I
A
E
S
I
n
t
e
r
n
at
io
n
al
Jou
r
n
al
of
A
r
t
if
ic
ia
l
I
n
t
e
ll
ig
e
n
c
e
(
I
J
-
AI
)
V
ol
.
14
, N
o.
3
,
J
une
2025
, pp.
1829
~
1838
I
S
S
N
:
2252
-
8938
,
D
O
I
:
10.11591/
ij
a
i.
v
14
.i
3
.pp
1829
-
1838
1829
Jou
r
n
al
h
om
e
page
:
ht
tp
:
//
ij
ai
.
ia
e
s
c
or
e
.c
om
T
h
e
i
n
f
l
u
e
n
c
e
of
se
n
t
i
m
e
n
t
an
al
ysi
s i
n
e
n
h
an
c
i
n
g
e
ar
l
y w
ar
n
i
n
g
sys
t
e
m
m
od
e
l
f
or
c
r
e
d
i
t
r
i
sk
m
i
t
i
gat
i
on
A
n
ge
l
K
ar
e
n
t
ia
1
, D
e
r
w
in
S
u
h
ar
t
on
o
2
1
C
om
put
e
r
S
c
i
e
nc
e
D
e
pa
r
t
m
e
nt
,
B
I
N
U
S
G
r
a
dua
t
e
P
r
ogr
a
m
-
M
a
s
t
e
r
of
C
om
put
e
r
S
c
i
e
nc
e
,
B
i
na
N
us
a
nt
a
r
a
U
ni
v
e
r
s
i
t
y, J
a
ka
r
t
a
, I
ndone
s
i
a
2
C
om
put
e
r
S
c
i
e
nc
e
D
e
pa
r
t
m
e
nt
, S
c
hool
of
C
om
put
e
r
S
c
i
e
nc
e
,
B
i
na
N
u
s
a
nt
a
r
a
U
ni
ve
r
s
i
t
y, J
a
ka
r
t
a
, I
ndone
s
i
a
A
r
t
ic
le
I
n
f
o
A
B
S
T
R
A
C
T
A
r
ti
c
le
h
is
to
r
y
:
R
e
c
e
iv
e
d
A
ug
8, 2024
R
e
vi
s
e
d
F
e
b 4, 2025
A
c
c
e
pt
e
d
M
a
r
15, 2025
One
important
source
of
bank
income
is
interest
income
from
credit
activit
ies,
another
part
of
which
is
obtained
from
fee
-
based
income.
Rapid
credit
growth
is
directly
proportional
to
an
increase
in
potential
cre
dit
risk
(counterparty
default).
In
addition
to
comprehensive
credit
assessment
at
the
initial
stage
of
credit
initiation,
banks
need
to
monitor
the
condit
ion
of
existin
g
debtors.
Empirical
ly,
difficul
ties
in
handlin
g
non
-
performing
loans
often
occur
due
to
delays
in
detection
and
preparation
of
act
ion
plans.
In
this
case,
losses
due
to
non
-
performing
loans
can
have
implications
f
or
the
bank'
s
reputation
and
worsen
its
financial
performance.
This
research
aims
to
determine
the
effect
of
sentiment
analysis
(external
sentiment
pre
diction
model
[positive,
neutral,
and
negative]
with
certain
keywords)
on
th
e
level
of
accuracy
of
the
early
warning
system
(EWS)
model
in
predicti
ng
the
credit
quality
of
bank
debtors
in
the
coming
months.
This
study
fou
nd
that
upgrading
EWS
with
sentiment
analysis
will
give
b
etter
ac
curacy
levels
compared
to
traditi
onal
EWS
models.
In
additio
n,
the
predicti
ve
power
of
EWS
(traditional
and
upgraded)
is
inversely
proportional
to
the
prediction
period,
the
longer
the
target
prediction
time,
and
the
less
predictive
po
wer
of
the EWS mode
l.
K
e
y
w
o
r
d
s
:
C
r
e
di
t
qua
li
ty
E
a
r
ly
w
a
r
ni
ng s
ys
te
m
F
or
w
a
r
d
-
lo
oki
ng
M
ode
l
a
c
c
ur
a
c
y
S
e
nt
im
e
nt
a
na
ly
s
is
This is an
open
acce
ss artic
le unde
r the
CC BY
-
SA
license.
C
or
r
e
s
pon
di
n
g A
u
th
or
:
A
nge
l
K
a
r
e
nt
ia
C
om
put
e
r
S
c
ie
nc
e
D
e
pa
r
tm
e
nt
, B
I
N
U
S
G
r
a
dua
te
P
r
ogr
a
m
-
M
a
s
te
r
of
C
om
put
e
r
S
c
ie
nc
e
B
in
a
N
us
a
nt
a
r
a
U
ni
v
e
r
s
it
y
J
a
ka
r
ta
11480, I
ndone
s
ia
E
m
a
il
:
a
nge
l
.ka
r
e
nt
ia
@
bi
nus
.a
c
.i
d
1.
I
N
T
R
O
D
U
C
T
I
O
N
T
he
c
or
ona
vi
r
us
di
s
e
a
s
e
2019
(
C
O
V
I
D
-
19)
pa
nde
m
ic
ha
s
ha
d
a
m
a
s
s
iv
e
im
pa
c
t
w
he
r
e
a
ll
c
ount
r
ie
s
ha
ve
im
pl
e
m
e
nt
e
d
a
c
ti
vi
ty
r
e
s
tr
ic
ti
ons
(
lo
c
kdown
s
)
to
pr
e
v
e
nt
th
e
r
a
pi
d
s
pr
e
a
d
of
th
e
vi
r
us
a
nd
th
e
c
ons
e
que
nc
e
s
of
th
is
lo
c
kdown
ha
ve
c
r
e
a
te
d
c
ha
ll
e
nge
s
in
th
e
he
a
lt
h,
e
c
onomi
c
,
a
nd
s
oc
ia
l
s
e
c
to
r
s
[
1]
.
A
c
c
or
di
ng t
o t
he
W
or
ld
B
a
nk i
n 2022,
t
hi
s
e
m
e
r
ge
nc
y r
e
s
pons
e
c
r
e
a
te
s
ne
w
r
is
ks
, na
m
e
ly
i
nc
r
e
a
s
in
g t
he
de
bt
r
a
ti
o
in
th
e
w
or
ld
e
c
onomy,
w
hi
c
h
c
a
n
th
r
e
a
te
n
th
e
r
e
c
ove
r
y
pr
oc
e
s
s
f
r
om
th
e
c
r
is
is
be
c
a
us
e
of
th
e
in
te
r
c
onne
c
ti
on
be
twe
e
n
th
e
f
in
a
nc
ia
l
he
a
lt
h
of
in
di
vi
dua
ls
/c
o
m
pa
ni
e
s
a
nd
f
in
a
nc
ia
l
in
s
ti
tu
ti
ons
[
2]
.
B
a
nks
a
s
f
in
a
nc
ia
l
in
s
ti
tu
ti
ons
ha
ve
a
c
r
it
ic
a
l
r
ol
e
in
th
e
e
c
onomy
in
c
l
udi
ng
th
e
pr
oc
e
s
s
of
di
s
tr
ib
ut
in
g
f
unds
to
th
e
publ
ic
[
3]
.
B
a
nks
a
r
e
obl
ig
e
d
to
m
a
na
ge
c
r
e
di
t
r
is
k
to
a
voi
d
lo
s
s
e
s
f
or
th
e
ba
nk
it
s
e
lf
a
nd
it
s
c
u
s
to
m
e
r
s
.
H
ow
e
ve
r
,
th
e
im
pl
e
m
e
nt
a
ti
on
of
r
is
k
m
a
na
ge
m
e
nt
f
or
e
a
c
h
ba
nk
c
a
n
va
r
y
a
c
c
or
di
ng
to
obj
e
c
ti
ve
s
,
bu
s
in
e
s
s
pol
ic
ie
s
,
bus
in
e
s
s
s
iz
e
,
a
nd
c
om
pl
e
xi
ty
,
f
in
a
nc
ia
l
c
a
pa
bi
li
ti
e
s
,
s
uppor
ti
ng
in
f
r
a
s
tr
uc
tu
r
e
,
a
nd
hum
a
n
r
e
s
our
c
e
c
a
pa
bi
li
ti
e
s
a
r
e
ta
ke
n
in
to
c
on
s
id
e
r
a
ti
on
a
nd
one
of
th
e
a
ppl
ic
a
ti
ons
is
c
r
e
a
ti
ng
a
n
e
a
r
ly
w
a
r
ni
ng
s
ys
te
m
(
E
W
S
)
w
hi
c
h i
s
a
n i
nnova
ti
on t
ha
t
w
a
s
bor
n a
s
a
r
e
s
ul
t
of
t
he
gl
oba
l
c
r
is
is
t
ha
t
oc
c
ur
r
e
d i
n 2007
-
2008.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
,
V
ol
. 14, No. 3, J
une
2025
:
1829
-
1838
1830
T
he
be
gi
nni
ng
of
E
W
S
in
ba
nki
ng
w
a
s
m
a
de
to
m
oni
to
r
th
e
ov
e
r
a
ll
s
ta
bi
li
ty
of
th
e
f
in
a
nc
ia
l
s
ys
te
m
to
pr
e
ve
nt
th
e
oc
c
ur
r
e
nc
e
of
s
ys
t
e
m
ic
r
is
k
w
h
e
r
e
th
e
e
c
onomy de
c
li
ne
s
s
o
th
a
t
th
e
gl
oba
l
f
in
a
nc
ia
l
c
r
is
i
s
doe
s
not
ha
ppe
n
a
ga
in
by
de
te
c
ti
ng
c
r
is
is
s
ig
na
ls
be
f
or
e
th
e
a
c
tu
a
l
c
r
is
is
oc
c
ur
s
[
4]
.
E
W
S
us
e
d
by
m
a
ny
ba
nk
s
te
nds
to
pr
oduc
e
f
a
ls
e
pos
it
iv
e
s
,
th
e
w
a
r
ni
ngs
gi
v
e
n
be
c
om
e
m
e
a
ni
ngl
e
s
s
a
nd
it
is
to
o
la
te
to
c
a
r
r
y
out
c
r
e
di
t
r
is
k
m
it
ig
a
ti
on.
L
os
s
e
s
a
r
e
in
e
vi
ta
bl
e
b
e
c
a
u
s
e
th
e
da
t
a
us
e
d
i
s
o
nl
y
ba
s
e
d
on
tr
a
di
ti
ona
l
da
ta
s
uc
h
a
s
f
in
a
nc
ia
l
da
ta
a
nd
m
one
y
m
a
r
ke
t
s
w
it
hout
c
ons
id
e
r
in
g
e
xt
e
r
na
l
e
ve
nt
s
th
a
t
a
r
e
ha
ppe
ni
ng.
T
hi
s
w
a
s
s
e
e
n
w
he
n
th
e
lo
c
kdown
w
a
s
im
pl
e
m
e
nt
e
d
dur
in
g
th
e
C
O
V
I
D
-
19
pa
nd
e
m
ic
in
2020,
e
c
onomi
c
a
c
ti
vi
ty
de
c
r
e
a
s
e
d
c
a
u
s
in
g
m
a
ny de
bt
or
s
t
o be
t
hr
e
a
te
ne
d w
it
h ba
nkr
upt
c
y (
de
f
a
ul
t)
w
hi
c
h w
a
s
de
te
c
t
e
d t
oo l
a
te
by t
he
E
W
S
[
5]
.
I
n
th
e
21
st
c
e
nt
ur
y
e
r
a
,
m
a
c
hi
ne
le
a
r
ni
ng
is
w
id
e
ly
us
e
d
by
th
e
ba
nki
ng
s
e
c
to
r
,
e
s
pe
c
ia
ll
y
r
is
k
m
a
na
ge
m
e
nt
,
to
c
r
e
a
t
e
m
or
e
a
c
c
ur
a
te
r
is
k
m
ode
ll
in
g
by
id
e
nt
if
yi
ng
c
om
pl
e
xi
ty
a
nd
nonl
in
e
a
r
pa
tt
e
r
ns
[
6]
.
T
he
pot
e
nt
ia
l
of
m
a
c
hi
ne
le
a
r
ni
ng
c
a
n
be
ut
il
iz
e
d
to
obt
a
in
da
t
a
r
e
la
te
d
to
e
xt
e
r
na
l
e
ve
nt
s
th
a
t
ha
ve
not
be
e
n
c
a
pt
ur
e
d
by
th
e
E
W
S
by
e
xt
r
a
c
ti
ng
s
e
nt
im
e
nt
r
e
la
te
d
to
c
ur
r
e
n
t
e
c
onomi
c
f
lu
c
tu
a
ti
ons
[
7]
.
N
a
tu
r
a
l
la
ngua
ge
pr
oc
e
s
s
in
g
(
N
L
P
)
is
u
s
e
d
a
s
a
n
e
f
f
ic
ie
nt
w
a
y
to
r
e
s
e
a
r
c
h
t
hi
ngs
th
a
t
a
r
e
im
por
ta
nt
w
he
n
not
e
xpl
ic
it
ly
m
e
nt
io
ne
d
by
K
im
e
t
al
.
[
8]
.
A
te
c
hni
que
in
N
L
P
th
a
t
w
a
s
pr
e
v
io
us
ly
popula
r
ly
us
e
d
to
obt
a
in
s
e
nt
im
e
nt
w
a
s
th
e
le
xi
c
on
a
ppr
oa
c
h
w
hi
c
h
c
r
e
a
te
s
a
di
c
ti
ona
r
y
of
pos
it
iv
e
a
nd
ne
ga
ti
ve
w
or
ds
,
but
th
i
s
a
ppr
oa
c
h
h
a
s
th
e
w
e
a
kne
s
s
of
not
be
in
g
a
bl
e
to
c
a
pt
ur
e
th
e
c
ont
e
xt
,
s
o
th
e
s
e
nt
im
e
nt
r
e
s
ul
ts
be
c
om
e
in
a
c
c
ur
a
te
[
9]
.
T
he
di
s
c
ove
r
y
of
tr
a
ns
f
or
m
e
r
s
in
th
e
w
or
ld
of
N
L
P
c
a
n
c
ove
r
th
e
s
hor
tc
om
in
gs
of
th
is
le
xi
c
on
by
c
odi
ng
a
s
e
r
ie
s
of
w
or
ds
in
to
one
pa
r
t
w
hi
c
h
is
c
a
ll
e
d
a
s
e
lf
-
a
tt
e
nt
io
n
m
e
c
ha
ni
s
m
[
10]
.
T
r
a
ns
f
or
m
e
r
s
a
r
e
pr
ove
n
to
be
th
e
m
os
t
f
r
e
que
nt
ly
us
e
d
a
r
c
hi
te
c
tu
r
e
in
N
L
P
be
c
a
us
e
th
e
ir
pe
r
f
or
m
a
nc
e
c
a
n
e
x
c
e
e
d
ne
ur
a
l
ne
twor
k
a
r
c
hi
te
c
tu
r
e
s
w
hi
c
h a
r
e
w
id
e
ly
us
e
d t
o c
r
e
a
te
m
ode
ls
f
r
om
uns
tr
uc
tu
r
e
d da
ta
[
11]
.
E
a
r
li
e
r
E
W
S
r
e
s
e
a
r
c
h
a
s
s
e
s
s
e
d
ove
r
a
ll
f
in
a
nc
ia
l
s
ta
bi
li
ty
[
12]
.
B
a
nks
’
s
m
oni
to
r
in
g
c
r
e
di
t
r
is
k
c
a
n
us
e
de
ve
lo
pe
d
in
te
r
na
l
m
ode
ls
th
a
t
a
r
e
a
dj
us
te
d
to
th
e
ba
nk'
s
s
tr
a
te
gy
a
nd
c
ondi
ti
ons
[
13]
.
O
th
e
r
th
a
n
th
a
t,
E
W
S
a
ls
o
he
a
vi
ly
r
e
li
e
d
on
f
in
a
nc
ia
l
r
e
por
ts
us
in
g
m
ode
ls
th
a
t
w
e
r
e
e
a
s
y
to
e
xpl
a
in
s
uc
h
a
s
lo
gi
s
ti
c
r
e
gr
e
s
s
io
n
[
14]
w
he
r
e
[
15]
f
in
d
th
e
be
s
t
us
in
g
C
a
t
B
oos
t
a
lg
or
i
th
m
.
O
n
th
e
ot
he
r
ha
nd,
to
id
e
nt
if
y
c
r
e
di
t
r
is
k
us
in
g
th
e
s
e
nt
im
e
nt
f
r
om
ne
w
s
a
lr
e
a
dy
c
onduc
te
d
to
pr
ove
it
c
a
n
s
uppor
t
th
e
tr
a
di
ti
ona
l
m
e
th
od
of
a
s
s
e
s
s
in
g
de
bt
or
r
is
k.
H
ow
e
ve
r
,
th
e
r
e
is
s
ti
ll
no
r
e
s
e
a
r
c
h
a
bout
th
e
in
te
gr
a
ti
on
of
a
s
s
e
s
s
in
g
d
e
bt
or
r
is
k
w
it
h
th
e
c
ur
r
e
nt
s
e
nt
im
e
nt
ne
w
s
[
16]
.
W
hi
le
th
e
be
s
t
m
e
th
od
f
or
e
xt
r
a
c
ti
ng
s
e
nt
im
e
nt
is
us
e
a
tr
a
ns
f
or
m
e
r
-
ba
s
e
d
pr
e
-
tr
a
in
e
d
m
ode
l
a
s
[
17]
pr
ove
d
th
a
t
I
ndoB
E
R
T
s
ur
pa
s
s
ot
he
r
m
e
th
ods
in
e
xt
r
a
c
ti
ng
I
ndone
s
ia
n
f
in
a
nc
ia
l
te
xt
s
.
T
hi
s
pa
pe
r
a
im
s
to
c
r
e
a
te
E
W
S
us
in
g
ba
nk
d
e
bt
or
’
s
da
ta
a
nd
th
e
n
a
dd
th
e
r
e
s
ul
ts
of
ne
w
s
s
e
nt
im
e
nt
a
n
a
ly
s
is
to
pr
e
di
c
t
th
e
c
r
e
di
t
qua
li
ty
pa
ym
e
nt
(
pa
s
s
,
s
pe
c
i
a
l
m
e
nt
io
n,
a
nd
non
-
pe
r
f
or
m
in
g
lo
a
n
)
.
T
he
m
a
in
c
ont
r
ib
ut
io
ns
of
th
is
pa
pe
r
a
r
e
i
)
im
p
r
ove
d
tr
a
di
ti
ona
l
m
ode
l
a
c
c
ur
a
c
y
by
a
ddi
ng
th
e
r
e
s
ul
ts
of
ne
w
s
s
e
nt
im
e
nt
a
nd
ii
)
pr
ovi
ng
by
a
ddi
ng
th
e
r
e
s
ul
ts
of
s
e
nt
im
e
nt
a
n
a
ly
s
is
da
ta
c
a
n
he
lp
th
e
c
r
e
di
t
r
is
k
m
it
ig
a
ti
on
pr
oc
e
s
s
by
pr
ovi
di
ng w
a
r
ni
ngs
be
f
or
e
de
bt
or
s
a
r
e
t
hr
e
a
te
ne
d t
o de
f
a
ul
t
(
f
or
w
a
r
d
-
lo
oki
ng)
.
T
he
s
tr
uc
tu
r
e
of
th
e
pa
pe
r
is
a
s
f
ol
lo
w
s
:
s
e
c
ti
on
2
s
how
s
th
e
li
te
r
a
tu
r
e
r
e
vi
e
w
o
f
th
is
r
e
s
e
a
r
c
h.
S
e
c
ti
on
3
pr
ovi
de
s
how
th
e
r
e
s
e
a
r
c
h
is
c
onduc
te
d.
S
e
c
ti
on
4
pr
e
s
e
nt
s
th
e
a
na
ly
s
i
s
of
th
e
r
e
s
e
a
r
c
h
r
e
s
ul
t.
S
e
c
ti
on 5 c
onc
lu
de
s
t
hi
s
r
e
s
e
a
r
c
h.
2.
L
I
T
E
R
A
T
U
R
E
R
E
V
I
E
W
2.1.
T
h
e
or
e
t
ic
al
f
ou
n
d
at
io
n
2.1.1.
E
ar
ly
w
ar
n
in
g s
ys
t
e
m
T
he
f
ir
s
t
r
e
s
e
a
r
c
h
r
e
la
te
d
to
E
W
S
is
pr
e
di
c
te
d
de
bt
or
ba
nkr
upt
c
y
us
in
g
r
ough
s
e
ts
a
nd
de
c
is
io
n
tr
e
e
s
to
f
in
d
pr
obl
e
m
a
ti
c
c
r
e
di
t
pa
tt
e
r
ns
s
o
th
a
t
th
e
y
c
a
n
pr
e
di
c
t
pr
obl
e
m
a
ti
c
c
r
e
di
t
be
f
or
e
c
us
to
m
e
r
s
be
c
om
e
de
bt
or
s
.
T
he
r
ough
s
e
t
a
c
c
ur
a
c
y
w
a
s
a
bl
e
to
r
e
a
c
h
87.42%
,
s
up
e
r
io
r
to
th
e
de
c
i
s
io
n
tr
e
e
w
ho
s
e
a
c
c
ur
a
c
y
w
a
s
83.33%
[
18]
.
S
e
c
ond
r
e
s
e
a
r
c
h
m
a
de
a
c
om
pa
r
is
on
of
s
e
ve
r
a
l
c
la
s
s
if
ic
a
ti
on
a
lg
or
it
hm
s
in
th
e
c
a
s
e
of
pr
e
di
c
ti
ng
ba
nkr
upt
c
y
of
c
or
por
a
te
s
e
gm
e
nt
c
om
pa
ni
e
s
us
in
g
2014
-
2016
f
in
a
nc
ia
l
r
e
por
t
da
ta
of
ba
nkr
upt
c
om
pa
ni
e
s
in
F
r
a
nc
e
.
A
s
a
r
e
s
ul
t,
C
a
tB
oos
t'
s
pe
r
f
or
m
a
nc
e
c
a
n
s
ur
pa
s
s
8
ot
he
r
popul
a
r
a
lg
or
it
hm
s
,
na
m
e
ly
di
s
c
r
im
in
a
nt
a
na
ly
s
is
,
lo
gi
s
ti
c
r
e
gr
e
s
s
io
n,
s
uppor
t
ve
c
to
r
m
a
c
hi
ne
,
ne
ur
a
l
ne
twor
k,
r
a
ndom
f
or
e
s
t,
gr
a
di
e
nt
boos
ti
ng, de
e
p ne
ur
a
l
ne
twor
ks
, a
nd X
G
B
oos
t
[
15]
.
T
hi
r
d
r
e
s
e
a
r
c
h
us
e
d
a
lo
gi
s
ti
c
r
e
gr
e
s
s
io
n
a
lg
or
it
hm
to
de
te
c
t
ba
nk
ba
nkr
upt
c
y
us
in
g
f
in
a
nc
ia
l
r
e
por
t
da
ta
publ
is
he
d
by
th
e
ba
nk
a
nd
m
a
c
r
oe
c
onomi
c
da
ta
.
L
ogi
s
ti
c
r
e
gr
e
s
s
io
n
pr
oduc
e
s
a
n
a
c
c
ur
a
c
y
of
89.76%
,
w
he
r
e
th
e
c
r
is
i
s
s
ig
n
a
l
th
a
t
w
a
s
s
uc
c
e
s
s
f
ul
ly
c
a
pt
ur
e
d
(
tr
u
e
po
s
it
iv
e
(
TP
)
)
w
a
s
67.64%
,
th
e
c
r
is
i
s
s
ig
n
a
l
th
a
t
w
a
s
lo
s
t
(
f
a
ls
e
ne
ga
ti
ve
(
FN
)
)
w
a
s
32.35%
,
a
nd
th
e
f
a
ls
e
c
r
is
is
s
ig
na
l
(
f
a
ls
e
pos
it
iv
e
(
FP
)
)
of
7.52%
[
14]
.
T
he
f
our
th
a
nd
f
i
f
th
r
e
s
e
a
r
c
h
pr
e
di
c
ts
de
bt
or
ba
nkr
upt
c
y
us
in
g
a
ne
ur
a
l
ne
twor
k
a
lg
or
it
hm
th
a
t
opt
im
iz
e
d
w
e
ig
ht
a
nd
th
r
e
s
hol
d
w
it
h
a
c
r
os
s
-
ove
r
of
0.4
a
nd
m
ut
a
ti
on
of
0.2
a
nd
a
c
c
ur
a
c
y
r
e
a
c
h
e
d
97%
w
it
h
e
r
r
or
r
e
duc
e
d
by
55.8%
w
hi
le
c
om
pa
r
in
g
ba
c
kpr
opa
ga
ti
on
ne
ur
a
l
ne
twor
ks
w
it
h
m
ode
ls
th
a
t
a
r
e
us
ua
ll
y
us
e
d
to
pr
e
di
c
t
f
ut
ur
e
c
r
e
di
t
r
is
k,
na
m
e
ly
a
ut
or
e
gr
e
s
s
iv
e
m
od
e
ls
a
nd
lo
gi
s
ti
c
r
e
gr
e
s
s
i
on,
it
w
a
s
f
ound
th
a
t
ba
c
kpr
opa
ga
ti
on
ne
ur
a
l
ne
twor
ks
i
nc
r
e
a
s
e
t
he
a
c
c
ur
a
c
y of
f
ut
ur
e
-
lo
oki
ng
c
r
e
di
t
r
is
k m
a
na
ge
m
e
nt
[
19]
, [
20
]
. S
ix
th
r
e
s
e
a
r
c
h m
ode
ll
in
g
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
T
he
i
nf
lu
e
nc
e
of
s
e
nt
ime
nt
analy
s
is
i
n e
nhanc
in
g e
a
r
ly
w
a
r
ni
ng
s
y
s
te
m
m
od
e
l
fo
r
c
r
e
di
t
…
(
A
nge
l
K
ar
e
nt
ia
)
1831
a
c
om
pr
e
he
ns
iv
e
E
W
S
f
or
C
hi
ne
s
e
E
nt
e
r
pr
is
e
s
by
ut
il
iz
in
g
qu
a
nt
it
a
ti
ve
a
nd
qua
li
ta
ti
ve
us
in
g
f
u
s
io
n
-
lo
gi
s
ti
c
a
lg
or
it
hm
r
e
a
c
he
d a
c
c
ur
a
c
y 89.7%
i
n pr
e
di
c
ti
ng s
pe
c
i
a
l
tr
e
a
tm
e
nt
c
om
pa
ni
e
s
[
21]
.
2
.1.2. S
e
n
t
im
e
n
t
an
al
ys
is
F
ir
s
t
r
e
s
e
a
r
c
h
r
e
la
te
d
to
s
e
nt
im
e
nt
a
na
ly
s
is
a
ppl
ie
d
in
th
e
f
in
a
nc
ia
l
s
e
c
to
r
w
a
s
c
onduc
te
d
us
in
g
I
ndoB
E
R
T
,
w
hi
c
h
to
id
e
nt
if
y
ne
w
s
r
e
la
te
d
to
s
to
c
k
pr
ic
e
m
ove
m
e
nt
s
,
obt
a
in
e
d
a
n
a
c
c
ur
a
c
y
of
68%
,
s
ur
pa
s
s
in
g
th
e
pe
r
f
or
m
a
nc
e
of
ot
he
r
a
lg
or
it
hm
s
s
uc
h
a
s
lo
gi
s
ti
c
r
e
gr
e
s
s
io
n,
di
s
c
r
im
in
a
nt
li
ne
a
r
a
na
ly
s
is
,
k
-
ne
a
r
e
s
t
ne
ig
hbor
s
,
d
e
c
is
io
n
tr
e
e
,
s
uppor
t
v
e
c
to
r
m
a
c
hi
ne
,
r
a
n
dom
f
or
e
s
t,
X
G
boos
t,
na
ïv
e
B
a
ye
s
,
lo
ng s
hor
t
-
te
r
m
m
e
m
or
y
,
a
nd
m
u
lt
i
-
la
ye
r
pe
r
c
e
pt
r
on
[
17]
.
T
he
s
e
c
ond
r
e
s
e
a
r
c
h
us
e
d
a
le
xi
c
on
a
ppr
oa
c
h
to
a
na
ly
z
e
s
e
nt
im
e
nt
f
r
om
G
e
r
m
a
n
twe
e
ts
r
e
la
te
d
to
c
om
pa
ni
e
s
by
c
om
bi
ni
ng
S
e
nt
iW
or
ts
c
ha
tz
(
G
e
r
m
a
n)
a
nd
S
e
nt
iW
or
dN
e
t
3.0
(
E
ngl
is
h)
e
nha
nc
e
d
w
it
h
G
e
r
m
a
N
e
t
r
e
s
ul
ti
ng
in
a
n
a
c
c
ur
a
c
y
of
59.19%
a
nd
s
ur
pa
s
s
in
g
r
a
ndom
c
la
s
s
if
ic
a
ti
on
p
e
r
f
or
m
a
nc
e
[
22]
.
T
hi
r
d
r
e
s
e
a
r
c
h
ha
s
c
r
e
a
te
d
a
m
od
e
l
to
id
e
nt
if
y
c
r
e
di
t
r
is
k
th
r
ough
f
in
a
nc
ia
l
ne
w
s
by e
xt
r
a
c
ti
ng
s
e
nt
im
e
nt
f
r
om
e
le
c
tr
a
-
ba
s
e
d
ta
r
ge
t
e
nt
it
y
s
e
nt
im
e
nt
(
T
E
S
)
w
hi
c
h
a
na
ly
s
e
s
n
e
w
s
r
e
la
te
d
to
c
om
pa
ni
e
s
th
a
t
a
r
e
in
th
e
pr
oc
e
s
s
of
a
ppl
yi
ng
f
or
c
r
e
di
t.
T
hi
s
m
ode
l
pr
oduc
e
s
a
n
F
1
-
s
c
or
e
of
77%
w
he
r
e
t
he
be
s
t
pe
r
f
or
m
a
nc
e
i
s
i
n t
he
or
ga
ni
z
a
ti
on t
a
g
[
23]
.
F
our
th
r
e
s
e
a
r
c
h
c
r
e
a
te
d
a
m
ode
l
f
or
id
e
nt
i
f
yi
ng
c
r
e
di
t
r
is
k
f
r
o
m
ne
w
s
th
a
t
c
om
bi
ne
s
s
our
c
e
-
la
te
nt
D
ir
ic
hl
e
t
a
ll
oc
a
ti
on
(
L
D
A
)
,
na
m
e
d
e
nt
it
y
r
e
c
ogni
ti
on
(
N
E
R
)
,
a
nd
ta
r
ge
t
a
s
pe
c
t
-
ba
s
e
d
s
e
nt
im
e
nt
(
T
A
B
S
A
)
us
in
g
bi
di
r
e
c
ti
ona
l
e
nc
ode
r
r
e
pr
e
s
e
nt
a
ti
ons
f
r
om
tr
a
ns
f
or
m
e
r
s
(
B
E
R
T
)
.
T
hi
s
c
om
bi
na
ti
on
m
ode
l
pr
ove
s
th
a
t
s
e
nt
im
e
nt
s
c
or
e
s
s
tr
e
ngt
he
n
tr
a
di
ti
ona
l
pr
e
di
c
ti
ons
th
a
t
onl
y
u
s
e
s
tr
uc
tu
r
e
d
f
in
a
nc
i
a
l
va
r
ia
bl
e
s
a
nd
in
c
r
e
a
s
e
th
e
pr
e
di
c
ti
ve
pow
e
r
of
c
om
pa
ni
e
s
le
a
vi
ng
ba
nks
ot
he
r
th
a
n
due
to
ba
nkr
upt
c
y
[
24]
.
T
he
f
if
th
r
e
s
e
a
r
c
h
id
e
nt
if
ie
s
c
r
e
di
t
r
is
k
us
in
g
ne
w
s
m
a
ki
ng
th
e
R
i
s
ky
N
e
w
s
I
nde
x
by
a
ggr
e
ga
ti
ng
ne
w
s
s
c
or
e
s
r
e
la
te
d
to
c
r
e
di
t,
in
ve
s
tm
e
nt
s
tr
a
te
gi
e
s
,
a
nd
f
undi
ng
ba
s
e
d
on
th
e
w
or
d
ve
c
to
r
d
is
ta
nc
e
r
e
s
ul
ti
ng
f
r
om
w
or
d
e
m
be
ddi
ngs
f
r
o
m
f
in
a
nc
ia
l
ne
w
s
c
r
e
a
te
d
u
s
in
g
th
e
W
or
d2V
e
c
m
e
th
od
to
o
bt
a
in
s
c
or
e
a
ggr
e
ga
ti
on.
N
e
w
s
is
pos
it
iv
e
ly
c
or
r
e
la
te
d w
it
h c
la
s
s
ic
f
in
a
nc
i
a
l
r
e
por
ts
[
16]
.
2.2. T
h
e
or
e
t
ic
al
gap
s
2.2.1. I
n
ac
c
u
r
ac
y o
f
E
WS
m
od
e
l
m
ad
e
w
it
h
s
t
r
u
c
t
u
r
e
d
f
in
a
n
c
ia
l
var
ia
b
le
s
on
ly
T
he
C
O
V
I
D
-
19
pa
nde
m
ic
ha
s
m
a
de
gl
oba
l
e
c
onomi
c
c
ondi
ti
ons
unc
e
r
ta
in
.
T
he
us
e
of
E
W
S
a
t
th
a
t
ti
m
e
be
c
a
m
e
le
s
s
a
c
c
ur
a
te
b
e
c
a
u
s
e
it
di
d
not
c
on
s
id
e
r
th
e
e
xt
e
r
na
l
e
ve
nt
s
th
a
t
w
e
r
e
ha
pp
e
ni
ng.
A
n
in
nova
ti
ve
a
ppr
oa
c
h i
s
ne
e
de
d
w
he
r
e
E
W
S
a
nd e
xt
e
r
na
l
s
e
nt
im
e
nt
do not r
un i
ndi
vi
dua
ll
y.
2.2.2. S
p
e
c
i
f
i
c
t
e
r
m
s
i
n
t
h
e
f
i
n
a
n
c
i
a
l
d
o
m
a
i
n
a
n
d
l
a
n
g
u
a
g
e
s
t
r
u
c
t
u
r
e
s
v
a
r
y
b
e
t
w
e
e
n
t
e
x
t
s
i
n
d
i
f
f
e
r
e
n
t
l
a
n
g
u
a
g
e
s
V
a
r
io
us
N
L
P
s
tu
di
e
s
ha
ve
be
e
n
c
onduc
te
d
to
im
pr
ove
s
e
nt
im
e
nt
a
na
ly
s
is
s
pe
c
if
ic
a
ll
y
f
or
th
e
f
in
a
nc
ia
l
dom
a
in
.
H
ow
e
ve
r
,
if
th
e
la
ngua
ge
of
th
e
te
xt
us
e
d
is
di
f
f
e
r
e
nt
,
a
m
ode
l
s
pe
c
if
ic
to
th
a
t
la
ngua
ge
is
ne
e
de
d.
A
ls
o,
th
e
un
a
va
il
a
bi
li
ty
of
da
ta
s
e
ts
s
pe
c
if
ic
to
th
e
f
in
a
nc
ia
l
dom
a
in
is
s
ti
ll
a
bi
g
c
ha
ll
e
ng
e
.
A
n
a
ppr
opr
ia
te
a
ppr
oa
c
h i
s
ne
e
de
d
s
o t
ha
t
th
e
r
e
s
ul
ti
ng s
e
nt
im
e
nt
a
na
ly
s
is
c
a
n b
e
us
e
f
ul
.
2.2.3. S
e
n
t
im
e
n
t
an
al
ys
is
r
e
s
e
ar
c
h
i
n
i
d
e
n
t
if
yi
n
g c
r
e
d
it
r
is
k
l
im
it
e
d
t
o as
s
e
s
s
n
e
w
d
e
b
t
or
S
e
nt
im
e
nt
a
na
ly
s
is
r
e
s
e
a
r
c
h i
n i
de
nt
if
yi
ng c
r
e
di
t
r
is
k only a
na
ly
z
e
d de
bt
or
s
w
ho w
il
l
a
ppl
y f
or
c
r
e
di
t
a
nd
pr
ove
d
th
a
t
th
e
r
e
w
a
s
a
c
or
r
e
la
ti
on
w
it
h
tr
a
di
ti
ona
l
f
in
a
nc
ia
l
da
ta
.
D
ue
to
th
e
s
hor
tc
om
in
gs
of
th
e
s
e
s
tu
di
e
s
,
it
is
ne
c
e
s
s
a
r
y
to
ha
ve
a
m
e
th
od
f
or
de
te
c
ti
ng
s
ig
na
ls
of
de
te
r
io
r
a
ti
on
f
r
om
e
xi
s
ti
ng
de
bt
or
s
.
I
t
c
a
n
c
a
pt
ur
e
f
ut
ur
e
-
lo
oki
ng
c
r
e
di
t
r
is
ks
by
c
r
e
a
ti
ng
a
n
E
W
S
th
a
t
c
o
m
bi
ne
s
tr
a
di
ti
ona
l
da
ta
f
r
om
f
in
a
nc
ia
l
r
e
por
ts
w
it
h t
he
l
a
te
s
t
f
in
a
nc
ia
l
ne
w
s
s
o i
t
c
a
n r
e
f
le
c
t
th
e
c
ur
r
e
nt
e
c
ono
m
ic
s
it
ua
ti
on.
2.3
.
C
on
c
e
p
t
u
al
m
od
e
l
W
he
n
th
e
ba
nk
gi
ve
s
c
r
e
di
t,
id
e
a
ll
y
th
e
de
bt
or
w
il
l
pa
y
a
c
c
or
di
ng
to
th
e
pa
ym
e
nt
s
c
he
dul
e
.
I
f
th
e
c
r
e
di
t
pa
ym
e
nt
f
ol
lo
w
s
th
e
pa
ym
e
nt
s
c
h
e
dul
e
,
th
e
n
th
e
c
r
e
di
t
q
ua
li
ty
is
'
pa
s
s
'
.
I
f
th
e
pa
ym
e
nt
is
d
e
li
nque
nt
by
1
-
90
da
ys
f
r
om
th
e
pa
ym
e
nt
s
c
he
dul
e
,
th
e
n
th
e
c
r
e
di
t
qua
li
ty
is
'
s
pe
c
ia
l
m
e
nt
io
n'
.
M
e
a
nw
hi
le
,
if
th
e
pa
ym
e
nt
is
de
li
nque
nt
by
m
or
e
th
a
n
90
d
a
ys
f
r
om
th
e
pa
ym
e
nt
s
c
h
e
dul
e
,
th
e
n
th
e
c
r
e
di
t
qu
a
li
ty
is
a
‘
non
-
pe
r
f
or
m
in
g
lo
a
n’
.
C
r
e
di
t
qua
li
ty
in
th
e
c
om
in
g
m
ont
hs
w
il
l
be
th
e
ta
r
ge
t
o
f
E
W
S
pr
e
di
c
ti
on
w
it
h
in
de
pe
nde
nt
va
r
ia
bl
e
s
f
r
om
th
e
in
te
r
na
l
m
ode
l.
I
n
a
ddi
ti
on,
f
r
om
th
os
e
E
W
S
s
tu
di
e
s
,
lo
gi
s
ti
c
r
e
gr
e
s
s
io
n
i
s
w
id
e
ly
us
e
d
be
c
a
us
e
of
it
s
e
a
s
e
of
e
xpl
a
na
ti
on
[
25]
.
A
f
te
r
ot
he
r
m
or
e
a
dva
n
c
e
d
t
e
c
h
ni
que
s
w
e
r
e
f
ound,
s
u
c
h
a
s
r
ough
s
e
t,
ne
ur
a
l
ne
twor
k,
a
nd
boos
te
d
m
ode
l,
th
e
ir
a
c
c
ur
a
c
y
out
pe
r
f
or
m
s
th
e
l
ogi
s
ti
c
r
e
gr
e
s
s
io
n,
but
th
e
ir
e
xpl
a
na
ti
on
is
s
ti
ll
not
a
s
e
a
s
y
a
s
l
ogi
s
ti
c
r
e
gr
e
s
s
io
n.
T
he
e
m
e
r
ge
nc
e
of
tr
a
ns
f
or
m
e
r
ha
s
m
a
de
it
e
a
s
ie
r
f
or
th
e
N
L
P
c
om
m
uni
ty
to
de
ve
lo
p
pr
e
-
tr
a
in
e
d
m
ode
ls
c
a
ll
e
d
la
r
ge
la
ngua
ge
m
ode
ls
(
L
L
M
)
w
hi
c
h
us
e
a
lo
t
of
te
xt
in
it
s
c
r
e
a
ti
on
p
r
oc
e
s
s
[
26]
.
L
L
M
c
a
n
s
im
pl
if
y
th
e
m
ode
l
c
r
e
a
ti
on
pr
oc
e
s
s
w
he
r
e
you
do
not
h
a
ve
to
t
r
a
in
f
r
om
s
c
r
a
tc
h
be
c
a
us
e
it
ju
s
t
ne
e
ds
to
f
in
e
-
tu
ne
/t
r
a
ns
f
e
r
le
a
r
ni
ng
f
r
om
L
L
M
to
th
e
da
ta
s
e
t
to
e
xt
r
a
c
t
f
e
a
tu
r
e
s
f
r
om
th
e
da
ta
s
e
t.
F
in
a
nc
ia
l
ne
w
s
in
th
e
I
ndone
s
ia
n
la
ngua
g
e
is
s
ti
ll
a
c
ha
ll
e
ng
e
due
to
th
e
una
va
il
a
bi
li
t
y
of
da
ta
s
e
ts
s
pe
c
if
ic
to
th
a
t
dom
a
in
[
27]
.
T
hi
s
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
,
V
ol
. 14, No. 3, J
une
2025
:
1829
-
1838
1832
li
m
it
a
ti
on
c
a
n
be
ove
r
c
om
e
by
f
in
e
-
tu
ni
ng/
t
r
a
ns
f
e
r
le
a
r
ni
ng
th
e
L
L
M
m
ode
l
w
it
h
a
s
pe
c
if
ic
da
ta
s
e
t
on
f
in
a
nc
ia
l
ne
w
s
to
be
a
bl
e
to
c
a
pt
ur
e
th
e
c
ont
e
xt
,
s
o
th
e
r
e
s
ul
t
of
s
e
nt
im
e
nt
r
e
la
te
d
to
e
c
onomi
c
f
lu
c
tu
a
ti
ons
r
e
f
le
c
ts
th
e
e
xt
e
r
na
l
e
v
e
nt
s
th
a
t
a
r
e
c
ur
r
e
nt
ly
oc
c
ur
r
in
g.
T
hi
s
ha
s
be
e
n
im
pl
e
m
e
nt
e
d
in
F
in
B
E
R
T
by
f
in
e
-
tu
ni
ng
B
E
R
T
u
s
in
g
a
d
a
ta
s
e
t
of
E
ngl
is
h
f
in
a
nc
ia
l
te
xt
s
w
hi
c
h
is
pr
ove
n
to
im
pr
ove
th
e
p
e
r
f
or
m
a
nc
e
of
r
e
gul
a
r
B
E
R
T
on f
in
a
nc
ia
l
te
xt
s
[
28]
, [
29]
.
2.4
.
I
n
n
ovat
io
n
T
he
m
a
in
c
ont
r
ib
ut
io
n
of
th
is
pa
pe
r
is
to
m
a
ke
a
s
e
nt
im
e
nt
a
na
ly
s
is
m
ode
l
f
r
om
th
e
I
ndone
s
ia
n
la
ngua
ge
L
L
M
w
hi
c
h
is
s
pe
c
if
ic
a
ll
y
f
or
th
e
e
c
onomi
c
dom
a
in
by
c
ont
in
ui
ng
th
e
le
a
r
ni
ng
pr
oc
e
s
s
of
th
e
or
ig
in
a
l
I
ndoB
E
R
T
us
in
g
I
ndone
s
ia
n
f
in
a
nc
ia
l
ne
w
s
.
I
ndoB
E
R
T
,
w
he
r
e
th
is
m
ode
l
c
a
n
s
ur
pa
s
s
th
e
m
ul
ti
li
ngua
l
pe
r
f
o
r
m
a
nc
e
of
B
E
R
T
a
nd
M
a
la
yB
E
R
T
in
e
x
tr
a
c
ti
ng
s
e
nt
im
e
nt
f
r
om
ge
ne
r
a
l
I
ndone
s
ia
n
la
ngua
ge
te
xt
[
30]
.
T
he
upgr
a
d
e
d
I
ndoB
E
R
T
is
u
s
e
d
to
e
xt
r
a
c
t
f
e
a
tu
r
e
s
f
r
om
th
e
f
in
a
n
c
ia
l
ne
w
s
s
e
nt
im
e
nt
a
na
ly
s
is
da
t
a
s
e
t.
A
f
te
r
th
e
s
e
nt
im
e
nt
is
ge
ne
r
a
te
d,
it
w
il
l
be
a
dde
d
to
th
e
ba
n
k
de
bt
or
da
ta
s
e
t
to
pr
e
di
c
t
th
e
c
r
e
di
t
qua
li
ty
in
th
e
ne
xt
2
y
e
a
r
s
u
s
in
g
th
e
C
a
tB
oos
t
a
lg
or
it
hm
.
C
a
t
B
oos
t
is
u
s
e
d
a
s
a
n
a
lg
or
it
hm
be
c
a
u
s
e
it
ha
s
f
e
a
tu
r
e
s
th
a
t
pr
e
ve
nt
ove
r
f
it
ti
ng
a
nd
is
e
f
f
ic
ie
nt
w
he
n
th
e
in
f
e
r
e
nc
e
pr
oc
e
s
s
is
c
a
r
r
ie
d
out
.
I
n
th
e
e
nd,
th
e
r
e
s
ul
t
of
de
c
is
io
n t
r
e
e
i
s
s
ym
m
e
tr
ic
a
l
[
31]
, [
32]
. T
hi
s
m
ode
l
w
il
l
be
t
he
E
W
S
.
E
W
S
t
e
s
ti
ng i
s
di
vi
de
d i
n
to
2
m
ode
s
w
he
r
e
E
W
S
w
it
hout
s
e
nt
im
e
nt
r
e
s
ul
ts
a
nd E
W
S
w
it
h
s
e
nt
im
e
nt
r
e
s
ul
ts
.
T
he
a
im
is
to
c
om
pa
r
e
th
e
pe
r
f
or
m
a
nc
e
of
th
e
tr
a
di
ti
ona
l
m
ode
l
w
it
h
th
e
ot
he
r
m
ode
l
th
a
t
ha
s
be
e
n
a
dde
d
w
it
h
s
e
nt
im
e
nt
a
n
a
ly
s
is
r
e
s
ul
ts
.
I
n
a
ddi
ti
on,
th
e
E
W
S
th
a
t
w
a
s
c
r
e
a
te
d
in
ve
s
ti
ga
te
d
w
he
th
e
r
it
us
e
s
s
e
nt
im
e
nt
a
na
ly
s
is
r
e
s
ul
ts
to
c
r
e
a
te
f
or
w
a
r
d
-
lo
oki
ng
us
in
g
S
h
a
pl
e
y
a
ddi
ti
ve
e
xpl
a
na
ti
ons
(
S
H
A
P
)
s
o
th
a
t
i
t
c
a
n c
r
e
a
te
a
b
e
tt
e
r
c
r
e
di
t
r
is
k m
it
ig
a
ti
on pr
oc
e
s
s
by pr
ovi
di
ng w
a
r
ni
ngs
be
f
or
e
t
he
de
bt
or
goe
s
ba
nkr
upt
.
3.
M
E
T
H
O
D
3.1. Dat
a c
ol
le
c
t
io
n
T
he
da
ta
c
ol
le
c
ti
on
is
c
a
r
r
ie
d
out
by
w
e
b
s
c
r
a
pi
ng
th
e
C
N
B
C
I
ndone
s
ia
ne
w
s
por
ta
l
w
it
h
th
e
S
c
r
a
py
pa
c
ka
ge
us
in
g
th
e
P
yt
hon
pr
ogr
a
m
m
in
g
la
ngua
ge
in
G
oogl
e
C
ol
la
bor
a
to
r
y
(
P
yt
hon
3.9
ve
r
s
io
n)
.
T
he
c
ol
le
c
te
d
ne
w
s
is
th
e
ne
w
s
th
a
t
w
a
s
publ
i
s
he
d
f
r
om
2018
-
2022.
T
he
r
e
s
ul
ts
a
r
e
s
a
v
e
d
e
ve
r
y
m
ont
h
in
a
f
il
e
in
C
S
V
f
or
m
a
t.
A
f
te
r
obt
a
in
in
g
th
e
ne
w
s
da
ta
,
it
w
a
s
r
a
ndomi
z
e
d
a
nd
th
e
n
a
nnot
a
te
d
in
to
3
la
be
ls
(
ne
ga
ti
ve
,
ne
ut
r
a
l,
a
nd
pos
it
iv
e
)
ba
s
e
d on the
de
ta
il
e
d c
ont
e
nt
of
t
he
ne
w
s
a
r
ti
c
le
w
it
h 500 ins
ta
nc
e
s
of
e
a
c
h l
a
be
l,
s
o t
he
to
ta
l
s
e
nt
im
e
nt
a
na
ly
s
i
s
da
ta
s
e
t
in
s
ta
n
c
e
s
a
r
e
1,500
in
s
ta
nc
e
s
.
M
e
a
nw
hi
le
,
E
W
S
da
ta
u
s
e
s
c
or
por
a
te
s
e
gm
e
nt
c
r
e
di
t
da
ta
f
r
om
one
of
th
e
pr
iv
a
te
ba
nks
in
I
ndone
s
i
a
f
r
om
th
e
2018
-
2022
pe
r
io
d.
F
ig
ur
e
1
s
how
s
th
e
w
hol
e
f
lo
w
of
t
he
pr
opos
e
d m
e
th
od.
F
ig
ur
e
1. P
r
opos
e
d
m
e
th
od f
lo
w
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
T
he
i
nf
lu
e
nc
e
of
s
e
nt
ime
nt
analy
s
is
i
n e
nhanc
in
g e
a
r
ly
w
a
r
ni
ng
s
y
s
te
m
m
od
e
l
fo
r
c
r
e
di
t
…
(
A
nge
l
K
ar
e
nt
ia
)
1833
3.2. P
r
e
-
p
r
oc
e
s
s
in
g
S
ta
r
t
f
r
om
p
r
e
-
pr
oc
e
s
s
in
g
th
e
te
xt
da
ta
by
c
a
s
e
-
f
ol
di
ng
to
e
qua
t
e
a
ll
c
ha
r
a
c
te
r
s
us
e
d
in
ne
w
s
a
r
ti
c
le
s
to
lo
w
e
r
c
a
s
e
us
in
g
th
e
s
tr
.l
ow
e
r
(
)
f
unc
ti
on
a
nd
r
e
m
ovi
ng
m
e
a
ni
ngl
e
s
s
r
e
pe
ti
ti
ve
s
e
nt
e
n
c
e
s
us
in
g
th
e
r
e
pl
a
c
e
(
)
f
unc
ti
on
f
r
o
m
th
e
P
a
nda
s
pa
c
ka
ge
(
2.2.0
ve
r
s
io
n)
.
T
he
ne
xt
s
te
p
is
r
e
m
ovi
ng
nu
m
be
r
s
a
nd
punc
tu
a
ti
on
us
in
g
th
e
R
e
ge
x pa
c
ka
ge
(
2023.12.25 ve
r
s
io
n)
. A
f
te
r
t
ha
t,
c
or
r
e
c
ti
ng
non
-
s
ta
nda
r
d w
or
ds
i
nt
o s
ta
nda
r
d w
or
ds
us
in
g
a
n
A
la
y
la
ngua
ge
di
c
ti
ona
r
y
f
r
om
th
e
r
e
s
e
a
r
c
h
[
33]
.
A
f
te
r
c
or
r
e
c
ti
on,
c
ont
in
ue
d
to
th
e
s
to
pw
or
d
s
r
e
m
ova
l
to
r
e
m
ove
m
e
a
ni
ngl
e
s
s
w
or
d
s
s
u
c
h
a
s
c
onj
unc
ti
ons
u
s
in
g
th
e
I
ndone
s
ia
n
s
to
pw
or
d
di
c
ti
ona
r
y
f
r
om
th
e
N
L
T
K
pa
c
ka
ge
(
3.8.1
ve
r
s
io
n)
.
T
he
la
s
t
s
te
p
i
s
e
xt
r
a
c
ti
ng
th
e
r
oot
w
or
ds
f
r
om
w
or
ds
in
ne
w
s
a
r
ti
c
le
s
us
in
g
th
e
S
a
s
tr
a
w
i
pa
c
ka
ge
(
1.0.1
ve
r
s
io
n)
.
P
r
e
-
pr
oc
e
s
s
in
g
th
e
da
ta
s
e
t
f
or
th
e
la
ngua
ge
m
ode
l
doe
s
not
r
e
qui
r
e
s
te
m
m
in
g t
o m
a
in
ta
in
pa
r
a
gr
a
phs
t
ha
t
c
a
n be
und
e
r
s
to
od by hu
m
a
ns
.
P
r
e
-
pr
oc
e
s
s
in
g
on
th
e
E
W
S
da
ta
s
e
t
w
a
s
c
a
r
r
ie
d
out
by
a
ddi
n
g
s
e
nt
im
e
nt
a
na
ly
s
is
r
e
s
ul
ts
in
to
th
e
E
W
S
da
ta
s
e
t
on R
S
tu
di
o
(
R
4.2.0
ve
r
s
io
n)
.
T
h
e
pr
oc
e
s
s
of
a
ggr
e
ga
ti
ng
s
e
nt
im
e
nt
a
n
a
ly
s
is
r
e
s
ul
ts
in
th
e
E
W
S
da
ta
s
e
t
us
in
g
th
e
dpl
yr
(
1.0.9
ve
r
s
io
n)
to
f
il
te
r
ne
ga
ti
ve
s
e
nt
im
e
nt
r
e
s
ul
ts
w
hi
c
h
ha
ve
a
pr
oba
bi
li
ty
pr
e
di
c
ti
on
a
bove
80%
a
nd
s
tr
_de
te
c
t
f
unc
ti
on
f
r
om
ti
dyve
r
s
e
pa
c
ka
g
e
s
(
1.3.1
ve
r
s
io
n)
to
de
te
c
t
de
bt
or
'
s
in
du
s
tr
ia
l
s
e
c
to
r
. T
a
bl
e
1
s
how
s
t
h
e
de
ta
il
s
of
E
W
S
va
r
ia
bl
e
s
.
T
a
bl
e
1. V
a
r
ia
bl
e
of
E
W
S
da
ta
s
e
t
V
a
r
i
a
bl
e
D
a
t
a
t
ype
E
xpl
a
na
t
i
on
I
nt
e
r
na
l
r
i
s
k gr
a
di
ng
N
um
e
r
i
c
L
ow
r
i
s
k
-
hi
gh r
i
s
k:
1
-
10
C
ur
r
e
nt
c
r
e
di
t
qua
l
i
t
y
C
a
t
e
gor
i
c
a
l
D
e
l
i
nque
nt
0 da
y:
‘
P
a
s
s
’
1
-
90 da
ys
:
‘
S
pe
c
i
a
l
M
e
nt
i
on’
> 90 da
ys
:
‘
N
on
-
pe
r
f
or
m
i
ng L
oa
n’
U
s
a
ge
N
um
e
r
i
c
C
r
e
di
t
out
s
t
a
ndi
ng di
vi
de
d by c
r
e
di
t
pl
a
f
ond
U
S
D
t
o I
D
R
N
um
e
r
i
c
E
xc
ha
nge
r
a
t
e
f
r
om
U
S
D
t
o I
D
R
B
i
r
a
t
e
N
um
e
r
i
c
T
he
pr
e
va
i
l
i
ng i
nt
e
r
e
s
t
r
a
t
e
of
t
he
c
e
nt
r
a
l
ba
nk
i
n I
ndone
s
i
a
I
nf
l
a
t
i
on r
a
t
e
N
um
e
r
i
c
I
ndone
s
i
a
’
s
i
nf
l
a
t
i
on r
a
t
e
H
a
s
ne
ga
t
i
ve
ne
w
s
C
a
t
e
gor
i
c
a
l
D
e
bt
or
s
ha
ve
ne
ga
t
i
ve
ne
w
s
ba
s
e
d on t
he
i
r
i
ndus
t
r
y s
e
c
t
or
I
f
a
ny:
‘
Y
e
s
’
O
n t
he
c
ont
r
a
r
y:
‘
N
o’
C
ount
ne
ga
t
i
ve
ne
w
s
N
um
e
r
i
c
T
ot
a
l
c
ount
of
de
bt
or
’
s
ne
ga
t
i
ve
ne
w
s
ba
s
e
d on i
t
s
i
ndu
s
t
r
y s
e
c
t
or
3.3. M
od
e
ll
in
g
T
h
e
m
od
e
l
li
n
g
pr
o
c
e
s
s
s
t
a
r
t
s
w
it
h
t
h
e
u
n
a
n
no
t
a
t
e
d
n
e
w
s
d
a
ta
to
c
r
e
a
t
e
a
n
I
nd
o
B
E
R
T
m
o
de
l
t
h
a
t
i
s
s
p
e
c
if
ic
t
o
t
h
e
e
c
o
no
m
i
c
do
m
a
i
n by f
i
n
e
-
t
u
ni
ng a
n e
x
i
s
t
i
ng I
n
doB
E
R
T
m
o
d
e
l
(
i
nd
o
be
n
c
hm
a
r
k/
in
d
ob
e
r
t
-
ba
s
e
-
p
2)
pr
ovi
de
d
by
I
ndoB
e
nc
hm
a
r
k
on
th
e
hugging
f
a
c
e
por
ta
l
in
to
a
ne
w
la
ngua
ge
m
ode
l.
N
e
xt
,
th
e
ne
w
la
ngua
ge
m
ode
l
be
c
a
m
e
a
m
ode
l
th
a
t
e
xt
r
a
c
t
s
f
e
a
tu
r
e
s
f
r
om
th
e
s
e
nt
im
e
nt
a
na
ly
s
i
s
da
ta
s
e
t
a
nd
th
e
r
e
s
ul
ts
b
e
c
a
m
e
a
m
ode
l
f
or
pr
e
di
c
ti
ng
ne
w
s
s
e
nt
im
e
nt
.
B
ot
h
pr
oc
e
s
s
e
s
a
r
e
c
a
r
r
ie
d
out
us
in
g
th
e
s
im
pl
e
tr
a
ns
f
or
m
e
r
s
pa
c
ka
ge
(
0.64.3 ve
r
s
io
n)
.
T
he
da
ta
te
s
ti
ng
m
e
th
od
us
e
d
f
or
m
ode
ll
in
g
ne
w
s
s
e
nt
im
e
nt
a
na
ly
s
is
us
e
s
a
p
e
r
c
e
nt
a
ge
s
pl
it
te
s
t
m
e
th
od of
90%
w
he
r
e
1,350 ins
ta
nc
e
s
a
s
t
r
a
in
in
g da
ta
a
r
e
t
a
ke
n
, 75 ins
ta
nc
e
s
of
w
hi
c
h a
r
e
t
a
k
e
n a
s
v
a
li
da
ti
on
da
ta
a
nd
150
in
s
ta
nc
e
s
a
s
t
e
s
t
da
t
a
.
H
ype
r
pa
r
a
m
e
te
r
tu
ni
ng
w
a
s
a
l
s
o
c
ondu
c
te
d
on
th
e
s
e
nt
im
e
nt
a
na
ly
s
i
s
us
in
g
le
a
r
ni
ng
r
a
te
,
e
poc
h,
a
nd
r
e
gul
a
r
iz
a
ti
on
va
lu
e
s
.
T
he
ba
t
c
h
s
iz
e
a
nd
m
a
x
s
e
que
nc
e
le
ngt
h
va
lu
e
s
a
r
e
m
a
de
f
ix
e
d
w
he
r
e
th
e
ba
tc
h
s
iz
e
us
e
d
is
16
w
hi
le
th
e
m
a
x
s
e
que
nc
e
le
ngt
h
i
s
512
w
hi
c
h
is
th
e
m
a
xi
m
um
c
a
pa
c
it
y
of
a
B
E
R
T
-
ba
s
e
d
a
r
c
hi
te
c
tu
r
e
to
r
e
c
e
iv
e
s
e
qu
e
nt
ia
l
da
ta
.
T
he
opt
im
iz
e
r
th
a
t
is
us
e
d
is
A
da
m
W
.
I
n
a
ddi
ti
on,
e
a
r
ly
s
to
ppi
ng
is
us
e
d
to
pr
e
ve
nt
ove
r
f
it
ti
ng
by
m
oni
to
r
in
g
th
e
in
c
r
e
a
s
e
of
M
a
th
e
w
s
c
or
r
e
la
ti
on
c
oe
f
f
ic
ie
nt
(
M
C
C
)
va
lu
e
w
hi
c
h ha
s
be
e
n pr
ove
n t
o be
r
e
li
a
bl
e
f
or
a
s
s
e
s
s
in
g t
he
m
ode
l
pr
opor
ti
ona
ll
y
f
or
e
a
c
h
c
la
s
s
e
l
e
m
e
nt
i
n t
he
da
ta
s
e
t
[
34]
. A
ll
pa
r
a
m
e
te
r
s
a
r
e
s
um
m
a
r
iz
e
d i
n T
a
bl
e
2.
T
a
bl
e
2. P
a
r
a
m
e
te
r
s
of
s
e
nt
im
e
nt
a
na
ly
s
i
s
m
ode
ll
in
g
P
a
r
a
m
e
t
e
r
V
a
l
ue
B
a
t
c
h
s
i
z
e
16
O
pt
i
m
i
z
e
r
A
da
m
W
E
a
r
l
y
s
t
oppi
ng
M
C
C
(
de
l
t
a
=0.01)
M
a
x
s
e
qu
e
nc
e
l
e
ngt
h
512
L
e
a
r
ni
ng
r
a
t
e
5e
-
5
-
1e
-
5
E
poc
h
2
-
5
R
e
gul
a
r
i
z
a
t
i
on
0
-
0.2
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
,
V
ol
. 14, No. 3, J
une
2025
:
1829
-
1838
1834
A
f
te
r
th
e
s
e
nt
im
e
nt
a
n
a
ly
s
is
m
ode
l
is
c
r
e
a
te
d,
th
e
E
W
S
da
ta
s
e
t
s
ta
r
ts
to
b
e
pr
e
-
pr
oc
e
s
s
e
d
by
a
ddi
ng
va
r
ia
bl
e
s
f
r
om
th
e
s
e
nt
im
e
nt
a
na
ly
s
is
r
e
s
ul
ts
.
E
W
S
m
ode
ll
in
g
is
c
a
r
r
ie
d
out
us
in
g
th
e
C
a
tB
oos
t
a
lg
or
it
hm
w
it
h
a
pe
r
c
e
nt
a
ge
s
pi
lt
te
s
ti
ng
m
e
th
od
of
80%
a
nd
100
it
e
r
a
ti
o
ns
f
or
pr
e
di
c
ti
ng
th
e
de
bt
or
’
s
c
r
e
di
t
qua
li
ty
f
or
th
e
ne
xt
3,
6,
9,
12,
15,
18,
21,
a
nd
24
m
ont
h
s
.
T
he
pr
oc
e
s
s
of
di
vi
di
ng
th
e
da
ta
s
e
t
f
or
m
ode
ll
in
g
u
s
e
d
th
e
s
c
ik
it
-
le
a
r
n
(
s
kl
e
a
r
n)
pa
c
ka
ge
(
1.0.2
ve
r
s
io
n)
a
nd
th
e
C
a
tB
oos
t
a
lg
or
it
hm
f
r
om
th
e
C
a
tb
oos
t
pa
c
ka
ge
(
1.2.3 ve
r
s
io
n)
on t
he
J
upyt
e
r
N
ot
e
book (
8.2.0 ve
r
s
io
n)
.
3.4. P
e
r
f
or
m
an
c
e
m
e
as
u
r
e
m
e
n
t
T
he
pr
oc
e
s
s
of
m
e
a
s
ur
in
g
la
ngua
ge
m
ode
l
c
a
n
u
s
e
lo
g
pr
ob
a
bi
li
ty
a
nd
pe
r
pl
e
xi
ty
a
s
e
v
a
lu
a
ti
on
m
e
tr
ic
s
w
he
r
e
is
a
to
ke
n
a
t
t
-
pos
it
io
n
a
nd
p
is
th
e
pr
oba
bi
li
ty
[
35]
.
T
he
s
e
nt
im
e
nt
a
na
ly
s
is
m
od
e
l
a
nd
th
e
E
W
S
m
ode
l
us
e
th
e
c
onf
us
io
n
m
a
tr
ix
to
obt
a
in
p
r
e
c
is
io
n,
r
e
c
a
ll
,
a
nd
a
c
c
ur
a
c
y
va
lu
e
s
f
or
a
c
la
s
s
if
ic
a
ti
on
m
ode
l.
T
a
bl
e
3 s
ho
w
s
t
he
c
onf
us
io
n m
a
tr
ix
of
3 c
la
s
s
e
s
.
=
∑
2
(
|
<
)
1
=
(
1)
=
(
1
,
2
,
…
,
)
=
2
−
1
∑
l
o
g
2
(
|
<
)
1
=
(
2)
T
a
bl
e
3. C
onf
us
io
n
m
a
tr
ix
of
3 c
la
s
s
e
s
A
c
t
ua
l
P
r
e
di
c
t
i
on
C
l
a
s
s
1
C
l
a
s
s
2
C
l
a
s
s
3
C
l
a
s
s
1
T
r
ue
C
1
F
a
l
s
e
C
2a
F
a
l
s
e
C
3a
C
l
a
s
s
2
F
a
l
s
e
C
1a
T
r
ue
C
2
F
a
l
s
e
C
3b
C
l
a
s
s
3
F
a
l
s
e
C
1b
F
a
l
s
e
C
2b
T
r
ue
C
3
=
1
1
+
1
+
1
+
2
2
+
2
+
2
+
3
3
+
3
+
3
3
(
3)
=
1
1
+
2
+
3
+
2
1
+
2
+
3
+
3
1
+
2
+
3
3
(
4)
=
1
+
2
+
3
1
+
1
…
1
+
2
+
2
…
2
+
3
+
3
…
3
(
5)
T
o
ha
ve
a
be
tt
e
r
unde
r
s
ta
ndi
ng
of
how
th
e
E
W
S
m
ode
l
w
or
ks
w
he
th
e
r
it
ut
il
iz
e
s
th
e
r
e
s
ul
ts
of
s
e
nt
im
e
nt
a
na
ly
s
is
or
not
,
a
s
e
a
r
c
h
f
or
f
e
a
tu
r
e
im
por
ta
nc
e
is
c
a
r
r
ie
d
out
to
know
how
th
e
m
ode
l
m
a
k
e
s
d
e
c
is
io
ns
us
in
g
T
r
e
e
S
H
A
P
f
r
om
th
e
S
H
A
P
pa
c
ka
ge
.
T
r
e
e
S
H
A
P
is
de
s
ig
ne
d
to
e
xt
r
a
c
t
S
ha
pl
e
y
va
lu
e
s
f
r
om
s
pe
c
if
ic
tr
e
e
-
ba
s
e
d m
ode
ls
s
o t
ha
t
it
i
s
s
ui
ta
bl
e
f
or
t
he
C
a
tB
oos
t
a
lg
or
it
hm
[
36]
, [
37]
.
4.
R
E
S
U
L
T
S
A
N
D
D
I
S
C
U
S
S
I
O
N
W
hi
le
e
a
r
li
e
r
r
e
s
e
a
r
c
h
s
uc
c
e
s
s
f
ul
ly
m
ode
le
d
th
e
c
la
s
s
ic
f
in
a
nc
ia
l
r
e
por
ts
to
c
a
tc
h
th
e
s
ig
na
l
of
de
te
r
io
r
a
ti
on, a
nd t
he
f
in
a
nc
ia
l
ne
w
s
ha
s
a
pos
it
iv
e
c
or
r
e
la
ti
on
w
it
h i
t,
no
r
e
s
e
a
r
c
h ha
s
c
om
bi
ne
d t
he
f
in
a
nc
ia
l
r
e
por
ts
w
it
h
th
e
f
in
a
nc
ia
l
ne
w
s
.
S
ta
r
t
by
m
a
ki
ng
a
s
e
nt
im
e
nt
a
na
ly
s
is
m
ode
l
f
r
om
f
in
a
nc
ia
l
ne
w
s
th
e
n
th
e
r
e
s
ul
t
is
a
dde
d t
o t
he
f
in
a
nc
ia
l
r
e
por
ts
. A
f
te
r
i
t
is
a
dde
d, i
t
w
il
l
be
t
he
da
ta
s
e
t
of
e
nha
n
c
e
d E
W
S
.
4.1. S
e
n
t
im
e
n
t
an
al
ys
is
m
od
e
l
F
r
om
w
e
b
s
c
r
a
pi
ng
f
r
om
th
e
C
N
B
C
I
ndone
s
ia
ne
w
s
por
ta
l,
th
e
r
e
a
r
e
258,678
ne
w
s
c
ol
le
c
te
d,
a
nd
it
w
a
s
us
e
d
a
s
a
da
t
a
s
e
t
to
f
in
e
-
tu
ne
I
ndoB
E
R
T
in
to
a
ne
w
la
ngua
ge
m
ode
l.
A
s
s
how
n
in
T
a
bl
e
4,
th
e
r
e
s
ul
ts
of
th
e
n
e
w
la
ngua
g
e
in
c
r
e
a
s
e
lo
g
pr
oba
bi
li
ty
b
y
7.04
a
nd
r
e
duc
e
p
e
r
pl
e
xi
ty
by
58,275.74,
in
di
c
a
ti
ng
th
a
t
th
e
r
e
s
ul
ts
of
f
in
e
-
tu
ni
ng
I
ndoB
E
R
T
in
to
a
ne
w
la
ngua
ge
m
ode
l
c
a
n
r
e
c
ogni
z
e
f
in
a
nc
i
a
l
ne
w
s
te
xt
be
tt
e
r
t
ha
n t
he
or
ig
in
a
l
I
ndoB
E
R
T
m
ode
l
w
it
hout
f
in
e
-
tu
ne
. T
he
s
e
nt
im
e
nt
a
na
ly
s
is
c
r
e
a
te
d f
r
om
t
he
ne
w
la
ngua
ge
m
ode
l
pr
oduc
e
s
th
e
b
e
s
t
r
e
s
ul
ts
on
hype
r
pa
r
a
m
e
te
r
s
le
a
r
ni
ng
r
a
te
=
3e
-
5,
r
e
gul
a
r
iz
a
ti
on
=
0.1,
a
nd
e
poc
h=
5
w
it
h
pr
e
c
is
io
n
of
80.89%
,
r
e
c
a
ll
of
80.67%
,
a
nd
a
c
c
ur
a
c
y
of
80.67%
on
te
s
t
d
a
ta
.
T
hi
s
r
e
s
ul
t
s
how
s
th
a
t
f
in
e
-
tu
ni
ng
th
e
ne
w
la
ngua
ge
m
ode
l
c
a
n
a
c
hi
e
v
e
be
tt
e
r
r
e
s
ul
ts
c
om
pa
r
e
d
to
th
e
or
ig
in
a
l
la
ngua
ge
m
ode
l.
A
s
s
how
n
in
T
a
bl
e
5,
th
is
r
e
s
ul
t
is
hi
ghe
r
c
om
pa
r
e
d
to
ot
he
r
s
im
il
a
r
pr
e
vi
ous
r
e
s
e
a
r
c
h
f
r
o
m
A
nde
r
ie
s
e
t
al
.
[
17]
w
hi
c
h
obt
a
in
e
d
a
n
a
c
c
ur
a
c
y
of
onl
y
68%
on
te
s
t
da
ta
.
T
hi
s
m
ode
l
is
us
e
d
to
ge
ne
r
a
te
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
T
he
i
nf
lu
e
nc
e
of
s
e
nt
ime
nt
analy
s
is
i
n e
nhanc
in
g e
a
r
ly
w
a
r
ni
ng
s
y
s
te
m
m
od
e
l
fo
r
c
r
e
di
t
…
(
A
nge
l
K
ar
e
nt
ia
)
1835
ne
w
s
s
e
nt
im
e
nt
a
nd
a
dd
it
to
th
e
E
W
S
da
ta
s
e
t.
H
ow
e
ve
r
,
t
hi
s
good
r
e
s
ul
t
c
a
n
be
obt
a
in
e
d
be
c
a
us
e
th
e
pr
opos
e
d
m
e
th
od
c
ont
in
ue
d
th
e
pr
oc
e
s
s
of
or
ig
in
a
l
la
ngua
ge
m
ode
l
le
a
r
ni
ng,
s
o
it
ha
s
a
b
e
tt
e
r
unde
r
s
ta
ndi
ng
of
th
e
da
ta
s
e
t.
T
he
pr
opo
s
e
d
m
e
th
od
a
ls
o
onl
y
u
s
e
s
one
s
our
c
e
of
f
in
a
nc
ia
l
n
e
w
s
w
hi
c
h
c
a
n
h
a
ve
th
e
s
a
m
e
w
r
it
in
g
s
ty
le
.
F
ut
ur
e
r
e
s
e
a
r
c
h
m
a
y
a
dd
m
or
e
r
e
s
our
c
e
s
of
da
ta
to
gi
ve
m
or
e
va
r
ia
ti
on
to
gi
ve
a
be
tt
e
r
ge
ne
r
a
li
z
a
ti
on.
T
a
bl
e
4. R
e
s
ul
t
of
f
in
e
-
tu
ne
I
ndoB
E
R
T
i
nt
o
la
ngua
ge
m
ode
l
F
i
ne
-
t
une
s
t
a
t
us
L
og
pr
oba
bi
l
i
t
y
P
e
r
pl
e
xi
t
y
B
e
f
or
e
-
10.97
58,326.80
A
f
t
e
r
-
3.93
51.06
T
a
bl
e
5. C
om
pa
r
is
on t
o
ot
he
r
s
s
im
il
a
r
r
e
s
e
a
r
c
h
R
e
s
e
a
r
c
he
r
R
e
s
ul
t
(
A
c
c
ur
a
c
y)
(%)
A
nde
r
i
e
s
e
t
al
.
[
17]
68
T
hi
s
R
e
s
e
a
r
c
h
81
4.2. E
WS
m
od
e
l
A
s
um
m
a
r
y
of
th
e
pe
r
f
or
m
a
nc
e
of
th
e
E
W
S
m
ode
l
c
a
n
be
s
e
e
n
in
T
a
bl
e
6.
F
r
om
th
e
r
e
s
ul
ts
of
th
e
E
W
S
m
ode
ll
in
g,
th
e
pr
e
c
is
io
n,
r
e
c
a
ll
,
a
nd
a
c
c
ur
a
c
y
va
lu
e
s
a
f
te
r
a
ddi
ng
th
e
s
e
nt
im
e
nt
a
na
ly
s
is
r
e
s
ul
t
s
a
r
e
m
os
tl
y
hi
ghe
r
th
a
n
be
f
or
e
a
ddi
ng
th
e
s
e
nt
im
e
nt
a
na
ly
s
is
r
e
s
ul
ts
s
o
th
e
E
W
S
pe
r
f
or
m
a
nc
e
a
f
te
r
a
ddi
ng
th
e
s
e
nt
im
e
nt
a
na
ly
s
is
r
e
s
ul
ts
pr
oduc
e
s
be
tt
e
r
pe
r
f
or
m
a
nc
e
.
H
ow
e
ve
r
,
ove
r
ti
m
e
th
e
pr
e
di
c
te
d
c
r
e
di
t
qua
li
ty
s
ti
ll
s
how
s
poor
r
e
s
ul
ts
.
I
n
a
ddi
ti
on,
E
W
S
be
f
or
e
a
nd
a
f
te
r
a
ddi
ng
s
e
nt
im
e
nt
a
na
ly
s
is
r
e
s
ul
ts
a
lw
a
ys
ha
s
hi
ghe
r
pr
e
c
is
io
n
c
om
pa
r
e
d
to
r
e
c
a
ll
,
in
di
c
a
ti
ng
th
a
t
E
W
S
ha
s
de
s
c
r
ib
e
d
th
e
in
f
or
m
a
ti
on
pr
ovi
de
d
e
ve
n
th
ough
th
e
le
ve
l
of
i
nf
or
m
a
ti
on r
e
tr
ie
va
l
is
qui
te
l
ow
.
T
a
bl
e
6. R
e
s
ul
t
of
f
in
e
-
tu
ne
I
ndoB
E
R
T
i
nt
o
la
ngua
ge
m
ode
l
C
r
e
di
t
qua
l
i
t
y ne
xt
W
i
t
hout
s
e
nt
i
m
e
nt
a
na
l
ys
i
s
r
e
s
ul
t
W
i
t
h s
e
nt
i
m
e
nt
a
na
l
ys
i
s
r
e
s
ul
t
D
i
f
f
e
r
e
nc
e
(
A
c
c
ur
a
c
y)
(%)
P
r
e
c
i
s
i
on
(%)
R
e
c
a
l
l
(%)
A
c
c
ur
a
c
y
(%)
P
r
e
c
i
s
i
on
(%)
R
e
c
a
l
l
(%)
A
c
c
ur
a
c
y
(%)
3 M
ont
hs
87.40
68.37
97.51
88.39
69.24
97.56
0.05
6 M
ont
hs
84.09
63.29
95.56
85.38
63.60
95.69
0.13
9 M
ont
hs
77.46
55.82
93.59
78.64
55.28
93.62
0.03
12 M
ont
hs
79.95
53.94
91.89
80.14
52.65
91.92
0.03
15 M
ont
hs
72.47
44.33
89.87
72.95
43.62
89.89
0.02
18 M
ont
hs
74.80
41.76
88.44
72.06
43.11
88.46
0.02
21 M
ont
hs
71.10
41.46
87.31
74.78
43.11
87.55
0.24
24 M
ont
hs
71.31
38.30
86.30
75.14
42.90
86.96
0.66
T
he
E
W
S
m
ode
l
a
dde
d
w
it
h
th
e
s
e
nt
im
e
nt
a
na
ly
s
is
r
e
s
ul
ts
w
a
s
th
e
n
a
na
ly
z
e
d
us
in
g
T
r
e
e
S
H
A
P
to
a
s
s
e
s
s
th
e
f
e
a
tu
r
e
im
por
ta
nc
e
of
th
e
s
e
nt
im
e
nt
a
na
ly
s
is
r
e
s
u
lt
s
ba
s
e
d
on
th
e
s
iz
e
of
th
e
S
ha
pl
e
y
v
a
lu
e
.
F
ig
ur
e
2
s
how
s
th
e
m
ove
m
e
nt
of
th
e
a
v
e
r
a
ge
S
ha
pl
e
y
va
lu
e
of
t
he
e
nt
ir
e
E
W
S
da
ta
s
e
t
a
s
th
e
pr
e
di
c
te
d
pe
r
io
d
in
c
r
e
a
s
e
s
. F
r
om
t
he
f
ig
ur
e
, t
he
hi
ghe
s
t
a
ve
r
a
ge
S
ha
pl
e
y va
lu
e
i
s
pr
e
di
c
ti
ng c
r
e
di
t
qua
li
ty
f
o
r
12 mont
hs
w
hi
c
h
m
e
a
ns
t
ha
t
E
W
S
i
s
m
os
t
e
f
f
e
c
ti
ve
i
n pr
e
di
c
ti
ng c
r
e
di
t
qua
li
ty
f
or
t
he
ne
xt
12 mont
hs
.
T
he
va
r
ia
bl
e
of
w
he
th
e
r
th
e
r
e
is
a
pr
e
s
e
n
c
e
or
a
bs
e
n
c
e
,
a
nd
th
e
to
ta
l
of
ne
ga
ti
ve
ne
w
s
ba
s
e
d
on
th
e
de
bt
or
'
s
in
dus
tr
ia
l
s
e
c
to
r
ha
s
a
s
m
a
ll
a
ve
r
a
ge
S
ha
pl
e
y
va
lu
e
w
he
n
c
om
pa
r
e
d
to
ot
he
r
va
r
ia
bl
e
s
.
I
t
m
e
a
ns
th
a
t
th
e
s
e
nt
im
e
nt
a
na
ly
s
is
r
e
s
ul
t
is
not
th
e
m
a
in
c
om
pone
nt
of
th
e
E
W
S
m
ode
l
to
pr
e
di
c
t
th
e
c
r
e
di
t
qua
li
ty
in
th
e
f
ut
ur
e
w
he
r
e
it
is
c
le
a
r
f
r
om
th
e
g
r
a
ph
th
a
t
th
e
a
ve
r
a
ge
S
ha
pl
e
y
va
lu
e
of
th
e
us
a
ge
a
nd
in
te
r
na
l
r
is
k
g
r
a
di
ng
va
r
ia
bl
e
s
is
gr
e
a
te
r
th
a
n
th
e
ot
h
e
r
va
r
ia
bl
e
s
.
H
ow
e
ve
r
,
th
e
a
v
e
r
a
ge
S
ha
pl
e
y
va
lu
e
of
th
e
va
r
ia
bl
e
w
he
th
e
r
th
e
r
e
is
pr
e
s
e
nc
e
or
a
bs
e
nc
e
a
nd
th
e
a
m
ount
of
ne
ga
ti
ve
ne
w
s
ba
s
e
d
on
th
e
de
bt
or
'
s
in
dus
tr
ia
l
s
e
c
to
r
te
nds
t
o
in
c
r
e
a
s
e
,
w
hi
c
h
m
e
a
n
s
th
e
im
por
ta
nc
e
of
th
e
va
r
ia
bl
e
r
e
s
ul
ti
ng
f
r
om
s
e
nt
im
e
nt
a
n
a
ly
s
is
in
c
r
e
a
s
e
s
a
s
th
e
pr
e
di
c
te
d
pe
r
io
d
in
c
r
e
a
s
e
s
,
w
hi
c
h
in
di
c
a
t
e
s
th
a
t
E
W
S
le
ve
r
a
ge
s
th
e
r
e
s
ul
ts
of
s
e
nt
im
e
nt
a
na
ly
s
i
s
f
r
om
ne
w
s
r
e
la
te
d
to
e
c
onomi
c
f
lu
c
tu
a
ti
ons
to
a
s
s
e
s
f
ut
ur
e
c
ondi
ti
ons
s
o
th
a
t
it
c
a
n
s
uppor
t
a
be
tt
e
r
c
r
e
di
t
r
is
k
m
it
ig
a
ti
on
pr
oc
e
s
s
by
pr
ovi
di
ng
w
a
r
ni
ngs
be
f
or
e
th
e
de
bt
or
’
s
c
r
e
di
t
qua
l
it
y
s
ta
r
t
to
de
te
r
io
r
a
ti
on.
T
he
s
e
r
e
s
ul
ts
pr
ove
th
a
t
a
ddi
ng
s
e
nt
im
e
nt
a
na
ly
s
is
r
e
s
ul
t
s
f
r
om
f
in
a
nc
ia
l
ne
w
s
to
f
in
a
nc
ia
l
r
e
por
ts
da
ta
c
a
n
c
r
e
a
te
a
f
or
w
a
r
d
-
lo
oki
ng
E
W
S
.
F
ut
ur
e
r
e
s
e
a
r
c
h
m
a
y
s
tu
dy
a
not
he
r
m
e
th
od
to
pr
oduc
e
E
W
S
w
it
h
r
e
li
a
bl
e
pe
r
f
or
m
a
nc
e
of
in
f
or
m
a
ti
on r
e
tr
ie
va
l
ove
r
t
i
m
e
t
he
pr
e
di
c
te
d c
r
e
di
t
qua
li
ty
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
,
V
ol
. 14, No. 3, J
une
2025
:
1829
-
1838
1836
F
ig
ur
e
2. A
ve
r
a
ge
S
ha
pl
e
y va
lu
e
of
E
W
S
m
ode
l
a
f
te
r
a
ddi
ng s
e
nt
im
e
nt
a
na
ly
s
is
r
e
s
ul
ts
5.
C
O
N
C
L
U
S
I
O
N
T
hi
s
r
e
s
e
a
r
c
h
s
u
c
c
e
s
s
f
ul
ly
im
pr
ove
d
E
W
S
m
ode
l
a
c
c
ur
a
c
y
by
a
ddi
ng
th
e
r
e
s
ul
ts
of
ne
w
s
s
e
nt
im
e
nt
in
di
c
a
te
d
by
th
e
m
os
tl
y
hi
ghe
r
pr
e
c
is
io
n,
r
e
c
a
ll
,
a
nd
a
c
c
ur
a
c
y
va
lu
e
s
th
a
n
be
f
or
e
a
ddi
ng
th
e
s
e
nt
im
e
nt
a
na
ly
s
is
.
A
pa
r
t
f
r
om
th
a
t,
th
e
a
ve
r
a
ge
S
ha
pl
e
y
va
lu
e
of
th
e
s
e
nt
im
e
nt
a
na
ly
s
is
r
e
s
ul
t
in
c
r
e
a
s
e
s
a
s
th
e
pr
e
di
c
te
d
pe
r
io
d
in
c
r
e
a
s
e
s
.
I
t
m
e
a
ns
th
a
t
th
e
E
W
S
ut
il
iz
e
s
th
e
r
e
s
ul
ts
of
s
e
nt
im
e
nt
a
na
ly
s
is
f
r
om
ne
w
s
r
e
la
te
d
to
e
c
onomi
c
f
lu
c
tu
a
ti
ons
t
o s
e
e
f
ut
ur
e
c
ondi
ti
ons
t
ha
t
m
a
ke
t
he
E
W
S
f
or
w
a
r
d
-
lo
oki
ng. T
he
f
ut
ur
e
di
r
e
c
ti
on o
f
th
is
r
e
s
e
a
r
c
h
w
oul
d
us
e
bi
g
da
t
a
te
c
hnol
ogy
to
s
im
pl
if
y
th
e
pr
oc
e
s
s
of
c
om
bi
ni
ng
f
in
a
nc
ia
l
r
e
por
ts
w
it
h
th
e
f
in
a
nc
ia
l
ne
w
s
i
n r
e
a
l
-
ti
m
e
t
o a
c
c
e
le
r
a
te
c
r
e
di
t
r
is
k m
it
ig
a
ti
on.
F
U
N
D
I
N
G
I
N
F
O
R
M
A
T
I
O
N
T
hi
s
r
e
s
e
a
r
c
h
r
e
c
e
iv
e
d
no
s
pe
c
if
ic
gr
a
nt
f
r
om
a
ny
f
undi
ng
a
ge
nc
y
in
th
e
publ
ic
,
c
om
m
e
r
c
ia
l,
or
not
-
f
or
-
pr
o
f
it
s
e
c
to
r
s
.
A
U
T
H
O
R
C
O
N
T
R
I
B
U
T
I
O
N
S
S
T
A
T
E
M
E
N
T
T
hi
s
jo
ur
na
l
us
e
s
th
e
C
ont
r
ib
ut
or
R
ol
e
s
T
a
xonomy
(
C
R
e
d
iT
)
to
r
e
c
ogni
z
e
in
di
vi
dua
l
a
ut
hor
c
ont
r
ib
ut
io
ns
, r
e
duc
e
a
ut
hor
s
hi
p di
s
put
e
s
,
a
nd f
a
c
il
it
a
te
c
ol
la
bo
r
a
ti
on.
N
am
e
o
f
A
u
t
h
or
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
A
nge
l
K
a
r
e
nt
ia
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
D
e
r
w
in
S
uha
r
to
no
✓
✓
✓
✓
✓
✓
✓
C
:
C
onc
e
pt
ua
l
i
z
a
t
i
on
M
:
M
e
t
hodol
ogy
So
:
So
f
t
w
a
r
e
Va
:
Va
l
i
da
t
i
on
Fo
:
Fo
r
m
a
l
a
na
l
ys
i
s
I
:
I
nve
s
t
i
ga
t
i
on
R
:
R
e
s
our
c
e
s
D
:
D
a
t
a
C
ur
a
t
i
on
O
:
W
r
i
t
i
ng
-
O
r
i
gi
na
l
D
r
a
f
t
E
:
W
r
i
t
i
ng
-
R
e
vi
e
w
&
E
di
t
i
ng
Vi
:
Vi
s
ua
l
i
z
a
t
i
on
Su
:
Su
pe
r
vi
s
i
on
P
:
P
r
oj
e
c
t
a
dm
i
ni
s
t
r
a
t
i
on
Fu
:
Fu
ndi
ng a
c
qui
s
i
t
i
on
C
O
N
F
L
I
C
T
O
F
I
N
T
E
R
E
S
T
S
T
A
T
E
M
E
N
T
T
he
a
ut
hor
s
de
c
la
r
e
t
ha
t
th
e
r
e
i
s
no c
onf
li
c
t
of
i
nt
e
r
e
s
t
r
e
ga
r
di
n
g t
he
publi
c
a
ti
on of
t
hi
s
pa
pe
r
.
D
A
T
A
A
V
A
I
L
A
B
I
L
I
T
Y
N
e
w
s
da
ta
a
r
e
a
va
il
a
bl
e
f
r
om
th
e
c
or
r
e
s
ponding
a
ut
hor
,
[
A
K
]
,
upon
r
e
a
s
ona
bl
e
r
e
que
s
t.
T
he
de
bt
or
da
ta
s
uppor
ti
ng
th
e
f
in
di
ngs
of
th
is
s
tu
dy
c
a
nnot
be
publ
i
s
he
d
due
to
th
e
s
tr
ic
tn
e
s
s
of
I
ndone
s
ia
n
ba
nki
ng
la
w
s
a
nd
r
e
gul
a
ti
ons
,
e
s
pe
c
ia
ll
y
r
e
ga
r
di
ng
th
e
pr
ot
e
c
ti
on
of
pe
r
s
ona
l
da
ta
a
nd
c
onf
id
e
nt
ia
li
ty
of
c
us
to
m
e
r
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
T
he
i
nf
lu
e
nc
e
of
s
e
nt
ime
nt
analy
s
is
i
n e
nhanc
in
g e
a
r
ly
w
a
r
ni
ng
s
y
s
te
m
m
od
e
l
fo
r
c
r
e
di
t
…
(
A
nge
l
K
ar
e
nt
ia
)
1837
da
ta
.
F
ur
th
e
r
m
or
e
,
w
e
ha
ve
c
onduc
te
d
a
s
c
r
e
e
ni
ng
pr
oc
e
s
s
of
th
e
de
bt
or
da
ta
/i
nf
or
m
a
ti
on
us
e
d
in
th
is
r
e
s
e
a
r
c
h
a
nd
e
ns
ur
e
d
th
a
t
th
e
da
ta
/i
nf
or
m
a
ti
on
doe
s
not
vi
o
la
te
th
e
de
bt
or
'
s
pr
iv
a
c
y/
pe
r
s
on
a
l
r
ig
ht
s
a
nd
c
om
pl
ia
nc
e
w
it
h ba
nki
ng r
e
gul
a
ti
ons
i
n I
ndone
s
ia
.
R
E
F
E
R
E
N
C
E
S
[
1]
S
.
N
a
s
e
e
r
,
S
.
K
ha
l
i
d,
S
.
P
a
r
ve
e
n,
K
.
A
bba
s
s
,
H
.
S
ong,
a
nd
M
.
V
.
A
c
hi
m
,
“
C
O
V
I
D
-
19
out
br
e
a
k:
i
m
pa
c
t
on
gl
oba
l
e
c
onom
y,”
F
r
ont
i
e
r
s
i
n P
ubl
i
c
H
e
al
t
h
, vol
. 10, 2023, doi
:
10.3389/
f
pubh.2022.1009393.
[
2]
W
or
l
d
B
a
nk,
“
W
or
l
d
de
ve
l
opm
e
nt
r
e
por
t
2022:
f
i
na
nc
e
f
or
a
n
e
qui
t
a
bl
e
r
e
c
ove
r
y,”
W
or
l
d
D
e
v
e
l
opm
e
nt
R
e
po
r
t
,
W
a
s
hi
ngt
on
:
U
ni
t
e
d S
t
a
t
e
s
, vol
. 1,
2022.
[
3]
A
.
N
.
B
e
r
ge
r
,
P
.
M
ol
yne
ux,
a
nd
J
.
O
.
S
.
W
i
l
s
on,
T
he
O
x
f
o
r
d
handbook
of
ban
k
i
n
g,
O
xf
or
d,
U
ni
t
e
d
K
i
ngdom
:
O
xf
or
d
U
ni
ve
r
s
i
t
y
P
r
e
s
s
, 2014
, doi
:
10.1093/
oxf
or
dhb/
9780199688500.001.0001
.
[
4]
E
.
J
.
C
a
s
a
bi
a
nc
a
,
M
.
C
a
t
a
l
a
no,
L
.
F
or
ni
,
E
.
G
i
a
r
da
,
a
nd
S
.
P
a
s
s
e
r
i
,
A
n
e
ar
l
y
w
ar
ni
ng
s
y
s
t
e
m
f
or
ban
k
i
ng
c
r
i
s
e
s
:
f
r
om
r
e
g
r
e
s
s
i
on
-
bas
e
d anal
y
s
i
s
t
o m
ac
hi
ne
l
e
ar
ni
ng t
e
c
hni
que
s
,
P
a
dova
, I
t
a
l
y:
M
a
r
c
o F
a
nno W
or
ki
ng P
a
pe
r
s
,
2019.
[
5]
G
a
l
yt
i
x
a
nd
P
w
C
,
“
B
a
nks
m
u
s
t
a
c
t
on
t
he
i
r
e
a
r
l
y
w
a
r
ni
ng
s
ys
t
e
m
s
or
r
i
s
k
R
O
E
dow
nt
ur
n
i
n
a
s
s
oc
i
a
t
i
on
w
i
t
h,”
G
al
t
i
x
L
i
m
i
t
e
d
,
L
ondon, E
ngl
a
nd,
2022.
[
6]
M
.
L
e
o,
S
.
S
ha
r
m
a
,
a
nd
K
.
M
a
ddul
e
t
y,
“
M
a
c
hi
ne
l
e
a
r
ni
ng
i
n
ba
nki
ng
r
i
s
k
m
a
na
ge
m
e
nt
:
a
l
i
t
e
r
a
t
ur
e
r
e
vi
e
w
,”
R
i
s
k
s
,
vol
.
7,
no.
1,
2019, doi
:
10.3390/
r
i
s
ks
7010029.
[
7]
M
.
E
.
K
.
A
gor
a
ki
,
N
.
A
s
l
a
ni
di
s
,
a
nd
G
.
P
.
K
our
e
t
a
s
,
“
U
.S
.
ba
nks
’
l
e
ndi
ng,
f
i
na
nc
i
a
l
s
t
a
bi
l
i
t
y,
a
nd
t
e
xt
-
ba
s
e
d
s
e
nt
i
m
e
nt
a
na
l
y
s
i
s
,
”
J
our
nal
of
E
c
onom
i
c
B
e
hav
i
or
and O
r
gani
z
at
i
on
, vol
. 197, pp. 73
–
90, 2022, doi
:
10.1016/
j
.j
e
bo.2022.02.025.
[
8]
S
.
K
i
m
,
S
.
C
hoi
,
a
nd
J
.
S
e
ok,
“
K
e
yw
or
d
e
xt
r
a
c
t
i
on
i
n
e
c
onom
i
c
s
l
i
t
e
r
a
t
ur
e
s
us
i
ng
na
t
ur
a
l
l
a
ngua
ge
pr
oc
e
s
s
i
ng,”
2021
T
w
e
l
f
t
h
I
nt
e
r
nat
i
onal
C
onf
e
r
e
n
c
e
on
U
bi
qui
t
ous
and
F
ut
ur
e
N
e
t
w
or
k
s
(
I
C
U
F
N
)
,
J
e
j
u
I
s
l
a
nd,
K
or
e
a
,
R
e
publ
i
c
of
,
2021,
pp.
75
-
77,
doi
:
10.1109/
I
C
U
F
N
49451.2021.9528546.
[
9]
B
.
L
i
u,
Se
nt
i
m
e
nt
anal
y
s
i
s
:
m
i
ni
ng
opi
ni
ons
,
s
e
nt
i
m
e
nt
s
,
and
e
m
ot
i
ons
,
s
e
c
ond
e
di
t
i
on
.
C
a
m
br
i
dge
,
U
ni
t
e
d
K
i
ngdom
:
C
a
m
br
i
dge
U
ni
ve
r
s
i
t
y P
r
e
s
s
, 2020.
[
10]
A
.
R
a
ha
l
i
a
nd
M
.
A
.
A
khl
ouf
i
,
“
E
nd
-
to
-
e
nd
t
r
a
ns
f
or
m
e
r
-
ba
s
e
d
m
ode
l
s
i
n
t
e
xt
ua
l
-
ba
s
e
d
N
L
P
,”
AI
,
vol
.
4,
no.
1,
pp.
54
–
110,
2023,
doi
:
10.3390/
a
i
4010004.
[
11]
T
.
W
ol
f
e
t
al
.
,
“
T
r
a
ns
f
or
m
e
r
s
:
s
t
a
t
e
-
of
-
t
he
-
a
r
t
na
t
ur
a
l
l
a
ngua
ge
pr
oc
e
s
s
i
ng,”
E
M
N
L
P
2020
-
C
onf
e
r
e
nc
e
on
E
m
pi
r
i
c
al
M
e
t
hods
i
n
N
at
ur
al
L
anguage
P
r
oc
e
s
s
i
ng,
P
r
oc
e
e
di
ng
s
of
Sy
s
t
e
m
s
D
e
m
ons
t
r
at
i
ons
,
A
s
s
o
c
i
a
t
i
on
f
or
C
om
put
a
t
i
ona
l
L
i
ngui
s
t
i
c
s
,
pp.
38
–
45,
2020, doi
:
10.18653/
v1/
2020.e
m
nl
p
-
de
m
os
.6.
[
12]
A
.
C
os
t
e
a
,
“
A
n
e
a
r
l
y
-
w
a
r
ni
ng
s
ys
t
e
m
f
or
f
i
na
nc
i
a
l
pe
r
f
or
m
a
nc
e
pr
e
di
c
t
i
ons
,”
E
c
onom
i
c
C
om
put
at
i
on
and
E
c
onom
i
c
C
y
be
r
ne
t
i
c
s
St
udi
e
s
and R
e
s
e
ar
c
h
, vol
. 56, no. 2, pp. 5
–
20, 2022, doi
:
10.24818/
18423264/
56.2.22.01.
[
13]
A
.
S
i
ddi
que
,
M
.
A
.
K
ha
n,
a
nd
Z
.
K
ha
n,
“
T
he
e
f
f
e
c
t
of
c
r
e
di
t
r
i
s
k
m
a
na
g
e
m
e
nt
a
nd
ba
nk
-
s
pe
c
i
f
i
c
f
a
c
t
or
s
on
t
he
f
i
na
nc
i
a
l
pe
r
f
or
m
a
nc
e
of
t
he
S
out
h A
s
i
a
n
c
om
m
e
r
c
i
a
l
ba
nk
s
,”
A
s
i
an
J
our
nal
of
A
c
c
ount
i
ng R
e
s
e
ar
c
h
,
vol
. 7, no.
2, pp.
182
–
194, 2022, doi
:
10.1108/
A
J
A
R
-
08
-
2020
-
0071.
[
14]
A.
-
I
.
S
t
r
ă
c
hi
na
r
u
,
“
E
a
r
l
y
w
a
r
ni
ng
s
ys
t
e
m
s
f
or
ba
nki
ng
c
r
i
s
i
s
a
nd
s
ove
r
e
i
gn
r
i
s
k,”
J
our
nal
of
F
i
nanc
i
al
St
udi
e
s
and
R
e
s
e
ar
c
h
,
pp.
1
–
9, 2022, doi
:
10.5171/
2022.441237.
[
15]
S
.
B
e
n
J
a
be
ur
,
C
.
G
ha
r
i
b,
S
.
M
e
f
t
e
h
-
W
a
l
i
,
a
nd
W
.
B
e
n
A
r
f
i
,
“
C
a
t
B
oos
t
m
ode
l
a
nd
a
r
t
i
f
i
c
i
a
l
i
nt
e
l
l
i
ge
nc
e
t
e
c
hni
que
s
f
or
c
or
por
a
t
e
f
a
i
l
ur
e
pr
e
di
c
t
i
on,”
T
e
c
hnol
ogi
c
al
F
or
e
c
as
t
i
ng
and
Soc
i
al
C
hange
,
vol
.
166,
pp.
1
-
19,
M
a
y.
2021,
doi
:
10.1016/
j
.t
e
c
hf
or
e
.2021.120658.
[
16]
P
. L
a
bonne
a
nd
L
. A
. T
hor
s
r
ud,
R
i
s
k
y
ne
w
s
and c
r
e
di
t
m
ar
k
e
t
s
e
nt
i
m
e
nt
,
O
s
l
o,
N
or
w
a
y:
B
I
N
or
w
e
gi
a
n B
us
i
ne
s
s
S
c
hool
,
2023.
[
17]
A
nde
r
i
e
s
,
R
.
R
a
hut
om
o,
a
nd
B
.
P
a
r
da
m
e
a
n,
“
F
i
ne
t
unni
ng
I
ndoB
E
R
T
t
o
un
de
r
s
t
a
nd
i
ndone
s
i
a
n
s
t
oc
k
t
r
a
de
r
s
l
a
ng
l
a
ngua
ge
,
”
P
r
oc
e
e
di
ngs
of
2021
1s
t
I
nt
e
r
nat
i
onal
C
onf
e
r
e
nc
e
on
C
om
put
e
r
Sc
i
e
nc
e
and
A
r
t
i
f
i
c
i
al
I
n
t
e
l
l
i
ge
nc
e
,
I
C
C
SA
I
2021
,
pp.
42
–
46,
2021, doi
:
10.1109/
I
C
C
S
A
I
53272.2021.9609746.
[
18]
N
.
L
ua
ngna
r
ue
dom
a
nd
S
.
P
r
a
ka
nc
ha
r
oe
n,
“
L
oa
n
r
e
pa
ym
e
nt
s
t
a
t
u
s
pr
e
di
c
t
i
on,”
I
nt
e
r
nat
i
onal
J
our
nal
of
I
nnov
at
i
on,
C
r
e
at
i
v
i
t
y
and C
hange
,
vol
. 15, no. 4, pp. 133
-
153, 2021.
[
19]
Y
.
W
a
ng,
“
R
e
s
e
a
r
c
h
on
c
om
m
e
r
c
i
a
l
ba
nk
r
i
s
k
e
a
r
l
y
w
a
r
ni
ng
m
ode
l
ba
s
e
d
on
dyna
m
i
c
pa
r
a
m
e
t
e
r
opt
i
m
i
z
a
t
i
on
ne
ur
a
l
ne
t
w
or
k,”
J
our
nal
of
M
at
he
m
at
i
c
s
, vol
. 2022, 2022, doi
:
10.1155/
2022/
9754428.
[
20]
Y
.
C
he
ng,
Q
.
Y
a
ng,
L
.
W
a
ng,
A
.
X
i
a
ng,
a
nd
J
.
Z
ha
ng,
“
R
e
s
e
a
r
c
h
on
c
r
e
di
t
r
i
s
k
e
a
r
l
y
w
a
r
ni
ng
m
ode
l
of
c
om
m
e
r
c
i
a
l
ba
nk
s
ba
s
e
d
on
ne
ur
a
l
ne
t
w
or
k
a
l
gor
i
t
hm
,”
P
r
oc
e
di
a
C
om
put
e
r
Sc
i
e
nc
e
,
vol
.
199,
pp.
1168
–
1176,
M
a
y
2024,
doi
:
10.1016/
j
.pr
oc
s
.2022.01.148.
[
21]
H
.
W
e
i
a
nd
X
.
W
a
ng,
“
F
i
na
nc
i
a
l
r
i
s
k
m
a
na
ge
m
e
nt
e
a
r
l
y
-
w
a
r
ni
ng
m
ode
l
f
or
C
hi
ne
s
e
e
nt
e
r
pr
i
s
e
s
,”
J
our
nal
of
R
i
s
k
and
F
i
nanc
i
al
M
anage
m
e
nt
, vol
. 17, no. 7, 2024, doi
:
10.3390/
j
r
f
m
17070255.
[
22]
A
.
M
e
nge
l
ka
m
p,
K
.
K
oc
h,
a
nd
M
.
S
c
hum
a
nn,
“
C
r
e
a
t
i
ng
s
e
nt
i
m
e
nt
di
c
t
i
ona
r
i
e
s
:
pr
oc
e
s
s
m
ode
l
a
nd
qua
nt
i
t
a
t
i
ve
s
t
udy
f
or
c
r
e
di
t
r
i
s
k,”
9t
h E
ur
ope
an C
onf
e
r
e
n
c
e
on Soc
i
al
M
e
di
a, E
C
SM
2022
, pp. 121
–
129, 2022, doi
:
10.34190/
e
c
s
m
.9.1.167.
[
23]
N
.
A
hba
l
i
,
X
.
L
i
u,
A
.
A
.
N
a
nd
a
,
J
.
S
t
a
r
k,
A
.
T
a
l
ukde
r
,
a
nd
R
.
P
.
K
ha
ndpur
,
“
I
de
nt
i
f
yi
ng
c
or
por
a
t
e
c
r
e
di
t
r
i
s
k
s
e
nt
i
m
e
nt
s
f
r
om
f
i
na
nc
i
a
l
ne
w
s
,”
N
A
A
C
L
2022
-
2022
C
onf
e
r
e
nc
e
of
t
he
N
o
r
t
h
A
m
e
r
i
c
an
C
hapt
e
r
of
t
he
A
s
s
o
c
i
at
i
on
f
or
C
o
m
put
at
i
onal
L
i
ngui
s
t
i
c
s
:
H
um
an L
anguage
T
e
c
hnol
ogi
e
s
, I
ndus
t
r
y
P
ape
r
s
, pp. 362
–
370, 20
22, doi
:
10.18653/
v1/
2022.na
a
c
l
-
i
ndus
t
r
y.40.
[
24]
J
. C
. D
ua
n a
nd X
. Y
a
o, “
M
e
di
a
s
e
nt
i
m
e
nt
s
f
or
e
nha
nc
e
d c
r
e
di
t
r
i
s
k a
s
s
e
s
s
m
e
nt
,
”
N
U
S C
r
e
di
t
R
e
s
e
ar
c
h I
ni
t
i
at
i
v
e
,
pp. 1
-
26,
2022.
[
25]
F
.
G
a
m
ba
r
ot
a
a
nd
G
.
A
l
t
oè
,
“
O
r
di
na
l
r
e
gr
e
s
s
i
on
m
ode
l
s
m
a
de
e
a
s
y:
A
t
ut
or
i
a
l
on
pa
r
a
m
e
t
e
r
i
nt
e
r
pr
e
t
a
t
i
on,
da
t
a
s
i
m
ul
a
t
i
on
a
nd
pow
e
r
a
na
l
ys
i
s
,”
I
nt
e
r
nat
i
onal
J
ou
r
nal
of
P
s
y
c
hol
ogy
,
vol
. 59, no. 4, pp. 1263
-
1292,
2024, doi
:
10.1002/
i
j
op.13243.
[
26]
A
.
D
a
s
h,
T
he
L
L
M
adv
ant
age
:
how
t
o
unl
oc
k
t
he
pow
e
r
of
l
anguage
m
od
e
l
s
f
o
r
bus
i
ne
s
s
s
uc
c
e
s
s
:
a
pr
ac
t
i
c
al
gui
de
t
o
l
e
v
e
r
agi
ng
A
I
f
or
i
nnov
at
i
on, gr
ow
t
h, and c
om
pe
t
i
t
i
v
e
e
dge
, 1s
t
e
d.
T
ha
ne
W
e
s
t
, I
ndi
a
:
G
r
a
z
i
ng M
i
nds
P
ubl
i
s
hi
ng, 2023.
[
27]
K
.
M
i
s
he
v,
A
.
G
j
or
gj
e
vi
kj
,
I
.
V
ode
ns
ka
,
L
.
T
.
C
hi
t
kus
he
v,
a
nd
D
.
T
r
a
j
a
nov,
“
E
va
l
ua
t
i
on
of
s
e
nt
i
m
e
nt
a
na
l
ys
i
s
i
n
f
i
na
nc
e
:
f
r
om
l
e
xi
c
ons
t
o t
r
a
ns
f
or
m
e
r
s
,”
I
E
E
E
A
c
c
e
s
s
, vol
. 8, pp. 131662
–
131682, 2020, doi
:
10.1109/
A
C
C
E
S
S
.2020.3009626.
[
28]
D
.
A
r
a
c
i
,
“
F
i
nB
E
R
T
:
f
i
na
nc
i
a
l
s
e
nt
i
m
e
nt
a
n
a
l
ys
i
s
w
i
t
h
pr
e
-
t
r
a
i
ne
d
l
a
ngua
ge
m
ode
l
s
,”
M
.Sc
.
T
he
s
i
s
,
F
a
c
ul
t
y
of
S
c
i
e
nc
e
,
U
ni
ve
r
s
i
t
y of
A
m
s
t
e
r
da
m
,
A
m
s
t
e
r
da
m
, N
e
t
he
r
l
a
nds
,
2019.
[
29]
Z
.
L
i
u,
D
.
H
ua
ng,
K
.
H
ua
ng,
Z
.
L
i
,
a
nd
J
.
Z
ha
o,
“
F
i
nB
E
R
T
:
A
pr
e
-
t
r
a
i
ne
d
f
i
n
a
nc
i
a
l
l
a
ngua
ge
r
e
pr
e
s
e
nt
a
t
i
on
m
ode
l
f
or
f
i
na
nc
i
a
l
t
e
xt
m
i
ni
ng,”
P
r
oc
e
e
di
ngs
of
t
h
e
T
w
e
nt
y
-
N
i
nt
h I
nt
e
r
nat
i
onal
J
oi
nt
C
onf
e
r
e
n
c
e
on A
r
t
i
f
i
c
i
al
I
nt
e
l
l
i
ge
nc
e
(
I
J
C
A
I
-
20)
:
Spe
c
i
al
T
r
ac
k
on A
I
i
n F
i
n
T
e
c
h
, pp. 4513
–
4519, 2020, doi
:
10.24963/
i
j
c
a
i
.2020/
622.
[
30]
F
.
K
ot
o,
A
.
R
a
hi
m
i
,
J
.
H
.
L
a
u,
a
nd
T
.
B
a
l
dw
i
n,
“
I
ndoL
E
M
a
nd
I
ndoB
E
R
T
:
a
be
nc
hm
a
r
k
da
t
a
s
e
t
a
nd
pr
e
-
t
r
a
i
ne
d
l
a
ngua
g
e
m
ode
l
f
or
i
ndone
s
i
a
n
N
L
P
,”
C
O
L
I
N
G
2020
-
28t
h
I
nt
e
r
nat
i
onal
C
onf
e
r
e
nc
e
on
C
om
put
at
i
onal
L
i
ngui
s
t
i
c
s
,
P
r
oc
e
e
di
ngs
of
t
h
e
C
onf
e
r
e
nc
e
, pp. 757
–
770, 2020, doi
:
10.18653/
v1/
2020.c
ol
i
ng
-
m
a
i
n.66.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
,
V
ol
. 14, No. 3, J
une
2025
:
1829
-
1838
1838
[
31]
J
.
T
.
H
a
nc
oc
k
a
nd
T
.
M
.
K
ho
s
hgof
t
a
a
r
,
“
C
a
t
B
oos
t
f
or
bi
g
da
t
a
:
a
n
i
nt
e
r
di
s
c
i
pl
i
na
r
y
r
e
vi
e
w
,”
J
our
nal
of
B
i
g
D
at
a
,
vol
.
7,
no.
1
,
2020, doi
:
10.1186/
s
40537
-
020
-
00369
-
8.
[
32]
L
. O
w
e
n,
H
y
pe
r
par
a
m
e
t
e
r
t
uni
ng w
i
t
h P
y
t
hon
.
B
i
r
m
i
ngha
m
, U
ni
t
e
d K
i
ngdom
:
P
a
c
kt
P
ubl
i
s
hi
ng,
2022.
[
33]
M
.
O
.
I
br
ohi
m
a
nd
I
.
B
udi
,
“
M
ul
t
i
-
l
a
be
l
ha
t
e
s
pe
e
c
h
a
nd
a
bus
i
ve
l
a
ngu
a
ge
d
e
t
e
c
t
i
on
i
n
i
ndone
s
i
a
n
t
w
i
t
t
e
r
,”
P
r
oc
e
e
di
ng
s
of
t
he
T
hi
r
d W
or
k
s
hop on A
bus
i
v
e
L
anguage
O
nl
i
ne
,
2019, pp. 46
–
57, doi
:
10.18653/
v1/
w
19
-
3506.
[
34]
D
.
C
hi
c
c
o
a
nd
G
.
J
u
r
m
a
n,
“
T
he
a
dva
nt
a
ge
s
of
t
he
M
a
t
t
he
w
s
c
or
r
e
l
a
t
i
on
c
oe
f
f
i
c
i
e
nt
(
M
C
C
)
ove
r
F
1
s
c
or
e
a
nd
a
c
c
ur
a
c
y
i
n
b
i
na
r
y
c
l
a
s
s
i
f
i
c
a
t
i
on e
va
l
ua
t
i
on,”
B
M
C
G
e
nom
i
c
s
, vol
. 21, no. 1, 2020, doi
:
10.1186/
s
12864
-
019
-
6413
-
7.
[
35]
Y
.
P
r
uks
a
c
ha
t
kun,
M
.
M
c
a
t
e
e
r
,
a
nd
S
.
M
a
j
um
da
r
,
P
r
ac
t
i
c
i
ng
t
r
us
t
w
or
t
hy
m
a
c
hi
ne
l
e
ar
ni
ng:
c
ons
i
s
t
e
nt
,
t
r
an
s
par
e
nt
,
and
f
ai
r
A
I
pi
pe
l
i
ne
s
, 1s
t
e
d.
C
a
l
i
f
or
ni
a
, U
ni
t
e
d S
t
a
t
e
s
:
O
’
R
e
i
l
l
y M
e
di
a
, I
nc
, 2023.
[
36]
S
.
M
.
L
undbe
r
g
e
t
al
.
,
“
F
r
om
l
oc
a
l
e
xpl
a
na
t
i
ons
t
o
gl
oba
l
unde
r
s
t
a
ndi
ng
w
i
t
h
e
xpl
a
i
na
bl
e
A
I
f
or
t
r
e
e
s
,”
N
at
ur
e
M
ac
hi
n
e
I
nt
e
l
l
i
ge
nc
e
, vol
. 2, no. 1, pp. 56
–
67, 2020, doi
:
10.1038/
s
42256
-
019
-
0138
-
9.
[
37]
C
.
M
ol
na
r
,
I
nt
e
r
p
r
e
t
abl
e
m
ac
hi
n
e
l
e
ar
ni
ng:
a
gui
d
e
f
or
m
a
k
i
ng
bl
ac
k
bo
x
m
ode
l
s
e
x
pl
ai
nabl
e
,
2
nd
ed
.
N
or
t
h
C
a
r
ol
i
na
,
U
ni
t
e
d
S
t
a
t
e
s
:
L
e
a
n P
ubl
i
s
hi
ng,
2020.
B
I
O
G
R
A
P
H
I
E
S
O
F
A
U
T
H
O
R
S
Angel
Karentia
is
a
graduate
student
of
computer
science
from
B
ina
Nusantara
University,
Indonesia.
Currently,
she
is
working
as
a
data
analyst
specializing
in
risk
manageme
nt
in
the
financial
industry
in
Indonesia.
Her
resear
ch
int
erest
includes
artificia
l
intelligence
,
machine
learning,
and
deep
learning.
She
can
be
contacted
at
email:
angel
.karentia@
binus.ac.id
.
Derwin Suha
rtono
is a fac
ulty member
of
Bina Nusanta
ra Unive
rsi
ty, Indonesia
.
He got his P
h
.
D
.
degree in computer
science from
Universitas Indones
ia in 2018. His
research
fields
are
natural
language
processing.
Recently,
he
has
been
contin
ually
doing
research
in
argumentat
ion
mining
and
personali
ty
recogniti
on.
He
is
actively
in
volved
in
the
Indonesi
a
Association
of
Computation
al
Linguistics
(INACL),
a
national
s
cientific
association
in
Indonesia.
He
has
his
professional
memberships
in
ACM,
INSTICC,
a
nd
IACT.
He
also
takes
the
role
of
reviewer
in
several
international
conferences
and
journals.
He
ca
n
be
contacted
at
email:
dsuhartono@
binus.edu
.
Evaluation Warning : The document was created with Spire.PDF for Python.