I
nd
o
ne
s
ia
n J
o
urna
l o
f
E
lect
rica
l En
g
ineering
a
nd
Co
m
pu
t
er
Science
Vo
l.
43
,
No
.
2
,
A
u
g
u
s
t
2
0
2
6
,
p
p
.
5
2
2
~
5
3
3
I
SS
N:
2
5
0
2
-
4
7
5
2
,
DOI
: 1
0
.
1
1
5
9
1
/ijeecs.v
43
.i
2
.
pp
522
-
5
3
3
522
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ee
cs.ia
esco
r
e.
co
m
Tra
nsfo
rmer
-
ba
s
ed sentim
ent
mo
deling
f
o
r ide
ntif
y
i
ng
cros
s
-
co
untry fin
tech
p
ercept
io
n
ga
p
s
K
a
y
la
Z
ha
f
ira
Ardino
v
,
M
uh
a
rdi Sa
pu
t
ra
,
Ris
k
a
Ya
nu
F
a
’
rif
a
h
I
n
f
o
r
mat
i
o
n
S
y
st
e
m St
u
d
y
P
r
o
g
r
a
m,
S
c
h
o
o
l
o
f
I
n
d
u
st
r
i
a
l
E
n
g
i
n
e
e
r
i
n
g
,
Te
l
k
o
m
U
n
i
v
e
r
s
i
t
y
,
B
a
n
d
u
n
g
,
I
n
d
o
n
e
s
i
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Feb
1
6
,
2
0
2
6
R
ev
is
ed
J
u
n
1
2
,
2
0
2
6
Acc
ep
ted
J
u
n
2
7
,
2
0
2
6
Co
n
v
e
n
ti
o
n
a
l
se
n
ti
m
e
n
t
a
n
a
ly
sis
l
a
c
k
s
g
ra
n
u
larity
fo
r
c
a
p
t
u
rin
g
d
e
t
a
il
e
d
u
se
r
e
x
p
e
rien
c
e
s
a
n
d
c
ro
ss
-
c
o
u
n
tr
y
c
o
m
p
a
ra
ti
v
e
i
n
sig
h
ts
i
n
d
i
g
it
a
l
fin
a
n
c
e
.
Th
is
stu
d
y
i
d
e
n
ti
fies
a
n
d
m
a
p
s
p
e
rc
e
p
ti
o
n
g
a
p
s
a
m
o
n
g
S
h
o
p
e
e
P
a
y
u
se
rs
in
In
d
o
n
e
sia
a
n
d
Th
a
il
a
n
d
u
sin
g
a
to
p
ic
-
i
n
fo
rm
e
d
se
n
ti
m
e
n
t
a
n
a
l
y
si
s
p
ip
e
li
n
e
in
sp
ired
b
y
a
sp
e
c
t
-
b
a
se
d
se
n
ti
m
e
n
t
a
n
a
ly
sis
(ABSA)
p
ri
n
c
ip
les
.
Ad
o
p
ti
n
g
th
e
k
n
o
wle
d
g
e
d
isc
o
v
e
r
y
in
d
a
tab
a
se
s
(KD
D) fram
e
wo
rk
,
o
v
e
r
1
7
0
,
0
0
0
we
b
-
sc
ra
p
e
d
re
v
iew
s
we
re
p
re
p
ro
c
e
ss
e
d
,
a
u
to
m
a
ti
c
a
ll
y
la
b
e
led
th
r
o
u
g
h
p
se
u
d
o
-
l
a
b
e
li
n
g
,
a
n
d
b
a
lan
c
e
d
u
sin
g
ra
n
d
o
m
o
v
e
rsa
m
p
li
n
g
.
To
m
it
i
g
a
te p
s
e
u
d
o
-
la
b
e
l
re
in
fo
rc
e
m
e
n
t,
m
a
n
u
a
l
v
a
li
d
a
ti
o
n
o
n
5
0
0
re
v
iew
s
p
e
r
c
o
u
n
try
a
c
h
iev
e
d
a
g
re
e
m
e
n
t
ra
tes
o
f
9
8
.
8
0
%
(In
d
o
n
e
sia
)
a
n
d
9
9
.
2
0
%
(Th
a
il
a
n
d
)
wit
h
Co
h
e
n
’
s
Ka
p
p
a
a
b
o
v
e
0
.
9
7
.
T
h
e
fin
e
-
t
u
n
e
d
Distil
B
ERT
m
o
d
e
l
a
c
h
iev
e
d
a
c
c
u
ra
c
ies
o
f
9
7
.
6
5
%
(
In
d
o
n
e
sia
)
a
n
d
9
8
.
3
8
%
(Th
a
il
a
n
d
)
,
th
o
u
g
h
th
e
se
f
ig
u
re
s
sh
o
u
l
d
b
e
in
terp
re
ted
wit
h
in
t
h
e
p
se
u
d
o
-
lab
e
led
e
v
a
l
u
a
ti
o
n
c
o
n
te
x
t.
S
ig
n
ifi
c
a
n
t
p
e
rc
e
p
ti
o
n
g
a
p
s
we
re
re
v
e
a
led
:
Th
a
i
u
se
rs
sh
o
we
d
l
o
we
r
sa
ti
sfa
c
ti
o
n
wi
th
t
ra
n
sa
c
ti
o
n
s
(6
7
.
6
%
n
e
g
a
ti
v
e
)
wh
il
e
In
d
o
n
e
sia
n
u
se
rs
we
re
m
o
re
p
o
siti
v
e
(9
3
.
0
%
p
o
siti
v
e
).
P
ro
c
e
ss
tim
e
re
c
e
iv
e
d
n
e
g
a
ti
v
e
d
o
m
i
n
a
n
c
e
in
b
o
t
h
c
o
u
n
tr
ies
,
with
In
d
o
n
e
sia
a
t
7
8
.
2
%
a
n
d
T
h
a
il
a
n
d
a
t
5
0
.
8
%
n
e
g
a
ti
v
e
.
T
h
e
se
fin
d
i
n
g
s
d
e
m
o
n
stra
te
th
a
t
u
se
r
sa
ti
sfa
c
ti
o
n
is
sh
a
p
e
d
b
y
l
o
c
a
l
in
fr
a
stru
c
tu
re
a
n
d
c
u
lt
u
ra
l
c
o
n
tex
ts,
p
r
o
v
i
d
in
g
stra
teg
ic
i
n
sig
h
ts
fo
r
re
g
io
n
a
l
fin
tec
h
d
e
v
e
lo
p
m
e
n
t.
K
ey
w
o
r
d
s
:
Asp
ec
t
-
b
ased
s
en
tim
en
t
a
n
aly
s
is
C
r
o
s
s
-
co
u
n
tr
y
co
m
p
ar
is
o
n
to
p
ic
m
o
d
elin
g
Dis
til
B
E
R
T
Fin
tech
Per
ce
p
tio
n
g
ap
Sh
o
p
ee
Pay
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
:
Mu
h
ar
d
i Sap
u
t
r
a
I
n
f
o
r
m
atio
n
Sy
s
tem
Stu
d
y
Pr
o
g
r
am
,
Sch
o
o
l o
f
I
n
d
u
s
tr
ial
E
n
g
in
ee
r
in
g
,
T
elk
o
m
Un
iv
er
s
ity
B
an
d
u
n
g
,
I
n
d
o
n
esia
E
m
ail:
m
u
h
ar
d
i@
telk
o
m
u
n
iv
e
r
s
ity
.
ac
.
id
1.
I
NT
RO
D
UCT
I
O
N
T
h
e
i
n
t
e
g
r
a
t
i
o
n
o
f
i
n
t
e
r
n
e
t
t
e
c
h
n
o
l
o
g
y
h
a
s
t
r
a
n
s
f
o
r
m
e
d
f
i
n
a
n
c
i
a
l
s
e
r
v
i
c
e
s
w
i
t
h
i
n
t
h
e
d
i
g
i
t
a
l
e
c
o
n
o
m
y
[
1
]
,
en
ter
in
g
th
e
f
in
an
cial
tec
h
n
o
l
o
g
y
(
f
in
tech
)
4
.
0
p
h
ase
with
a
f
o
cu
s
o
n
f
u
ll
ass
et
d
i
g
itizatio
n
an
d
p
ay
m
en
t
in
f
r
astru
ctu
r
e
s
tr
en
g
th
e
n
in
g
[
2
]
.
T
h
ese
in
n
o
v
atio
n
s
s
p
an
elec
tr
o
n
ic
p
ay
m
e
n
t
s
y
s
tem
s
to
d
ig
ital
wallets
(e
-
wallets)
[
3
]
,
s
u
p
p
o
r
ted
b
y
r
esp
o
n
s
iv
e
r
e
g
u
lato
r
y
p
o
licies
an
d
in
c
r
ea
s
ed
f
in
an
cial
liter
ac
y
[
4
]
.
I
n
So
u
th
ea
s
t
Asi
a,
I
n
d
o
n
esia
an
d
T
h
ailan
d
lead
f
in
tech
in
v
estme
n
t
g
r
o
wth
[
5
]
,
d
r
iv
in
g
m
o
r
e
eq
u
itab
le
f
in
an
cial
in
clu
s
io
n
ac
r
o
s
s
ASEA
N
[
6
]
th
r
o
u
g
h
f
in
tech
-
b
an
k
in
g
c
o
llab
o
r
atio
n
[
7
]
.
E
-
wallets
h
av
e
b
ec
o
m
e
e
s
s
en
tial
to
o
ls
with
8
0
%
ad
o
p
tio
n
r
ates
in
b
o
th
m
ar
k
ets
[
8
]
–
[
1
0
]
,
an
d
Sh
o
p
e
ePay
h
as
em
er
g
ed
as
th
e
m
ar
k
et
lead
er
in
e
-
co
m
m
er
ce
tr
an
s
ac
tio
n
s
[
1
1
]
.
Desp
ite
th
is
lead
er
s
h
ip
,
Sh
o
p
ee
Pay
f
ac
es
p
er
s
is
ten
t
u
s
er
s
atis
f
ac
tio
n
ch
allen
g
e
s
[
1
1
]
,
in
cl
u
d
in
g
ac
c
o
u
n
t
v
er
i
f
icatio
n
co
n
s
tr
ain
ts
an
d
cy
b
er
s
ec
u
r
ity
th
r
ea
ts
[
1
2
]
.
T
h
e
p
r
es
s
u
r
e
o
f
in
f
o
r
m
atio
n
tech
n
o
lo
g
y
u
s
e
o
f
ten
tr
ig
g
er
s
tech
n
o
s
tr
ess
,
h
in
d
er
in
g
s
er
v
ic
e
ad
o
p
tio
n
[
1
3
]
,
[
1
4
]
a
n
d
m
a
n
if
esti
n
g
as
n
eg
ativ
e
r
ev
iews
co
n
tain
in
g
tech
n
ical
c
o
m
p
lain
ts
an
d
p
r
iv
ac
y
co
n
ce
r
n
s
[
1
5
]
.
Asp
ec
t
-
b
ased
s
en
tim
e
n
t
an
aly
s
is
(
AB
SA)
o
f
f
er
s
a
r
ele
v
an
t
f
r
a
m
ewo
r
k
f
o
r
u
n
d
er
s
tan
d
in
g
th
ese
p
er
ce
p
tio
n
s
b
y
m
ea
s
u
r
i
n
g
s
en
tim
en
t
o
n
s
p
ec
if
ic
r
e
v
iew
asp
ec
ts
[
1
6
]
,
[
1
7
]
.
Un
lik
e
c
o
n
v
en
tio
n
al
s
en
tim
en
t
an
aly
s
is
,
wh
ich
ass
ig
n
s
a
s
in
g
le
p
o
lar
ity
to
en
tire
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4
7
5
2
Tr
a
n
s
fo
r
mer
-
b
a
s
ed
s
en
timen
t
mo
d
elin
g
fo
r
id
en
tifyin
g
cro
s
s
-
co
u
n
tr
y
fin
tech
…
(
K
a
yla
Zh
a
fir
a
A
r
d
in
o
v)
523
d
o
cu
m
e
n
ts
[
1
8
]
,
[
1
9
]
,
AB
SA
ev
alu
ates
s
en
tim
en
t
b
ased
o
n
asp
ec
t
ca
teg
o
r
ies
an
d
o
p
in
i
o
n
ter
m
s
[
1
6
]
,
[
2
0
]
,
with
ad
v
an
ce
s
d
r
iv
en
b
y
tr
a
n
s
f
o
r
m
er
-
b
ased
m
o
d
els
an
d
to
p
ic
m
o
d
elin
g
s
u
ch
as
L
aten
t
Dir
ich
let
Allo
ca
tio
n
(
L
DA
)
[
2
1
]
–
[
2
3
]
.
Ho
wev
er
,
ca
n
o
n
ical
AB
SA
as
d
ef
in
ed
in
s
em
an
tic
ev
alu
atio
n
(
Sem
E
v
al
)
s
h
ar
ed
task
s
in
v
o
lv
es
jo
in
t
ex
tr
ac
tio
n
o
f
asp
ec
t
ter
m
s
an
d
p
o
lar
ities
at
th
e
s
en
ten
ce
le
v
el.
Ma
n
y
a
p
p
lied
s
tu
d
ies
in
s
tead
ad
o
p
t
a
r
elax
e
d
o
p
er
atio
n
aliza
tio
n
wh
er
e
L
D
A
-
d
er
iv
ed
to
p
ics
s
er
v
e
as
p
r
o
x
y
asp
ec
ts
an
d
s
en
tim
en
t
i
s
class
if
ied
at
th
e
d
o
cu
m
e
n
t
lev
el
[
2
4
]
,
[
2
5
]
.
T
h
is
s
tu
d
y
f
o
llo
ws
th
e
latter
ap
p
r
o
ac
h
.
W
h
ile
we
u
s
e
th
e
ter
m
“
AB
S
A
”
to
s
i
tu
ate
o
u
r
wo
r
k
with
in
th
e
b
r
o
ad
e
r
li
ter
atu
r
e,
we
ex
p
licitly
ac
k
n
o
w
led
g
e
th
at
o
u
r
im
p
lem
en
tatio
n
d
o
es
n
o
t
p
er
f
o
r
m
f
in
e
-
g
r
ain
e
d
asp
ec
t
-
o
p
i
n
io
n
p
air
ex
tr
ac
tio
n
;
in
s
tead
,
L
DA
-
d
er
iv
ed
to
p
ics
f
u
n
ctio
n
as
s
e
r
v
ice
-
lev
el
asp
ec
ts
with
s
en
tim
en
t
class
if
icatio
n
ap
p
lied
p
er
r
ev
iew
with
in
ea
ch
to
p
ic
c
lu
s
ter
.
Mo
s
t
AB
SA
s
tu
d
ies
r
em
ain
r
estricte
d
to
s
in
g
le
-
co
u
n
tr
y
o
r
s
in
g
le
-
lan
g
u
ag
e
d
atasets
,
o
v
er
lo
o
k
in
g
r
eg
io
n
al
co
m
p
a
r
ativ
e
n
u
an
ce
s
[
2
6
]
.
C
r
o
s
s
-
co
u
n
tr
y
AB
SA
r
esear
ch
in
f
in
tech
is
p
ar
ticu
lar
l
y
s
ca
r
ce
,
leav
in
g
a
s
ig
n
if
ica
n
t
g
a
p
:
ex
is
tin
g
s
tu
d
ies
ca
n
n
o
t
ex
p
lain
wh
y
t
h
e
s
am
e
d
ig
ita
l
s
er
v
ice
r
ec
ei
v
es
f
u
n
d
am
en
tally
d
if
f
er
e
n
t
p
er
ce
p
tio
n
s
ac
r
o
s
s
m
ar
k
ets.
T
h
is
g
ap
is
cr
itical
i
n
So
u
th
ea
s
t
Asi
a,
wh
er
e
r
ap
id
f
in
tech
ad
o
p
tio
n
co
ex
is
ts
with
d
iv
er
s
e
r
eg
u
lato
r
y
en
v
ir
o
n
m
en
ts
,
in
f
r
astru
ct
u
r
e
m
atu
r
ity
lev
els,
an
d
s
o
ci
o
-
cu
lt
u
r
al
n
o
r
m
s
.
I
n
d
o
n
esia
a
n
d
T
h
ailan
d
wer
e
s
elec
ted
f
o
r
t
h
r
ee
r
ea
s
o
n
s
:
(
1
)
b
o
th
r
ep
r
esen
t
th
e
lar
g
est
e
-
wallet
m
ar
k
ets
in
So
u
th
ea
s
t
Asi
a
b
y
tr
a
n
s
ac
tio
n
v
o
lu
m
e
[
5
]
,
[
9
]
,
[
1
0
]
;
(
2
)
th
ey
o
p
er
ate
u
n
d
er
d
if
f
e
r
en
t
r
e
g
u
lato
r
y
f
r
am
ew
o
r
k
s
I
n
d
o
n
esia
’
s
B
an
k
I
n
d
o
n
e
s
ia
licen
s
in
g
v
er
s
u
s
T
h
ailan
d
’
s
B
an
k
o
f
T
h
ailan
d
s
an
d
b
o
x
m
o
d
el;
a
n
d
(
3
)
p
r
io
r
liter
atu
r
e
h
as
f
o
c
u
s
ed
o
n
ea
ch
m
ar
k
et
in
d
ep
en
d
en
tly
[
1
2
]
,
[
1
3
]
with
o
u
t sy
s
tem
atic
cr
o
s
s
-
co
u
n
tr
y
as
p
ec
t
-
lev
el
co
m
p
ar
is
o
n
.
T
h
is
s
tu
d
y
ad
d
r
ess
es th
is
g
ap
th
r
o
u
g
h
a
u
n
if
ie
d
to
p
ic
-
in
f
o
r
m
ed
s
en
tim
en
t p
ip
elin
e
in
s
p
ir
ed
b
y
AB
SA
p
r
in
cip
les,
co
m
b
in
in
g
L
DA
f
o
r
au
to
m
atic
to
p
ic
ex
tr
ac
tio
n
as
p
r
o
x
y
s
er
v
ice
asp
ec
ts
[
2
7
]
with
a
f
in
e
-
tu
n
ed
Dis
til
B
E
R
T
m
o
d
el
f
o
r
co
m
p
u
tatio
n
al
ef
f
icien
c
y
[
2
8
]
,
[
2
9
]
.
T
h
is
k
n
o
wled
g
e
d
is
till
atio
n
m
ec
h
an
is
m
ca
p
tu
r
es
th
e
u
n
iq
u
e
c
h
ar
ac
ter
is
tics
o
f
e
-
wallet
r
ev
iews
[
2
4
]
,
[
3
0
]
.
T
h
e
f
r
am
ewo
r
k
’
s
im
p
o
r
tan
ce
lies
in
u
n
co
v
er
in
g
g
r
an
u
lar
p
er
ce
p
tio
n
g
ap
s
t
h
at
g
en
er
ic
s
en
tim
en
t
m
etr
ics
o
v
er
l
o
o
k
,
p
r
o
v
id
i
n
g
a
s
tr
ateg
ic
b
lu
e
p
r
in
t
f
o
r
n
av
ig
atin
g
r
eg
io
n
al
s
o
ci
o
-
cu
ltu
r
al
co
m
p
lex
ities
in
f
in
tec
h
.
T
h
e
m
ai
n
co
n
tr
i
b
u
tio
n
s
ar
e
f
o
u
r
f
o
ld
:
(
i
)
a
s
tr
u
ctu
r
ed
cr
o
s
s
-
co
u
n
tr
y
to
p
ic
-
in
f
o
r
m
e
d
s
en
tim
en
t
p
ip
elin
e
with
ex
p
licit
d
is
tin
ctio
n
f
r
o
m
ca
n
o
n
ical
AB
SA;
(
ii
)
au
to
m
atic
to
p
ic
ex
tr
ac
tio
n
u
s
in
g
L
DA
o
p
tim
ized
with
co
h
e
r
en
ce
s
co
r
es
as
p
r
o
x
y
s
er
v
ice
asp
ec
ts
;
(
iii
)
s
en
tim
en
t
class
if
icatio
n
with
p
s
eu
d
o
-
lab
el
r
eliab
ilit
y
v
al
id
ated
th
r
o
u
g
h
m
an
u
al
an
n
o
tat
io
n
u
s
in
g
C
o
h
en
’
s
Kap
p
a
(
κ
>
0
.
9
7
)
;
an
d
(
iv
)
p
er
ce
p
tio
n
g
ap
q
u
an
tific
atio
n
s
u
p
p
o
r
ted
b
y
a
d
o
cu
m
en
ted
cr
o
s
s
-
co
u
n
tr
y
to
p
ic
al
ig
n
m
en
t
p
r
o
to
c
o
l
f
o
r
s
tr
ate
g
ic
in
s
ig
h
ts
in
So
u
th
ea
s
t
Asi
an
f
in
tech
.
Sectio
n
2
d
escr
ib
es
th
e
m
eth
o
d
o
lo
g
y
;
s
ec
tio
n
3
p
r
esen
ts
r
esu
lts
an
d
d
is
cu
s
s
io
n
;
s
ec
tio
n
4
p
r
esen
ts
co
n
clu
s
io
n
s
.
2.
M
E
T
H
O
D
2
.
1
.
Resea
rc
h f
ra
m
ewo
r
k
o
v
er
v
iew
T
h
is
s
tu
d
y
ad
o
p
ts
th
e
k
n
o
wled
g
e
d
is
co
v
er
y
i
n
d
atab
a
s
es
(
KDD)
f
r
am
ewo
r
k
,
en
c
o
m
p
ass
in
g
s
y
s
tem
atic
s
tag
es
f
r
o
m
d
ata
a
cq
u
is
itio
n
to
k
n
o
wled
g
e
in
ter
p
r
etatio
n
.
T
h
e
KDD
ap
p
r
o
ac
h
en
s
u
r
es
s
tr
u
ctu
r
ed
in
f
o
r
m
atio
n
ex
tr
ac
tio
n
th
r
o
u
g
h
s
elec
tio
n
,
p
r
e
p
r
o
ce
s
s
in
g
,
tr
a
n
s
f
o
r
m
atio
n
,
d
ata
m
i
n
in
g
,
an
d
ev
alu
atio
n
s
tag
es
[
3
1
]
.
T
h
e
f
r
a
m
ewo
r
k
is
iter
ati
v
e,
allo
win
g
r
esear
ch
e
r
s
to
r
et
u
r
n
to
p
r
ev
io
u
s
s
tag
es
if
ev
alu
atio
n
r
esu
lts
d
o
n
o
t
m
ee
t q
u
ality
s
tan
d
ar
d
s
.
T
h
e
en
tire
r
esear
ch
f
lo
w
is
s
h
o
wn
in
Fig
u
r
e
1
.
Fig
u
r
e
1
.
T
h
e
p
r
o
p
o
s
ed
r
esear
ch
f
r
am
ewo
r
k
f
o
r
cr
o
s
s
-
co
u
n
tr
y
s
en
tim
en
t a
n
aly
s
is
o
f
f
in
tec
h
s
er
v
ices u
s
in
g
KDD
an
d
L
DA
m
eth
o
d
o
lo
g
y
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
5
2
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
43
,
No
.
2
,
Au
g
u
s
t
20
26
:
522
-
5
3
3
524
2
.
2
.
Da
t
a
c
o
llect
io
n
T
h
e
d
ata
co
llectio
n
p
r
o
ce
s
s
u
s
ed
web
s
cr
a
p
in
g
in
Py
th
o
n
with
th
e
Go
o
g
le
Play
Scr
ap
e
r
lib
r
ar
y
to
ex
tr
ac
t
d
ata
f
r
o
m
Sh
o
p
ee
Pay
a
p
p
licatio
n
s
in
I
n
d
o
n
esia
an
d
T
h
ailan
d
.
T
h
e
d
ata
co
llected
o
n
Octo
b
er
2
0
,
2
0
2
5
,
y
ield
ed
a
m
ass
iv
e
v
o
lu
m
e
o
f
r
ev
iews,
r
an
g
in
g
f
r
o
m
8
0
,
0
0
0
to
9
0
,
0
0
0
p
er
c
o
u
n
t
r
y
.
T
h
e
d
ata
attr
ib
u
tes
co
llected
in
clu
d
ed
r
ev
iewI
d
,
u
s
er
Nam
e,
co
n
ten
t,
s
co
r
e,
an
d
r
ev
iew
d
ate
(
at)
.
H
o
wev
er
,
th
i
s
s
tu
d
y
f
o
cu
s
ed
its
an
aly
s
is
o
n
th
e
co
n
ten
t
attr
ib
u
te
d
u
e
to
p
o
ten
tial
b
ias
in
n
u
m
er
ical
r
atin
g
s
,
wh
ich
o
f
ten
d
o
n
o
t
alig
n
with
u
s
er
s
’
tex
tu
al
ex
p
r
ess
io
n
s
[
3
2
]
.
All
r
aw
r
ev
iews
wer
e
th
en
s
av
ed
in
XL
SX
f
o
r
m
at
f
o
r
p
r
o
ce
s
s
in
g
in
th
e
p
r
ep
r
o
ce
s
s
in
g
s
tag
e.
2
.
3
.
T
ra
ns
la
t
io
n a
nd
t
ex
t
pr
epro
ce
s
s
ing
All
r
ev
iews
wer
e
tr
an
s
lated
in
to
E
n
g
lis
h
to
lev
er
a
g
e
th
e
Dis
tilB
E
R
T
-
b
ase
-
u
n
ca
s
ed
m
o
d
el,
wh
ich
h
as
b
ee
n
ex
ten
s
iv
ely
p
r
e
-
tr
ain
ed
o
n
h
ig
h
-
q
u
ality
E
n
g
lis
h
co
r
p
o
r
a.
T
h
is
cr
o
s
s
-
lan
g
u
ag
e
ap
p
r
o
ac
h
allo
ws
an
aly
s
is
o
n
v
ar
i
o
u
s
tar
g
et
lan
g
u
ag
es
wi
th
o
u
t
r
eq
u
ir
in
g
m
ass
iv
e
lab
eled
d
ata
[
3
3
]
.
T
ex
t
clea
n
in
g
in
v
o
lv
ed
ca
s
e
-
f
o
l
d
in
g
,
r
em
o
v
in
g
em
p
ty
v
alu
es,
HT
ML
en
titi
es,
s
y
m
b
o
ls
,
em
o
jis
,
an
d
n
u
m
b
er
s
.
Data
s
et
s
tatis
ti
cs
ar
e
p
r
esen
ted
in
T
ab
le
1
,
an
d
clea
n
in
g
ex
am
p
le
s
in
T
ab
le
2
.
T
ab
le
1
.
Su
m
m
a
r
y
o
f
r
aw
a
n
d
clea
n
ed
Sh
o
p
ee
Pay
r
ev
iew
d
at
asets
f
o
r
I
n
d
o
n
esia a
n
d
T
h
ailan
d
C
o
u
n
t
r
y
R
a
w
d
a
t
a
C
l
e
a
n
e
d
d
a
t
a
I
n
d
o
n
e
si
a
9
1
2
3
4
8
9
3
1
1
Th
a
i
l
a
n
d
8
4
1
0
0
8
2
2
6
0
T
ab
le
2
.
E
x
am
p
les o
f
tr
a
n
s
lated
r
ev
iews a
n
d
c
o
r
r
esp
o
n
d
in
g
c
lean
ed
tex
t a
f
ter
p
r
ep
r
o
ce
s
s
in
g
s
tep
s
C
o
u
n
t
r
y
Tr
a
n
s
l
a
t
e
d
r
e
v
i
e
w
s
C
l
e
a
n
e
d
r
e
v
i
e
w
s
I
n
d
o
n
e
si
a
C
h
e
a
p
i
t
e
ms..
d
i
s
c
o
u
n
t
s
t
o
s
a
t
i
sf
y
c
u
st
o
mers
I
&#
3
9
;
m st
i
l
l
h
a
v
i
n
g
t
r
o
u
b
l
e
v
e
r
i
f
y
i
n
g
c
h
e
a
p
i
t
e
ms
d
i
sc
o
u
n
t
s
t
o
sa
t
i
sf
y
c
u
s
t
o
mers
i
m st
i
l
l
h
a
v
i
n
g
t
r
o
u
b
l
e
v
e
r
i
f
y
i
n
g
T
h
a
i
l
a
n
d
Ex
c
e
l
l
e
n
t
.
T
h
i
s
a
p
p
i
s
t
h
e
b
e
st
.
G
o
o
d
p
r
o
d
u
c
t
s,
e
v
e
r
y
t
h
i
n
g
a
v
a
i
l
a
b
l
e
,
e
x
c
e
l
l
e
n
t
t
h
i
s
a
p
p
i
s
t
h
e
b
e
st
g
o
o
d
p
r
o
d
u
c
t
s
e
v
e
r
y
t
h
i
n
g
a
v
a
i
l
a
b
l
e
a
f
f
o
r
d
a
b
l
e
2
.
4
.
T
o
k
eniza
t
io
n a
nd
lemma
t
iza
t
io
n
T
e
x
t
was
l
em
m
a
tiz
ed
u
s
i
n
g
t
h
e
s
p
aC
y
li
b
r
ar
y
,
a
n
d
s
t
a
n
d
a
r
d
E
n
g
l
is
h
s
t
o
p
w
o
r
d
s
w
er
e
r
em
o
v
e
d
to
r
e
d
u
ce
s
em
an
tic
n
o
is
e
[
3
4
]
.
T
h
e
f
in
al
r
esu
lt
o
f
t
h
is
s
ta
g
e
is
a
c
o
ll
ec
t
io
n
o
f
f
i
n
a
l t
o
k
e
n
s
r
e
a
d
y
to
b
e
u
s
e
d
as i
n
p
u
t
f
o
r
f
o
r
m
in
g
a
t
o
p
ic
m
o
d
el
in
g
c
o
r
p
u
s
,
as
p
r
ese
n
t
ed
i
n
T
ab
le
3
.
T
ab
le
3
.
Sam
p
le
lem
m
atize
d
r
ev
iew
r
ep
r
esen
tatio
n
s
u
s
ed
as
in
p
u
t f
o
r
to
p
ic
m
o
d
elin
g
Le
mm
a
t
i
z
e
d
r
e
v
i
e
w
s
[
‘
c
h
e
a
p
’
,
‘
i
t
e
m
’
,
‘
d
i
s
c
o
u
n
t
’
,
‘
sat
i
sf
y
’
,
‘
c
u
st
o
m
e
r
’
]
[
‘
im
’
,
‘
st
i
l
l
’
,
‘
h
a
v
i
n
g
’
,
‘
t
r
o
u
b
l
e
’
,
‘
v
e
r
i
f
y
i
n
g
]
[
‘
e
x
c
e
l
l
e
n
t
’
,
‘
a
p
p
’
,
‘
g
o
o
d
’
]
[
‘
g
o
o
d
’
,
‘
p
r
o
d
u
c
t
’
,
‘
a
v
a
i
l
a
b
l
e
’
,
‘
a
f
f
o
r
d
a
b
l
e
’
,
‘
p
r
i
c
e
’
,
‘
a
p
p
’
]
2
.
5
.
Asp
ec
t
ex
t
r
a
ct
io
n
C
lea
n
t
e
x
t
was
c
o
n
v
er
te
d
to
n
u
m
er
i
ca
l
f
o
r
m
at
u
s
i
n
g
B
a
g
-
of
-
W
o
r
d
s
(
B
o
W
)
.
T
o
p
ic
e
x
t
r
ac
ti
o
n
was
p
e
r
f
o
r
m
e
d
u
s
i
n
g
L
D
A
t
o
i
d
en
t
if
y
m
ai
n
r
e
v
i
ew
th
em
es
[
3
5
]
.
L
DA
-
e
x
t
r
a
cte
d
to
p
i
cs
a
r
e
t
r
e
at
ed
as
p
r
o
x
y
s
e
r
v
i
ce
asp
ec
ts
r
at
h
e
r
t
h
a
n
f
i
n
e
-
g
r
ai
n
e
d
as
p
ec
t
te
r
m
s
as
d
e
f
i
n
e
d
i
n
ca
n
o
n
ica
l
AB
S
A
f
r
a
m
ew
o
r
k
s
(
e
.
g
.
,
Se
m
E
v
a
l
t
as
k
s
)
,
f
o
ll
o
wi
n
g
ap
p
l
ie
d
r
es
ea
r
c
h
th
at
em
p
l
o
y
s
u
n
s
u
p
er
v
is
ed
t
o
p
i
c
m
o
d
e
ls
t
o
a
p
p
r
o
x
im
at
e
asp
ec
t
ca
te
g
o
r
ies
wh
e
n
lab
el
ed
as
p
e
ct
-
l
ev
el
d
ata
is
u
n
a
v
ai
la
b
l
e
[
2
4
]
,
[
2
5
]
.
T
h
e
o
p
t
im
al
n
u
m
b
e
r
o
f
t
o
p
ics
(
K
)
was
d
et
e
r
m
i
n
ed
f
r
o
m
t
h
e
p
e
ak
o
f
t
h
e
t
o
p
ic
c
o
h
er
e
n
ce
s
c
o
r
e
g
r
ap
h
.
T
h
e
L
DA
p
r
o
ce
d
u
r
e
p
r
o
d
u
ce
s
t
o
p
i
c
d
is
t
r
i
b
u
ti
o
n
s
p
e
r
wo
r
d
a
n
d
p
e
r
d
o
c
u
m
e
n
t
,
e
n
a
b
li
n
g
s
tr
u
c
tu
r
ed
id
e
n
t
if
ica
ti
o
n
o
f
p
u
b
lic
o
p
in
io
n
p
a
tte
r
n
s
[
2
7
]
.
T
h
e
I
n
d
o
n
esia
n
d
a
tas
et
ac
h
i
ev
e
d
h
i
g
h
est
c
o
h
er
en
ce
at
K
=9
,
wh
ile
th
e
T
h
ai
d
atas
et
’
s
o
p
ti
m
a
l
p
o
i
n
t
was
K=
6
(
F
i
g
u
r
e
2
an
d
Fi
g
u
r
e
3
)
.
T
h
is
v
a
r
ia
n
ce
r
ef
lec
ts
d
if
f
e
r
i
n
g
s
em
a
n
tic
d
i
v
e
r
s
it
y
:
I
n
d
o
n
esia
n
r
ev
iews
ex
h
i
b
it
e
d
b
r
o
ad
er
t
h
e
m
es
(
e.
g
.
,
l
o
a
n
f
e
at
u
r
es,
a
d
m
in
f
ee
c
o
m
p
lai
n
ts
)
r
e
q
u
i
r
i
n
g
h
i
g
h
e
r
K,
w
h
i
le
T
h
ai
r
e
v
iews
c
o
n
c
en
tr
ate
d
o
n
c
o
r
e
f
u
n
ct
io
n
al
iti
es
li
k
e
r
e
g
is
t
r
a
ti
o
n
a
n
d
p
ay
m
e
n
t
p
r
o
ce
s
s
i
n
g
,
r
e
q
u
i
r
i
n
g
f
ewe
r
t
o
p
i
cs
f
o
r
s
e
m
an
t
ic
co
n
s
is
te
n
c
y
.
T
o
p
i
cs
we
r
e
la
b
e
le
d
b
as
ed
o
n
d
o
m
in
a
n
t
k
ey
w
o
r
d
s
,
r
e
f
l
ec
t
in
g
k
e
y
d
ig
ital
w
all
et
s
e
r
v
ice
ar
ea
s
in
cl
u
d
i
n
g
t
r
a
n
s
ac
ti
o
n
s
/
p
a
y
m
e
n
ts
,
u
s
e
r
e
x
p
er
ie
n
c
e,
p
r
o
m
o
ti
o
n
s
/i
n
ce
n
t
iv
es,
s
er
v
i
ce
p
r
o
c
ess
es,
an
d
ac
c
o
u
n
t
/s
ec
u
r
it
y
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4
7
5
2
Tr
a
n
s
fo
r
mer
-
b
a
s
ed
s
en
timen
t
mo
d
elin
g
fo
r
id
en
tifyin
g
cro
s
s
-
co
u
n
tr
y
fin
tech
…
(
K
a
yla
Zh
a
fir
a
A
r
d
in
o
v)
525
Fig
u
r
e
2
.
T
o
p
ic
co
h
er
en
ce
f
o
r
th
e
T
h
ai
d
ataset.
T
h
e
p
ea
k
at
K=
6
in
d
icate
s
o
p
t
im
al
s
em
an
tic
co
n
s
is
ten
cy
f
o
r
f
o
cu
s
ed
a
n
d
r
e
p
r
esen
tativ
e
asp
ec
t
id
en
tific
atio
n
Fig
u
r
e
3
.
T
o
p
ic
co
h
er
en
ce
f
o
r
th
e
I
n
d
o
n
esian
d
ataset.
T
h
e
p
ea
k
at
K
=9
e
n
s
u
r
es o
p
ti
m
al
s
em
an
tic
co
n
s
is
ten
cy
f
o
r
d
is
tin
ct
an
d
r
e
p
r
esen
tativ
e
asp
ec
t
id
en
tific
atio
n
Af
t
er
d
et
er
m
i
n
i
n
g
t
h
e
o
p
tim
al
n
u
m
b
e
r
o
f
t
o
p
i
cs
(
K
)
,
t
h
e
n
ex
t
s
te
p
is
t
o
a
n
n
o
t
ate
e
ac
h
t
o
p
i
c
g
e
n
e
r
a
te
d
b
y
th
e
L
DA
m
o
d
el
t
h
r
o
u
g
h
t
h
e
t
o
p
ic
l
a
b
el
in
g
s
t
ag
e.
T
h
is
p
r
o
ce
s
s
is
d
o
n
e
m
a
n
u
al
l
y
b
y
an
al
y
zi
n
g
t
h
e
m
o
s
t
r
e
p
r
ese
n
t
ati
v
e
k
ey
wo
r
d
s
(
t
o
p
i
c
k
e
y
w
o
r
d
s
)
an
d
e
x
am
i
n
i
n
g
t
h
e
r
e
v
i
ew
e
x
am
p
l
es
wit
h
t
h
e
h
i
g
h
est
p
r
o
b
a
b
i
lit
y
i
n
ea
c
h
to
p
i
c
t
o
en
s
u
r
e
ac
cu
r
ate
s
em
an
tic
i
n
te
r
p
r
eta
ti
o
n
.
T
h
e
r
es
u
lts
o
f
asp
ec
t
e
x
t
r
a
cti
o
n
f
o
r
e
ac
h
co
u
n
t
r
y
a
r
e
p
r
ese
n
t
ed
in
T
a
b
l
es
4
a
n
d
5
.
B
ase
d
o
n
t
h
es
e
r
es
u
lts
,
9
s
e
r
v
ice
as
p
ec
ts
w
e
r
e
i
d
e
n
t
if
ie
d
f
o
r
t
h
e
I
n
d
o
n
esi
an
r
eg
io
n
an
d
6
f
o
r
t
h
e
T
h
ai
r
e
g
i
o
n
.
T
ab
le
4
.
L
DA
-
b
ased
t
o
p
ic
k
e
y
wo
r
d
s
an
d
m
a
n
u
ally
ass
i
g
n
ed
asp
ec
t la
b
els f
o
r
I
n
d
o
n
esian
S
h
o
p
ee
Pay
r
e
v
iews
T
ab
le
5
.
L
DA
-
b
ased
t
o
p
ic
k
e
y
wo
r
d
s
an
d
m
a
n
u
ally
ass
ig
n
ed
asp
ec
t la
b
els f
o
r
T
h
ailan
d
Sh
o
p
ee
Pay
r
ev
i
ews
Fo
r
cr
o
s
s
-
c
o
u
n
tr
y
c
o
m
p
ar
is
o
n
a
n
a
ly
s
is
,
t
h
is
s
t
u
d
y
f
o
c
u
s
es
i
ts
e
v
al
u
at
io
n
o
n
o
v
e
r
la
p
p
i
n
g
asp
ec
ts
i
n
b
o
t
h
co
u
n
t
r
i
es.
T
h
ese
as
p
e
cts
i
n
cl
u
d
e
t
h
e
d
i
m
e
n
s
i
o
n
s
o
f
t
r
a
n
s
a
cti
o
n
&
p
a
y
m
e
n
t,
p
r
o
m
o
ti
o
n
/
v
o
u
ch
e
r
,
a
n
d
p
r
o
ce
s
s
ti
m
e
.
T
h
is
ali
g
n
m
en
t
is
i
n
te
n
d
e
d
t
o
e
n
s
u
r
e
t
h
at
t
h
e
p
er
ce
p
ti
o
n
g
ap
is
m
ea
s
u
r
e
d
u
s
i
n
g
s
er
v
i
ce
p
a
r
a
m
e
te
r
s
t
h
at
a
r
e
e
q
u
i
v
a
le
n
t
a
n
d
f
u
n
cti
o
n
al
ly
v
ali
d
a
cr
o
s
s
b
o
t
h
f
in
tec
h
ec
o
s
y
s
te
m
s
.
T
h
e
s
e
s
el
ec
t
ed
as
p
e
cts
will
t
h
en
b
e
p
r
o
c
ess
e
d
f
o
r
a
u
t
o
m
ati
c
s
e
n
t
im
en
t
la
b
eli
n
g
b
ef
o
r
e
e
n
te
r
i
n
g
th
e
c
lass
i
f
i
ca
t
i
o
n
p
h
ase
u
s
i
n
g
t
h
e
Dis
tilB
E
R
T
m
o
d
el
Sin
ce
L
DA
was
co
n
d
u
ct
e
d
s
ep
ar
ate
ly
o
n
t
h
e
I
n
d
o
n
esi
a
n
an
d
T
h
ai
c
o
r
p
o
r
a
,
a
n
as
p
e
ct
ali
g
n
m
e
n
t
p
r
o
ce
s
s
was
r
e
q
u
ir
e
d
to
i
d
e
n
t
if
y
c
o
m
p
a
r
a
b
l
e
s
e
r
v
ice
d
i
m
e
n
s
i
o
n
s
ac
r
o
s
s
b
o
t
h
co
u
n
t
r
ies
.
Fi
r
s
t,
J
ac
ca
r
d
S
im
i
la
r
it
y
was
ca
l
c
u
la
te
d
b
ase
d
o
n
th
e
to
p
1
0
k
ey
wo
r
d
s
o
f
ea
c
h
t
o
p
ic,
r
es
u
lt
in
g
i
n
a
m
a
x
i
m
u
m
s
im
il
ar
it
y
s
c
o
r
e
o
f
0
.
2
0
.
T
h
is
r
el
ati
v
el
y
lo
w
s
co
r
e
i
n
d
ic
at
es
l
e
x
ic
al
d
if
f
e
r
e
n
ce
s
a
r
is
i
n
g
f
r
o
m
t
h
e
u
s
e
o
f
d
i
f
f
e
r
e
n
t
la
n
g
u
a
g
es
a
n
d
ter
m
i
n
o
l
o
g
ies
,
s
u
g
g
est
in
g
t
h
at
k
ey
wo
r
d
o
v
er
la
p
al
o
n
e
is
i
n
s
u
f
f
i
ci
en
t
to
e
s
ta
b
l
is
h
to
p
ic
eq
u
i
v
ale
n
ce
.
T
h
er
ef
o
r
e
,
a
m
a
n
u
al
al
ig
n
m
e
n
t
p
r
o
ce
s
s
was
p
e
r
f
o
r
m
e
d
b
y
e
x
a
m
i
n
i
n
g
to
p
i
c
k
ey
wo
r
d
s
,
r
e
p
r
es
en
tati
v
e
r
e
v
iews
,
To
p
i
c
n
o.
To
p
i
c
l
a
b
e
l
i
n
g
I
n
d
o
n
e
s
i
a
To
p
i
c
k
e
y
w
o
r
d
s
La
b
e
l
s
0
e
a
s
y
,
t
r
a
n
sac
t
i
o
n
,
f
a
st
,
p
a
y
me
n
t
,
si
mp
l
e
,
p
r
a
c
t
i
c
a
l
,
s
a
f
e
,
p
r
o
c
e
ss
Tr
a
n
s
a
c
t
i
o
n
a
n
d
p
a
y
me
n
t
1
u
se,
w
e
l
l
,
d
o
w
n
l
o
a
d
,
a
p
p
,
g
e
t
,
a
p
p
l
i
c
a
t
i
o
n
,
b
a
l
a
n
c
e
,
c
o
i
n
A
p
p
u
s
a
g
e
2
g
r
e
a
t
,
p
a
y
,
a
w
e
so
m
e
,
s
h
o
p
e
e
,
l
o
a
n
,
l
i
mi
t
,
sp
i
n
j
a
m,
p
a
y
l
a
t
e
r
,
s
p
a
y
l
a
t
e
r
Lo
a
n
f
e
a
t
u
r
e
3
e
x
c
e
l
l
e
n
t
,
l
i
k
e
,
t
r
a
n
sf
e
r
,
a
d
mi
n
,
f
r
e
e
,
b
a
n
k
,
f
e
e
,
w
a
l
l
e
t
A
d
mi
n
f
e
e
s
a
n
d
b
a
n
k
t
r
a
n
sf
e
r
4
c
o
o
l
,
l
o
t
,
s
a
t
i
sf
i
e
d
,
st
a
r
,
s
h
o
p
e
e
,
b
u
y
,
d
i
sc
o
u
n
t
,
s
h
o
p
e
e
p
a
y
,
v
o
u
c
h
e
r
P
r
o
mo
,
d
i
sc
o
u
n
t
,
a
n
d
v
o
u
c
h
e
r
5
t
r
y
,
w
a
n
t
,
a
c
c
o
u
n
t
,
t
i
m
e
,
b
a
d
,
a
p
p
l
i
c
a
t
i
o
n
,
m
o
n
e
y
,
l
o
n
g
P
r
o
c
e
ss
t
i
me
6
h
e
l
p
f
u
l
,
o
k
a
y
,
a
p
p
l
i
c
a
t
i
o
n
,
s
l
o
w
,
sm
o
o
t
h
l
y
,
r
u
n
,
a
p
p
,
b
u
g
A
p
p
p
e
r
f
o
r
ma
n
c
e
(
b
u
g
/
s
l
o
w
)
7
t
h
a
n
k
,
sh
o
p
e
e
p
a
y
,
s
h
o
p
e
e
,
h
e
l
p
,
e
a
s
y
,
mak
e
,
h
o
p
e
f
u
l
l
y
,
s
h
o
p
p
i
n
g
G
e
n
e
r
a
l
p
o
si
t
i
v
e
f
e
e
d
b
a
c
k
8
g
o
o
d
,
a
p
p
l
i
c
a
t
i
o
n
,
u
s
e
f
u
l
,
o
k
,
j
o
b
,
a
p
p
,
ser
v
i
c
e
G
o
o
d
a
p
p
/
s
e
r
v
i
c
e
To
p
i
c
no
.
To
p
i
c
la
b
e
l
i
n
g
Th
a
i
l
a
n
d
To
p
i
c
k
e
y
w
o
r
d
s
La
b
e
l
s
0
mo
n
e
y
,
a
c
c
o
u
n
t
,
b
a
n
k
,
t
r
a
n
sf
e
r
,
a
d
d
,
b
a
h
t
,
g
a
m
e
,
a
p
p
,
w
a
n
t
,
d
e
d
u
c
t
Tr
a
n
s
a
c
t
i
o
n
a
n
d
p
a
y
me
n
t
1
l
i
k
e
,
c
a
r
d
,
n
u
m
b
e
r
,
a
p
p
,
t
r
y
,
e
n
t
e
r
,
v
e
r
i
f
y
,
i
d
e
n
t
i
t
y
,
sa
y
,
c
o
d
e
R
e
g
i
s
t
r
a
t
i
o
n
2
c
o
n
v
e
n
i
e
n
t
,
p
a
y
,
g
r
e
a
t
,
l
i
k
e
,
b
i
l
l
,
s
h
o
p
e
e
,
t
h
a
n
k
,
o
k
,
b
u
y
,
a
p
p
B
i
l
l
p
a
y
m
e
n
t
a
n
d
m
e
r
c
h
a
n
t
3
u
se,
e
a
sy
,
c
o
n
v
e
n
i
e
n
t
,
f
a
s
t
,
w
o
r
k
,
l
i
k
e
,
l
o
t
,
d
i
sc
o
u
n
t
,
p
r
o
mo
t
i
o
n
,
a
p
p
U
ser
e
x
p
e
r
i
e
n
c
e
(
U
X
)
a
n
d
p
r
o
m
o
t
i
o
n
4
t
i
m
e
,
l
o
n
g
,
p
r
o
b
l
e
m,
st
a
r
,
b
a
d
,
t
a
k
e
,
s
e
r
v
i
c
e
,
s
y
s
t
e
m
,
w
a
i
t
,
c
o
n
t
a
c
t
S
e
r
v
i
c
e
p
r
o
c
e
ss
t
i
me
5
g
o
o
d
,
a
p
p
,
ser
v
i
c
e
,
w
o
r
t
h
,
g
a
me
,
m
a
st
e
r
c
a
r
d
,
a
c
t
i
v
i
t
y
,
u
sefu
l
,
e
x
c
h
a
n
g
e
G
e
n
e
r
a
l
p
o
si
t
i
v
e
f
e
e
d
b
a
c
k
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
5
2
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
43
,
No
.
2
,
Au
g
u
s
t
20
26
:
522
-
5
3
3
526
an
d
t
h
e
f
u
n
ct
io
n
a
l
m
ea
n
i
n
g
o
f
e
ac
h
t
o
p
ic
.
T
h
e
r
esu
lti
n
g
as
p
e
ct
p
ai
r
s
w
er
e
s
u
b
s
eq
u
en
tl
y
v
al
id
at
ed
th
r
o
u
g
h
ex
p
e
r
t
ju
d
g
m
e
n
t
b
y
a
d
o
m
ai
n
ex
p
er
t
t
o
e
n
s
u
r
e
co
n
c
e
p
t
u
al
a
n
d
f
u
n
c
ti
o
n
al
eq
u
i
v
ale
n
c
e
a
c
r
o
s
s
t
h
e
tw
o
f
i
n
te
c
h
ec
o
s
y
s
te
m
s
.
T
h
r
o
u
g
h
t
h
is
p
r
o
ce
s
s
,
t
h
r
ee
o
v
er
la
p
p
in
g
as
p
e
ct
s
we
r
e
id
e
n
ti
f
i
ed
,
n
a
m
e
ly
t
r
a
n
s
ac
ti
o
n
&
p
a
y
m
e
n
t
,
p
r
o
m
o
ti
o
n
/
v
o
u
c
h
er
,
a
n
d
p
r
o
ce
s
s
ti
m
e
.
2
.
6
.
Sentim
ent
ps
eudo
-
la
belin
g
I
n
itial
s
en
tim
en
t
lab
elin
g
was
p
er
f
o
r
m
e
d
u
s
in
g
a
p
s
eu
d
o
-
lab
elin
g
ap
p
r
o
ac
h
with
th
e
p
r
etr
ain
ed
Dis
til
B
E
R
T
S
ST
-
2
m
o
d
el
to
g
en
er
ate
b
in
ar
y
p
o
s
itiv
e/n
eg
ativ
e
class
es.
T
h
is
s
tag
e
en
ab
led
au
to
m
atic
lab
elin
g
o
f
lar
g
e
-
s
ca
le
d
atasets
with
o
u
t
r
eq
u
ir
in
g
co
m
p
r
eh
en
s
iv
e
m
an
u
al
a
n
n
o
tatio
n
.
T
h
is
m
o
d
el
was
d
ev
elo
p
ed
th
r
o
u
g
h
k
n
o
wled
g
e
d
is
till
atio
n
tech
n
iq
u
es
to
tr
an
s
f
er
k
n
o
wl
ed
g
e
f
r
o
m
lar
g
e
B
E
R
T
m
o
d
el
s
to
s
m
aller
m
o
d
els
with
o
u
t sig
n
if
ican
t lo
s
s
o
f
ac
c
u
r
ac
y
[
2
8
]
.
Alth
o
u
g
h
th
e
n
u
m
b
er
o
f
lay
e
r
s
was
r
ed
u
ce
d
,
th
e
Mu
lti
-
H
ea
d
Atten
tio
n
b
lo
ck
s
tr
u
ctu
r
e
,
wh
ich
is
cr
u
cial
f
o
r
la
n
g
u
a
g
e
p
r
o
ce
s
s
in
g
,
was
r
etain
ed
[
3
6
]
.
T
h
e
m
o
d
el
was
tr
ain
ed
u
s
in
g
a
tr
an
s
f
o
r
m
er
ar
c
h
itectu
r
e
th
at
ca
p
tu
r
es
b
id
ir
ec
tio
n
al
lan
g
u
ag
e
co
n
tex
t
[
2
3
]
.
I
n
th
e
cl
ass
if
icatio
n
p
r
o
ce
s
s
,
th
e
m
o
d
el
ex
tr
ac
ts
f
ea
tu
r
es
f
r
o
m
th
e
s
p
ec
ial
class
if
icatio
n
to
k
en
[
C
L
S],
wh
ich
ar
e
th
en
p
r
o
ce
s
s
ed
th
r
o
u
g
h
th
e
So
f
tm
ax
f
u
n
ctio
n
to
g
en
er
ate
class
p
r
o
b
a
b
ilit
ies
[
3
7
]
.
T
h
e
f
o
llo
win
g
eq
u
atio
n
d
ef
in
es
in
(
1
)
:
(
|
)
=
{
{
}
}
∑
{
}
{
}
{
=
1
}
(
1
)
w
h
er
e
(
|
)
r
ep
r
esen
ts
th
e
lik
elih
o
o
d
o
f
in
p
u
t
b
ein
g
class
if
ied
in
to
class
,
is
th
e
m
o
d
el
’
s
o
u
tp
u
t
lo
g
i
t
s
co
r
e
f
o
r
th
at
class
,
an
d
is
th
e
to
tal
n
u
m
b
er
o
f
s
en
tim
en
t
class
es.
T
h
e
p
s
eu
d
o
-
la
b
elin
g
r
esu
lts
(
T
ab
le
6
)
s
h
o
w
class
im
b
alan
ce
in
b
o
th
d
atasets
:
I
n
d
o
n
esian
p
o
s
itiv
e
s
en
tim
en
t
at
7
8
.
7
%
(
7
0
,
2
8
8
r
ev
iews)
v
er
s
u
s
2
1
.
3
% n
e
g
ativ
e
(
1
9
,
0
2
3
)
,
a
n
d
T
h
ailan
d
at
7
2
.
1
% p
o
s
itiv
e
(
5
6
,
1
9
0
)
v
er
s
u
s
2
7
.
9
% n
e
g
ativ
e
(
2
1
,
7
5
6
)
.
T
ab
le
6
.
Su
m
m
a
r
y
o
f
s
en
tim
e
n
t p
o
lar
ity
d
is
tr
ib
u
tio
n
ac
r
o
s
s
co
u
n
tr
ies h
ig
h
lig
h
tin
g
class
im
b
alan
ce
C
o
u
n
t
r
y
P
o
si
t
i
v
e
c
o
u
n
t
N
e
g
a
t
i
v
e
c
o
u
n
t
P
o
si
t
i
v
e
(
%)
N
e
g
a
t
i
v
e
(
%)
I
n
d
o
n
e
si
a
7
0
,
2
8
8
1
9
,
0
2
3
7
8
.
7
2
1
.
3
T
h
a
i
l
a
n
d
5
6
,
1
9
0
2
1
,
7
5
6
7
2
.
1
2
7
.
9
T
h
e
s
ig
n
if
ican
t
im
b
alan
ce
in
d
is
tr
ib
u
tio
n
b
etwe
en
th
e
p
o
s
itiv
e
an
d
n
eg
ativ
e
class
es
in
b
o
th
co
u
n
tr
ies
co
n
f
ir
m
s
th
e
p
r
esen
ce
o
f
d
ata
im
b
alan
ce
.
I
f
th
is
d
ata
is
u
s
ed
d
ir
ec
tly
f
o
r
f
in
e
-
tu
n
in
g
with
o
u
t
s
p
ec
ial
h
an
d
lin
g
,
th
er
e
is
a
co
n
ce
r
n
th
at
th
e
m
o
d
el
will
b
e
b
iased
to
war
d
s
th
e
m
ajo
r
ity
class
(
p
o
s
itiv
e)
an
d
p
er
f
o
r
m
p
o
o
r
ly
at
r
ec
o
g
n
izin
g
u
s
er
co
m
p
lain
ts
(
n
eg
ativ
e
class
)
.
T
h
er
e
f
o
r
e,
t
h
ese
lab
elin
g
r
esu
lts
p
r
o
v
id
e
a
s
tr
o
n
g
b
asis
f
o
r
co
n
d
u
ctin
g
th
e
I
m
b
alan
ce
d
Data
Han
d
lin
g
s
tag
e
u
s
in
g
t
h
e
o
v
er
s
am
p
lin
g
tech
n
i
q
u
e
b
ef
o
r
e
s
tar
tin
g
m
o
d
el
tr
ain
in
g
.
T
o
m
itig
ate
th
e
r
is
k
o
f
cir
cu
lar
ev
alu
atio
n
,
a
s
tr
atif
ied
r
an
d
o
m
s
am
p
le
o
f
5
0
0
r
ev
iews
p
er
co
u
n
tr
y
was
in
d
ep
en
d
en
tly
an
n
o
tated
b
y
th
e
a
u
th
o
r
s
.
R
esu
lts
(
T
ab
l
e
7
)
s
h
o
w
C
o
h
e
n
’
s
Kap
p
a
o
f
0
.
9
7
6
0
(
I
n
d
o
n
esia)
an
d
0
.
9
8
4
0
(
T
h
ailan
d
)
,
in
d
icat
in
g
“
Alm
o
s
t
Per
f
ec
t
”
ag
r
ee
m
e
n
t.
Dis
ag
r
ee
m
en
ts
o
cc
u
r
r
ed
p
r
im
ar
ily
in
s
h
o
r
t
o
r
s
ar
ca
s
tic
r
ev
iews
with
in
h
er
en
tly
am
b
ig
u
o
u
s
p
o
lar
ity
.
T
h
ese
f
in
d
in
g
s
co
n
f
ir
m
th
at
th
e
p
s
eu
d
o
-
lab
els
g
en
er
a
ted
b
y
t
h
e
Dis
tilB
E
R
T
SS
T
-
2
m
o
d
el
a
r
e
ex
ce
p
tio
n
all
y
r
eliab
le
f
o
r
tr
ai
n
in
g
p
u
r
p
o
s
es.
W
h
ile
th
e
f
in
al
class
if
icatio
n
ac
cu
r
ac
ies
(
9
7
–
9
8
%)
r
ep
o
r
ted
in
Sectio
n
3
p
ar
tly
r
ef
lect
th
e
m
o
d
el
’
s
alig
n
m
en
t
with
th
e
p
s
eu
d
o
-
lab
eler
’
s
o
u
tp
u
t,
th
is
m
an
u
al
v
alid
atio
n
p
r
o
v
i
d
es
r
o
b
u
s
t
ev
id
e
n
ce
t
h
at
s
u
ch
a
lig
n
m
en
t
is
cl
o
s
ely
s
y
n
ch
r
o
n
ize
d
with
h
u
m
an
e
m
o
tio
n
al
in
ter
p
r
etatio
n
.
T
h
e
r
esu
lts
o
f
th
e
m
a
n
u
al
v
alid
at
io
n
,
co
m
p
ar
in
g
th
e
au
to
m
ated
p
s
eu
d
o
-
lab
els ag
ai
n
s
t h
u
m
an
ju
d
g
m
e
n
t,
ar
e
s
u
m
m
ar
ized
in
T
a
b
le
7
.
T
ab
le
7
.
Ma
n
u
al
v
alid
atio
n
r
esu
lts
o
f
p
s
eu
d
o
-
lab
el
r
elia
b
ilit
y
C
o
u
n
t
r
y
S
u
b
s
e
t
si
z
e
A
g
r
e
e
m
e
n
t
c
o
u
n
t
A
g
r
e
e
m
e
n
t
r
a
t
e
(
%)
C
o
h
e
n
’
s
K
a
p
p
a
(
κ
)
A
g
r
e
e
m
e
n
t
l
e
v
e
l
I
n
d
o
n
e
si
a
5
0
0
4
9
4
9
8
.
8
0
0
.
9
7
6
0
A
l
mo
s
t
p
e
r
f
e
c
t
Th
a
i
l
a
n
d
5
0
0
4
9
6
9
9
.
2
0
0
.
9
8
4
0
A
l
mo
s
t
p
e
r
f
e
c
t
2
.
7
.
Da
t
a
s
pli
t
t
ing
a
nd
cla
s
s
im
ba
la
nce
ha
nd
lin
g
T
h
e
d
ataset
was
s
p
lit
8
0
:2
0
f
o
r
tr
ain
in
g
an
d
test
in
g
,
with
t
h
e
f
ix
ed
s
p
lit
s
elec
ted
d
u
e
to
th
e
s
ca
le
(
1
7
0
,
0
0
0
+
r
ev
iews)
a
n
d
co
m
p
u
tatio
n
al
c
o
s
t
(
~2
4
h
o
u
r
s
p
e
r
f
in
e
-
t
u
n
in
g
cy
cle)
.
R
an
d
o
m
ove
r
s
am
p
lin
g
was
ap
p
lied
ex
cl
u
s
iv
ely
to
th
e
tr
a
in
in
g
s
et;
th
e
test
s
et
r
etain
e
d
its
n
atu
r
al
im
b
alan
ce
d
d
is
tr
ib
u
tio
n
.
E
x
ac
t
d
ata
co
u
n
ts
ar
e
in
T
a
b
le
8
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4
7
5
2
Tr
a
n
s
fo
r
mer
-
b
a
s
ed
s
en
timen
t
mo
d
elin
g
fo
r
id
en
tifyin
g
cro
s
s
-
co
u
n
tr
y
fin
tech
…
(
K
a
yla
Zh
a
fir
a
A
r
d
in
o
v)
527
T
ab
le
8
.
Dis
tr
ib
u
tio
n
o
f
p
o
s
itiv
e
an
d
n
eg
ativ
e
s
en
tim
en
t sam
p
les ac
r
o
s
s
tr
ain
in
g
an
d
test
in
g
s
ets
C
o
u
n
t
r
y
S
e
n
t
i
me
n
t
D
a
t
a
sp
l
i
t
s
u
mm
a
r
y
D
a
t
a
t
o
t
a
l
D
a
t
a
t
r
a
i
n
D
a
t
a
t
e
s
t
I
n
d
o
n
e
si
a
P
o
si
t
i
v
e
7
0
2
8
8
5
6
2
3
0
1
4
0
5
8
N
e
g
a
t
i
v
e
1
9
0
2
3
1
5
2
1
8
3
8
0
5
Th
a
i
l
a
n
d
P
o
si
t
i
v
e
5
6
1
9
0
4
4
9
5
1
1
1
2
3
9
N
e
g
a
t
i
v
e
2
1
7
5
6
1
7
4
0
5
4
3
5
1
T
h
e
lab
elin
g
r
esu
lts
r
ev
ea
led
s
tr
ik
in
g
class
im
b
alan
ce
in
b
o
th
d
atasets
(
e.
g
.
,
I
n
d
o
n
esian
r
ev
iews:
7
8
.
7
%
p
o
s
itiv
e
v
s
.
2
1
.
3
%
n
e
g
ativ
e)
.
T
o
ad
d
r
ess
th
is
,
r
an
d
o
m
o
v
er
s
am
p
lin
g
was
ap
p
lied
s
p
ec
if
ically
to
th
e
tr
ain
in
g
s
et,
r
e
p
licatin
g
m
in
o
r
ity
class
(
n
eg
ativ
e)
s
am
p
les
t
o
e
q
u
alize
d
is
tr
ib
u
tio
n
.
T
h
is
m
eth
o
d
was
s
elec
ted
o
v
er
s
y
n
th
etic
m
in
o
r
ity
o
v
er
-
s
am
p
lin
g
tech
n
iq
u
e
(
SMOT
E
)
,
wh
ich
g
e
n
er
ates
s
y
n
th
eti
c
s
am
p
les
th
r
o
u
g
h
v
ec
to
r
-
s
p
ac
e
i
n
ter
p
o
latio
n
th
at
o
f
ten
p
r
o
d
u
ce
s
‘
h
allu
cin
a
ted
’
s
em
an
tic
m
ea
n
in
g
s
a
b
s
en
t
in
r
ea
l
-
wo
r
l
d
lan
g
u
ag
e.
Fo
r
a
tr
an
s
f
o
r
m
er
-
b
ased
m
o
d
el
lik
e
Dis
tilB
E
R
T
th
at
r
elies
o
n
b
id
ir
ec
tio
n
al
co
n
tex
t,
d
u
p
licated
b
u
t
au
th
en
tic
r
ev
iews
p
r
eser
v
e
n
atu
r
al
lin
g
u
is
tic
s
tr
u
ctu
r
es
m
o
r
e
ef
f
ec
tiv
el
y
th
an
d
is
to
r
ted
s
y
n
th
etic
d
ata.
R
an
d
o
m
ove
r
s
am
p
lin
g
was
al
s
o
p
r
ef
er
r
e
d
o
v
e
r
lo
s
s
-
ad
ju
s
tm
en
t
tech
n
iq
u
es
s
u
ch
as
class
weig
h
tin
g
o
r
f
o
ca
l
lo
s
s
,
as
it
p
r
o
v
id
es
a
m
o
r
e
s
tab
le
tr
ain
in
g
d
is
tr
ib
u
tio
n
f
o
r
f
i
n
e
-
tu
n
i
n
g
lar
g
e
-
s
ca
le
tr
a
n
s
f
o
r
m
er
s
,
d
ir
ec
tl
y
o
p
tim
izin
g
r
ec
all
f
o
r
n
eg
ativ
e
s
en
tim
en
t
to
en
s
u
r
e
h
ig
h
s
en
s
itiv
ity
in
d
etec
tin
g
tech
n
o
s
tr
ess
a
n
d
cr
itical
u
s
er
co
m
p
lain
ts
.
2
.
8
.
F
ine
-
t
un
ing
Dis
t
ilB
E
R
T
f
o
r
s
ent
im
ent
cla
s
s
if
ica
t
io
n
T
h
e
Dis
tilB
E
R
T
-
b
ase
-
u
n
ca
s
e
d
m
o
d
el,
p
r
e
-
tr
ain
e
d
o
n
th
e
SS
T
-
2
d
ataset,
was
f
in
e
-
tu
n
ed
f
o
r
d
o
m
ain
-
s
p
ec
if
ic
s
en
tim
en
t
class
if
icat
io
n
.
Dis
tilB
E
R
T
was
ch
o
s
e
n
o
v
e
r
lar
g
e
r
m
o
d
e
ls
s
u
ch
as
B
E
R
T
-
b
ase
o
r
R
o
B
E
R
T
a
d
u
e
to
co
m
p
u
tatio
n
al
co
n
s
tr
ain
ts
:
with
o
v
er
1
7
0
,
0
0
0
r
ev
iews
an
d
ap
p
r
o
x
im
ately
2
4
h
o
u
r
s
p
e
r
tr
ain
in
g
cy
cle,
a
lig
h
ter
ar
ch
it
ec
tu
r
e
was
ess
en
tial.
Dis
til
B
E
R
T
o
f
f
er
s
a
4
0
%
s
ize
r
ed
u
cti
o
n
an
d
6
0
%
s
p
ee
d
in
cr
ea
s
e
wh
ile
r
etain
in
g
~
9
7
%
o
f
B
E
R
T
’
s
p
er
f
o
r
m
an
ce
[
2
8
]
.
Fin
e
-
tu
n
in
g
was c
o
n
d
u
cted
u
s
in
g
th
e
Sh
o
p
ee
Pay
r
ev
iew
d
ataset
to
ad
ju
s
t
th
e
m
o
d
el
’
s
lan
g
u
ag
e
r
e
p
r
esen
t
atio
n
to
e
-
wallet
s
er
v
ice
ter
m
in
o
lo
g
y
an
d
u
s
er
co
m
p
lain
t
p
atter
n
s
.
T
h
e
co
n
f
i
g
u
r
atio
n
i
n
clu
d
ed
a
lear
n
in
g
r
ate
o
f
1
×
1
0
⁻⁵,
3
e
p
o
ch
s
to
p
r
ev
en
t
o
v
er
f
itti
n
g
wh
ile
en
s
u
r
in
g
c
o
n
v
er
g
en
ce
,
an
d
a
v
alid
atio
n
-
s
et
-
b
ased
ev
alu
atio
n
s
tr
ateg
y
to
s
elec
t
th
e
b
est
m
o
d
e
l
ch
ec
k
p
o
i
n
t.
2
.
9
.
E
v
a
lua
t
i
o
n m
et
ho
ds
T
h
e
ev
alu
atio
n
s
tag
e
ass
ess
e
s
th
e
class
if
icat
io
n
m
o
d
el
’
s
r
e
liab
ilit
y
an
d
th
e
q
u
ality
o
f
th
e
f
ea
tu
r
es
d
er
iv
ed
[
3
8
]
.
T
h
e
f
ir
s
t
co
m
p
o
n
en
t
is
th
e
ev
alu
atio
n
o
f
asp
ec
t
ex
tr
ac
tio
n
r
esu
lts
u
s
in
g
th
e
to
p
ic
co
h
er
en
ce
m
etr
ic.
T
h
is
m
etr
ic
m
ea
s
u
r
es
s
em
an
tic
co
n
s
is
ten
cy
with
in
e
ac
h
to
p
ic
g
en
er
ated
b
y
th
e
L
DA
m
o
d
el
;
a
h
ig
h
er
co
h
er
en
ce
v
al
u
e
in
d
icate
s
th
at
th
e
wo
r
d
s
in
th
at
to
p
ic
ar
e
s
tr
o
n
g
ly
s
em
an
tically
co
n
n
ec
ted
.
T
h
e
o
p
tim
al
n
u
m
b
er
o
f
asp
ec
ts
,
K
is
d
eter
m
in
ed
f
r
o
m
th
e
p
ea
k
o
f
th
e
c
o
h
er
en
ce
s
co
r
e
g
r
ap
h
b
ef
o
r
e
a
s
ig
n
if
ican
t
d
ec
lin
e
.
T
h
e
s
ec
o
n
d
co
m
p
o
n
e
n
t
is
th
e
ev
alu
atio
n
o
f
th
e
f
in
e
-
tu
n
e
d
s
en
tim
en
t
class
if
icatio
n
m
o
d
el
’
s
p
er
f
o
r
m
an
ce
.
T
esti
n
g
was
co
n
d
u
cted
o
n
t
est
d
ata
m
ea
s
u
r
ed
u
s
in
g
a
c
o
n
f
u
s
io
n
m
at
r
ix
to
o
b
tai
n
ac
cu
r
ac
y
,
p
r
ec
is
i
o
n
,
r
ec
all,
an
d
F1
-
s
co
r
e
v
alu
es
[
3
9
]
.
Giv
e
n
th
e
d
ata
b
alan
cin
g
e
f
f
o
r
ts
in
th
e
p
r
ev
i
o
u
s
s
tag
e,
th
e
F1
-
s
co
r
e
is
cr
u
cial
as
a
m
o
r
e
b
alan
ce
d
m
etr
ic
f
o
r
ass
ess
in
g
m
o
d
el
p
er
f
o
r
m
a
n
ce
ac
r
o
s
s
p
o
s
itiv
e
an
d
n
eg
ativ
e
class
es
[
4
0
]
.
T
h
e
f
o
r
m
u
la
is
d
ef
in
e
d
as
(
2
)
.
1
=
2
×
×
+
(2
)
T
o
en
s
u
r
e
ev
alu
atio
n
tr
an
s
p
a
r
en
cy
:
(
i
)
ev
alu
atio
n
was
co
n
d
u
cted
ex
clu
s
iv
ely
o
n
t
h
e
u
n
m
o
d
if
ied
test
s
et
(
2
0
%
s
p
lit)
r
etain
in
g
its
n
atu
r
al
class
d
is
tr
ib
u
tio
n
;
(
ii
)
th
e
r
e
p
o
r
ted
ac
cu
r
ac
y
r
ef
lects
alig
n
m
en
t
with
p
s
eu
d
o
-
lab
els,
n
o
t
h
u
m
an
g
r
o
u
n
d
tr
u
th
,
th
o
u
g
h
m
an
u
al
v
alid
atio
n
o
n
a
5
0
0
-
r
e
v
iew
s
u
b
s
et
(
s
ec
tio
n
2
.
6
)
p
r
o
v
id
es
an
ex
ter
n
al
r
eliab
ilit
y
ch
ec
k
;
an
d
(
iii
)
h
ig
h
class
if
icatio
n
ac
cu
r
ac
y
alo
n
e
d
o
e
s
n
o
t
g
u
ar
a
n
tee
r
ea
l
-
wo
r
l
d
r
o
b
u
s
tn
ess
in
th
is
tr
an
s
late
d
,
cr
o
s
s
-
co
u
n
t
r
y
s
ettin
g
,
th
o
u
g
h
th
e
m
o
d
el
’
s
r
eliab
ilit
y
is
f
u
r
t
h
er
s
u
p
p
o
r
ted
b
y
c
o
n
s
is
ten
t
F1
-
s
co
r
es
ac
r
o
s
s
b
o
th
s
en
tim
en
t
class
es
an
d
t
h
e
n
ar
r
o
w
co
n
f
id
en
ce
in
ter
v
als
,
as
d
etai
led
in
th
e
s
ec
tio
n
3
.
B
ey
o
n
d
g
lo
b
al
m
et
r
ics,
asp
ec
t
-
lev
el
ev
alu
atio
n
ass
ess
es
m
o
d
el
co
n
s
is
ten
cy
ac
r
o
s
s
d
ef
in
ed
Sh
o
p
ee
Pay
s
er
v
ice
asp
e
cts.
I
f
r
esu
lts
m
ee
t
q
u
ality
s
tan
d
ar
d
s
,
t
h
e
p
r
o
ce
s
s
p
r
o
ce
ed
s
to
k
n
o
wled
g
e
in
ter
p
r
etatio
n
;
o
th
er
wis
e,
th
e
s
y
s
te
m
iter
ates
b
ac
k
to
d
ata
m
in
in
g
o
r
tr
a
n
s
f
o
r
m
atio
n
s
tag
es.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
5
2
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
43
,
No
.
2
,
Au
g
u
s
t
20
26
:
522
-
5
3
3
528
3.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
3
.
1
.
G
ener
a
l sent
im
ent
dis
t
ributio
n
Af
ter
f
in
e
-
tu
n
in
g
th
e
Dis
tilB
E
R
T
-
b
ase
-
u
n
ca
s
ed
m
o
d
el,
s
e
n
tim
en
t
class
if
icatio
n
p
er
f
o
r
m
an
ce
was
ev
alu
ated
o
n
test
d
atasets
f
r
o
m
b
o
th
co
u
n
t
r
ies.
E
v
alu
atio
n
m
etr
ics
in
clu
d
ed
ac
cu
r
ac
y
,
p
r
ec
is
io
n
,
r
ec
all,
an
d
F1
-
s
co
r
e
to
ass
ess
th
e
m
o
d
el
’
s
ab
ilit
y
to
class
if
y
r
ev
iews
a
s
p
o
s
itiv
e
o
r
n
eg
ativ
e.
A
s
u
m
m
ar
y
o
f
t
h
e
g
l
o
b
al
class
if
icatio
n
p
er
f
o
r
m
an
ce
f
o
r
th
e
I
n
d
o
n
esian
an
d
T
h
ai
d
ata
s
ets
i
s
p
r
esen
ted
in
T
ab
le
9
.
Fo
r
th
e
I
n
d
o
n
esi
a
n
d
ataset,
th
e
m
o
d
el
ac
h
iev
ed
an
ac
cu
r
ac
y
o
f
9
7
.
6
5
%
with
s
tr
o
n
g
d
etec
tio
n
o
f
b
o
t
h
n
eg
ativ
e
s
en
tim
en
t
(
p
r
ec
is
io
n
=
0
.
9
5
1
7
,
r
ec
all
=
0
.
9
3
7
2
)
an
d
p
o
s
itiv
e
s
en
tim
en
t
(
p
r
ec
is
io
n
=
0
.
9
8
3
1
,
r
ec
a
ll
=
0
.
9
8
7
1
)
.
T
h
e
T
h
ailan
d
d
ataset
ac
h
iev
ed
9
8
.
3
8
%
ac
cu
r
ac
y
with
wel
l
-
b
ala
n
ce
d
m
etr
ics
ac
r
o
s
s
b
o
th
class
es
(
n
eg
ativ
e
F1
=
0
.
9
7
0
9
,
p
o
s
itiv
e
F1
=
0
.
9
8
8
7
)
.
Ho
wev
er
,
th
ese
f
ig
u
r
es
r
eq
u
ir
e
ca
v
ea
ts
:
th
e
9
7
–
9
8
%
ac
cu
r
a
cy
p
r
im
ar
ily
r
e
f
lects
ag
r
ee
m
e
n
t
with
th
e
p
s
eu
d
o
-
lab
el
teac
h
e
r
m
o
d
el
r
ath
er
th
an
h
u
m
an
g
r
o
u
n
d
t
r
u
th
,
th
o
u
g
h
m
a
n
u
al
v
alid
at
io
n
(
Sectio
n
2
.
6
)
co
n
f
ir
m
e
d
~9
9
%
h
u
m
a
n
a
g
r
e
em
en
t
(
κ
>
0
.
9
7
)
.
T
r
a
n
s
latio
n
to
E
n
g
lis
h
m
a
y
also
r
ed
u
c
e
s
en
tim
en
t
-
b
ea
r
in
g
n
u
an
ce
,
p
o
ten
tially
s
im
p
lify
in
g
class
if
icatio
n
f
o
r
an
E
n
g
lis
h
-
ce
n
tr
ic
m
o
d
el.
No
n
eth
eless
,
th
is
p
er
f
o
r
m
a
n
ce
is
co
n
s
is
ten
t
with
s
im
il
ar
tr
an
s
f
o
r
m
er
f
in
e
-
tu
n
in
g
s
tu
d
ies
[
4
0
]
,
[
4
1
]
,
wh
er
e
b
in
a
r
y
clas
s
if
icatio
n
ty
p
ically
ex
ce
ed
s
9
5
%.
M
o
d
el
p
er
f
o
r
m
an
ce
is
f
u
r
th
er
v
alid
ated
th
r
o
u
g
h
C
o
n
f
u
s
io
n
Ma
tr
ix
an
al
y
s
is
(
Fig
u
r
e
4
a
n
d
Fig
u
r
e
5
)
,
c
o
m
p
a
r
in
g
p
r
ed
icte
d
lab
els ag
ain
s
t a
ctu
al
lab
els o
n
th
e
test
d
ata.
T
ab
le
9
.
Glo
b
al
class
if
icatio
n
p
er
f
o
r
m
an
ce
m
etr
ics o
f
th
e
f
in
e
-
tu
n
ed
Dis
tilB
E
R
T
m
o
d
el
f
o
r
I
n
d
o
n
esian
an
d
T
h
ai
d
atasets
C
o
u
n
t
r
y
C
l
a
s
s
P
r
e
c
i
s
i
o
n
R
e
c
a
l
l
F1
-
sc
o
r
e
S
u
p
p
o
r
t
I
n
d
o
n
e
si
a
N
e
g
a
t
i
v
e
0
.
9
5
1
7
0
.
9
3
7
2
0
.
9
4
4
4
3
8
0
5
P
o
si
t
i
v
e
0
.
9
8
3
1
0
.
9
8
7
1
0
.
9
8
5
1
1
4
0
5
8
A
c
c
u
r
a
c
y
(
g
l
o
b
a
l
)
0
.
9
7
6
5
1
7
8
6
3
T
h
a
i
l
a
n
d
N
e
g
a
t
i
v
e
0
.
9
7
0
8
0
.
9
7
1
0
.
9
9
0
5
4
3
5
1
P
o
si
t
i
v
e
0
.
9
8
8
8
0
.
9
8
8
7
0
.
9
9
0
5
1
1
2
3
9
Fig
u
r
e
4
.
C
o
n
f
u
s
io
n
m
atr
i
x
f
o
r
th
e
I
n
d
o
n
esian
d
ataset.
T
h
e
d
o
m
i
n
an
t d
iag
o
n
a
l v
alu
es v
alid
ate
th
e
m
o
d
el
’
s
r
eliab
ilit
y
in
d
is
t
in
g
u
is
h
in
g
s
en
tim
en
t
p
o
lar
ities
Fig
u
r
e
5
.
C
o
n
f
u
s
io
n
m
atr
i
x
f
o
r
th
e
T
h
ai
d
ataset.
Min
im
al
o
f
f
-
d
iag
o
n
al
v
alu
es e
n
s
u
r
e
class
if
icatio
n
ac
cu
r
ac
y
,
s
u
p
p
o
r
tin
g
th
e
v
alid
ity
o
f
th
e
p
er
ce
p
tio
n
g
ap
f
in
d
in
g
s
B
ey
o
n
d
g
lo
b
al
m
etr
ics,
to
f
u
lf
ill
th
e
s
tan
d
ar
d
s
o
f
Asp
ec
t
-
B
ased
Sen
tim
en
t
An
aly
s
is
(
AB
SA)
,
class
if
icatio
n
p
er
f
o
r
m
a
n
ce
w
as
ev
alu
ated
at
th
e
asp
ec
t
lev
el.
T
o
m
ain
tain
alig
n
m
e
n
t
with
th
e
s
u
b
s
eq
u
e
n
t
cr
o
s
s
-
co
u
n
tr
y
co
m
p
ar
is
o
n
,
T
ab
le
1
0
p
r
esen
ts
th
e
d
etaile
d
class
if
icatio
n
r
ep
o
r
t
s
p
ec
if
ically
f
o
r
th
e
th
r
ee
o
v
er
lap
p
i
n
g
s
er
v
ice
asp
ec
ts
ac
r
o
s
s
b
o
th
d
atasets
.
T
h
e
m
o
d
el
d
em
o
n
s
tr
ates
h
ig
h
co
n
s
is
ten
cy
an
d
r
o
b
u
s
tn
ess
in
b
o
th
co
u
n
tr
ies.
Fo
r
ex
am
p
le,
in
th
e
“
p
r
o
m
o
tio
n
/v
o
u
ch
er
”
as
p
ec
t
,
th
e
m
o
d
el
ac
h
iev
ed
ex
ce
p
tio
n
al
F1
-
s
co
r
es o
f
0
.
9
8
5
(
I
n
d
o
n
esia)
an
d
0
.
9
9
3
(
T
h
ailan
d
)
.
W
h
ile
th
e
“
p
r
o
ce
s
s
tim
e
”
asp
ec
t
ex
h
ib
ited
a
s
lig
h
tly
lo
wer
p
r
ec
is
io
n
(
0
.
9
3
0
)
in
th
e
I
n
d
o
n
esian
d
ata
s
et,
lik
ely
d
u
e
t
o
th
e
h
ig
h
l
y
n
u
an
ce
d
o
r
s
ar
ca
s
tic
lan
g
u
ag
e
o
f
ten
em
p
lo
y
ed
b
y
u
s
er
s
wh
en
ex
p
r
ess
in
g
laten
cy
f
r
u
s
tr
atio
n
s
,
th
e
o
v
er
all
a
s
p
ec
t
-
lev
el
m
etr
ics
co
n
f
ir
m
t
h
at
th
e
f
in
e
-
tu
n
ed
Dis
til
B
E
R
T
m
o
d
el
’
s
p
r
ed
ictiv
e
ca
p
ab
ilit
y
is
h
ig
h
ly
r
eliab
le
f
o
r
id
en
tif
y
in
g
th
e
cr
o
s
s
-
co
u
n
tr
y
p
er
ce
p
tio
n
g
a
p
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4
7
5
2
Tr
a
n
s
fo
r
mer
-
b
a
s
ed
s
en
timen
t
mo
d
elin
g
fo
r
id
en
tifyin
g
cro
s
s
-
co
u
n
tr
y
fin
tech
…
(
K
a
yla
Zh
a
fir
a
A
r
d
in
o
v)
529
T
ab
le
1
0
.
Asp
ec
t
-
lev
el
class
if
ica
tio
n
p
er
f
o
r
m
an
ce
o
n
o
v
er
la
p
p
in
g
asp
ec
ts
A
sp
e
c
t
C
o
u
n
t
r
y
S
a
mp
l
e
s
A
c
c
u
r
a
c
y
P
r
e
c
i
s
i
o
n
R
e
c
a
l
l
F1
-
sc
o
r
e
Tr
a
n
s
a
c
t
i
o
n
a
n
d
p
a
y
me
n
t
ID
1
,
8
9
1
0
.
9
7
7
0
.
9
8
5
0
.
9
9
0
.
9
8
7
TH
2
,
6
9
5
0
.
9
8
8
0
.
9
7
4
0
.
9
4
7
0
.
9
6
1
P
r
o
mo
t
i
o
n
/
v
o
u
c
h
e
r
ID
1
,
2
5
4
0
.
9
7
6
0
.
9
8
3
0
.
9
8
8
0
.
9
8
5
TH
1
,
7
5
7
0
.
9
8
9
0
.
9
9
2
0
.
9
9
3
0
.
9
9
3
P
r
o
c
e
ss
t
i
me
ID
1
,
7
8
7
0
.
9
6
8
0
.
9
3
0
.
9
2
5
0
.
9
2
8
TH
1
,
3
5
2
0
.
9
9
3
0
.
9
8
9
0
.
9
8
6
0
.
9
8
7
3
.
2
.
Cro
s
s
-
co
un
t
ry
perc
ept
i
o
n g
a
p a
na
ly
s
is
T
h
e
f
in
al
s
tag
e
o
f
k
n
o
wled
g
e
in
ter
p
r
etatio
n
with
in
th
e
KDD
f
r
am
ewo
r
k
is
to
co
n
d
u
ct
a
co
m
p
ar
ativ
e
s
en
tim
en
t
an
aly
s
is
o
n
o
v
er
la
p
p
in
g
asp
ec
ts
o
f
s
er
v
ice
b
et
wee
n
I
n
d
o
n
esia
(
I
D)
an
d
T
h
ailan
d
(
T
H)
.
T
h
is
an
aly
s
is
aim
s
to
id
en
tify
th
e
p
er
ce
p
tio
n
g
ap
am
o
n
g
u
s
er
s
r
e
g
ar
d
in
g
th
e
q
u
ality
o
f
Sh
o
p
ee
Pay
s
er
v
ices
ac
r
o
s
s
two
m
ar
k
ets.
A
v
is
u
aliza
tio
n
o
f
th
e
co
m
p
a
r
is
o
n
o
f
s
en
tim
e
n
t
p
r
o
p
o
r
tio
n
s
u
s
in
g
a
Gr
o
u
p
e
d
1
0
0
%
s
tack
ed
b
ar
ch
ar
t
is
p
r
esen
ted
in
Fig
u
r
e
6.
I
n
tr
an
s
ac
tio
n
an
d
p
ay
m
e
n
t
,
a
s
ig
n
if
ican
t
g
ap
was
o
b
s
er
v
ed
:
I
n
d
o
n
esian
u
s
er
s
s
h
o
w
ed
9
3
.
0
%
p
o
s
itiv
e
s
en
tim
en
t,
wh
ile
T
h
ai
u
s
er
s
ex
p
r
ess
ed
o
n
ly
3
2
.
4
%
p
o
s
itiv
e
s
en
tim
en
t
(
6
7
.
6
%
n
eg
ativ
e)
.
T
h
is
in
d
icate
s
n
o
tab
le
f
r
ictio
n
with
p
a
y
m
en
t
f
ea
tu
r
es
i
n
T
h
ailan
d
,
alig
n
in
g
with
r
esear
ch
o
n
d
ig
ital
p
a
y
m
en
t
f
r
ictio
n
in
d
ev
elo
p
in
g
m
ar
k
ets
w
h
er
e
e
-
wallet
an
d
b
an
k
in
g
i
n
f
r
astru
ctu
r
e
in
ter
o
p
er
a
b
ilit
y
s
ig
n
if
ican
tly
af
f
ec
ts
s
atis
f
ac
tio
n
[
4
2
]
.
T
h
aila
n
d
’
s
f
r
ag
m
en
te
d
p
ay
m
en
t
lan
d
s
ca
p
e,
d
esp
ite
th
e
n
atio
n
al
Pro
m
p
tPay
s
y
s
tem
,
m
a
y
cr
ea
te
in
teg
r
atio
n
f
r
ictio
n
th
at
I
n
d
o
n
esian
u
s
er
s
o
p
er
atin
g
with
in
a
m
o
r
e
co
n
s
o
lid
ated
e
-
c
o
m
m
er
ce
ec
o
s
y
s
tem
d
o
n
o
t
ex
p
er
ie
n
ce
to
th
e
s
am
e
d
eg
r
ee
.
I
n
p
r
o
m
o
t
i
o
n
/
v
o
u
c
h
e
r
,
b
o
t
h
co
u
n
t
r
i
e
s
s
h
o
w
e
d
p
r
e
d
o
m
i
n
a
n
tly
p
o
s
i
ti
v
e
s
e
n
t
i
m
e
n
t
,
wi
t
h
T
h
a
i
u
s
e
r
s
a
t
9
0
.
6
%
a
n
d
I
n
d
o
n
e
s
i
a
n
u
s
e
r
s
a
t
8
0
.
8
%
p
o
s
i
t
i
v
e
.
T
h
e
r
e
m
a
i
n
i
n
g
g
a
p
m
a
y
r
e
f
l
e
c
t
I
n
d
o
n
e
s
i
a
’
s
h
i
g
h
l
y
c
o
m
p
e
t
i
t
i
v
e
e
-
w
a
l
l
et
e
c
o
s
y
s
t
e
m
(
G
o
P
a
y
,
OV
O
,
D
AN
A
,
S
h
o
p
e
e
P
a
y
)
,
w
h
ic
h
g
e
n
e
r
a
t
e
s
h
i
g
h
e
r
p
r
o
m
o
t
i
o
n
a
l
e
x
p
e
c
t
a
ti
o
n
s
[
4
3
]
.
I
n
p
r
o
ce
s
s
tim
e
,
I
n
d
o
n
esia
s
h
o
wed
7
8
.
2
% n
eg
ativ
e
s
en
tim
en
t
co
m
p
a
r
ed
to
T
h
ailan
d
’
s
m
o
r
e
b
alan
ce
d
s
p
lit
(
5
0
.
8
%
n
e
g
ativ
e,
4
9
.
2
%
p
o
s
itiv
e)
.
T
h
is
m
ay
b
e
li
n
k
ed
to
v
ar
ia
b
le
in
ter
n
et
c
o
n
n
ec
tiv
ity
ac
r
o
s
s
th
e
I
n
d
o
n
esian
ar
ch
ip
elag
o
an
d
h
ig
h
er
s
er
v
er
lo
ad
s
d
u
r
in
g
p
ea
k
s
h
o
p
p
i
n
g
ev
e
n
ts
[
4
4
]
.
Ov
er
all,
th
ese
f
in
d
in
g
s
p
r
o
v
id
e
em
p
ir
ical
e
v
id
en
ce
th
at
u
s
er
ex
p
ec
tatio
n
s
ar
e
d
e
ep
ly
r
o
o
ted
in
lo
ca
l
in
f
r
astru
ctu
r
e
an
d
cu
ltu
r
al
co
n
tex
ts
.
B
y
b
r
id
g
i
n
g
th
e
g
ap
b
etwe
en
h
ig
h
-
lev
el
s
en
tim
en
t
an
d
s
p
ec
if
ic
s
er
v
ice
attr
ib
u
tes,
th
is
s
tu
d
y
o
f
f
er
s
a
d
ata
-
d
r
iv
e
n
f
r
am
ewo
r
k
f
o
r
r
e
g
io
n
al
f
in
tech
p
r
o
v
id
er
s
to
r
ef
i
n
e
o
p
er
atio
n
al
s
tr
ateg
ies
ac
r
o
s
s
d
iv
er
s
e
So
u
th
ea
s
t
Asi
an
ec
o
s
y
s
tem
s
.
Fig
u
r
e
6
.
C
o
m
p
a
r
is
o
n
o
f
p
o
s
it
iv
e
an
d
n
eg
ativ
e
s
en
tim
en
t
p
r
o
p
o
r
tio
n
s
o
n
o
v
e
r
lap
p
in
g
s
er
v
ic
e
asp
ec
ts
b
etwe
en
I
n
d
o
n
esian
an
d
T
h
ai
Sh
o
p
ee
P
ay
u
s
er
s
3.
3
.
T
ec
hn
ica
l st
re
ng
t
h o
f
D
is
t
ilB
E
R
T
a
nd
m
o
del e
v
a
lua
t
io
n
T
h
e
Dis
tilB
E
R
T
m
o
d
el
ac
h
iev
ed
9
7
–
9
8
%
ac
cu
r
ac
y
,
d
em
o
n
s
tr
atin
g
th
e
t
r
an
s
f
o
r
m
er
ar
c
h
itectu
r
e
’
s
ef
f
ec
tiv
en
ess
f
o
r
in
f
o
r
m
al
f
in
tech
r
ev
iew
lan
g
u
a
g
e.
Ge
n
er
aliza
tio
n
was
en
s
u
r
ed
b
y
lim
itin
g
f
in
e
-
t
u
n
in
g
to
th
r
ee
ep
o
ch
s
with
a
co
n
s
er
v
ativ
e
lear
n
in
g
r
ate
(
1
×1
0
⁻⁵)
an
d
s
elec
tin
g
ch
ec
k
p
o
in
ts
b
ased
o
n
lo
west
v
alid
atio
n
lo
s
s
.
C
o
n
s
is
ten
t
F1
-
s
co
r
es
ac
r
o
s
s
all
s
en
tim
en
t
class
es
co
n
f
ir
m
t
h
at
r
a
n
d
o
m
o
v
er
s
am
p
l
in
g
ad
d
r
ess
ed
class
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
5
2
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
43
,
No
.
2
,
Au
g
u
s
t
20
26
:
522
-
5
3
3
530
im
b
alan
ce
s
with
o
u
t
ar
tific
ial
b
ias.
T
h
e
n
ar
r
o
w
co
n
f
id
e
n
c
e
in
ter
v
als
in
T
ab
le
1
1
f
u
r
t
h
er
p
r
o
v
id
e
s
tr
o
n
g
ev
id
en
ce
th
at
th
e
p
er
f
o
r
m
a
n
ce
is
s
tati
s
tically
r
o
b
u
s
t a
n
d
n
o
t
an
ar
tifa
ct
o
f
a
f
av
o
r
ab
l
e
d
ata
s
p
lit.
T
ab
le
1
1
.
Statis
tical
p
er
f
o
r
m
a
n
ce
an
d
r
o
b
u
s
tn
ess
m
etr
ics
C
o
u
n
t
r
y
Te
st
sam
p
l
e
s
i
z
e
(
n
)
Te
st
i
n
g
a
c
c
u
r
a
c
y
9
5
%
C
o
n
f
i
d
e
n
c
e
i
n
t
e
r
v
a
l
(
C
I
)
I
n
d
o
n
e
si
a
1
7
,
8
6
3
9
7
.
6
5
%
[
9
7
.
4
3
%
,
9
7
.
8
7
%]
Th
a
i
l
a
n
d
1
5
,
5
9
0
9
8
.
3
8
%
[
9
8
.
1
8
%
,
9
8
.
5
8
%]
Alth
o
u
g
h
k
-
f
o
ld
cr
o
s
s
-
v
alid
atio
n
an
d
ab
latio
n
s
tu
d
ies
(
e.
g
.
,
co
m
p
ar
in
g
ag
ai
n
s
t
B
E
R
T
-
b
ase,
R
o
B
E
R
T
a,
o
r
lo
g
is
tic
r
eg
r
ess
io
n
)
wer
e
n
o
t
co
n
d
u
cted
d
u
e
to
co
m
p
u
tatio
n
al
co
n
s
tr
ain
ts
(
~2
4
h
o
u
r
s
p
er
cy
cle)
,
th
e
p
ip
elin
e
’
s
r
eliab
ilit
y
is
s
u
p
p
o
r
ted
b
y
co
n
s
is
ten
t
p
er
f
o
r
m
a
n
ce
ac
r
o
s
s
t
wo
lin
g
u
is
tically
d
is
tin
ct
d
ataset
s
an
d
alig
n
m
en
t
with
c
r
o
s
s
-
lin
g
u
al
s
en
tim
en
t
an
aly
s
is
s
tan
d
ar
d
s
[
3
3
]
,
[
4
1
]
.
Fu
tu
r
e
wo
r
k
s
h
o
u
ld
i
n
co
r
p
o
r
ate
b
aselin
e
co
m
p
ar
is
o
n
s
to
b
etter
co
n
te
x
tu
alize
th
e
m
o
d
el
’
s
co
n
tr
ib
u
tio
n
.
3.
4
.
Str
a
t
eg
ic
bu
s
ines
s
im
pli
ca
t
io
ns
a
nd
re
co
m
m
enda
t
i
o
ns
T
h
e
f
o
llo
win
g
ar
e
d
ir
ec
tio
n
al
h
y
p
o
th
eses
in
f
o
r
m
e
d
b
y
s
en
tim
en
t
p
atter
n
s
,
n
o
t
ca
u
s
al
claim
s
.
T
h
e
s
en
tim
en
t d
ata
id
e
n
tifie
s
wh
at
u
s
er
s
ar
e
d
is
s
atis
f
ied
wit
h
b
u
t c
an
n
o
t e
s
tab
lis
h
wh
y
wit
h
o
u
t su
p
p
lem
en
tar
y
d
ata.
3.
4
.
1
.
O
ptim
i
za
t
io
n o
f
T
ha
ila
nd
’
s
pa
y
m
ent
ec
o
s
y
s
t
em
T
h
e
tr
an
s
ac
tio
n
an
d
p
a
y
m
en
t
g
ap
,
6
7
.
6
%
n
e
g
ativ
e
in
T
h
ail
an
d
v
er
s
u
s
7
.
0
%
in
I
n
d
o
n
esia
in
d
icate
s
s
ig
n
if
ican
t
tr
an
s
ac
tio
n
ex
p
er
i
en
ce
o
b
s
tacle
s
in
T
h
ailan
d
.
T
h
e
n
eg
ativ
e
s
en
tim
en
t
s
u
g
g
ests
p
o
ten
tial
f
r
ictio
n
p
o
in
ts
in
p
ay
m
en
t
o
r
k
n
o
w
y
o
u
r
c
u
s
to
m
er
(
KYC
)
p
r
o
ce
s
s
es
,
th
o
u
g
h
ca
u
s
al
attr
ib
u
tio
n
ca
n
n
o
t
b
e
co
n
clu
s
iv
ely
estab
lis
h
ed
.
T
h
is
r
ef
lects
tech
n
o
s
tr
ess
,
wh
er
e
u
s
er
s
p
er
ce
iv
e
tech
n
o
lo
g
y
b
u
r
d
en
as
ex
ce
e
d
in
g
th
eir
ad
a
p
tiv
e
ca
p
ac
ity
[
4
5
]
.
Dir
ec
tio
n
al
r
ec
o
m
m
en
d
atio
n
:
Sh
o
p
ee
Pay
s
h
o
u
ld
in
v
esti
g
ate
wh
eth
er
d
ee
p
er
Pro
m
p
tPay
in
teg
r
atio
n
an
d
lo
ca
l
b
an
k
i
n
g
ap
p
licatio
n
p
r
o
g
r
am
m
in
g
in
ter
f
ac
e
(
API
)
co
n
n
ec
tiv
it
y
co
u
ld
m
in
im
ize
tr
an
s
ac
tio
n
f
ailu
r
es,
an
d
wh
eth
er
s
im
p
lify
in
g
th
e
e
-
KYC
jo
u
r
n
ey
th
r
o
u
g
h
au
to
m
ate
d
o
p
tical
ch
ar
ac
te
r
r
ec
o
g
n
itio
n
(
OC
R
)
f
o
r
T
h
ai
I
D
ca
r
d
s
m
ay
r
e
d
u
ce
f
r
u
s
tr
atio
n
an
d
l
o
wer
ch
u
r
n
r
ates to
war
d
lo
ca
l c
o
m
p
etito
r
s
.
3
.
4
.
2
.
E
nh
a
ncing
v
o
ucher
co
m
pet
it
iv
eness
in I
n
do
nes
ia
I
n
ter
m
s
o
f
Pro
m
o
tio
n
/Vo
u
c
h
er
s
,
b
o
t
h
co
u
n
tr
ies
s
h
o
w
p
o
s
itiv
e
s
en
tim
en
t,
th
o
u
g
h
T
h
ai
u
s
er
s
ar
e
m
o
r
e
s
atis
f
ied
(
9
0
.
6
%
p
o
s
itiv
e)
th
an
I
n
d
o
n
esian
u
s
er
s
(
8
0
.
8
%
p
o
s
itiv
e)
.
Dir
ec
tio
n
al
r
ec
o
m
m
en
d
atio
n
:
t
o
b
r
id
g
e
th
e
s
atis
f
ac
tio
n
g
ap
,
S
h
o
p
ee
Pay
I
n
d
o
n
esia
c
o
u
ld
co
n
s
id
er
ad
o
p
tin
g
a
‘
tr
an
s
p
ar
e
n
cy
-
f
ir
s
t
’
p
r
o
m
o
tio
n
s
tr
ateg
y
.
T
h
is
m
ig
h
t
in
v
o
lv
e
s
im
p
lify
in
g
th
e
ter
m
s
an
d
co
n
d
itio
n
s
(
T
&
C
)
v
is
ib
le
to
u
s
er
s
an
d
im
p
lem
en
tin
g
r
ea
l
-
tim
e
v
o
u
ch
e
r
av
ailab
ilit
y
n
o
tific
atio
n
s
.
T
h
ese
r
ec
o
m
m
en
d
atio
n
s
ar
e
b
ased
o
n
th
e
o
b
s
er
v
ed
s
en
tim
en
t
p
atter
n
s
an
d
wo
u
ld
b
en
ef
it
f
r
o
m
f
u
r
th
er
v
alid
atio
n
t
h
r
o
u
g
h
u
s
er
s
u
r
v
e
y
s
o
r
A
/B
tes
tin
g
b
ef
o
r
e
im
p
lem
en
tatio
n
.
3
.
4
.
3
.
I
nfr
a
s
t
ruct
ure
s
ca
lin
g
a
nd
la
t
ency
m
a
na
g
em
ent
pr
o
ce
s
s
t
im
e
Pro
ce
s
s
tim
e
is
a
cr
itical
is
s
u
e
f
o
r
b
o
th
co
u
n
tr
ies,
b
u
t
th
e
u
r
g
e
n
cy
is
h
i
g
h
er
i
n
I
n
d
o
n
esia,
wh
er
e
7
8
.
2
%
o
f
r
esp
o
n
d
en
ts
ex
p
r
ess
ed
n
eg
ativ
e
s
en
tim
en
t
co
m
p
ar
ed
to
T
h
ailan
d
’
s
5
0
.
8
%.
Dir
ec
tio
n
al
r
ec
o
m
m
en
d
atio
n
:
th
e
tec
h
n
ic
al
team
s
h
o
u
ld
in
v
esti
g
ate
b
a
ck
en
d
s
er
v
e
r
s
ca
lab
ilit
y
an
d
laten
cy
m
itig
atio
n
,
s
p
ec
if
ically
f
o
r
to
p
-
u
p
a
n
d
r
e
al
-
tim
e
p
ay
m
en
t
co
n
f
ir
m
atio
n
s
.
T
h
e
s
en
tim
en
t
d
ata
alo
n
e
c
an
n
o
t
id
en
tif
y
th
e
s
p
ec
if
ic
tech
n
ical
r
o
o
t
ca
u
s
es,
b
u
t
th
e
v
o
lu
m
e
a
n
d
in
ten
s
ity
o
f
c
o
m
p
lain
ts
s
u
g
g
est
th
at
i
n
f
r
astru
ctu
r
e
ca
p
ac
ity
an
d
q
u
er
y
p
er
f
o
r
m
an
ce
war
r
a
n
t p
r
io
r
ity
in
v
esti
g
atio
n
.
3.
5
.
T
heo
re
t
ica
l i
m
pli
ca
t
io
ns
T
h
eo
r
etica
lly
,
th
is
s
tu
d
y
o
f
f
er
s
a
n
ew
p
er
s
p
ec
tiv
e
o
n
th
e
tr
an
s
ac
tio
n
al
m
o
d
el
o
f
tec
h
n
o
s
t
r
ess
.
Ou
r
f
in
d
i
n
g
s
s
u
g
g
est
th
at
tec
h
n
o
s
tr
ess
is
n
o
t
a
u
n
iv
e
r
s
al
is
s
u
e;
r
ath
er
,
it
is
s
h
ap
ed
b
y
r
e
g
io
n
al
in
f
r
astru
ct
u
r
e
an
d
lo
ca
l
cu
ltu
r
al
co
n
tex
ts
.
Fo
r
in
s
tan
ce
,
wh
ile
T
h
ai
u
s
er
s
ex
p
er
ien
ce
h
i
g
h
er
s
tr
e
s
s
lev
els
r
eg
ar
d
in
g
tr
an
s
ac
tio
n
r
eliab
ilit
y
,
I
n
d
o
n
e
s
ian
u
s
er
s
ar
e
m
o
r
e
af
f
ec
ted
b
y
s
er
v
ice
s
p
ee
d
an
d
laten
c
y
.
T
h
is
d
e
m
o
n
s
tr
ates
th
at
a
u
s
er
’
s
ex
p
er
ien
ce
with
tech
n
o
lo
g
y
-
r
elate
d
s
tr
ess
is
d
ee
p
ly
tied
to
th
eir
s
p
ec
if
ic
d
ig
ital
en
v
ir
o
n
m
en
t.
B
y
co
m
p
ar
in
g
th
ese
two
m
ar
k
ets,
th
is
s
tu
d
y
h
ig
h
lig
h
ts
th
at
u
s
er
b
eh
av
io
r
th
e
o
r
ies
ca
n
n
o
t
f
o
llo
w
a
‘
one
-
s
ize
-
f
its
-
all
’
ap
p
r
o
ac
h
,
p
ar
ticu
la
r
ly
with
in
th
e
d
iv
er
s
e
So
u
t
h
ea
s
t A
s
ian
f
in
tech
lan
d
s
ca
p
e.
3.
6
.
L
im
it
a
t
io
ns
a
nd
f
uture
wo
rk
T
h
is
s
tu
d
y
h
as
s
ev
er
al
lim
itatio
n
s
.
First,
th
e
p
ip
elin
e
em
p
lo
y
s
to
p
ic
-
in
f
o
r
m
ed
s
en
tim
en
t
class
if
icatio
n
r
ath
er
th
an
ca
n
o
n
ical
asp
ec
t
-
ter
m
-
lev
el
AB
SA,
lim
itin
g
g
r
an
u
lar
ity
.
Seco
n
d
,
p
s
eu
d
o
-
lab
elin
g
in
tr
o
d
u
ce
s
cir
cu
lar
ev
al
u
atio
n
r
is
k
,
th
o
u
g
h
m
itig
ated
b
y
m
an
u
al
v
alid
atio
n
(
κ
>
0
.
9
7
)
.
T
h
ir
d
,
E
n
g
lis
h
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4
7
5
2
Tr
a
n
s
fo
r
mer
-
b
a
s
ed
s
en
timen
t
mo
d
elin
g
fo
r
id
en
tifyin
g
cro
s
s
-
co
u
n
tr
y
fin
tech
…
(
K
a
yla
Zh
a
fir
a
A
r
d
in
o
v)
531
tr
an
s
latio
n
m
ay
f
latten
r
eg
io
n
al
n
u
an
ce
s
s
u
ch
as
I
n
d
o
n
e
s
ian
s
lan
g
o
r
T
h
ai
em
o
tio
n
a
l
p
ar
ticles.
Fo
u
r
th
,
s
em
i
-
m
an
u
al
to
p
ic
alig
n
m
en
t
(
κ
=
0
.
9
2
)
m
ay
n
o
t
g
en
er
ali
ze
with
o
u
t
ad
ap
tatio
n
.
Fifth
,
n
o
b
aselin
e
m
o
d
el
co
m
p
ar
i
s
o
n
s
wer
e
co
n
d
u
cte
d
d
u
e
to
c
o
m
p
u
tatio
n
al
co
n
s
tr
ai
n
ts
.
Fu
tu
r
e
r
esear
ch
s
h
o
u
ld
ex
p
lo
r
e
m
u
ltil
in
g
u
al
m
o
d
els
(
m
B
E
R
T
,
XL
M
-
R
o
B
E
R
T
a)
f
o
r
d
i
r
ec
t
an
aly
s
is
o
n
o
r
ig
in
al
tex
ts
,
in
c
o
r
p
o
r
ate
lar
g
er
h
u
m
an
-
a
n
n
o
tated
ev
alu
atio
n
s
ets
as
p
r
im
ar
y
b
en
ch
m
ar
k
s
,
an
d
ex
p
a
n
d
to
ot
h
er
So
u
th
ea
s
t
Asi
an
m
ar
k
ets
(
Vietn
am
,
Ma
lay
s
ia)
t
o
v
alid
ate
g
en
er
aliza
b
ilit
y
.
Ad
d
itio
n
a
lly
,
in
teg
r
atin
g
th
is
p
ip
elin
e
in
to
liv
e
f
in
tec
h
d
a
s
h
b
o
ar
d
s
an
d
ex
p
lo
r
in
g
lar
g
e
lan
g
u
a
g
e
m
o
d
el
(
LLM
)
-
b
a
s
ed
p
r
o
m
p
tin
g
f
o
r
s
im
u
ltan
eo
u
s
asp
ec
t
-
p
o
lar
ity
e
x
tr
ac
tio
n
[
2
0
]
r
ep
r
esen
t stra
te
g
ic
o
p
p
o
r
tu
n
ities
.
4.
CO
NCLU
SI
O
N
T
h
is
s
tu
d
y
id
en
tifie
d
Sh
o
p
ee
P
ay
p
er
ce
p
tio
n
g
ap
s
b
etwe
en
I
n
d
o
n
esian
an
d
T
h
ai
u
s
er
s
b
y
ap
p
ly
in
g
a
KDD
f
r
am
ewo
r
k
in
teg
r
ated
with
f
in
e
-
tu
n
ed
Dis
tilB
E
R
T
,
ac
h
iev
in
g
ac
cu
r
ac
ies
o
f
9
7
.
6
5
%
an
d
9
8
.
3
8
%
r
esp
ec
tiv
ely
.
T
h
ese
r
esu
lts
d
e
m
o
n
s
tr
ate
th
at
tr
a
n
s
f
o
r
m
er
a
r
ch
itectu
r
e
with
r
an
d
o
m
o
v
er
s
am
p
lin
g
p
r
o
v
i
d
es
a
co
n
s
is
ten
t
ap
p
r
o
ac
h
f
o
r
s
en
tim
en
t
class
if
icatio
n
in
d
ig
ital
f
in
an
ce
,
th
o
u
g
h
r
elian
ce
o
n
p
s
eu
d
o
-
lab
elin
g
an
d
E
n
g
lis
h
tr
an
s
latio
n
m
ay
in
tr
o
d
u
ce
s
em
an
tic
b
iases
.
Ma
n
u
a
l
v
ali
d
atio
n
co
n
f
ir
m
ed
ap
p
r
o
x
im
ately
9
9
%
lab
el
ag
r
ee
m
en
t
(
C
o
h
e
n
’
s
Kap
p
a
>
0
.
9
7
)
,
estab
lis
h
in
g
r
o
b
u
s
t
r
eliab
ilit
y
.
B
ey
o
n
d
tech
n
ica
l
p
er
f
o
r
m
a
n
ce
,
th
e
p
r
im
ar
y
co
n
tr
ib
u
tio
n
is
em
p
ir
ical
ev
id
en
ce
th
at
T
ec
h
n
o
s
tr
ess
is
a
lo
ca
lized
p
h
en
o
m
e
n
o
n
m
o
d
er
ated
b
y
r
eg
io
n
al
in
f
r
a
s
tr
u
ctu
r
e
m
atu
r
ity
an
d
s
o
cio
-
c
u
ltu
r
al
e
x
p
ec
t
atio
n
s
,
n
o
t
a
u
n
if
o
r
m
r
esp
o
n
s
e.
B
y
r
e
v
ea
lin
g
d
iv
er
g
en
t
p
er
ce
p
tio
n
s
o
f
th
e
s
am
e
p
latf
o
r
m
,
th
is
s
tu
d
y
ad
d
s
a
cr
itical
co
m
p
ar
ativ
e
d
im
en
s
io
n
to
b
e
h
av
io
r
al
tech
n
o
lo
g
y
ad
o
p
tio
n
m
o
d
els,
d
em
o
n
s
tr
atin
g
th
at
s
in
g
le
-
m
ar
k
et
ev
alu
atio
n
s
ar
e
i
n
s
u
f
f
i
cien
t
f
o
r
em
er
g
i
n
g
So
u
th
ea
s
t A
s
ian
m
ar
k
ets.
ACK
NO
WL
E
DG
M
E
N
T
S
T
h
e
au
th
o
r
s
wo
u
ld
lik
e
to
e
x
p
r
ess
th
eir
s
in
ce
r
e
g
r
atitu
d
e
to
th
e
I
n
f
o
r
m
atio
n
Sy
s
tem
Stu
d
y
Pro
g
r
am
,
Sch
o
o
l
o
f
I
n
d
u
s
tr
ial
E
n
g
in
e
er
in
g
,
T
elk
o
m
Un
iv
er
s
ity
,
f
o
r
p
r
o
v
id
in
g
th
e
ess
en
tial
f
ac
ilit
ies,
ac
ad
em
ic
en
v
ir
o
n
m
en
t,
an
d
ad
m
i
n
is
tr
ativ
e
s
u
p
p
o
r
t th
at
m
ad
e
th
is
r
esear
ch
p
o
s
s
ib
le.
F
UNDING
I
NF
O
R
M
A
T
I
O
N
T
h
is
r
esear
ch
r
ec
ei
v
ed
n
o
s
p
e
cif
ic
g
r
an
t
f
r
o
m
an
y
f
u
n
d
in
g
a
g
en
cy
i
n
th
e
p
u
b
lic,
co
m
m
er
ci
al,
o
r
n
o
t
-
f
o
r
-
p
r
o
f
it secto
r
s
.
AUTHO
R
CO
NT
RI
B
UT
I
O
NS ST
A
T
E
M
E
N
T
T
h
is
jo
u
r
n
al
u
s
es
th
e
C
o
n
tr
ib
u
to
r
R
o
les
T
ax
o
n
o
m
y
(
C
R
ed
iT)
to
r
ec
o
g
n
ize
in
d
iv
id
u
al
au
th
o
r
co
n
tr
ib
u
tio
n
s
,
r
ed
u
ce
au
th
o
r
s
h
ip
d
is
p
u
tes,
an
d
f
ac
ilit
ate
co
llab
o
r
atio
n
.
Na
m
e
o
f
Aut
ho
r
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
Kay
la
Z
h
af
ir
a
Ar
d
in
o
v
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
Mu
h
ar
d
i Sap
u
t
r
a
✓
✓
✓
✓
✓
✓
R
is
k
a
Yan
u
Fa
’
r
if
ah
✓
✓
✓
✓
✓
✓
C
:
C
o
n
c
e
p
t
u
a
l
i
z
a
t
i
o
n
M
:
M
e
t
h
o
d
o
l
o
g
y
So
:
So
f
t
w
a
r
e
Va
:
Va
l
i
d
a
t
i
o
n
Fo
:
Fo
r
mal
a
n
a
l
y
s
i
s
I
:
I
n
v
e
s
t
i
g
a
t
i
o
n
R
:
R
e
so
u
r
c
e
s
D
:
D
a
t
a
C
u
r
a
t
i
o
n
O
:
W
r
i
t
i
n
g
-
O
r
i
g
i
n
a
l
D
r
a
f
t
E
:
W
r
i
t
i
n
g
-
R
e
v
i
e
w
&
E
d
i
t
i
n
g
Vi
:
Vi
su
a
l
i
z
a
t
i
o
n
Su
:
Su
p
e
r
v
i
s
i
o
n
P
:
P
r
o
j
e
c
t
a
d
mi
n
i
st
r
a
t
i
o
n
Fu
:
Fu
n
d
i
n
g
a
c
q
u
i
si
t
i
o
n
CO
NF
L
I
C
T
O
F
I
N
T
E
R
E
S
T
ST
A
T
E
M
E
NT
T
h
e
au
t
h
o
r
s
d
ec
lar
e
th
at
th
e
y
h
av
e
n
o
k
n
o
wn
c
o
m
p
etin
g
f
in
an
cial
in
ter
ests
o
r
p
er
s
o
n
al
r
el
atio
n
s
h
ip
s
th
at
co
u
ld
h
av
e
ap
p
ea
r
ed
t
o
in
f
lu
en
ce
th
e
wo
r
k
r
e
p
o
r
te
d
in
t
h
is
p
ap
er
.
Au
th
o
r
s
s
tate
n
o
co
n
f
lict o
f
in
ter
est.
DATA AV
AI
L
AB
I
L
I
T
Y
T
h
e
d
ataset
u
s
ed
in
th
is
s
tu
d
y
co
n
s
is
ts
o
f
u
s
er
r
ev
iews
s
cr
ap
ed
f
r
o
m
th
e
Go
o
g
le
Play
Sto
r
e
p
latf
o
r
m
o
n
Octo
b
er
2
0
,
2
0
2
5
.
T
h
e
r
a
w
d
ata
an
d
clea
n
e
d
to
k
e
n
s
ar
e
ar
ch
iv
e
d
an
d
ca
n
b
e
m
ad
e
av
ailab
le
b
y
t
h
e
co
r
r
esp
o
n
d
in
g
a
u
th
o
r
u
p
o
n
r
ea
s
o
n
ab
le
r
eq
u
est f
o
r
ac
ad
e
m
ic
p
u
r
p
o
s
es.
Evaluation Warning : The document was created with Spire.PDF for Python.