e
n
c
e
,
a
n
d
ima
g
e
p
ro
c
e
ss
in
g
.
Sh
e
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
d
r.
k
a
l
p
a
n
a
@m
lri
n
stit
u
ti
o
n
s.a
c
.
in
.
J
y
o
th
i
Chi
n
n
a
B
a
b
u
h
a
s
re
c
e
iv
e
d
h
is
B.
Tec
h
.
d
e
g
re
e
fro
m
JN
TU,
Hy
d
e
ra
b
a
d
.
M
.
Tec
h
.
d
e
g
re
e
in
VLS
I
sy
ste
m
d
e
sig
n
a
n
d
P
h
.
D.
in
VL
S
I
sig
n
a
l
p
ro
c
e
ss
in
g
d
e
g
re
e
s
fro
m
JN
TUA,
An
a
n
th
a
p
u
ra
m
u
.
C
u
rre
n
tl
y
,
h
e
is
wo
r
k
in
g
a
s
a
ss
o
c
iate
p
r
o
fe
ss
o
r
in
th
e
De
p
a
rtme
n
t
o
f
El
e
c
tro
n
ics
a
n
d
Co
m
m
u
n
ica
ti
o
n
En
g
in
e
e
rin
g
a
t
A
n
n
a
m
a
c
h
a
ry
a
In
stit
u
te
o
f
Tec
h
n
o
lo
g
y
a
n
d
S
c
ien
c
e
s
-
Au
to
n
o
m
o
u
s,
Ra
jam
p
e
t,
Ka
d
a
p
a
,
A.P
,
a
n
d
In
d
ia.
He
h
a
s
p
u
b
li
sh
e
d
n
u
m
b
e
r
o
f
re
se
a
rc
h
p
a
p
e
rs
in
v
a
rio
u
s
i
n
te
rn
a
ti
o
n
a
l
jo
u
rn
a
ls
a
n
d
v
a
rio
u
s
n
a
ti
o
n
a
l
a
n
d
i
n
tern
a
ti
o
n
a
l
c
o
n
fe
re
n
c
e
s.
He
is
a
re
v
iew
m
e
m
b
e
r
o
f
v
a
rio
u
s
i
n
tern
a
ti
o
n
a
l
jo
u
rn
a
ls.
He
wa
s
j
o
in
e
d
a
t
AITS
,
Ra
jam
p
e
t
a
s
a
n
a
ss
istan
t
p
r
o
fe
ss
o
r
in
th
e
De
p
a
rtme
n
t
o
f
El
e
c
t
ro
n
ics
a
n
d
C
o
m
m
u
n
ica
ti
o
n
En
g
i
n
e
e
rin
g
i
n
t
h
e
y
e
a
r
2
0
0
8
.
H
e
is
h
a
v
i
n
g
1
5
y
e
a
rs
o
f
Ex
p
e
rien
c
e
in
Tea
c
h
in
g
UG
a
n
d
P
G
c
o
u
rse
s.
He
is
a
li
fe
m
e
m
b
e
r
in
IS
TE
a
n
d
m
e
m
b
e
r
in
IACSIT
,
M
I
RED
a
n
d
CS
TA.
His
c
u
rre
n
t
re
se
a
rc
h
in
tere
sts
a
r
e
VLS
I
si
g
n
a
l
p
r
o
c
e
ss
in
g
,
m
icro
p
ro
c
e
ss
o
r,
e
m
b
e
d
d
e
d
sy
ste
m
s
a
n
d
c
o
m
m
u
n
ica
ti
o
n
,
Io
T,
m
a
c
h
in
e
le
a
rn
in
g
a
n
d
sig
n
a
l
p
r
o
c
e
ss
in
g
.
He
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
jch
in
n
a
b
a
b
u
@
g
m
a
il
.
c
o
m
.
Evaluation Warning : The document was created with Spire.PDF for Python.