T
E
L
K
O
M
N
I
K
A
T
elec
o
m
m
un
ica
t
io
n,
Co
m
pu
t
ing
,
E
lect
ro
nics
a
nd
Co
ntr
o
l
Vo
l.
19
,
No
.
3
,
J
u
n
e
2
0
2
1
,
p
p
.
7
5
4
~7
6
1
I
SS
N:
1
6
9
3
-
6
9
3
0
,
ac
cr
ed
ited
First Gr
ad
e
b
y
Kem
en
r
is
tek
d
i
k
ti,
Dec
r
ee
No
: 2
1
/E/KPT
/2
0
1
8
DOI
: 1
0
.
1
2
9
2
8
/TE
L
KOM
NI
K
A.
v
1
9
i
3
.
1
8
3
5
6
754
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//jo
u
r
n
a
l.u
a
d
.
a
c.
id
/in
d
ex
.
p
h
p
/TELK
OM
N
I
K
A
Summ
a
riza
tion o
f
C
O
VID
-
19
news
docum
ents
deep learning
-
ba
sed usin
g
t
ra
nsfo
r
mer archit
ectu
re
Nur
H
a
y
a
t
in,
K
ha
rism
a
M
uza
k
i G
hu
f
ro
n,
G
a
lih
Wa
s
is
Wica
k
s
o
no
De
p
a
rtme
n
t
o
f
In
fo
rm
a
ti
c
s,
F
a
c
u
l
ty
o
f
En
g
i
n
e
e
rin
g
,
Un
i
v
e
rsity
o
f
M
u
h
a
m
m
a
d
iy
a
h
M
a
lan
g
,
I
n
d
o
n
e
s
ia
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
J
u
l 2
5
,
2
0
2
0
R
ev
is
ed
Oct
1
2
,
2
0
2
0
Acc
ep
ted
Oct
2
3
,
2
0
2
0
F
a
c
in
g
t
h
e
n
e
ws
o
n
th
e
i
n
tern
e
t
a
b
o
u
t
th
e
s
p
re
a
d
in
g
o
f
C
o
ro
n
a
v
ir
u
s
d
ise
a
se
2
0
1
9
(COV
ID
-
1
9
)
is
c
h
a
ll
e
n
g
i
n
g
b
e
c
a
u
se
it
is
re
q
u
ire
d
a
lo
n
g
t
ime
to
g
e
t
v
a
lu
a
b
le
in
f
o
rm
a
ti
o
n
fr
o
m
th
e
n
e
ws
.
De
e
p
lea
rn
i
n
g
h
a
s
a
sig
n
ifi
c
a
n
t
imp
a
c
t
o
n
NLP
re
se
a
rc
h
.
Ho
we
v
e
r,
t
h
e
d
e
e
p
lea
rn
i
n
g
m
o
d
e
ls
u
se
d
in
se
v
e
r
a
l
stu
d
ies
,
e
sp
e
c
ially
i
n
d
o
c
u
m
e
n
t
su
m
m
a
ry
,
stil
l
h
a
v
e
a
d
e
fici
e
n
c
y
.
F
o
r
e
x
a
m
p
le,
t
h
e
m
a
x
imu
m
o
u
t
p
u
t
o
f
lo
n
g
te
x
t
p
ro
v
id
e
s
i
n
c
o
rre
c
tl
y
.
Th
e
o
th
e
r
re
su
lt
s
a
re
re
d
u
n
d
a
n
t,
o
r
th
e
c
h
a
ra
c
ters
re
p
e
a
t
e
d
ly
a
p
p
e
a
re
d
so
t
h
a
t
th
e
re
su
lt
i
n
g
se
n
ten
c
e
s
we
re
les
s
o
rg
a
n
ize
d
,
a
n
d
th
e
re
c
a
ll
v
a
lu
e
o
b
tain
e
d
wa
s
lo
w.
Th
is
stu
d
y
a
ims
t
o
s
u
m
m
a
rize
u
sin
g
a
d
e
e
p
lea
rn
in
g
m
o
d
e
l
imp
lem
e
n
ted
to
COV
I
D
-
1
9
n
e
ws
d
o
c
u
m
e
n
ts.
We
p
ro
p
o
se
d
tran
sfo
rm
e
r
a
s
b
a
se
lan
g
u
a
g
e
m
o
d
e
ls
with
a
rc
h
it
e
c
tu
ra
l
m
o
d
ifi
c
a
ti
o
n
a
s
th
e
b
a
sis
f
o
r
d
e
sig
n
in
g
th
e
m
o
d
e
l
t
o
imp
ro
v
e
re
su
lt
s
sig
n
i
fica
n
tl
y
in
d
o
c
u
m
e
n
t
su
m
m
a
riza
ti
o
n
.
We
m
a
k
e
a
tran
sfo
rm
e
r
-
b
a
se
d
a
rc
h
it
e
c
tu
re
m
o
d
e
l
wit
h
e
n
c
o
d
e
r
a
n
d
d
e
c
o
d
e
r
t
h
a
t
c
a
n
b
e
d
o
n
e
se
v
e
ra
l
ti
m
e
s
re
p
e
a
ted
ly
a
n
d
m
a
k
e
a
c
o
m
p
a
riso
n
o
f
la
y
e
r
m
o
d
ifi
c
a
ti
o
n
s
b
a
se
d
o
n
sc
o
rin
g
.
F
r
o
m
th
e
re
su
lt
in
g
e
x
p
e
rime
n
t
u
se
d
,
RO
UG
E
-
1
a
n
d
ROU
G
E
-
2
sh
o
w
t
h
e
g
o
o
d
p
e
rfo
r
m
a
n
c
e
fo
r
th
e
p
ro
p
o
se
d
m
o
d
e
l
with
sc
o
re
s
0
.
5
8
a
n
d
0
.
4
2
,
re
sp
e
c
ti
v
e
l
y
,
with
a
train
in
g
ti
m
e
o
f
1
1
4
3
8
se
c
o
n
d
s.
Th
e
m
o
d
e
l
p
ro
p
o
se
d
wa
s
e
v
id
e
n
tl
y
e
ffe
c
ti
v
e
in
imp
r
o
v
i
n
g
re
su
lt
p
e
rfo
r
m
a
n
c
e
in
a
b
stra
c
ti
v
e
d
o
c
u
m
e
n
t
su
m
m
a
riza
ti
o
n
.
K
ey
w
o
r
d
s
:
C
OVI
D
-
19
Dee
p
lear
n
in
g
New
s
s
u
m
m
ar
izatio
n
T
r
an
s
f
o
r
m
e
r
ar
ch
itectu
r
e
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
:
Kh
ar
is
m
a
Mu
za
k
i G
h
u
f
r
o
n
Dep
ar
tm
en
t o
f
I
n
f
o
r
m
atics
Un
iv
er
s
ity
o
f
Mu
h
am
m
ad
i
y
ah
Ma
lan
g
C
am
p
u
s
I
I
I
UM
M
,
2
4
6
R
ay
a
T
lo
g
o
m
as
St.
,
Ma
lan
g
,
J
awa
T
im
u
r
6
5
1
4
4
,
I
n
d
o
n
esia
E
m
ail:
k
h
ar
is
m
a.
m
u
za
k
i@
we
b
m
ail.
u
m
m
.
ac
.
i
d
1.
I
NT
RO
D
UCT
I
O
N
I
n
ea
r
ly
2
0
2
0
th
e
wo
r
ld
was
h
it
b
y
C
o
r
o
n
a
v
ir
u
s
d
is
ea
s
e
2
0
1
9
(
C
OVI
D
-
1
9
)
p
an
d
e
m
ic,
wh
i
ch
af
f
ec
ted
g
lo
b
al
life
.
All
p
e
o
p
le
with
v
ar
io
u
s
b
ac
k
g
r
o
u
n
d
s
an
d
f
ield
s
o
f
s
cien
ce
d
is
cu
s
s
C
OVI
D
-
1
9
p
a
n
d
em
ic,
b
o
t
h
th
r
o
u
g
h
s
o
cial
m
ed
ia
an
d
web
n
ews.
L
ik
ewise,
with
d
ata
s
cien
tis
ts
,
s
ev
er
al
s
tu
d
ies p
r
im
ar
ily
r
elate
d
to
n
atu
r
al
lan
g
u
ag
e
p
r
o
ce
s
s
in
g
(
NL
P)
o
n
C
OVI
D
-
1
9
also
b
eg
an
to
b
e
ca
r
r
ied
o
u
t
.
S
u
c
h
as
p
r
ed
ict
iv
e
m
o
d
els
th
at
ca
n
esti
m
ate
r
etu
r
n
s
o
n
s
to
ck
s
f
r
o
m
co
u
n
tr
ies
m
o
s
t
af
f
ec
ted
b
y
t
h
e
C
OVI
D
-
1
9
p
an
d
em
ic
[
1
]
,
also
m
o
d
eled
ca
u
s
ality
u
s
in
g
n
e
u
r
al
n
etwo
r
k
s
t
o
e
x
p
l
o
r
e
m
is
in
f
o
r
m
atio
n
o
n
s
o
cial
m
ed
ia
d
u
r
in
g
th
e
C
OVI
D
-
1
9
p
an
d
em
ic
[
2
]
.
As
f
a
r
as
o
u
r
o
b
s
er
v
atio
n
,
n
o
n
e
o
f
th
ese
s
tu
d
ies
r
elate
d
to
th
e
C
OV
I
D
-
1
9
n
ews
d
o
cu
m
en
t
h
as
b
ee
n
d
o
n
e
p
r
ev
i
o
u
s
ly
.
Me
an
wh
ile,
o
n
e
o
f
th
e
n
ews
m
ed
ia
tr
en
d
s
r
ec
en
tly
is
th
e
p
u
b
licatio
n
o
f
n
ews
d
o
c
u
m
en
ts
ab
o
u
t
C
OVI
D
-
19
,
w
h
ich
was r
elea
s
ed
q
u
ick
ly
an
d
u
p
d
ated
ev
e
r
y
d
ay
,
s
o
th
at
th
e
r
elea
s
e
r
esu
lted
in
a
lo
t o
f
d
a
ta
an
d
n
ews a
b
o
u
t
th
e
C
OVI
D
-
1
9
.
T
h
is
was
p
r
o
o
f
ed
b
y
t
h
e
s
ea
r
ch
r
esu
lts
f
r
o
m
th
e
Go
o
g
le
s
ea
r
ch
e
n
g
in
e
th
at
g
av
e
a
lo
t
o
f
o
u
t
p
u
t
ab
o
v
e
6
b
i
llio
n
d
o
cu
m
e
n
ts
f
o
r
C
OVI
D
-
1
9
k
ey
wo
r
d
s
th
at
ar
e
ac
ce
s
s
ed
i
n
J
u
ly
2
0
2
0
.
T
h
is
is
a
ch
allen
g
e
in
h
o
w
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KOM
NI
KA
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l Co
n
tr
o
l
S
u
mma
r
iz
a
tio
n
o
f COVI
D
-
1
9
n
ews
d
o
cu
men
ts
d
ee
p
lea
r
n
in
g
-
b
a
s
ed
u
s
in
g
… (
N
u
r
Ha
ya
tin
)
755
to
b
e
ab
le
to
f
in
d
th
e
r
elev
an
t
in
f
o
r
m
atio
n
f
r
o
m
th
e
d
o
cu
m
en
t
co
llectio
n
,
esp
ec
ially
f
r
o
m
th
e
n
ews
d
o
cu
m
en
ts
.
New
s
d
o
cu
m
en
ts
ar
e
o
n
e
o
f
t
h
e
m
an
y
u
n
s
tr
u
ctu
r
ed
d
ata
th
at
ar
e
f
o
u
n
d
a
n
d
ea
s
ily
ac
ce
s
s
ed
o
n
th
e
i
n
ter
n
et.
Mu
ltip
le
ty
p
es
o
f
d
ata
th
at
ar
e
s
ca
tter
ed
o
n
th
e
in
ter
n
et,
in
clu
d
i
n
g
n
ews
d
o
cu
m
e
n
ts
,
in
cr
ea
s
e
r
ap
id
g
r
o
wth
[
3
,
4
]
an
d
in
cr
ea
s
e
ex
p
o
n
en
tially
[
4
]
.
W
ith
th
e
r
ap
id
g
r
o
wth
o
f
d
ata
,
th
e
s
u
m
m
ar
iza
tio
n
o
f
i
n
f
o
r
m
atio
n
b
ec
o
m
es
im
p
o
r
tan
t
to
m
ee
t
t
h
e
n
ee
d
s
o
f
in
ter
n
et
u
s
er
s
.
I
n
s
u
m
m
ar
izin
g
d
o
c
u
m
en
ts
,
two
t
y
p
es
ca
n
b
e
u
s
ed
,
n
am
ely
ab
s
tr
ac
tiv
e
an
d
ex
tr
ac
tiv
e.
An
ex
tr
ac
tiv
e
s
u
m
m
ar
y
is
a
co
n
cise
tex
t
m
eth
o
d
wh
ic
h
co
n
s
is
ts
o
f
th
r
ee
s
tag
es:
tex
t
r
ep
r
esen
tatio
n
,
s
en
ten
ce
ev
alu
atio
n
,
an
d
s
en
ten
ce
s
elec
tio
n
u
s
in
g
a
s
tatis
tical
m
o
d
el
[
5
]
.
Ab
s
tr
ac
tiv
e
s
u
m
m
ar
izatio
n
wo
r
k
s
b
y
p
r
o
d
u
cin
g
n
ew
s
en
ten
ce
s
in
s
u
m
m
ar
y
b
ased
o
n
an
ex
is
tin
g
te
x
t
b
y
r
ep
ea
tin
g
n
ew
wo
r
d
s
as a
n
ex
tr
ac
tio
n
ac
t
[
6
]
.
Sev
er
al
s
tu
d
ies
o
n
ab
s
tr
ac
tiv
e
n
ews
s
u
m
m
ar
ies
u
s
in
g
d
ee
p
lear
n
in
g
h
av
e
b
ee
n
d
o
n
e
p
r
ev
io
u
s
ly
.
Ab
s
tr
ac
tiv
e
s
u
m
m
ar
izatio
n
s
wer
e
co
n
d
u
cted
o
n
th
e
C
h
in
ese
n
ews
d
ataset
u
s
in
g
p
u
b
l
ic
o
p
in
io
n
[
7
]
.
Me
an
wh
ile,
o
th
er
r
esear
ch
f
o
cu
s
ed
o
n
th
e
k
ey
wo
r
d
s
th
at
ex
is
t
in
t
h
e
te
x
t
s
en
ten
ce
s
th
at
ca
n
wo
r
k
ef
f
ec
tiv
ely
to
p
r
o
d
u
ce
in
ter
p
r
eted
tex
ts
[
8
]
.
T
h
e
o
t
h
er
m
o
d
el
th
at
was
u
s
ed
f
o
r
ab
s
tr
ac
tiv
e
n
ews
s
u
m
m
ar
y
is
s
eq
u
en
ce
m
o
d
elin
g
,
s
u
ch
as
lo
n
g
s
h
o
r
t
-
ter
m
m
em
o
r
y
(
L
STM
)
an
d
r
ec
u
r
r
e
n
t
n
eu
r
al
n
etwo
r
k
(
R
NN)
.
T
h
e
s
eq
u
en
ce
to
s
eq
u
en
ce
th
e
R
NNs
m
o
d
el
h
as
s
u
cc
ess
f
u
lly
r
ed
u
c
ed
th
e
tr
ain
in
g
lo
s
s
f
o
r
ab
s
tr
ac
tiv
e
s
u
m
m
ar
y
u
s
ed
am
az
o
n
f
in
e
f
o
o
d
r
ev
iews
d
ataset
[
9
]
.
Un
f
o
r
tu
n
ately
,
t
h
e
m
ax
im
u
m
o
u
t
p
u
t
o
f
lo
n
g
te
x
t
p
r
o
v
id
es
in
co
r
r
ec
tly
.
T
h
e
r
esear
ch
p
r
o
v
id
e
s
a
co
r
r
ec
t
s
u
m
m
a
r
y
o
n
ly
f
o
r
s
h
o
r
t
tex
t.
An
o
th
er
s
tu
d
y
also
c
o
n
d
u
cted
ex
p
e
r
im
en
ts
b
y
d
o
i
n
g
a
c
o
m
b
in
atio
n
o
f
lo
ca
l
atten
tio
n
an
d
L
STM
in
wh
ich
th
e
r
esu
lts
o
f
t
h
e
s
u
m
m
ar
izatio
n
o
f
t
h
e
ch
ar
ac
te
r
s
r
ep
ea
ted
ly
ap
p
ea
r
ed
s
o
th
at
th
e
r
esu
ltin
g
s
en
ten
ce
s
w
er
e
less
o
r
g
an
ized
an
d
t
h
e
r
e
ca
ll
v
alu
e
o
b
tain
e
d
was
lo
w
[
1
0
]
.
Ho
wev
er
,
th
e
r
ep
etitiv
e
wo
r
k
in
g
s
f
o
u
n
d
in
t
h
e
r
ec
u
r
r
en
t
m
o
d
el
lik
e
R
NN
an
d
L
STM
,
p
r
ev
e
n
t
th
e
m
o
d
el
f
r
o
m
co
n
d
u
cti
n
g
p
ar
allel
tr
ain
in
g
a
n
d
lim
it th
e
ab
ilit
y
to
k
n
o
w
co
n
tex
t w
ith
l
o
n
g
er
i
n
p
u
t seq
u
en
ce
s
[
1
1
]
.
T
r
an
s
f
o
r
m
e
r
,
as
b
ase
lan
g
u
ag
e
m
o
d
els,
h
as
s
ig
n
if
ica
n
tly
im
p
ac
ted
th
e
NL
P
r
esear
c
h
f
ield
to
r
ep
lace
th
e
d
ef
icien
c
y
o
f
b
o
th
L
STM
,
C
NN
an
d
R
NN
b
ased
as
a
d
ee
p
lear
n
i
n
g
a
r
ch
itectu
r
e
[
1
2
,
1
3
]
,
s
o
th
at
m
a
n
y
r
ea
s
o
n
s
wh
y
th
e
tr
an
s
f
o
r
m
er
was
ch
o
s
en
as
b
ase
m
o
d
el
a
r
ch
itectu
r
e.
Var
i
o
u
s
s
tu
d
ies
a
p
p
lied
to
tr
an
s
f
o
r
m
er
ar
ch
itectu
r
e
h
av
e
b
ee
n
ca
r
r
ie
d
o
u
t
an
d
h
a
v
e
im
p
r
o
v
ed
r
esu
lts
s
ig
n
if
ican
tly
in
d
o
cu
m
en
t
s
u
m
m
ar
izatio
n
[
1
4
]
.
I
n
p
r
ev
io
u
s
s
tu
d
ies,
tr
an
s
f
o
r
m
er
s
w
as
u
s
ed
a
s
a
d
etec
tio
n
ir
o
n
y
g
r
o
u
p
in
g
in
Sp
an
is
h
u
s
in
g
p
r
e
-
tr
ain
in
g
T
witter
wo
r
d
r
esear
ch
r
esu
lts
co
m
p
ar
ed
to
L
STM
atten
tio
n
,
an
d
th
e
d
ee
p
av
er
ag
i
n
g
n
etwo
r
k
s
h
o
wed
an
in
cr
ea
s
e
s
ig
n
if
ican
tly
o
n
p
er
f
o
r
m
an
ce
s
[
1
5
]
.
T
h
e
t
r
an
s
f
o
r
m
e
r
also
s
u
cc
ee
d
ed
in
m
ak
in
g
C
h
in
ese
s
to
r
y
-
g
en
er
atio
n
b
y
cr
ea
tin
g
two
la
y
er
s
o
f
s
elf
-
atte
n
tio
n
a
n
d
r
ed
u
ci
n
g
t
h
e
n
u
m
b
e
r
o
f
en
co
d
er
an
d
d
ec
o
d
er
lay
er
s
to
id
e
n
tic
o
n
e.
T
h
e
r
esu
lts
s
h
o
wed
a
lo
w
l
o
s
s
an
d
an
i
n
cr
ea
s
e
s
ig
n
if
ican
tly
f
r
o
m
th
e
b
ase
lay
er
o
f
th
e
tr
a
n
s
f
o
r
m
er
m
o
d
el
[
1
6
]
.
Me
an
wh
ile,
th
e
u
s
e
o
f
th
e
tr
an
s
f
o
r
m
er
was
ca
r
r
ied
o
u
t
s
u
cc
ess
f
u
lly
u
s
in
g
a
co
m
b
in
ed
m
o
d
if
icatio
n
o
f
t
h
e
b
i
d
i
r
e
c
t
i
o
n
a
l
e
n
c
o
d
e
r
r
e
p
r
es
e
n
ta
t
i
o
n
s
f
r
o
m
t
r
a
n
s
f
o
r
m
e
r
s
(
B
E
R
T
)
a
s
a
t
r
a
n
s
f
o
r
m
e
r
-
b
a
s
e
d
e
n
c
o
d
e
r
a
n
d
d
e
c
o
d
e
r
i
n
J
a
p
a
n
e
s
e
a
b
s
t
r
a
ct
i
v
e
s
u
m
m
a
r
i
za
t
i
o
n
,
w
h
i
c
h
h
a
s
r
e
s
u
lt
e
d
i
n
g
o
o
d
a
v
e
r
a
g
e
a
c
c
u
r
a
c
y
a
n
d
t
h
e
l
o
w
e
s
t
l
o
s
s
v
a
l
u
e
[
1
7
]
.
I
n
th
is
s
tu
d
y
,
we
p
r
o
p
o
s
e
a
tr
a
n
s
f
o
r
m
er
-
b
ased
m
o
d
el
to
s
u
m
m
ar
ize
C
OVI
D
-
1
9
n
ews
with
s
ev
er
al
m
eth
o
d
s
a
n
d
s
tag
es.
An
o
th
er
d
is
cu
s
s
io
n
f
r
o
m
th
e
r
esu
lt
is
p
r
esen
t
b
y
m
ak
e
s
u
b
lay
er
m
o
d
if
icatio
n
s
to
d
et
er
m
in
e
th
e
ef
f
ec
t
o
f
p
ar
am
eter
s
o
n
t
h
e
en
co
d
er
a
n
d
d
ec
o
d
er
lay
er
s
.
2.
DATAS
E
T
T
h
e
d
ataset
we
u
s
ed
ca
m
e
f
r
o
m
n
ews d
o
cu
m
en
ts
ab
o
u
t COVI
D
-
1
9
th
at
was p
u
b
lis
h
ed
o
n
th
e
Kag
g
le
p
latf
o
r
m
[
1
8
]
f
r
o
m
t
h
e
C
an
ad
i
an
b
r
o
a
d
ca
s
tin
g
co
r
p
o
r
atio
n
(
C
B
C
)
n
ews
s
ite,
w
ith
a
to
tal
n
u
m
b
er
o
f
d
o
c
u
m
en
ts
th
at
w
er
e
u
s
ed
to
b
u
ild
th
e
m
o
d
el
is
2
7
5
5
d
o
c
u
m
en
ts
.
T
h
e
r
elev
an
ce
o
f
th
e
n
ews
in
th
e
d
ataset
co
n
tain
i
n
g
v
ar
iatio
n
s
co
m
b
i
n
ed
to
p
ics
r
elate
d
to
C
OVI
D
-
1
9
ar
e
p
r
o
ce
s
s
ed
u
s
in
g
th
e
cr
awle
r
with
th
e
k
ey
wo
r
d
C
OVI
D
-
1
9
p
u
b
lis
h
ed
f
r
o
m
J
an
u
ar
y
0
8
,
2
0
2
0
,
u
n
til
Ma
r
ch
0
3
,
2
0
2
0
.
I
n
t
h
e
d
ataset,
th
er
e
ar
e
tex
t
d
escr
ip
tio
n
co
n
tain
s
th
e
n
ews c
o
n
te
n
t,
a
n
d
th
e
d
escr
ip
tio
n
f
ea
tu
r
e
is
a
s
u
m
m
ar
y
o
f
th
e
n
ews c
o
n
ten
t.
S
p
ec
if
ic
k
e
y
wo
r
d
s
i
n
n
ews
co
n
ten
t
a
r
e
lis
t
ed
in
th
e
wo
r
d
c
o
r
o
n
a
v
ir
u
s
.
T
h
er
e
is
a
v
ar
iety
o
f
m
ix
ed
n
ews
co
n
ten
t,
b
u
t
th
e
n
ew
s
co
n
te
n
t
is
m
o
r
e
im
p
o
r
ta
n
t in
th
e
a
m
o
u
n
t o
f
C
OVI
D
-
1
9
g
r
o
wth
in
ea
ch
r
eg
io
n
.
3.
P
RE
P
RO
CE
SS
I
NG
Sev
er
al
p
r
ev
i
o
u
s
s
tu
d
ies
h
av
e
s
h
o
wn
th
e
r
esu
lts
o
f
th
eir
r
ese
ar
ch
b
y
d
o
in
g
p
r
ep
r
o
ce
s
s
in
g
c
an
in
cr
ea
s
e
ac
cu
r
ac
y
r
esu
lts
b
y
a
p
er
ce
n
ta
g
e
o
f
2
%
[
1
9
]
,
p
r
ep
r
o
ce
s
s
in
g
is
also
u
s
ed
in
s
o
m
e
wo
r
d
s
th
at
h
av
e
th
e
f
o
r
m
o
f
m
is
s
p
ellin
g
s
[
2
0
]
.
At
th
e
p
r
e
p
r
o
ce
s
s
in
g
s
tag
e,
s
ev
er
al
p
r
o
ce
s
s
es
o
cc
u
r
.
i.e
.
,
co
n
tr
ac
tio
n
s
,
lo
wer
ca
s
in
g
&
p
r
in
tab
le
c
h
ec
k
s
,
s
p
litt
in
g
d
ata,
to
k
en
izatio
n
,
a
n
d
w
o
r
d
em
b
e
d
d
in
g
.
We
d
iv
id
e
th
e
d
atas
et
in
to
th
r
ee
f
o
r
m
s
,
i.e
.
,
tr
ain
in
g
,
v
alid
atio
n
,
an
d
test
in
g
,
with
p
er
ce
n
ta
g
es
o
f
7
0%
,
1
0
%,
an
d
2
0
%,
r
esp
ec
tiv
ely
.
C
o
n
tr
ac
tio
n
s
an
d
p
r
in
tab
le
c
h
ec
k
s
,
m
ap
p
e
d
o
u
t
co
n
tr
ac
tio
n
wo
r
d
s
f
r
o
m
t
h
e
lis
t
o
f
wo
r
d
co
n
tr
ac
tio
n
s
.
T
h
ese
wo
r
d
s
wer
e
d
ef
in
e
d
b
y
o
u
r
s
elv
es
to
g
et
th
e
o
r
ig
in
al
f
o
r
m
ter
m
s
s
u
ch
as
"d
o
n
'
t"
b
e
co
m
e
"d
o
n
o
t
,"
th
e
n
p
r
in
tab
le
ch
ec
k
u
s
ed
to
d
elete
ch
ar
ac
ter
s
o
th
er
th
an
p
u
n
ctu
a
tio
n
m
ar
k
s
,
an
d
ASC
I
I
letter
s
.
Af
ter
th
at,
d
u
e
to
m
em
o
r
y
lim
itatio
n
s
we
d
id
d
is
tr
ib
u
te
co
n
t
r
o
l
to
lim
it
tex
t
wh
ich
was
n
ee
d
ed
as
a
tr
ain
i
n
g
m
o
d
el.
Fu
r
th
er
m
o
r
e,
t
o
k
en
iz
atio
n
is
th
e
p
r
o
ce
s
s
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
1
6
9
3
-
6
9
3
0
T
E
L
KOM
NI
KA
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l Co
n
tr
o
l
,
Vo
l.
19
,
No
.
3
,
J
u
n
e
2
0
2
1
:
7
5
4
-
7
61
756
o
f
b
r
ea
k
in
g
tex
t
in
t
o
s
ep
ar
ate
wo
r
d
s
an
d
a
d
d
in
g
u
n
i
q
u
e
t
o
k
en
s
.
I
n
th
e
m
o
d
if
icatio
n
o
f
t
h
e
m
o
d
el,
we
u
s
e
a
p
r
e
-
tr
ain
ed
w
o
r
d
em
b
ed
d
i
n
g
g
lo
b
al
v
ec
to
r
(
Glo
Ve)
with
a
v
o
ca
b
u
lar
y
o
f
2
.
2
M
to
p
r
es
en
t
ea
ch
wo
r
d
i
n
a
300
-
d
im
en
s
io
n
al
v
ec
to
r
s
ize
[
2
1
]
.
Pre
v
io
u
s
s
tu
d
ies
h
av
e
s
h
o
wn
th
at
u
n
s
u
p
er
v
is
ed
co
m
p
ar
i
s
o
n
r
esu
lts
b
ased
o
n
tex
t su
m
m
ar
izatio
n
u
s
in
g
wo
r
d
em
b
ed
d
in
g
a
r
e
m
o
r
e
ef
f
ec
tiv
e
th
an
u
s
in
g
a
b
ag
o
f
wo
r
d
s
[
2
2
]
.
4.
T
RANSF
O
R
M
E
R
M
O
D
E
L
W
e
an
aly
ze
d
in
Fig
u
r
e
1
,
a
n
d
th
er
e
a
r
e
en
co
d
er
an
d
d
e
co
d
er
lay
er
s
th
at
h
av
e
th
e
d
r
o
p
o
u
t
an
d
No
r
m
aliza
tio
n
in
ea
ch
s
u
b
lay
er
.
We
u
s
e
o
f
g
au
s
s
ian
er
r
o
r
lin
ea
r
u
n
it
(
GE
L
U)
in
th
e
f
ee
d
-
f
o
r
war
d
n
etwo
r
k
is
u
s
ed
o
n
ly
o
n
ce
o
n
ea
ch
en
co
d
er
o
r
d
e
co
d
er
lay
er
,
d
u
e
to
GE
L
U
h
as
h
ig
h
co
m
p
lex
ity
i
n
th
e
NL
P
f
ield
b
u
t
th
e
p
er
f
o
r
m
an
ce
p
r
o
d
u
ce
d
is
s
u
p
e
r
io
r
co
m
p
ar
ed
to
o
t
h
er
ac
tiv
at
io
n
f
u
n
ctio
n
s
s
u
ch
as
E
L
U
a
n
d
R
eL
U
[
2
3
]
.
T
h
e
m
u
lti
-
h
ea
d
atten
tio
n
f
o
r
m
u
lati
o
n
ca
n
b
e
s
ee
n
f
o
llo
win
g
in
(
1
)
[
1
4
]
.
h
is
th
e
to
tal
atten
tio
n
c
ar
r
ied
o
u
t
in
p
ar
a
llel
s
o
th
at
ev
er
y
ℎ
is
ca
r
r
ied
o
u
t th
e
atten
tio
n
f
u
n
ctio
n
c
o
n
tain
ed
in
(
2
)
[
1
4
]
.
Fig
u
r
e
1
.
T
r
an
s
f
o
r
m
e
r
m
o
d
el
p
r
o
p
o
s
ed
(
,
,
)
=
(
ℎ
1
,
…
,
ℎ
ℎ
)
(
1
)
ℎ
=
(
,
,
)
(
2
)
(
,
,
)
=
(
√
)
(
3
)
T
h
e
atten
tio
n
f
u
n
ctio
n
ca
n
b
e
d
ef
in
ed
as
a
f
u
n
ctio
n
th
at
p
er
f
o
r
m
s
th
e
m
ap
p
in
g
o
f
th
e
q
u
er
y
Q
is
th
e
tar
g
et
s
eq
u
en
ce
;
th
e
k
e
y
p
air
K
an
d
th
e
v
alu
e
V
ar
e
d
er
iv
e
d
f
r
o
m
th
e
s
eq
u
en
ce
.
E
ac
h
Q
,
K
,
V
,
a
n
d
o
u
tp
u
t
m
ap
p
in
g
a
r
e
d
ef
in
e
d
in
v
ec
to
r
f
o
r
m
.
T
h
e
weig
h
t
o
f
ea
c
h
ca
l
cu
lated
v
alu
e
is
a
r
ep
r
esen
tati
o
n
o
f
a
d
ju
s
tin
g
th
e
q
u
er
y
to
th
e
k
e
y
.
T
h
e
q
u
e
r
y
an
d
k
ey
d
im
en
s
io
n
s
ar
e
d
ef
in
e
d
as
,
an
d
th
e
v
alu
es
d
im
en
s
io
n
is
u
s
ed
as
th
e
Atten
tio
n
p
ar
am
eter
f
o
u
n
d
in
(
3
)
[
1
4
]
.
Mu
lti
-
h
ea
d
atten
tio
n
co
m
b
in
es
s
ev
er
al
atten
ti
o
n
m
o
d
els
to
ea
ch
o
f
th
e
,
,
m
o
d
els.
T
h
e
weig
h
tin
g
d
i
m
e
n
s
io
n
o
f
a
s
eq
u
e
n
ce
is
d
ef
in
e
d
as
s
o
t
h
at
its
r
e
p
r
esen
tatio
n
is
in
m
u
lti
-
h
ea
d
∈
ℝ
×
.
T
h
e
p
r
im
ar
y
d
if
f
er
en
ce
b
etwe
en
a
m
ask
ed
m
u
lti
-
h
ea
d
atten
tio
n
an
d
a
m
u
lti
-
h
ea
d
atten
tio
n
is
t
h
at
s
o
m
e
to
k
e
n
s
co
n
tain
e
d
in
a
s
eq
u
en
ce
ar
e
r
an
d
o
m
ly
r
em
o
v
ed
t
o
tr
ain
th
e
m
o
d
el
to
u
n
d
er
s
tan
d
t
h
e
co
n
tex
t
c
o
n
tain
ed
in
th
e
s
eq
u
en
ce
.
T
r
an
s
f
o
r
m
er
also
p
er
f
o
r
m
s
p
o
s
itio
n
al
en
co
d
in
g
(
)
,
wh
ich
is
th
e
in
jectio
n
o
f
s
o
m
e
i
n
f
o
r
m
atio
n
o
n
ea
ch
wo
r
d
p
o
s
itio
n
co
n
tai
n
ed
in
a
s
eq
u
en
ce
.
PE
h
as
th
e
s
am
e
d
im
en
s
io
n
s
as
.
I
n
th
is
p
ap
er
,
we
u
s
e
s
in
e
-
co
s
in
e
p
o
s
itio
n
al
en
co
d
in
g
,
wh
er
e
th
e
f
o
r
m
u
la
eq
u
atio
n
ca
n
b
e
s
ee
n
in
(
4
)
an
d
(
5
)
,
th
e
pos
is
a
p
o
s
itio
n
,
an
d
i
is
d
im
e
n
s
io
n
.
(
,
2
)
=
s
in
(
10000
2
)
(
4
)
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KOM
NI
KA
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l Co
n
tr
o
l
S
u
mma
r
iz
a
tio
n
o
f COVI
D
-
1
9
n
ews
d
o
cu
men
ts
d
ee
p
lea
r
n
in
g
-
b
a
s
ed
u
s
in
g
… (
N
u
r
Ha
ya
tin
)
757
(
,
2
+
1
)
=
c
os
(
10000
2
)
(
5
)
Dr
o
p
o
u
t
r
ed
u
c
es
th
e
lo
s
s
va
lu
e
d
u
r
i
n
g
th
e
tr
ain
in
g
p
r
o
ce
s
s
,
also
h
elp
s
p
r
ev
en
t
o
v
er
f
itti
n
g
[
2
4
]
.
T
h
e
n
o
r
m
aliza
tio
n
lay
e
r
n
o
r
m
alize
d
v
alu
es
co
m
e
f
r
o
m
th
e
h
id
d
e
n
lay
er
.
P
er
f
o
r
m
o
n
s
m
all
b
atch
s
izes
d
ep
en
d
en
t
t
o
r
ed
u
ce
m
em
o
r
y
c
o
s
t,
n
o
r
m
ali
za
tio
n
ca
n
r
ely
f
o
r
in
cr
ea
s
e
t
r
ain
in
g
ac
cu
r
ac
y
[
2
5
]
.
T
h
e
n
o
r
m
aliza
tio
n
lay
er
m
in
im
ize
p
ar
am
eter
c
h
an
g
e
d
u
r
in
g
p
r
o
p
a
g
ated
th
r
o
u
g
h
th
e
d
ee
p
n
etwo
r
k
s
[
2
6
]
.
5.
SCO
RING
T
h
e
s
u
m
m
a
r
izatio
n
r
esu
lt
m
ea
s
u
r
em
en
ts
ar
e
p
er
f
o
r
m
e
d
u
s
in
g
r
ec
all
-
o
r
ie
n
ted
u
n
d
e
r
s
tan
d
in
g
f
o
r
g
is
tin
g
ev
alu
atio
n
(
R
OUGE
)
[
2
7
]
.
W
e
ch
o
s
e
th
e
R
OUGE
-
N
m
e
th
o
d
,
w
h
ich
was
r
ep
r
esen
ted
in
(
6
)
,
in
wh
ich
th
e
ca
lcu
latio
n
is
b
ased
o
n
n
-
g
r
am
r
ec
all
[
2
7
]
.
W
h
er
e
n
is
th
e
len
g
th
o
f
n
-
g
r
am
,
R
ef
is
a
s
et
o
f
r
e
f
er
en
ce
s
u
m
m
ar
ies.
ℎ
(
)
is
th
e
ca
lcu
latio
n
o
f
th
e
m
ax
im
u
m
n
u
m
b
er
o
f
n
-
g
r
am
s
c
o
-
o
cc
u
r
r
in
g
o
n
th
e
g
e
n
er
ated
s
u
m
m
ar
ies
m
o
d
el
an
d
t
h
e
s
et
o
f
r
ef
er
e
n
ce
s
u
m
m
ar
ies.
(
)
is
th
e
n
u
m
b
er
o
f
n
-
g
r
am
s
in
r
ef
er
e
n
ce
s
u
m
m
ar
ies.
−
=
Σ
∈
Σ
∈
ℎ
(
)
Σ
∈
Σ
∈
(
)
(
6
)
6.
E
XP
E
R
I
M
E
N
T
AND
D
I
SC
USSI
O
N
I
n
ac
co
r
d
an
ce
with
Fig
u
r
e
2
,
th
e
ex
p
er
im
e
n
t
f
r
o
m
all
tr
an
s
f
o
r
m
er
s
u
m
m
ar
izatio
n
m
o
d
el
s
th
at
we
b
u
ild
f
ir
s
t
p
er
f
o
r
m
ed
p
r
ep
r
o
c
ess
in
g
o
n
th
e
tex
t.
T
h
e
u
s
e
o
f
p
r
e
p
r
o
ce
s
s
in
g
is
to
r
ed
u
ce
less
r
elev
an
t
f
ea
tu
r
es,
an
d
th
e
am
o
u
n
t
o
f
m
e
m
o
r
y
n
ee
d
e
d
to
ca
r
r
y
o
u
t
th
e
tr
ain
in
g
p
r
o
ce
s
s
[
2
8
]
.
T
h
en
af
ter
p
r
ep
r
o
ce
s
s
in
g
,
we
p
er
f
o
r
m
all
m
o
d
el
s
tr
an
s
f
o
r
m
er
-
b
ased
ar
c
h
itectu
r
e
with
en
c
o
d
er
an
d
d
ec
o
d
er
th
at
ca
n
b
e
d
o
n
e
s
ev
er
al
ti
m
es
r
ep
ea
ted
ly
an
d
m
ak
e
a
co
m
p
ar
is
o
n
o
f
lay
er
m
o
d
if
icatio
n
s
b
ased
o
n
s
co
r
in
g
.
Fig
u
r
e
2
.
T
h
e
n
ews su
m
m
ar
iz
atio
n
ex
p
er
im
e
n
t
o
f
C
OVI
D
-
19
6
.
1
.
So
f
t
wa
re
a
nd
s
pli
t
t
ing
T
h
e
s
p
ec
if
icatio
n
s
in
th
e
e
x
p
e
r
im
en
t
u
s
ed
Go
o
g
le
co
lab
clo
u
d
co
m
p
u
tin
g
with
d
ata
as
f
o
l
lo
ws:
I
n
tel
Xeo
n
C
PU
2
.
2
0
GHz
,
1
4
GB
R
AM
,
T
esla
P1
0
0
1
6
GB
GPU.
T
h
e
Py
th
o
n
T
en
s
o
r
Flo
w
lib
r
ar
y
is
u
s
ed
as
a
d
ee
p
b
ac
k
en
d
lear
n
in
g
wh
er
e
ca
lcu
l
atio
n
s
a
r
e
p
er
f
o
r
m
e
d
o
n
th
e
GPU.
T
h
e
r
esu
lts
o
f
s
p
litt
in
g
o
n
th
e
C
OVI
D
-
1
9
n
ews
d
o
cu
m
e
n
t
d
ataset
wer
e
1
9
2
8
d
o
cu
m
e
n
ts
to
co
n
d
u
ct
tr
ai
n
in
g
,
2
7
5
d
o
cu
m
e
n
ts
as
v
alid
atio
n
d
u
r
in
g
t
h
e
tr
ain
in
g
p
r
o
ce
s
s
,
an
d
5
5
2
d
o
c
u
m
en
ts
to
test
th
e
r
esu
lts
o
f
th
e
tr
ain
i
n
g
m
o
d
el,
f
o
r
ea
c
h
tr
ain
in
g
m
o
d
el
.
V
alid
atin
g
p
e
r
f
o
r
m
u
n
d
er
a
b
atch
o
f
3
2
d
o
cu
m
e
n
t
s
to
ca
lcu
late
th
e
av
er
ag
e
R
O
UGE
-
1
s
co
r
e
an
d
lo
s
s
v
alu
e.
T
h
e
m
ax
im
u
m
len
g
th
o
f
th
e
n
ews te
x
t c
o
n
ten
t o
f
th
e
en
tire
tr
ain
in
g
d
o
c
u
m
en
t
was
eq
u
al
to
6
0
0
a
n
d
2
5
in
th
e
tex
t
d
escr
ip
tio
n
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
1
6
9
3
-
6
9
3
0
T
E
L
KOM
NI
KA
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l Co
n
tr
o
l
,
Vo
l.
19
,
No
.
3
,
J
u
n
e
2
0
2
1
:
7
5
4
-
7
61
758
6
.
2
.
E
x
perim
ent
s
ce
na
rio
I
n
th
is
r
esear
ch
ex
p
e
r
im
en
t,
w
e
u
s
e
Ad
am
w
h
ich
is
th
e
s
to
c
h
asti
c
b
ased
o
p
tim
izatio
n
m
eth
o
d
to
u
p
d
ate
th
e
weig
h
t
v
alu
e
o
f
th
e
lo
s
s
v
alu
e
m
ea
s
u
r
em
en
t
r
esu
lts
[
2
9
]
,
wh
er
e
th
e
ca
lcu
latio
n
o
f
t
h
e
lo
s
s
v
alu
e
o
f
th
e
weig
h
t
v
alu
e
u
s
ed
s
p
ar
s
e
s
o
f
t
m
ax
cr
o
s
s
-
en
tr
o
p
y
.
T
a
b
le
1
s
h
o
wn
t
h
e
e
x
p
er
im
e
n
t
s
ce
n
ar
io
o
f
s
ev
er
al
d
if
f
er
e
n
t
p
ar
am
eter
s
u
s
ed
to
b
u
ild
th
e
tr
an
s
f
o
r
m
er
d
ee
p
lear
n
in
g
m
o
d
el.
A
d
am
o
p
tim
izatio
n
p
ar
am
eter
s
u
s
ed
ar
e
b
eta1
=0
.
9
,
b
eta2
=0
.
9
8
,
ep
s
ilo
n
=1
e
-
9
.
All
m
o
d
els
h
ad
ca
r
r
i
ed
o
u
t
f
air
ly
ite
r
atio
n
s
o
f
4
0
ep
o
ch
s
o
n
tr
ain
in
g
ex
p
er
im
en
t
s
.
Fro
m
th
e
ex
is
ti
n
g
tr
an
s
f
o
r
m
er
d
esig
n
m
o
d
el
in
p
r
ev
io
u
s
s
tu
d
ies,
we
ch
o
s
e
th
e
tr
an
s
f
o
r
m
er
C
m
o
d
el
(
T
C
M)
[
1
4
]
as
a
co
m
p
ar
is
o
n
test
with
s
o
m
e
o
f
t
h
e
m
o
d
els
th
at
we
p
r
o
p
o
s
ed
.
T
h
e
s
elec
tio
n
is
b
ec
au
s
e
T
C
M
h
as
th
e
m
o
s
t
s
tr
aig
h
tf
o
r
war
d
d
esig
n
an
d
th
e
r
esu
lts
o
f
tr
ials
with
o
th
er
ar
ch
itectu
r
es
th
at
ar
e
m
o
r
e
c
o
m
p
lex
b
y
1
%.
T
h
e
m
o
d
el
th
at
we
p
r
o
p
o
s
ed
in
clu
d
es
to
k
e
n
izatio
n
a
n
d
wo
r
d
e
m
b
ed
d
in
g
,
a
n
d
in
th
e
f
o
r
m
o
f
p
ar
am
eter
ch
an
g
es
m
o
d
i
f
icatio
n
o
f
th
e
e
n
co
d
er
-
d
ec
o
d
er
lay
er
,
o
r
ca
n
b
e
ca
lled
a
m
o
d
if
ie
d
b
ase
m
o
d
el
tr
an
s
f
o
r
m
er
with
d
is
tr
ib
u
te
c
o
n
tr
o
l
to
k
e
n
izatio
n
an
d
Glo
Ve
wo
r
d
em
b
ed
d
in
g
.
T
h
e
p
o
s
tf
ix
n
u
m
b
er
is
a
r
ep
r
esen
tatio
n
o
f
th
e
n
u
m
b
er
o
f
id
e
n
tical
lay
er
en
co
d
er
d
ec
o
d
er
s
(
MT
DT
G
Nx
)
.
T
h
e
ac
tiv
atio
n
f
u
n
ctio
n
in
t
h
e
M
T
DT
G
m
o
d
el
wh
ich
was
u
s
ed
in
th
i
s
r
esear
ch
is
t
h
e
GE
L
U
f
u
n
ctio
n
to
ca
lcu
late
th
e
weig
h
t
o
f
th
e
s
eq
u
en
ce
in
th
e
f
ee
d
-
f
o
r
war
d
lay
er
,
wh
er
ea
s
in
p
r
ev
io
u
s
s
tu
d
ies u
s
in
g
R
eL
U
as a
n
ac
tiv
atio
n
f
u
n
ctio
n
.
T
ab
le
1
.
Scen
ar
i
o
ex
p
e
r
im
en
t
P
a
r
a
me
t
e
r
TC
M
[
1
4
]
O
u
r
p
r
o
p
o
s
e
d
m
o
d
e
l
s
M
TD
TG
2
M
TD
TG
5
M
TD
TG
6
h
e
a
d
s
8
10
8
10
l
e
a
r
n
i
n
g
r
a
t
e
1e
-
3
1e
-
3
1e
-
5
1e
-
2
n
o
d
e
f
e
e
d
-
f
o
r
w
a
r
d
2
5
6
5
1
2
2
5
6
5
1
2
d
r
o
p
o
u
t
r
a
t
e
0
.
1
0
.
2
0
.
1
0
.
2
a
t
t
e
n
t
i
o
n
d
r
o
p
o
u
t
r
a
t
e
0
.
1
0
.
2
0
.
1
0
.
2
e
n
c
o
d
e
r
l
a
y
e
r
2
2
5
6
d
e
c
o
d
e
r
l
a
y
e
r
2
2
5
6
a
c
t
i
v
a
t
i
o
n
f
u
n
c
t
i
o
n
R
e
LU
G
ELU
G
ELU
G
ELU
6
.
3
.
E
x
perim
ent
re
s
ult
Du
r
in
g
t
h
e
tr
ain
in
g
p
r
o
ce
s
s
,
a
lo
s
s
v
alu
e
an
d
R
OUGE
-
1
is
o
b
tain
ed
,
as
s
h
o
wn
in
Fig
u
r
e
3
.
T
C
M
h
as
d
ec
r
ea
s
ed
lo
s
s
in
ep
o
ch
4
0
,
s
o
th
at
a
lo
s
s
v
alu
e
o
f
6
.
3
is
o
b
t
ain
ed
.
I
n
o
t
h
er
m
o
d
els
r
an
g
in
g
f
r
o
m
e
p
o
ch
1
8
-
27
lo
s
s
in
MT
DT
G
2
g
r
ad
u
ally
d
e
cr
ea
s
e
d
at
5
.
5
,
th
en
clim
b
e
d
b
a
ck
u
p
b
ec
au
s
e
o
f
t
h
e
v
ar
i
an
t
b
a
tch
in
th
e
d
o
cu
m
e
n
t
v
ar
y
,
s
o
t
h
at
th
e
m
o
d
el
m
u
s
t
r
e
co
g
n
ize
n
ew
d
ata
ag
ain
.
Oth
er
m
o
d
els
as
s
h
o
wn
in
Fig
u
r
e
3
(
a
)
,
th
e
g
r
ap
h
d
ep
icts
th
at
th
e
MT
DT
G
5
d
ec
r
ea
s
ed
g
r
ad
u
ally
in
lo
s
s
at
ep
o
ch
1
0
an
d
2
5
with
o
b
tain
ed
lo
s
s
v
al
u
e
o
f
4
.
8
.
T
h
e
latest
r
esu
lts
o
f
o
u
r
ex
p
e
r
im
en
t
o
n
M
T
DT
G
6
h
av
e
d
ec
r
ea
s
ed
a
lo
s
s
v
alu
e
to
4
.
7
wh
ic
h
is
n
o
t
m
u
ch
d
if
f
er
en
t
c
o
m
p
ar
e
d
to
MT
DT
G
5
with
a
d
if
f
e
r
en
c
e
o
f
0
.
1
%.
Du
r
i
n
g
th
e
tr
ain
in
g
p
h
ase,
ea
ch
ep
o
c
h
v
alid
atio
n
b
ased
o
n
m
ax
im
u
m
R
OUGE
-
1
s
co
r
e
is
co
n
s
id
er
ed
to
s
av
e
th
e
weig
h
t
m
o
d
el.
T
h
is
ex
p
er
im
en
t
u
s
ed
R
OUGE
-
1
to
s
u
m
m
ar
izatio
n
r
esu
lt
m
ea
s
u
r
e,
th
e
r
esu
lt
o
f
R
OUGE
-
1
was
s
h
o
wn
b
y
Fig
u
r
e
3
(
b
)
.
T
h
e
g
r
ap
h
ex
p
lain
ed
th
e
v
alid
atio
n
p
r
o
ce
s
s
wh
er
e
ea
ch
ep
o
ch
T
C
M
h
as
a
m
ax
im
u
m
s
co
r
e
0
.
2
0
.
Ou
r
p
r
o
p
o
s
ed
m
o
d
el
s
h
o
wn
b
y
MT
DT
G
2
,
MD
T
G
5
,
an
d
MT
DT
G
6
,
th
o
s
e
m
o
d
els
h
a
v
e
an
o
u
tp
er
f
o
r
m
ed
r
esu
lt
m
ea
s
u
r
ed
b
y
R
OUGE
-
1
with
s
co
r
e
0
.
5
4
,
0
.
5
9
,
an
d
0
.
6
0
r
esp
ec
tiv
ely
.
(
a)
(
b
)
Fig
u
r
e
3
.
C
o
m
p
a
r
is
o
n
;
(
a
)
m
o
d
el
tr
ain
in
g
lo
s
s
an
d
(
b
)
R
OUGE
-
1
s
co
r
e
v
alid
atio
n
6
.
4
.
Su
mm
a
riza
t
io
n
m
o
del
W
e
tr
y
to
ex
p
lo
r
e
th
e
r
esu
lts
o
f
MT
DT
G
6
,
o
n
th
e
wo
r
d
clo
u
d
,
as
s
ee
n
in
Fig
u
r
e
4
.
W
o
r
d
clo
u
d
d
escr
ib
es
th
e
wo
r
d
f
r
eq
u
e
n
c
y
r
ep
r
esen
tatio
n
o
f
th
e
wh
o
l
e
d
o
cu
m
en
t
f
r
o
m
th
e
g
en
er
a
ted
s
u
m
m
ar
izatio
n
co
n
d
u
cte
d
b
y
MT
DT
G
6
.
T
o
ta
l a
ll u
n
i
q
u
e
w
o
r
d
s
wh
ile
p
er
f
o
r
m
in
g
su
m
m
ar
izatio
n
ap
p
ea
r
e
d
in
Fig
u
r
e
4
is
2
0
0
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KOM
NI
KA
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l Co
n
tr
o
l
S
u
mma
r
iz
a
tio
n
o
f COVI
D
-
1
9
n
ews
d
o
cu
men
ts
d
ee
p
lea
r
n
in
g
-
b
a
s
ed
u
s
in
g
… (
N
u
r
Ha
ya
tin
)
759
wo
r
d
s
th
at
u
s
ed
th
e
MT
DT
G
6
m
o
d
el.
Mo
s
t
wo
r
d
g
e
n
er
ated
o
n
s
u
m
m
a
r
izatio
n
is
‘
s
ay
’
,
th
is
r
esu
lt
is
ca
u
s
ed
b
y
th
e
d
ataset
u
s
ed
r
elate
d
to
n
ew
s
co
n
ten
t
to
g
et
in
f
o
r
m
atio
n
f
r
o
m
an
in
ter
v
iew
o
n
ex
p
e
r
ts
o
r
g
o
v
er
n
m
e
n
t
o
f
f
ici
als.
T
h
e
wo
r
d
"Can
ad
ian
"
also
ap
p
ea
r
ed
,
b
ec
au
s
e
th
is
wo
r
d
r
ela
ted
to
th
e
n
ews
th
at
h
as
b
ee
n
r
ep
o
r
ted
m
o
s
tly
o
n
C
an
ad
ian
d
u
r
i
n
g
th
e
c
o
r
o
n
av
ir
u
s
s
itu
atio
n
.
W
e
class
if
ied
k
ey
wo
r
d
b
ased
o
n
f
r
e
q
u
en
t
o
f
a
p
p
ea
r
an
ce
.
On
T
ab
le
2
,
t
h
er
e
ar
e
two
ty
p
es
o
f
k
ey
wo
r
d
:
"m
o
s
t"
(
k
ey
w
o
r
d
s
h
av
e
a
h
ig
h
f
r
e
q
u
en
t
)
,
a
n
d
"lea
s
t"
(
k
ey
wo
r
d
s
ap
p
ea
r
ed
r
ar
ely
)
.
T
h
e
p
er
ce
n
tag
e
o
f
o
v
er
all
wo
r
d
clo
u
d
s
u
m
m
ar
izatio
n
r
es
u
lts
f
r
o
m
Fig
u
r
e
4
is
s
ee
n
th
e
t
ab
le.
T
h
e
m
o
s
t
5
wo
r
d
s
th
at
h
av
e
h
ig
h
an
ap
p
ea
r
f
r
eq
u
e
n
cy
ar
e
"say
,
will,
co
r
o
n
av
ir
u
s
,
p
eo
p
le,
p
an
d
em
ic".
T
h
ese
k
ey
wo
r
d
s
p
r
o
v
e
th
at
th
e
m
ain
to
p
ic
o
f
n
ews
r
ep
o
r
ted
wid
ely
d
u
r
in
g
th
is
p
e
r
io
d
is
a
b
o
u
t
th
e
co
r
o
n
a
v
ir
u
s
p
an
d
em
ic.
I
t
ca
u
s
ed
th
e
g
o
v
er
n
m
en
t o
f
C
an
a
d
a
t
o
tak
e
p
o
licies
f
o
c
u
s
ed
o
n
p
u
b
li
c
h
ea
lth
wh
ic
h
co
n
ce
r
n
e
d
with
th
e
s
af
ety
o
f
citizen
s
.
T
h
e
o
th
er
h
an
d
,
th
e
f
ewe
s
t
5
wo
r
d
s
f
o
u
n
d
:
"f
a
m
ilies
,
d
is
ea
s
e,
g
r
o
ce
r
y
s
to
r
e
,
f
ir
s
t n
atio
n
,
s
to
ck
m
ar
k
et
".
T
h
e
m
ain
th
in
g
th
at
h
ad
b
ee
n
less
co
n
ce
r
n
f
r
o
m
r
ep
o
r
t
n
ews
is
t
h
at
th
e
c
o
r
o
n
a
v
ir
u
s
p
an
d
e
m
ic
af
f
ec
ted
th
e
wea
k
o
f
m
an
y
s
ec
to
r
s
,
in
clu
d
in
g
tr
a
d
in
g
an
d
ec
o
n
o
m
y
,
an
d
also
less
r
ep
o
r
ted
a
b
o
u
t th
e
p
o
licy
is
s
u
e
t
o
clo
s
e
p
u
b
lic
p
lace
s
i.e
.
p
ar
k
an
d
s
to
r
e.
Fig
u
r
e
4
.
W
o
r
d
clo
u
d
s
u
m
m
ar
izatio
n
MT
DT
G
6
T
ab
le
2
.
Mo
s
t a
n
d
least wo
r
d
r
ep
r
esen
tatio
n
Ty
p
e
R
a
n
k
K
e
y
w
o
r
d
s
P
e
r
c
e
n
t
a
g
e
mo
st
1
-
5
say
,
w
i
l
l
,
c
o
r
o
n
a
v
i
r
u
s,
p
e
o
p
l
e
,
p
a
n
d
e
mi
c
0
.
8
3
6
-
10
p
r
o
v
i
n
c
e
,
n
e
w
,
c
a
se
,
c
a
n
a
d
i
a
n
,
c
o
r
o
n
a
v
i
r
u
s
o
u
t
b
r
e
a
k
0.
56
11
-
15
p
u
b
l
i
c
h
e
a
l
t
h
,
g
o
v
e
r
n
m
e
n
t
,
c
i
t
y
,
c
a
n
a
d
a
,
s
a
i
d
0.
45
16
-
20
d
a
y
,
mo
n
d
a
y
,
s
p
r
e
a
d
,
n
o
v
e
l
c
o
r
o
n
a
v
i
r
u
s,
t
w
o
0.
38
21
-
25
o
u
t
b
r
e
a
k
,
c
h
i
n
a
,
a
n
n
o
u
n
c
e
d
,
h
o
me
,
n
o
w
0.
34
l
e
a
s
t
1
9
6
-
200
d
i
s
e
a
se
,
f
a
m
i
l
i
e
s
,
m
a
sk
s
,
m
a
j
o
r
,
c
a
u
se
d
0
.
0
8
1
9
1
-
195
d
e
a
t
h
,
t
h
r
e
a
t
,
o
p
e
r
a
t
i
o
n
s
,
w
a
r
n
i
n
g
,
p
l
a
c
e
0
.
0
8
1
8
6
-
190
p
r
i
c
e
,
p
o
t
e
n
t
i
a
l
,
st
o
r
e
,
n
u
r
se
,
p
a
r
k
0
.
0
8
1
8
1
-
185
p
r
o
v
i
n
c
i
a
l
h
e
a
l
t
h
,
c
a
r
e
w
o
r
k
e
r
s,
s
e
l
f
i
s
o
l
a
t
i
o
n
,
sl
o
w
s
p
r
e
a
d
,
st
o
c
k
m
a
r
k
e
t
0
.
0
8
1
7
6
-
180
f
i
r
st
n
a
t
i
o
n
,
g
r
o
c
e
r
y
st
o
r
e
,
e
x
p
e
r
t
,
p
u
b
l
i
c
,
i
n
c
l
u
d
i
n
g
0
.
0
9
6
.
5
.
T
esting
m
o
del
T
h
e
d
is
tr
ib
u
tio
n
o
f
R
OUGE
-
1
an
d
R
OUGE
-
2
s
co
r
es o
n
MT
DT
G
6
is
s
h
o
wn
in
Fig
u
r
e
5
,
t
h
e
n
u
m
b
er
o
f
s
u
m
m
ar
izatio
n
r
esu
lts
is
at
m
o
s
t
8
0
-
1
0
0
%
with
a
to
tal
o
f
2
0
9
d
o
cu
m
e
n
ts
,
wh
ile
at
least
1
0
%
is
s
co
r
ed
with
a
to
tal
o
f
3
3
d
o
cu
m
e
n
ts
.
I
n
v
er
s
ely
p
r
o
p
o
r
tio
n
al
t
o
R
OUGE
-
2
,
wh
er
e
t
h
e
m
o
s
t
d
is
tr
ib
u
ti
o
n
in
th
e
r
esu
lt
o
f
th
e
s
co
r
e
is
in
th
e
0
-
2
0
%
b
ec
au
s
e
R
OUGE
-
2
u
s
es
th
e
b
ig
r
am
m
ec
h
an
is
m
.
So
,
th
er
e
ar
e
s
ev
er
al
o
v
er
lap
p
i
n
g
wo
r
d
s
b
etwe
e
n
s
u
m
m
ar
izatio
n
an
d
r
ef
er
en
ce
s
.
T
h
e
co
m
p
a
r
is
o
n
r
e
s
u
lts
te
s
t
al
s
o
s
h
o
wn
in
T
ab
le
3
.
T
h
e
r
esu
lts
o
f
MT
DT
G
5
an
d
MT
DT
G
6
o
n
th
e
R
OUGE
-
1
s
co
r
e
h
av
e
a
d
if
f
er
en
ce
o
f
1
%
wh
en
u
s
ed
o
n
th
e
R
OUGE
-
2
s
co
r
e
.
T
h
er
e
is
q
u
ite
a
d
if
f
er
en
ce
o
f
3
%.
T
h
e
ev
alu
atio
n
r
esu
lts
s
h
o
w
th
at
MT
DT
G
6
with
a
R
O
UGE
-
1
s
co
r
e
o
f
0
.
5
8
an
d
a
R
OUGE
-
2
s
co
r
e
o
f
0
.
4
2
i
s
th
e
b
est m
o
d
el
th
at
ca
n
b
e
u
s
ed
in
th
e
test
d
ata.
Fig
u
r
e
5
.
Dis
tr
ib
u
tio
n
R
OUG
E
s
co
r
e
o
n
test
d
o
cu
m
en
ts
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
1
6
9
3
-
6
9
3
0
T
E
L
KOM
NI
KA
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l Co
n
tr
o
l
,
Vo
l.
19
,
No
.
3
,
J
u
n
e
2
0
2
1
:
7
5
4
-
7
61
760
F
r
o
m
t
h
e
r
e
s
u
l
t
s
o
f
t
h
e
e
n
t
i
r
e
t
r
i
a
l
o
f
t
h
e
t
e
s
t
d
o
c
u
m
e
n
t
,
t
h
e
r
e
i
s
o
n
l
y
M
T
D
T
G
6
a
r
c
h
i
t
e
c
t
u
r
e
t
h
a
t
c
a
n
b
e
s
u
p
p
l
i
e
d
w
i
t
h
t
h
e
m
a
x
i
m
u
m
d
u
e
t
o
t
h
e
m
e
m
o
r
y
l
i
m
i
t
a
t
i
o
n
s
h
a
n
d
l
e
d
b
y
t
h
e
G
P
U
.
I
n
t
h
e
e
x
p
e
r
i
m
e
n
t
,
t
h
e
r
e
s
e
a
r
c
h
e
r
o
f
t
e
n
g
o
t
c
o
n
s
t
r
a
i
n
t
s
o
n
o
u
t
o
f
m
e
m
o
r
y
(
O
O
M
)
r
e
s
o
u
r
c
e
s
.
T
h
i
s
o
b
s
t
a
c
l
e
c
a
n
b
e
o
v
e
r
c
o
m
e
b
y
r
e
d
u
c
i
n
g
t
h
e
a
r
c
h
i
t
e
c
t
u
r
a
l
d
e
s
i
g
n
m
o
d
e
l
,
e
s
p
e
c
i
a
l
l
y
t
h
e
m
o
s
t
i
m
p
o
r
t
a
n
t
t
h
i
n
g
s
,
i
.
e
.
,
t
h
e
n
u
m
b
e
r
o
f
f
e
e
d
-
f
o
r
w
a
r
d
n
e
t
w
o
r
k
s
,
b
a
t
c
h
s
i
z
e
,
a
n
d
t
h
e
n
u
m
b
e
r
o
f
e
n
c
o
d
e
r
-
d
e
c
o
d
e
r
s
.
T
h
e
d
i
s
a
d
v
a
n
t
a
g
e
o
f
M
T
D
T
G
6
i
s
t
h
a
t
t
h
e
m
e
m
o
r
y
n
e
e
d
e
d
t
o
c
o
n
d
u
c
t
t
r
a
i
n
i
n
g
i
s
m
o
r
e
s
i
g
n
i
f
i
c
a
n
t
b
e
c
a
u
s
e
w
e
u
s
e
3
0
0
-
d
i
m
e
n
s
i
o
n
a
l
G
l
o
V
e
a
s
w
o
r
d
e
m
b
e
d
d
i
n
g
.
H
o
w
e
v
e
r
,
t
h
e
r
e
s
u
l
t
s
o
f
t
h
e
t
e
s
t
m
o
d
e
l
c
a
n
i
n
c
r
e
a
s
e
b
y
a
p
e
r
c
e
n
t
a
g
e
o
f
1
3
%
i
n
R
O
U
G
E
-
1
a
n
d
1
6
%
R
O
U
G
E
-
2
c
o
m
p
a
r
e
d
t
o
T
C
M
.
T
ab
le
3
.
Ov
e
r
all
co
m
p
a
r
is
o
n
s
co
r
e
m
o
d
el
M
o
d
e
l
V
a
l
i
d
a
t
i
o
n
Te
st
Tr
a
i
n
i
n
g
t
i
me
(
sec
o
n
d
)
M
a
x
i
m
u
m
R
O
U
G
E
-
1
R
O
U
G
E
-
1
R
O
U
G
E
-
2
TC
M
[
1
4
]
0
.
2
0
0
.
4
5
0
.
2
6
1
7
2
3
M
TD
TG
2
0
.
5
4
0
.
5
1
0
.
3
4
5
0
4
6
M
TD
TG
5
0
.
5
9
0
.
5
6
0
.
3
8
7
6
0
6
M
TD
TG
6
0
.
6
0
0
.
5
8
0
.
4
2
1
1
4
3
8
7.
CO
NCLU
SI
O
N
Su
m
m
ar
izatio
n
o
f
n
ews
d
o
cu
m
en
ts
C
OVI
D
-
1
9
b
ased
o
n
d
e
ep
lear
n
in
g
u
s
in
g
tr
an
s
f
o
r
m
e
r
ar
ch
itectu
r
e
ca
n
b
e
d
o
n
e
b
y
co
m
p
ilin
g
v
a
r
io
u
s
m
o
d
els
an
d
m
eth
o
d
s
o
f
a
ctiv
atio
n
f
u
n
ctio
n
s
.
W
e
p
r
o
p
o
s
ed
th
e
tr
an
s
f
o
r
m
er
with
ar
ch
itectu
r
al
m
o
d
if
icatio
n
as
th
e
b
asis
f
o
r
d
esig
n
i
n
g
th
e
m
o
d
el
in
a
b
s
tr
ac
tiv
e
d
o
cu
m
en
t
s
u
m
m
ar
izatio
n
,
wh
ich
was
ev
id
e
n
tly
ef
f
ec
tiv
e
in
im
p
r
o
v
in
g
r
esu
lt p
er
f
o
r
m
an
ce
.
T
h
e
b
est
m
o
d
el
MT
DT
G
6
p
er
f
o
r
m
s
th
at
was
m
ea
s
u
r
ed
u
s
in
g
th
e
R
OUGE
-
1
,
an
d
R
OUGE
-
2
h
as
o
b
tain
e
d
a
g
o
o
d
s
co
r
e
o
f
0
.
5
8
an
d
0
.
4
2
,
r
esp
ec
tiv
ely
,
wi
th
a
tr
ain
in
g
tim
e
o
f
1
1
4
3
8
s
ec
o
n
d
s
.
B
ased
o
n
wo
r
d
clo
u
d
s
f
r
o
m
all
d
o
cu
m
e
n
ts
with
th
e
m
o
s
t
d
is
cu
s
s
io
n
,
we
f
o
u
n
d
th
at
th
e
m
o
s
t
r
ep
o
r
ted
in
n
e
ws
r
elate
d
to
th
e
p
o
licy
a
n
d
r
eg
u
latio
n
f
r
o
m
th
e
g
o
v
er
n
m
en
t
o
n
p
u
b
lic
h
ea
lt
h
s
er
v
ices
p
r
io
r
itizatio
n
d
u
r
in
g
C
OVI
D
-
19
p
a
n
d
em
ic.
Sin
ce
th
e
r
esear
ch
o
f
C
OVI
D
-
1
9
n
ews
d
o
cu
m
en
t
ab
s
tr
ac
tiv
e
s
u
m
m
ar
ies
is
m
in
im
al,
m
an
y
r
esear
ch
o
p
p
o
r
tu
n
it
ies
ca
n
b
e
d
o
n
e
f
u
r
th
e
r
b
y
m
o
d
if
y
in
g
th
e
en
c
o
d
e
r
an
d
d
ec
o
d
e
r
lay
er
to
g
et
b
et
ter
m
o
d
el
q
u
ality
r
esu
lts
,
an
d
th
ey
also
wo
r
k
with
f
aster
tr
ain
in
g
tim
e.
T
h
e
in
teg
r
atio
n
o
f
s
ev
er
al
tr
a
n
s
f
o
r
m
er
ar
ch
itectu
r
e
m
o
d
els
ca
n
also
b
e
d
o
n
e,
s
u
ch
as
th
e
u
s
e
o
f
th
e
T
5
o
r
B
AR
T
m
o
d
els to
s
u
m
m
ar
ize
u
n
til th
e
q
u
ality
o
f
ex
is
tin
g
r
esear
ch
ca
n
b
e
ev
al
u
ated
o
r
co
m
p
a
r
ed
.
RE
F
E
R
E
NC
E
S
[1
]
A.
A.
S
a
li
su
a
n
d
X.
V.
Vo
,
“
P
re
d
ictin
g
sto
c
k
re
tu
r
n
s
in
t
h
e
p
re
se
n
c
e
o
f
COV
ID
-
1
9
p
a
n
d
e
m
ic:
Th
e
ro
le
o
f
h
e
a
lt
h
n
e
ws
,
”
In
t.
Rev
.
Fi
n
a
n
c
.
An
a
l.
,
v
o
l.
7
1
,
2
0
2
0
.
[2
]
A.
K.
M
.
N.
Isla
m
,
S
.
Laa
to
,
S
.
T
a
lu
k
d
e
r,
a
n
d
E.
S
u
ti
n
e
n
,
“
M
isin
f
o
rm
a
ti
o
n
sh
a
rin
g
a
n
d
so
c
ial
m
e
d
ia
fa
ti
g
u
e
d
u
rin
g
COV
ID
-
1
9
:
An
a
ff
o
rd
a
n
c
e
a
n
d
c
o
g
n
it
iv
e
lo
a
d
p
e
rsp
e
c
ti
v
e
,
”
T
e
c
h
n
o
l.
F
o
re
c
a
st.
S
o
c
.
C
h
a
n
g
e
,
v
o
l.
1
5
9
,
2
0
2
0
.
[3
]
A.
Kh
a
n
a
n
d
N.
S
a
li
m
,
“
A
re
v
iew
o
n
a
b
stra
c
ti
v
e
su
m
m
a
riza
ti
o
n
m
e
th
o
d
s,”
J
.
T
h
e
o
r.
Ap
p
l.
I
n
f.
T
e
c
h
n
o
l
.
,
v
o
l.
5
9
,
no
.
1
,
p
p
.
6
4
-
7
2
,
2
0
1
4
.
[4
]
M
.
M
a
rjan
i
,
e
t
a
l.
,
“
Bi
g
Io
T
Da
t
a
An
a
ly
ti
c
s:
Arc
h
it
e
c
tu
re
,
Op
p
o
r
tu
n
it
ies
,
a
n
d
Op
e
n
Re
se
a
rc
h
Ch
a
ll
e
n
g
e
s,”
IEE
E
Acc
e
ss
,
v
o
l.
5
,
p
p
.
5
2
4
7
-
5
2
6
1
,
2
0
1
7
.
[5
]
T.
Uç
k
a
n
a
n
d
A.
Ka
rc
ı,
“
Ex
trac
ti
v
e
m
u
lt
i
-
d
o
c
u
m
e
n
t
tex
t
s
u
m
m
a
riza
ti
o
n
b
a
se
d
o
n
g
ra
p
h
in
d
e
p
e
n
d
e
n
t
se
ts,”
Eg
y
p
t.
In
fo
rm
a
t
ics
J
.
,
v
o
l.
2
1
,
n
o
.
3
,
2
0
2
0
.
[6
]
S
.
G
u
p
t
a
,
e
t
a
l
.
,
“
A
b
s
t
r
a
c
t
i
v
e
s
u
m
m
a
r
i
z
a
t
i
o
n
:
A
n
o
v
e
r
v
i
e
w
o
f
t
h
e
s
t
a
t
e
o
f
t
h
e
a
r
t
,
”
E
x
p
e
r
t
S
y
s
t
.
A
p
p
l
.
,
v
o
l
.
1
2
1
,
p
p
.
4
9
-
6
5
,
2
0
1
9
.
[7
]
Y.
Hu
a
n
g
,
Z.
Yu
,
J.
G
u
o
,
Z
.
Y
u
,
a
n
d
Y.
Xia
n
,
“
Le
g
a
l
p
u
b
li
c
o
p
i
n
io
n
n
e
ws
a
b
stra
c
ti
v
e
su
m
m
a
riza
ti
o
n
b
y
in
c
o
rp
o
ra
ti
n
g
to
p
ic i
n
fo
rm
a
ti
o
n
,
”
I
n
t.
J
.
M
a
c
h
.
L
e
a
rn
.
Cy
b
e
rn
.
,
v
o
l
.
1
1
,
p
p
.
2
0
3
9
-
2
0
5
0
,
2
0
2
0
.
[8
]
C.
Yu
a
n
,
Z.
Ba
o
,
M
.
S
a
n
d
e
r
so
n
,
a
n
d
Y.
Tan
g
,
“
In
c
o
r
p
o
ra
ti
n
g
w
o
rd
a
tt
e
n
ti
o
n
with
c
o
n
v
o
lu
t
io
n
a
l
n
e
u
r
a
l
n
e
two
rk
s fo
r
a
b
stra
c
ti
v
e
su
m
m
a
riza
ti
o
n
,
”
W
o
rl
d
W
id
e
W
e
b
,
v
o
l
.
2
3
,
n
o
.
1
,
p
p
.
2
6
7
-
2
8
7
,
2
0
2
0
.
[9
]
A.
K.
M
o
h
a
m
m
a
d
M
a
su
m
,
e
t
a
l.
,
“
Ab
stra
c
ti
v
e
m
e
th
o
d
o
f
te
x
t
su
m
m
a
riza
ti
o
n
with
se
q
u
e
n
c
e
to
se
q
u
e
n
c
e
RNN
s,”
2
0
1
9
1
0
t
h
I
n
t.
C
o
n
f
.
Co
m
p
u
t
.
Co
mm
u
n
.
Ne
tw.
T
e
c
h
n
o
l.
ICCCN
T
2
0
1
9
,
2
0
1
9
,
p
p
.
1
-
5
.
[1
0
]
P
.
M
.
Ha
n
u
n
g
g
u
l
a
n
d
S
.
S
u
y
a
n
to
,
“
Th
e
Im
p
a
c
t
o
f
L
o
c
a
l
Atten
ti
o
n
in
LS
TM
fo
r
Ab
stra
c
ti
v
e
Tex
t
S
u
m
m
a
riza
ti
o
n
,
”
2
0
1
9
2
n
d
In
t.
S
e
min
.
Res
.
In
f.
T
e
c
h
n
o
l.
I
n
tell.
S
y
st.
I
S
RIT
I
2
0
1
9
,
p
p
.
5
4
-
5
7
,
2
0
1
9
[1
1
]
B.
M
y
a
g
m
a
r,
J.
Li
,
a
n
d
S
.
Kim
u
ra
,
“
Cro
ss
-
Do
m
a
in
S
e
n
ti
m
e
n
t
Clas
sifica
ti
o
n
wit
h
B
id
irec
ti
o
n
a
l
Co
n
tex
t
u
a
li
z
e
d
Tran
sfo
rm
e
r
Lan
g
u
a
g
e
M
o
d
e
ls,”
IEE
E
Acc
e
ss
,
v
o
l
.
7
,
p
p
.
1
6
3
2
1
9
-
1
6
3
2
3
0
,
2
0
1
9
.
[1
2
]
Y.
Ch
e
n
a
n
d
H.
Li
,
“
DA
M
:
Tran
sfo
rm
e
r
-
b
a
se
d
re
latio
n
d
e
tec
ti
o
n
fo
r
Qu
e
stio
n
A
n
sw
e
rin
g
o
v
e
r
K
n
o
wle
d
g
e
Ba
se
,
”
Kn
o
wled
g
e
-
Ba
se
d
S
y
st.
,
v
o
l.
2
0
1
-
2
0
2
,
2
0
2
0
.
[1
3
]
T.
A.
F
u
a
d
,
M
.
T
.
Na
y
e
e
m
,
A.
M
a
h
m
u
d
,
a
n
d
Y.
Ch
a
li
,
“
Ne
u
ra
l
se
n
ten
c
e
fu
si
o
n
fo
r
d
i
v
e
rsity
d
ri
v
e
n
a
b
stra
c
ti
v
e
m
u
lt
i
-
d
o
c
u
m
e
n
t
su
m
m
a
riza
ti
o
n
,
”
Co
mp
u
t
.
S
p
e
e
c
h
L
a
n
g
.
,
v
o
l
.
5
8
,
p
p
.
2
1
6
-
2
3
0
,
2
0
1
9
.
[1
4
]
A.
Va
sw
a
n
i
,
e
t
a
l.
,
“
Atten
ti
o
n
is all
y
o
u
n
e
e
d
,
”
A
d
v
.
Ne
u
ra
l
In
f
.
Pro
c
e
ss
.
S
y
st.
,
v
o
l.
2
0
1
7
,
p
p
.
5
9
9
9
-
6
0
0
9
,
2
0
1
7
.
[1
5
]
J.
Á.
G
o
n
z
á
lez
,
L.
F
.
Hu
r
tad
o
,
a
n
d
F
.
P
la,
“
Tran
sfo
rm
e
r
b
a
se
d
c
o
n
t
e
x
tu
a
li
z
a
ti
o
n
o
f
p
re
-
trai
n
e
d
w
o
rd
e
m
b
e
d
d
in
g
s fo
r
iro
n
y
d
e
tec
ti
o
n
i
n
Twi
tt
e
r,
”
In
f
.
P
ro
c
e
ss
.
M
a
n
a
g
.
,
v
o
l
.
5
7
,
n
o
.
4
,
2
0
2
0
.
[1
6
]
J.
W.
Li
n
,
Y.
C
.
G
a
o
,
a
n
d
R.
G
.
Ch
a
n
g
,
“
Ch
in
e
se
S
t
o
ry
G
e
n
e
ra
ti
o
n
with
F
a
stTex
t
Tra
n
sfo
rm
e
r
Ne
two
rk
,
”
1
st
In
t
.
Co
n
f.
Arti
f.
In
tell
.
In
f
.
Co
mm
u
n
.
I
CAIIC
2
0
1
9
,
p
p
.
3
9
5
-
3
9
8
,
2
0
1
9
.
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KOM
NI
KA
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l Co
n
tr
o
l
S
u
mma
r
iz
a
tio
n
o
f COVI
D
-
1
9
n
ews
d
o
cu
men
ts
d
ee
p
lea
r
n
in
g
-
b
a
s
ed
u
s
in
g
… (
N
u
r
Ha
ya
tin
)
761
[1
7
]
Y.
Iwa
sa
k
i,
A.
Ya
m
a
sh
it
a
,
Y.
Ko
n
n
o
,
a
n
d
K.
M
a
tsu
b
a
y
a
sh
i
,
“
Ja
p
a
n
e
se
a
b
stra
c
ti
v
e
tex
t
s
u
m
m
a
riza
ti
o
n
u
si
n
g
BERT
,
”
Pro
c
.
-
2
0
1
9
I
n
t.
C
o
n
f.
T
e
c
h
n
o
l
.
Ap
p
l
.
Arti
f
.
In
tel
l.
T
A
AI
2
0
1
9
,
2
0
1
9
.
[1
8
]
Ry
a
n
Ha
n
,
“
COV
ID
-
1
9
Ne
ws
Article
s
Op
e
n
Re
se
a
rc
h
Da
tas
e
t
|
Ka
g
g
le
,
”
2
0
2
0
.
[On
li
n
e
].
Av
a
il
a
b
le:
h
tt
p
s:/
/www
.
k
a
g
g
le.co
m
/ry
a
n
x
j
h
a
n
/cb
c
-
n
e
ws
-
c
o
ro
n
a
v
ir
u
s
-
a
rti
c
les
-
m
a
rc
h
-
2
6
(a
c
c
e
ss
e
d
Ju
l.
2
2
,
2
0
2
0
).
[1
9
]
L.
Ru
h
win
a
n
i
n
g
si
h
a
n
d
T.
Dja
tn
a
,
“
A
S
e
n
ti
m
e
n
t
Kn
o
wle
d
g
e
Dis
c
o
v
e
ry
M
o
d
e
l
in
Twit
ter’s
TV
Co
n
ten
t
Us
in
g
S
to
c
h
a
stic G
ra
d
ien
t
De
sc
e
n
t
Al
g
o
rit
h
m
,
”
T
E
L
KOM
NIKA
T
e
lec
o
m
mu
n
ica
t
io
n
C
o
mp
u
ti
n
g
E
lec
trica
l
El
e
c
tro
n
ics
a
n
d
Co
n
tro
l
,
v
o
l.
1
4
,
n
o
.
3
,
p
p
.
1
0
6
7
-
1
0
7
6
,
2
0
1
6
.
[2
0
]
M
.
A.
F
a
u
z
i,
R.
F
.
N.
F
irma
n
sy
a
h
,
a
n
d
T.
Afirian
t
o
,
“
Im
p
r
o
v
i
n
g
se
n
ti
m
e
n
t
a
n
a
ly
sis
o
f
s
h
o
rt
i
n
fo
r
m
a
l
In
d
o
n
e
sia
n
p
ro
d
u
c
t
re
v
iew
s
u
sin
g
sy
n
o
n
y
m
b
a
se
d
fe
a
tu
re
e
x
p
a
n
sio
n
,
”
T
E
L
KOM
NIKA
T
e
lec
o
mm
u
n
ic
a
ti
o
n
Co
m
p
u
ti
n
g
El
e
c
trica
l
El
e
c
tro
n
ics
a
n
d
C
o
n
tro
l
,
v
o
l.
1
6
,
n
o
.
3
,
p
p
.
1
3
4
5
-
1
3
5
0
,
2
0
1
8
.
[2
1
]
J.
P
e
n
n
i
n
g
t
o
n
,
R.
S
o
c
h
e
r,
a
n
d
C.
D.
M
a
n
n
i
n
g
,
“
G
lo
Ve
:
G
lo
b
a
l
v
e
c
to
rs
fo
r
wo
rd
re
p
re
se
n
t
a
ti
o
n
,
”
2
0
1
4
.
[O
n
li
n
e
].
Av
a
il
a
b
le:
h
tt
p
s://
n
l
p
.
sta
n
f
o
rd
.
e
d
u
/p
r
o
jec
ts/g
l
o
v
e
/
.
[2
2
]
N.
Ala
m
i,
M
.
M
e
k
n
a
ss
i,
a
n
d
N.
En
-
n
a
h
n
a
h
i
,
“
En
h
a
n
c
i
n
g
u
n
s
u
p
e
r
v
ise
d
n
e
u
ra
l
n
e
two
rk
s
b
a
se
d
tex
t
su
m
m
a
riza
ti
o
n
with
wo
r
d
e
m
b
e
d
d
in
g
a
n
d
e
n
se
m
b
le l
e
a
rn
in
g
,
”
Exp
e
rt
S
y
st.
A
p
p
l
.
,
v
o
l
.
1
2
3
,
p
p
.
1
9
5
-
2
1
1
,
2
0
1
9
.
[2
3
]
D.
He
n
d
r
y
c
k
s
a
n
d
K.
G
imp
e
l,
“
G
a
u
ss
ian
Err
o
r
Li
n
e
a
r
Un
i
ts
(G
EL
Us
),
”
Co
rn
e
ll
Un
ive
rs
it
y
,
p
p
.
1
-
9
,
2
0
1
6
.
[O
n
li
n
e
].
Av
a
il
a
b
le:
h
tt
p
:/
/arx
i
v
.
o
r
g
/ab
s/
1
6
0
6
.
0
8
4
1
5
.
[2
4
]
B.
D.
S
a
to
to
,
I.
Ut
o
y
o
,
R.
Ru
l
a
n
in
g
ty
a
s,
a
n
d
E.
B.
K
h
o
e
n
d
o
ri
,
“
An
imp
ro
v
e
m
e
n
t
o
f
G
ra
m
-
n
e
g
a
ti
v
e
b
a
c
teri
a
id
e
n
ti
fica
ti
o
n
u
sin
g
c
o
n
v
o
l
u
ti
o
n
a
l
n
e
u
ra
l
n
e
two
r
k
with
fi
n
e
tu
n
in
g
,
”
T
EL
KOM
NIKA
T
e
lec
o
mm
u
n
ica
ti
o
n
Co
m
p
u
t
in
g
El
e
c
trica
l
El
e
c
tro
n
ics
a
n
d
Co
n
tro
l
,
v
o
l.
1
8
,
n
o
.
3
,
p
p
.
1
3
9
7
-
1
4
0
5
,
2
0
2
0
.
[2
5
]
E.
G
ib
so
n
e
t
a
l.
,
“
Nifty
Ne
t:
a
d
e
e
p
-
lea
rn
in
g
p
latfo
rm
fo
r
m
e
d
ica
l
ima
g
in
g
,
”
Co
m
p
u
t
.
M
e
th
o
d
s
Pro
g
ra
ms
Bi
o
me
d
.
,
v
o
l.
1
5
8
,
p
p
.
1
1
3
-
1
2
2
,
2
0
1
8
.
[2
6
]
I.
Nu
r
h
a
id
a
,
V.
A
y
u
m
i,
D.
F
it
ria
n
a
h
,
R.
A.
M
.
Ze
n
,
H
.
No
p
risso
n
,
a
n
d
H.
Wei,
“
Im
p
lem
e
n
tati
o
n
o
f
d
e
e
p
n
e
u
ra
l
n
e
two
rk
s (DNN
)
wi
th
b
a
tch
n
o
rm
a
li
z
a
ti
o
n
f
o
r
b
a
ti
k
p
a
tt
e
rn
re
c
o
g
n
i
ti
o
n
,
”
In
t
.
J
.
El
e
c
tr.
Co
m
p
u
t.
E
n
g
.
,
v
o
l.
1
0
,
n
o
.
2
,
p
p
.
2
0
4
5
-
2
0
5
3
,
2
0
2
0
,
d
o
i:
1
0
.
1
1
5
9
1
/i
jec
e
.
v
1
0
i2
.
p
p
2
0
4
5
-
2
0
5
3
.
[2
7
]
C.
Y.
Li
n
,
“
Ro
u
g
e
:
A p
a
c
k
a
g
e
fo
r
a
u
to
m
a
ti
c
e
v
a
lu
a
ti
o
n
o
f
su
m
m
a
ri
e
s,”
Pro
c
.
W
o
rk
.
tex
t
S
u
mm
.
b
ra
n
c
h
e
s
o
u
t
(W
AS
2
0
0
4
)
,
n
o
.
1
,
2
0
0
4
,
p
p
.
2
5
-
26
.
[2
8
]
S
.
A.
Alsa
id
i,
A.
T.
S
a
d
iq
,
a
n
d
H.
S
.
Ab
d
u
ll
a
h
,
“
En
g
l
ish
p
o
e
m
s
c
a
te
g
o
riza
ti
o
n
u
si
n
g
te
x
t
m
in
i
n
g
a
n
d
r
o
u
g
h
se
t
th
e
o
r
y
,
”
Bu
ll
.
El
e
c
tr.
En
g
.
I
n
f
o
rm
a
ti
c
s
,
v
o
l
.
9
,
n
o
.
4
,
p
p
.
1
7
0
1
-
1
7
1
0
,
2
0
2
0
.
[2
9
]
D.
P
.
Kin
g
m
a
a
n
d
J.
L
.
Ba
,
“
Ad
a
m
:
A
m
e
th
o
d
f
o
r
st
o
c
h
a
stic
o
p
ti
m
iza
ti
o
n
,
”
3
r
d
I
n
t.
Co
n
f.
L
e
a
rn
.
Rep
re
se
n
t.
ICL
R
2
0
1
5
-
Co
n
f.
T
r
a
c
k
Pro
c
.
,
2
0
1
5
,
p
p
.
1
-
15
.
B
I
O
G
RAP
H
I
E
S O
F
A
UT
H
O
RS
Nur
H
a
y
a
tin
is
a
l
e
c
tu
re
r
a
t
th
e
Un
i
v
e
rsit
y
o
f
M
u
h
a
m
m
a
d
iy
a
h
M
a
la
n
g
.
S
h
e
re
c
e
iv
e
d
h
e
r
M
a
ste
r
in
I
n
fo
rm
a
ti
c
s
En
g
in
e
e
ri
n
g
fr
o
m
th
e
I
n
s
ti
tu
te
o
f
Tec
h
n
o
lo
g
y
S
e
p
u
l
u
h
N
o
p
e
m
b
e
r
S
u
ra
b
a
y
a
,
In
d
o
n
e
sia
,
with
a
n
a
re
a
o
f
in
tere
st
i
n
d
a
ta
sc
ien
c
e
,
wh
e
re
sh
e
tea
c
h
e
s
c
o
u
rse
s
re
late
d
to
Na
tu
ra
l
Lan
g
u
a
g
e
P
ro
c
e
ss
in
g
.
He
r
m
a
in
r
e
se
a
rc
h
in
tere
st
is
tex
t
a
n
a
l
y
ti
c
s,
so
c
ial
m
e
d
ia
a
n
a
ly
ti
c
s,
d
a
ta
m
in
in
g
,
a
n
d
i
n
fo
rm
a
ti
o
n
re
tri
e
v
a
l.
Ema
il
:
n
o
o
r
h
a
y
a
ti
n
@
u
m
m
.
a
c
.
id
K
h
a
r
ism
a
Mu
z
a
k
i
G
h
u
fr
o
n
is
c
u
rre
n
tl
y
c
o
m
p
leti
n
g
a
Ba
c
h
e
lo
r
's
d
e
g
re
e
fro
m
t
h
e
In
f
o
rm
a
ti
c
s
De
p
a
rtme
n
t,
F
a
c
u
lt
y
o
f
En
g
in
e
e
r
in
g
,
a
t
t
h
e
Un
iv
e
rsit
y
o
f
M
u
h
a
m
m
a
d
iy
a
h
M
a
lan
g
,
In
d
o
n
e
sia
.
His
in
tere
sts
in
c
lu
d
e
n
a
tu
ra
l
lan
g
u
a
g
e
p
r
o
c
e
ss
in
g
a
n
d
d
e
e
p
le
a
rn
in
g
a
rc
h
it
e
c
tu
re
.
Ema
il
:
k
h
a
rism
a
.
m
u
z
a
k
i@we
b
m
a
il
.
u
m
m
.
a
c
.
id
G
a
li
h
Wa
sis
W
ica
k
so
n
o
is
a
l
e
c
tu
re
r
in
th
e
I
n
fo
rm
a
ti
c
s
d
e
p
a
r
tme
n
t
a
t
th
e
Un
i
v
e
rsity
o
f
M
u
h
a
m
m
a
d
iy
a
h
M
a
lan
g
,
I
n
d
o
n
e
sia
.
He
tea
c
h
e
s
lo
g
ic
&
c
o
m
p
u
ti
n
g
a
n
d
c
o
m
p
u
ter
re
a
so
n
i
n
g
c
o
u
rse
s
with
a
n
a
re
a
o
f
in
tere
st
in
d
a
ta
sc
i
e
n
c
e
.
Th
e
fo
c
u
s
o
f
h
is
re
se
a
rc
h
is
c
a
se
-
b
a
se
d
re
a
so
n
in
g
a
n
d
o
n
li
n
e
lea
rn
in
g
.
Ema
il
:
g
a
li
h
.
w.w@u
m
m
.
a
c
.
id
Evaluation Warning : The document was created with Spire.PDF for Python.