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.
2389
~
2401
I
S
S
N
:
2252
-
8938
,
D
O
I
:
10.11591/
ij
a
i.
v
14
.i
3
.pp
2389
-
2401
2389
Jou
r
n
al
h
om
e
page
:
ht
tp
:
//
ij
ai
.
ia
e
s
c
or
e
.c
om
M
od
e
l
i
n
g se
n
t
i
m
e
n
t
an
al
y
si
s of
In
d
o
n
e
si
an
b
i
o
d
i
ve
r
si
t
y p
ol
i
c
y
T
w
e
e
t
s u
si
n
g In
d
oB
E
R
T
w
e
e
t
M
oh
am
m
ad
T
e
d
u
h
U
li
n
ia
n
s
yah
, A
s
r
il
Jar
in
, A
gu
n
g S
an
t
os
a, G
u
n
ar
s
o
R
e
s
e
a
r
c
h
C
e
nt
e
r
f
or
D
a
t
a
a
nd I
nf
or
m
a
t
i
on S
c
i
e
nc
e
s
,
R
e
s
e
a
r
c
h O
r
ga
ni
z
a
t
i
on f
or
E
l
e
c
t
r
oni
c
s
a
nd I
nf
or
m
a
t
i
c
s
,
N
a
t
i
ona
l
R
e
s
e
a
r
c
h a
nd
I
nnova
t
i
on
A
ge
nc
y (
B
R
I
N
)
, K
S
T
S
a
m
a
un S
a
m
a
di
kun B
a
ndung,
B
a
ndung,
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
28
,
2024
R
e
vi
s
e
d
F
e
b
12
,
2025
A
c
c
e
pt
e
d
M
a
r
15
,
2025
This
study
develops
and
evaluates
a
sentiment
analysis
model
using
IndoBERTweet
to
analyze
Twitter
data
on
Indonesia’s
biodiversi
ty
policy.
Twitter
data
focusing
on
topics
such
as
food
se
curity,
healt
h,
and
environm
ental
management
were
collected,
with
a
r
epresentati
ve
su
bset
of
13,435
tweets
annotated
from
a
larger
dataset
of
500,000
to
ensure
r
eliable
sentiment
labels
through
majority
voting.
IndoBERTweet
was
comp
ared
to
seven
traditional
machine
-
learning
classifier
s
using
TF
-
IDF
and
BERT
embeddin
gs
for
f
eature
extraction.
Model
performance
was
assessed
using
mean
accuracy
,
mean
F1
score
,
and
statistical
significance
(p
-
v
alues).
Additionally
,
sentiment
analysis
included
word
attribution
techniques
with
BERT
embeddings,
enhancing
relevance,
interpretability,
and
con
sistent
attribut
ion
to
deliver
accurate
insigh
ts.
IndoBER
Tweet
models
consi
stently
outperformed
traditional
methods
in
both
accuracy
and
F1
score.
While
BERT
embeddin
gs
boosted
performance
for
conventi
onal
models,
IndoBER
Tweet
delivered
superior
r
esults,
with
p
-
values
below
0.05
confirming
sta
tistical
significance.
This
approach
demonstrates
that
the
model’s
outpu
ts
are
explainab
le
and
align
with
human
understan
ding.
Findi
ngs
und
erscore
IndoBERTweet
’s
substanti
al
impact
on
advancing
sentiment
analy
sis
technology,
showcasing
its potential
to drive
innovation a
nd
elevate
pr
actices
in the field
.
K
e
y
w
o
r
d
s
:
B
E
R
T
e
m
be
ddi
ngs
B
io
di
ve
r
s
it
y poli
c
y
I
ndoB
E
R
T
w
e
e
t
S
e
nt
im
e
nt
a
na
ly
s
is
T
w
it
te
r
da
ta
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
s
r
il
J
a
r
in
R
e
s
e
a
r
c
h
C
e
nt
e
r
f
or
D
a
ta
a
nd I
nf
or
m
a
ti
on S
c
ie
nc
e
, N
a
ti
ona
l
R
e
s
e
a
r
c
h a
nd I
nnova
ti
on A
ge
nc
y
B
a
ndung, I
ndone
s
ia
E
m
a
il
:
a
s
r
i0
03@
br
in
.go.i
d
1.
I
N
T
R
O
D
U
C
T
I
O
N
I
ndone
s
ia
,
r
e
c
ogni
z
e
d
a
s
one
of
th
e
w
or
ld
’
s
m
os
t
s
ig
ni
f
ic
a
n
t
bi
odi
ve
r
s
it
y
hot
s
pot
s
w
it
h
a
hi
gh
c
onc
e
nt
r
a
ti
on
of
e
nde
m
ic
s
pe
c
ie
s
[
1]
,
is
f
a
c
in
g
in
c
r
e
a
s
in
g
th
r
e
a
ts
f
r
om
ha
bi
ta
t
de
s
tr
uc
ti
on,
c
li
m
a
te
c
ha
nge
,
a
nd
e
nvi
r
onm
e
nt
a
l
de
gr
a
da
ti
on.
T
he
s
e
c
ha
ll
e
nge
s
unde
r
s
c
or
e
th
e
ur
ge
nt
ne
e
d
f
or
e
f
f
e
c
ti
ve
c
ons
e
r
va
ti
on
pol
ic
ie
s
,
w
hi
c
h
m
us
t
be
in
f
or
m
e
d
by
publ
ic
s
e
nt
im
e
nt
to
e
n
s
ur
e
th
e
ir
r
e
s
pons
iv
e
ne
s
s
a
nd
e
f
f
e
c
ti
ve
ne
s
s
.
H
ow
e
ve
r
,
a
na
ly
z
in
g
publ
ic
s
e
nt
im
e
nt
f
r
om
la
r
ge
-
s
c
a
le
s
oc
ia
l
m
e
di
a
pl
a
tf
or
m
s
li
ke
T
w
it
te
r
pr
e
s
e
nt
s
s
ig
ni
f
ic
a
nt
c
ha
ll
e
nge
s
due
to
th
e
in
f
o
r
m
a
l,
di
ve
r
s
e
na
tu
r
e
of
th
e
la
ngua
ge
us
e
d.
T
o
a
ddr
e
s
s
th
e
s
e
c
ha
ll
e
ng
e
s
,
th
is
s
tu
dy
ut
il
iz
e
s
I
ndoB
E
R
T
w
e
e
t
[
2]
,
a
pr
e
-
tr
a
in
e
d
la
ngua
ge
m
ode
l
opt
im
iz
e
d
f
o
r
I
ndone
s
ia
n
T
w
it
te
r
da
ta
,
to
pe
r
f
or
m
s
e
nt
im
e
nt
a
na
ly
s
is
on
th
e
publ
ic
di
s
c
our
s
e
s
ur
r
ounding
I
ndone
s
ia
’
s
bi
o
di
ve
r
s
it
y
pol
ic
ie
s
.
B
y
le
ve
r
a
gi
ng
c
ont
e
xt
-
r
ic
h
e
m
be
ddi
ngs
,
I
ndoB
E
R
T
w
e
e
t
e
nha
nc
e
s
bot
h
th
e
a
c
c
ur
a
c
y
a
nd
in
t
e
r
pr
e
ta
bi
li
ty
of
s
e
nt
im
e
nt
a
na
ly
s
is
[
3]
,
of
f
e
r
in
g
a
m
or
e
e
f
f
e
c
ti
ve
to
ol
th
a
n
tr
a
di
ti
ona
l
m
a
c
hi
ne
le
a
r
ni
ng
a
ppr
o
a
c
he
s
.
T
hi
s
s
tu
dy
a
im
s
to
pr
ovi
de
a
c
ti
ona
bl
e
in
s
ig
ht
s
to
pol
ic
ym
a
ke
r
s
by
a
ppl
yi
ng
th
e
s
e
a
dva
nc
e
d
m
e
th
ods
to
r
e
f
in
e
bi
odi
ve
r
s
it
y
c
ons
e
r
va
ti
on
s
tr
a
te
gi
e
s
a
nd
im
pr
ove
publi
c
e
nga
ge
m
e
nt
[
4]
.
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
, N
o.
3
,
J
une
20
25
:
2389
-
2401
2390
B
ui
ld
in
g
on
th
e
de
m
ons
tr
a
te
d
s
tr
e
ngt
hs
of
I
ndoB
E
R
T
w
e
e
t,
th
is
s
tu
dy
f
ur
th
e
r
e
va
lu
a
te
s
th
e
pe
r
f
or
m
a
nc
e
of
it
s
di
f
f
e
r
e
nt
va
r
ia
nt
s
(
G
E
L
U
,
T
a
nh,
a
nd
N
one
)
a
s
de
s
c
r
ib
e
d
by
S
a
nt
os
a
e
t
al
.
[
5]
,
c
om
pa
r
in
g
th
e
m
w
it
h
tr
a
di
ti
ona
l
m
a
c
hi
ne
le
a
r
ni
ng
c
la
s
s
if
ie
r
s
s
uc
h
a
s
lo
g
is
ti
c
r
e
gr
e
s
s
io
n,
s
uppor
t
ve
c
to
r
m
a
c
hi
ne
,
a
nd
r
a
ndom f
or
e
s
t.
S
e
ve
r
a
l
s
tu
di
e
s
ha
ve
hi
ghl
ig
ht
e
d t
he
e
f
f
e
c
ti
ve
ne
s
s
of
B
E
R
T
a
nd i
ts
va
r
ia
nt
s
i
n t
w
e
e
t
s
e
nt
im
e
nt
a
na
ly
s
is
,
de
m
on
s
tr
a
ti
ng
s
ig
ni
f
ic
a
nt
ga
in
s
in
a
c
c
ur
a
c
y
a
nd
c
ont
e
x
tu
a
l
unde
r
s
ta
ndi
ng
w
he
n
c
om
bi
ne
d
w
it
h
ne
ur
a
l
ne
twor
k
a
r
c
hi
te
c
tu
r
e
s
.
F
or
in
s
ta
nc
e
,
B
e
ll
o
e
t
al
.
[
6]
pr
opos
e
d
a
B
E
R
T
f
r
a
m
e
w
or
k
f
or
s
e
nt
im
e
nt
a
na
ly
s
i
s
th
a
t
in
te
gr
a
te
s
B
E
R
T
w
it
h
c
onvolut
io
na
l
ne
ur
a
l
ne
twor
k
(
C
N
N
)
,
r
e
c
ur
r
e
nt
ne
ur
a
l
ne
twor
k
(
R
N
N
)
,
a
nd
bi
di
r
e
c
ti
ona
l
lo
ng
s
hor
t
-
te
r
m
m
e
m
or
y
(
B
iL
S
T
M
)
,
a
c
hi
e
vi
ng
im
pr
ove
m
e
nt
s
i
n
a
c
c
ur
a
c
y,
pr
e
c
is
io
n,
a
nd
r
e
c
a
ll
c
om
pa
r
e
d
to
c
onve
nt
io
na
l
m
e
th
ods
.
T
he
s
e
tr
a
di
ti
ona
l
m
ode
ls
,
w
hi
c
h
ty
pi
c
a
ll
y
r
e
ly
on
f
e
a
tu
r
e
e
xt
r
a
c
ti
on
m
e
th
ods
li
ke
TF
-
I
D
F
[
7]
,
of
te
n
f
a
ll
s
hor
t
in
ha
ndl
in
g
th
e
c
om
pl
e
xi
ty
a
nd
i
nf
or
m
a
l
la
ngua
ge
ty
pi
c
a
l
of
s
oc
ia
l
m
e
di
a
.
I
n
c
ont
r
a
s
t,
I
ndoB
E
R
T
w
e
e
t,
le
ve
r
a
gi
ng
B
E
R
T
e
m
b
e
ddi
ngs
[
8]
,
pr
ovi
de
s
a
m
or
e
nua
nc
e
d
a
na
ly
s
is
.
T
hi
s
s
tu
dy
e
nha
nc
e
s
pr
io
r
e
va
lu
a
ti
ons
by
in
te
gr
a
ti
ng
a
dva
n
c
e
d
te
c
hni
que
s
s
uc
h
a
s
w
or
d
a
tt
r
ib
ut
io
n
a
nd
te
n
-
f
ol
d
c
r
os
s
-
va
li
da
ti
on,
us
in
g
m
e
tr
ic
s
li
ke
m
e
a
n
a
c
c
ur
a
c
y
a
nd
m
e
a
n
F1
-
s
c
or
e
,
w
it
h
s
ta
ti
s
ti
c
a
l
s
ig
ni
f
ic
a
nc
e
c
onf
ir
m
e
d
th
r
ough
p
-
va
lu
e
s
.
T
h
e
s
e
c
om
pr
e
he
n
s
iv
e
e
v
a
lu
a
ti
ons
d
e
m
ons
tr
a
t
e
th
e
s
upe
r
io
r
pe
r
f
or
m
a
nc
e
of
I
ndoB
E
R
T
w
e
e
t
a
nd pr
ovi
de
m
or
e
r
e
li
a
bl
e
t
ool
s
f
or
pol
ic
y de
c
is
io
n
-
m
a
ki
ng.
T
he
a
r
c
hi
te
c
tu
r
e
of
th
e
I
ndoB
E
R
T
w
e
e
t
va
r
ia
nt
s
,
a
s
s
how
n
i
n
F
ig
ur
e
1,
il
lu
s
tr
a
te
s
a
de
e
p
ne
ur
a
l
ne
twor
k
(
D
N
N
)
s
tr
uc
tu
r
e
be
gi
nni
ng
w
it
h
a
pr
e
-
tr
a
in
e
d
I
ndoB
E
R
T
w
e
e
t
m
ode
l
de
s
ig
ne
d
to
c
a
pt
ur
e
c
ont
e
xt
ua
l
e
m
be
ddi
ngs
a
nd s
e
m
a
nt
ic
r
e
pr
e
s
e
nt
a
ti
ons
uni
que
to
B
a
ha
s
a
I
nd
one
s
ia
twe
e
ts
. T
hi
s
m
ode
l
is
th
e
n
f
ol
lo
w
e
d
by
two
f
ul
ly
c
onne
c
te
d
la
ye
r
s
:
a
256
-
uni
t
la
ye
r
th
a
t
m
a
tc
he
s
th
e
I
n
doB
E
R
T
w
e
e
t
pool
e
d
out
put
ve
c
to
r
a
nd
a
th
r
e
e
-
uni
t
out
put
la
ye
r
r
e
pr
e
s
e
nt
in
g
s
e
nt
im
e
nt
c
la
s
s
if
ic
a
ti
on
out
c
om
e
s
.
T
hr
e
e
di
f
f
e
r
e
nt
a
c
ti
va
ti
on
f
unc
ti
o
ns
-
N
one
,
G
E
L
U
,
a
nd
T
a
nh
-
a
r
e
te
s
te
d
in
th
e
f
ir
s
t
f
ul
ly
c
onne
c
te
d
la
y
e
r
to
a
s
s
e
s
s
th
e
ir
im
pa
c
t
on
s
e
nt
im
e
nt
a
na
ly
s
is
pe
r
f
or
m
a
nc
e
.
T
hi
s
s
tu
dy
f
ur
th
e
r
de
m
ons
tr
a
te
s
th
a
t
I
ndoB
E
R
T
w
e
e
t
s
ur
pa
s
s
e
s
s
e
ve
n
tr
a
di
ti
ona
l
c
la
s
s
if
ie
r
s
,
w
hi
c
h
r
e
ly
on
m
e
th
ods
li
ke
T
F
-
I
D
F
a
nd
B
E
R
T
e
m
be
ddi
ngs
,
in
a
na
ly
z
in
g
twe
e
ts
a
bout
I
ndone
s
ia
'
s
bi
odi
ve
r
s
it
y
pol
ic
ie
s
.
W
it
h
a
f
oc
us
e
d
da
ta
s
e
t
of
13,435
twe
e
ts
,
I
ndoB
E
R
T
w
e
e
t
a
c
hi
e
ve
s
s
upe
r
io
r
m
e
a
n
a
c
c
ur
a
c
y
a
nd
F
1
s
c
or
e
s
,
w
it
h
s
ta
ti
s
ti
c
a
ll
y
s
ig
ni
f
ic
a
nt
p
-
va
lu
e
s
(
be
lo
w
0.05)
c
o
m
pa
r
e
d
to
c
onve
nt
io
na
l
m
ode
ls
.
A
ddi
ti
ona
ll
y,
w
or
d
a
tt
r
ib
ut
io
n
te
c
hni
que
s
e
nha
nc
e
in
te
r
pr
e
ta
bi
li
ty
,
s
uppor
ti
ng
th
e
m
ode
l’
s
a
ppl
ic
a
ti
on
in
hi
gh
-
pr
e
c
is
io
n
pol
ic
y
-
m
a
ki
ng f
or
bi
odi
ve
r
s
it
y a
nd e
nvi
r
onm
e
nt
a
l
m
a
na
ge
m
e
nt
.
F
ig
ur
e
1.
D
N
N
a
r
c
hi
te
c
tu
r
e
of
t
he
I
ndoB
E
R
T
w
e
e
t
m
ode
l
w
it
h t
w
o f
ul
ly
c
onne
c
te
d l
a
ye
r
s
[
5]
T
he
pa
pe
r
is
or
ga
ni
z
e
d
a
s
f
ol
lo
w
s
:
in
th
e
m
e
th
od
s
e
c
ti
on,
w
e
d
e
s
c
r
ib
e
th
e
pr
oc
e
s
s
of
c
ol
le
c
ti
ng
a
nd
pr
e
pa
r
in
g
T
w
it
te
r
da
ta
on
bi
odi
ve
r
s
it
y
pol
ic
ie
s
in
I
ndone
s
ia
,
f
o
ll
ow
e
d
by
th
e
de
ve
lo
pm
e
nt
a
nd
e
va
lu
a
ti
on
of
th
e
s
e
nt
im
e
nt
a
na
ly
s
is
m
ode
ls
us
in
g
I
ndoB
E
R
T
w
e
e
t
a
nd
tr
a
d
it
io
na
l
c
la
s
s
if
ie
r
s
.
T
he
r
e
s
ul
ts
a
nd
di
s
c
us
s
io
n
s
e
c
ti
on
pr
e
s
e
nt
s
a
d
e
ta
il
e
d
c
om
p
a
r
is
on
of
th
e
pe
r
f
or
m
a
nc
e
o
f
I
ndoB
E
R
T
w
e
e
t
a
ga
in
s
t
tr
a
di
ti
ona
l
m
ode
ls
,
s
uppor
te
d
by
w
or
d
-
a
tt
r
ib
ut
io
n
te
c
hni
que
s
a
nd
s
ta
ti
s
ti
c
a
l
a
na
ly
s
i
s
.
F
in
a
ll
y,
th
e
c
onc
lu
s
io
n
s
um
m
a
r
iz
e
s
th
e
ke
y
f
in
di
ngs
,
hi
ghl
ig
ht
in
g
th
e
pr
a
c
ti
c
a
l
im
pl
ic
a
ti
ons
f
or
bi
odi
ve
r
s
it
y
pol
ic
y
-
m
a
ki
ng
a
nd
of
f
e
r
in
g
di
r
e
c
ti
ons
f
or
f
ut
ur
e
r
e
s
e
a
r
c
h t
o e
nha
nc
e
s
e
nt
im
e
nt
a
na
ly
s
i
s
m
e
th
odol
ogi
e
s
i
n
s
im
il
a
r
c
ont
e
xt
s
.
2.
M
E
T
H
O
D
A
s
i
ll
us
tr
a
te
d i
n F
ig
ur
e
2, t
he
di
a
gr
a
m
pr
e
s
e
nt
s
t
he
s
te
p
s
f
or
de
ve
lo
pi
ng a
nd e
va
lu
a
ti
ng t
he
s
e
nt
im
e
nt
a
na
ly
s
is
m
ode
l
us
in
g
I
ndoB
E
R
T
w
e
e
t
on
I
ndone
s
ia
n
T
w
it
te
r
da
t
a
a
bout
bi
odi
ve
r
s
it
y
pol
ic
y.
T
he
pr
oc
e
s
s
be
gi
n
s
w
it
h
da
ta
pr
e
pa
r
a
ti
on
, w
hi
c
h
i
nc
lu
de
s
c
ol
le
c
ti
ng, c
le
a
ni
ng, s
e
le
c
ti
ng, a
nd l
a
be
li
ng t
he
d
a
ta
f
r
om
T
w
it
te
r
. N
e
xt
,
da
ta
pr
e
pr
oc
e
s
s
in
g
in
vol
ve
s
nor
m
a
li
z
in
g
th
e
da
ta
a
nd
e
n
s
ur
in
g
c
ons
is
te
nt
da
ta
f
or
m
a
ts
.
T
he
m
ode
li
ng
ph
a
s
e
in
vol
ve
s
bot
h
de
e
p
l
e
a
r
ni
ng
a
nd
non
-
de
e
p
le
a
r
ni
ng
te
c
hni
que
s
.
F
or
de
e
p
le
a
r
ni
ng,
th
e
I
ndoB
E
R
T
w
e
e
t
m
ode
l
is
ut
il
iz
e
d
w
it
h
va
r
io
us
a
c
ti
va
ti
on
f
unc
ti
ons
.
F
e
a
tu
r
e
ve
c
to
r
iz
a
ti
on
is
pe
r
f
or
m
e
d
us
in
g
T
F
-
I
D
F
a
nd
B
E
R
T
e
m
be
ddi
ngs
f
or
non
-
de
e
p
l
e
a
r
ni
ng
te
c
hni
que
s
.
F
in
a
ll
y,
m
ode
l
pe
r
f
or
m
a
nc
e
e
va
lu
a
ti
on
u
s
e
s
10
-
f
ol
d
c
r
os
s
-
va
li
da
ti
on,
m
e
a
s
u
r
in
g
a
c
c
ur
a
c
y
a
nd
F
1
-
s
c
or
e
,
a
s
s
e
s
s
in
g
s
ta
ti
s
ti
c
a
l
s
ig
ni
f
ic
a
nc
e
w
it
h
p
-
va
lu
e
s
,
a
nd
e
va
lu
a
ti
ng
w
or
d
a
tt
r
ib
ut
io
n.
T
hi
s
c
om
pr
e
he
ns
iv
e
a
ppr
oa
c
h
e
ns
ur
e
s
a
r
obu
s
t
a
s
s
e
s
s
m
e
nt
of
th
e
m
ode
l’
s
a
bi
li
ty
to
a
na
ly
z
e
s
e
nt
im
e
nt
i
n t
he
c
ont
e
xt
of
I
ndone
s
ia
n bi
odi
ve
r
s
it
y pol
ic
y.
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
M
ode
li
ng s
e
nt
ime
nt
analy
s
is
of
I
ndone
s
ia
n bi
odi
v
e
r
s
it
y
pol
ic
y
T
w
e
e
ts
…
(
M
ohamm
ad T
e
duh Ulini
ans
y
ah
)
2391
F
ig
ur
e
2. W
or
kf
lo
w
of
s
e
nt
im
e
nt
a
na
ly
s
is
m
ode
l
de
ve
lo
pm
e
nt
a
nd e
va
lu
a
ti
on
us
in
g I
ndoB
E
R
T
w
e
e
t
on
I
ndone
s
ia
n T
w
it
te
r
da
ta
2.1.
D
at
as
e
t
p
r
e
p
ar
at
io
n
f
r
om
T
w
it
t
e
r
T
he
da
ta
s
e
t
pr
e
pa
r
a
ti
on
pha
s
e
in
vol
ve
d
c
r
e
a
ti
ng
a
la
be
le
d
da
ta
s
e
t
of
twe
e
ts
f
r
om
T
w
it
te
r
,
c
a
te
gor
iz
in
g
th
e
m
a
s
pos
it
iv
e
,
n
e
ga
ti
ve
,
or
ne
ut
r
a
l
f
or
s
e
nt
im
e
nt
a
na
ly
s
i
s
.
T
w
e
e
ts
w
e
r
e
c
ol
le
c
te
d
u
s
in
g
th
e
T
w
it
te
r
A
P
I
[
9]
f
r
om
J
a
nua
r
y
2020
to
M
a
r
c
h
2023,
f
oc
us
in
g
on
to
pi
c
s
r
e
la
te
d
t
o
f
ood
s
e
c
ur
it
y,
he
a
lt
hc
a
r
e
,
a
nd e
nvi
r
onm
e
nt
a
l
m
a
na
ge
m
e
nt
in
I
ndone
s
ia
.
T
hi
s
pr
oc
e
s
s
in
it
ia
ll
y
ga
th
e
r
e
d
500,000
twe
e
ts
.
A
f
te
r
c
le
a
ni
ng
to
r
e
m
ove
dupl
ic
a
te
s
,
non
-
I
ndone
s
ia
n
c
ont
e
nt
,
a
nd
ir
r
e
le
va
n
t
in
f
or
m
a
ti
on
s
uc
h
a
s
a
dv
e
r
ti
s
e
m
e
nt
s
a
nd
nove
l
quot
e
s
,
th
e
da
ta
s
e
t
w
a
s
r
e
duc
e
d
to
200,015
twe
e
ts
.
D
u
e
to
li
m
it
a
ti
ons
in
a
nnot
a
to
r
a
va
il
a
bi
li
ty
,
15,323
twe
e
ts
w
e
r
e
s
e
le
c
t
e
d
f
or
a
nnot
a
ti
on.
T
o
e
ns
ur
e
c
on
s
is
te
nc
y
a
nd
r
e
li
a
bi
li
ty
in
th
e
a
nnot
a
ti
o
n
pr
oc
e
s
s
,
18
tr
a
in
e
d
a
nnot
a
to
r
s
w
e
r
e
pr
ovi
de
d
w
it
h
de
ta
il
e
d
gui
de
li
ne
s
f
or
s
e
nt
im
e
nt
c
l
a
s
s
if
ic
a
ti
on
ba
s
e
d
on
e
s
t
a
bl
is
he
d
s
ta
nda
r
ds
[
4]
f
r
om
pr
io
r
s
tu
di
e
s
.
E
a
c
h
twe
e
t
w
a
s
r
e
vi
e
w
e
d
in
d
e
pe
nde
nt
ly
by
th
r
e
e
a
nnot
a
to
r
s
,
a
nd
i
nt
e
r
-
a
nnot
a
to
r
a
gr
e
e
m
e
nt
w
a
s
a
s
s
e
s
s
e
d
us
in
g
F
le
is
s
’
K
a
ppa
w
it
h
a
va
lu
e
of
0.62187,
w
hi
c
h
s
how
e
d
a
s
ub
s
ta
nt
ia
l
a
gr
e
e
m
e
nt
.
A
m
a
jo
r
it
y
vot
in
g
a
ppr
oa
c
h
w
a
s
e
m
pl
oye
d
to
f
in
a
li
z
e
th
e
s
e
nt
im
e
nt
la
be
l
f
or
e
a
c
h
twe
e
t,
r
e
s
ol
vi
ng
a
ny
di
s
a
gr
e
e
m
e
nt
s
.
T
hi
s
a
ppr
oa
c
h
e
ns
ur
e
d
th
a
t
th
e
a
nnot
a
ti
ons
w
e
r
e
c
ons
is
t
e
nt
a
nd
r
e
f
le
c
te
d
a
r
e
li
a
bl
e
c
on
s
e
ns
u
s
a
m
ong
th
e
a
nnot
a
to
r
s
.
A
f
te
r
e
xc
lu
di
ng
twe
e
ts
l
a
be
le
d
a
s
“
none
,
”
t
he
f
in
a
l
da
ta
s
e
t
u
s
e
d i
n t
he
r
e
s
e
a
r
c
h c
om
pr
is
e
d 13,435 twe
e
ts
. D
e
ta
il
e
d i
nf
or
m
a
ti
on
on
th
e
la
be
li
ng
pr
oc
e
s
s
a
nd
th
e
f
in
a
l
da
ta
s
e
t
is
pr
ovi
de
d
in
th
e
r
e
la
te
d
pa
pe
r
[
10]
,
a
nd
th
e
da
ta
s
e
t
is
a
va
il
a
bl
e
a
t
th
e
M
e
nde
le
y D
a
ta
r
e
pos
it
or
y
[
11]
.
2.2.
D
at
a
p
r
e
p
r
oc
e
s
s
in
g
F
ol
lo
w
in
g
th
e
in
it
ia
l
da
ta
s
e
t
pr
e
pa
r
a
ti
on,
s
e
ve
r
a
l
pr
e
pr
oc
e
s
s
in
g
s
te
ps
w
e
r
e
pe
r
f
or
m
e
d
to
r
e
a
dy
th
e
da
ta
f
or
m
ode
li
ng. I
n l
in
e
w
it
h t
he
a
ppr
oa
c
h by P
e
bi
a
na
e
t
al
.
[
12]
, t
he
t
e
xt
w
a
s
t
r
a
ns
f
or
m
e
d t
o l
ow
e
r
c
a
s
e
t
o e
n
s
ur
e
uni
f
or
m
it
y. U
R
L
s
w
e
r
e
r
e
m
ove
d, punc
tu
a
ti
on (
e
xc
e
pt
a
pos
tr
ophe
s
)
w
a
s
r
e
pl
a
c
e
d w
it
h s
pa
c
e
s
, a
nd non
-
A
S
C
I
I
c
ha
r
a
c
te
r
s
w
e
r
e
s
ub
s
ti
tu
te
d
w
it
h
th
e
ir
ne
a
r
e
s
t
A
S
C
I
I
e
qui
va
le
nt
s
.
I
nf
or
m
a
l
la
ngua
ge
a
nd
ty
pogr
a
phi
c
a
l
e
r
r
or
s
w
e
r
e
nor
m
a
li
z
e
d,
w
or
d
va
r
ia
ti
ons
w
e
r
e
s
ta
nda
r
di
z
e
d,
a
nd
n
um
e
r
ic
da
ta
a
nd
non
-
A
S
C
I
I
c
ha
r
a
c
te
r
s
w
e
r
e
e
li
m
in
a
te
d.
C
ons
e
c
ut
iv
e
s
pa
c
e
s
w
e
r
e
c
on
s
ol
id
a
te
d,
a
nd
s
to
pw
or
ds
w
e
r
e
r
e
m
ove
d
us
in
g
th
e
S
a
s
tr
a
w
i
P
yt
ho
n
li
br
a
r
y
[
13]
,
th
ough
a
dve
r
bs
w
e
r
e
r
e
ta
in
e
d
du
e
to
th
e
ir
s
i
gni
f
ic
a
nt
r
ol
e
in
s
e
nt
im
e
nt
a
n
a
ly
s
is
.
T
he
s
e
pr
e
pr
oc
e
s
s
in
g s
te
p
s
pr
e
pa
r
e
d t
he
da
ta
s
e
t
f
or
m
ode
li
ng by e
ns
ur
i
ng c
le
a
ne
r
a
nd mor
e
uni
f
or
m
t
e
xt
da
ta
.
O
ne
of
t
he
m
a
jo
r
c
ha
ll
e
nge
s
e
nc
ount
e
r
e
d dur
in
g t
he
s
tu
dy w
a
s
t
he
i
nf
or
m
a
l
a
nd dive
r
s
e
na
tu
r
e
of
t
he
la
ngua
ge
us
e
d i
n I
ndone
s
ia
n T
w
it
te
r
, w
hi
c
h i
nc
lu
de
s
s
la
ng, a
bbr
e
vi
a
ti
ons
, a
nd mi
xe
d l
a
ngua
ge
s
. T
hi
s
m
a
de
i
t
di
f
f
ic
ul
t
to
c
le
a
n
a
nd
pr
e
pr
oc
e
s
s
th
e
da
ta
e
f
f
e
c
ti
ve
ly
.
A
ddi
ti
ona
l
ly
,
obt
a
in
in
g
r
e
li
a
bl
e
s
e
nt
im
e
nt
l
a
be
ls
r
e
qui
r
e
d
c
a
r
e
f
ul
c
ur
a
ti
on
a
nd
a
nnot
a
ti
on
of
twe
e
ts
,
a
s
w
e
ll
a
s
r
e
s
ol
vi
ng
di
s
a
gr
e
e
m
e
nt
s
be
tw
e
e
n
a
nnot
a
to
r
s
,
w
hi
c
h
a
dde
d
c
om
pl
e
xi
ty
t
o t
he
da
ta
s
e
t
pr
e
pa
r
a
ti
on pr
oc
e
s
s
.
2.3.
M
od
e
li
n
g
T
hi
s
s
tu
d
y
c
om
p
a
r
e
s
s
e
nt
im
e
nt
a
na
ly
s
i
s
m
od
e
ls
by
e
m
pl
oyi
n
g
bot
h
d
e
e
p
l
e
a
r
ni
ng
a
nd
tr
a
di
ti
on
a
l
m
a
c
hi
n
e
l
e
a
r
n
in
g
t
e
c
hni
qu
e
s
,
e
a
c
h
of
f
e
r
i
ng
di
s
ti
n
c
t
a
dv
a
nt
a
g
e
s
f
or
te
xt
c
la
s
s
if
ic
a
t
io
n.
T
he
de
e
p
l
e
a
r
ni
ng
a
ppr
oa
c
h
us
e
s
a
B
E
R
T
a
r
c
hi
te
c
tu
r
e
,
s
pe
c
if
ic
a
ll
y
I
nt
r
ove
r
t
e
d
,
a
m
ode
l
pr
e
-
tr
a
in
e
d
on
I
n
don
e
s
i
a
n
T
w
i
tt
e
r
da
t
a
a
nd
th
us
hi
ghl
y
e
f
f
e
c
ti
v
e
a
t
h
a
ndl
in
g
in
f
or
m
a
l
l
a
ngu
a
ge
c
om
m
o
nl
y
f
oun
d
on
s
o
c
ia
l
m
e
di
a
.
I
ndo
B
E
R
T
w
e
e
t'
s
c
ont
e
xt
ua
l
e
m
b
e
ddi
n
gs
pr
ovi
d
e
nu
a
nc
e
d
s
e
m
a
nt
i
c
r
e
pr
e
s
e
nt
a
ti
o
ns
th
a
t
m
a
ke
i
t
pa
r
ti
c
ul
a
r
ly
s
ui
te
d
to
s
e
nt
im
e
n
t
a
na
ly
s
i
s
i
n I
ndon
e
s
i
a
n
T
w
it
t
e
r
d
a
ta
,
w
h
e
r
e
l
a
ngu
a
ge
pa
tt
e
r
ns
c
a
n
be
c
om
pl
e
x a
nd c
ont
e
xt
-
s
e
ns
it
iv
e
.
T
he
tr
a
di
ti
ona
l
m
a
c
hi
ne
le
a
r
ni
ng
m
e
th
od
s
s
e
le
c
te
d
in
c
lu
de
lo
gi
s
ti
c
r
e
gr
e
s
s
io
n,
s
uppor
t
v
e
c
to
r
c
la
s
s
if
ie
r
,
r
a
ndom
f
or
e
s
t
,
L
G
B
M
C
la
s
s
if
ie
r
,
e
xt
r
e
m
e
gr
a
di
e
nt
boos
ti
ng
(
X
G
B
oos
t)
,
a
da
pt
iv
e
boos
ti
ng
(
A
da
B
oos
t)
,
a
nd
de
c
is
io
n
tr
e
e
,
c
hos
e
n
f
or
th
e
ir
pr
ove
n
e
f
f
e
c
ti
ve
ne
s
s
in
te
xt
c
la
s
s
if
ic
a
ti
on
a
nd
s
e
nt
im
e
nt
a
na
ly
s
is
ta
s
ks
.
T
he
s
e
m
ode
l
s
of
f
e
r
a
ba
la
n
c
e
of
in
te
r
pr
e
ta
bi
li
ty
,
r
obus
tn
e
s
s
,
a
nd
pe
r
f
or
m
a
nc
e
:
lo
gi
s
ti
c
r
e
gr
e
s
s
io
n a
nd
s
uppor
t
ve
c
to
r
c
la
s
s
if
ie
r
s
a
r
e
s
tr
a
ig
ht
f
or
w
a
r
d a
nd e
f
f
ic
ie
nt
, r
a
ndom f
or
e
s
t
a
nd de
c
i
s
io
n t
r
e
e
a
dd
e
ns
e
m
bl
e
s
ta
bi
li
ty
to
r
e
duc
e
ove
r
f
it
ti
ng,
a
nd
gr
a
di
e
nt
boos
ti
ng
m
e
th
ods
li
ke
L
G
B
M
C
la
s
s
if
ie
r
a
nd
X
G
B
oos
t
pr
ovi
de
hi
gh a
c
c
ur
a
c
y f
or
c
om
pl
e
x da
t
a
s
e
t
s
. W
hi
le
s
im
pl
e
r
m
ode
ls
s
uc
h
a
s
n
a
iv
e
B
a
ye
s
w
e
r
e
c
on
s
id
e
r
e
d, t
he
y
w
e
r
e
e
xc
lu
de
d
a
f
te
r
in
it
ia
l
te
s
ts
in
di
c
a
te
d
li
m
it
e
d
pe
r
f
or
m
a
nc
e
on
th
is
c
ont
e
xt
-
he
a
vy
T
w
it
te
r
da
ta
s
e
t.
T
o
e
ns
ur
e
a
th
or
ough
c
om
pa
r
is
on,
w
e
e
m
pl
oye
d
bot
h
T
F
-
I
D
F
a
nd
B
E
R
T
e
m
be
ddi
ng
s
f
or
f
e
a
tu
r
e
r
e
pr
e
s
e
nt
a
ti
on
in
a
ll
tr
a
di
ti
on
a
l
m
ode
ls
,
a
ll
ow
in
g
a
di
r
e
c
t
e
va
lu
a
ti
on
o
f
tr
a
di
t
io
na
l
ve
c
to
r
iz
a
ti
on
ve
r
s
us
B
E
R
T
-
ba
s
e
d
c
ont
e
xt
ua
l
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
, N
o.
3
,
J
une
20
25
:
2389
-
2401
2392
e
m
be
ddi
ngs
.
R
e
s
ul
ts
a
r
e
s
um
m
a
r
iz
e
d
in
t
a
bl
e
s
in
th
e
r
e
s
ul
ts
a
nd
di
s
c
u
s
s
io
n
s
e
c
ti
on,
w
it
h
ke
y
p
e
r
f
or
m
a
nc
e
m
e
tr
ic
s
, i
nc
lu
di
ng me
a
n a
c
c
ur
a
c
y
a
nd F
1
-
s
c
or
e
, pr
e
s
e
nt
e
d
c
onc
is
e
ly
f
or
c
la
r
it
y.
2.3.1.
M
od
e
li
n
g
s
e
n
t
im
e
n
t
an
al
ys
is
w
it
h
B
E
R
T
-
b
as
e
d
d
e
e
p
l
e
ar
n
in
g t
e
c
h
n
iq
u
e
s
T
he
pr
e
-
tr
a
in
e
d
I
ndoB
E
R
T
w
e
e
t
m
ode
l
[
2]
ha
s
pr
ove
n
hi
ghl
y
e
f
f
e
c
ti
ve
f
or
s
e
nt
im
e
nt
a
n
a
ly
s
is
on
I
ndone
s
ia
n
T
w
it
te
r
da
ta
due
to
it
s
s
pe
c
if
ic
tr
a
in
in
g
in
th
e
in
f
or
m
a
l
la
ngua
ge
ty
pi
c
a
l
of
s
oc
ia
l
m
e
di
a
.
U
nl
ik
e
I
ndoB
E
R
T
[
3]
,
w
hi
c
h
w
a
s
tr
a
in
e
d
on
ove
r
220
m
il
li
on
w
or
ds
f
r
om
va
r
io
us
f
o
r
m
a
l
I
ndone
s
ia
n
s
our
c
e
s
li
ke
W
ik
ip
e
di
a
a
nd
ne
w
s
a
r
ti
c
le
s
,
I
ndoB
E
R
T
w
e
e
t’
s
tr
a
in
in
g
c
or
pu
s
c
ons
i
s
ts
of
a
ppr
oxi
m
a
te
ly
409
m
il
li
on
w
or
d
to
ke
ns
s
ol
e
ly
f
r
om
I
ndone
s
ia
n
T
w
it
te
r
da
ta
.
T
hi
s
f
oc
us
on
i
nf
or
m
a
l
la
ngua
ge
e
n
a
bl
e
s
I
ndoB
E
R
T
w
e
e
t
t
o be
tt
e
r
c
a
pt
ur
e
th
e
nua
nc
e
s
of
I
ndone
s
ia
n
s
oc
ia
l
m
e
di
a
di
s
c
our
s
e
,
m
a
k
in
g
it
hi
ghl
y
s
ui
ta
bl
e
f
or
a
na
ly
z
in
g
s
e
nt
im
e
nt
s
on
T
w
it
te
r
.
T
h
e
m
ode
l’
s
tr
a
in
in
g
e
m
pl
oy
s
uns
up
e
r
vi
s
e
d
le
a
r
ni
ng
te
c
hni
que
s
,
pr
e
di
c
ti
ng
m
a
s
ke
d
to
ke
ns
in
in
put
te
xt
,
c
ons
is
te
nt
w
it
h
B
E
R
T
’
s
a
r
c
hi
te
c
tu
r
e
,
w
hi
c
h
e
nha
n
c
e
s
it
s
a
bi
li
ty
to
unde
r
s
ta
nd
c
ont
e
xt
u
a
l
r
e
la
ti
ons
hi
ps
w
it
hi
n s
oc
ia
l
m
e
di
a
t
e
xt
.
B
ui
ld
in
g
upon
I
ndoB
E
R
T
w
e
e
t’
s
s
tr
e
ngt
hs
,
our
s
e
nt
im
e
nt
a
n
a
ly
s
is
s
ys
te
m
in
c
or
por
a
te
s
two
f
ul
ly
c
onne
c
te
d
la
ye
r
s
th
a
t
ut
il
iz
e
s
e
nt
e
n
c
e
e
m
be
ddi
ng
s
ge
ne
r
a
te
d
by
I
ndoB
E
R
T
w
e
e
t
a
s
in
put
f
e
a
tu
r
e
s
.
T
hi
s
a
r
c
hi
te
c
tu
r
e
,
il
lu
s
tr
a
te
d
in
F
ig
ur
e
1,
is
s
tr
uc
tu
r
e
d
a
s
a
D
N
N
t
o
c
a
pt
ur
e
c
om
pl
e
x
c
ont
e
xt
ua
l
r
e
la
ti
ons
hi
ps
in
he
r
e
nt
in
I
ndone
s
ia
n
la
ngua
ge
te
xt
,
e
s
s
e
nt
i
a
l
f
or
a
c
c
ur
a
te
s
e
nt
im
e
nt
a
na
ly
s
is
.
A
ddi
ti
ona
ll
y,
w
e
e
va
lu
a
te
d
th
e
pe
r
f
or
m
a
nc
e
im
pa
c
t
of
di
f
f
e
r
e
nt
a
c
ti
va
ti
on
f
unc
ti
ons
-
s
pe
c
if
ic
a
ll
y,
hype
r
bol
ic
ta
nge
nt
(
ta
nh)
,
G
a
us
s
ia
n
e
r
r
or
li
ne
a
r
uni
t
(
G
E
L
U
)
,
a
nd
no
a
c
ti
va
ti
on
-
in
th
e
f
ir
s
t
f
ul
ly
c
onne
c
te
d
la
ye
r
to
opt
im
iz
e
th
e
m
ode
l’
s
e
f
f
e
c
ti
ve
ne
s
s
.
T
he
s
e
c
om
pa
r
is
ons
pr
ovi
de
in
s
ig
ht
s
in
to
how
I
ndoB
E
R
T
w
e
e
t
e
m
be
ddi
ngs
in
te
r
a
c
t
w
it
hi
n
a
de
e
p
le
a
r
ni
ng
f
r
a
m
e
w
or
k,
c
ont
r
ib
ut
in
g
to
m
or
e
in
f
or
m
e
d
c
hoi
c
e
s
in
m
od
e
l
c
onf
ig
ur
a
ti
ons
f
or
im
pr
o
ve
d
s
e
nt
im
e
nt
a
n
a
ly
s
is
out
c
om
e
s
.
2.3.2.
M
od
e
li
n
g
s
e
n
t
im
e
n
t
an
al
ys
is
w
it
h
t
r
ad
it
io
n
al
m
ac
h
in
e
l
e
ar
n
in
g m
e
t
h
od
s
I
n
th
is
s
e
c
ti
on,
w
e
in
ve
s
ti
ga
te
th
e
e
f
f
e
c
ti
ve
ne
s
s
of
s
e
ve
n
tr
a
di
ti
ona
l
m
a
c
hi
ne
le
a
r
ni
ng
m
e
th
ods
in
s
e
nt
im
e
nt
a
na
ly
s
is
by
ut
il
iz
in
g
two
di
s
ti
nc
t
ve
c
to
r
iz
a
ti
on
te
c
hn
iq
ue
s
:
T
F
-
I
D
F
a
nd
B
E
R
T
e
m
be
ddi
ngs
.
W
hi
le
TF
-
I
D
F
pr
ovi
de
s
a
s
ta
ti
s
ti
c
a
l
m
e
a
s
ur
e
of
w
or
d
im
por
t
a
nc
e
w
it
hi
n
th
e
c
or
pus
,
B
E
R
T
e
m
be
ddi
ngs
of
f
e
r
r
ic
h,
c
ont
e
xt
ua
li
z
e
d w
or
d r
e
pr
e
s
e
nt
a
ti
ons
t
ha
t
c
a
pt
ur
e
s
e
m
a
nt
ic
nua
n
c
e
s
. T
he
t
r
a
di
ti
ona
l
m
a
c
hi
ne
l
e
a
r
ni
ng me
th
ods
w
e
e
m
pl
oy
a
r
e
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
,
r
a
ndom
f
or
e
s
t
,
L
G
B
M
C
la
s
s
if
ie
r
,
X
G
B
oos
t
,
A
da
B
oos
t
,
a
nd
de
c
i
s
io
n
tr
e
e
.
T
o
th
or
oughly
a
s
s
e
s
s
th
e
pe
r
f
or
m
a
nc
e
a
nd
pr
e
c
i
s
io
n
of
th
e
s
e
m
e
th
ods
in
s
e
nt
im
e
nt
a
na
ly
s
is
,
w
e
di
vi
de
our
in
ve
s
ti
ga
ti
on
in
to
two
a
ppr
oa
c
he
s
:
T
F
-
I
D
F
-
ba
s
e
d
tr
a
di
ti
ona
l
m
ode
ls
a
nd
B
E
R
T
e
m
be
ddi
ng
-
ba
s
e
d
tr
a
di
ti
ona
l
m
ode
ls
.
T
hi
s
a
na
ly
s
is
pr
ovi
de
s
in
s
ig
ht
s
in
to
th
e
s
tr
e
ngt
hs
a
nd
li
m
it
a
ti
ons
of
tr
a
di
ti
ona
l
te
c
hni
que
s
in
c
a
pt
ur
in
g
s
e
nt
im
e
nt
f
r
om
te
xt
ua
l
da
t
a
,
c
om
pa
r
e
d
w
it
h
B
E
R
T
-
ba
s
e
d
de
e
p
le
a
r
ni
ng
a
ppr
oa
c
he
s
.
‒
TF
-
I
D
F
-
ba
s
e
d
tr
a
di
ti
ona
l
m
ode
ls
:
t
he
T
F
-
I
D
F
te
c
hni
que
[
7]
is
a
n
e
s
s
e
nt
ia
l
to
ol
f
or
f
e
a
tu
r
e
e
xt
r
a
c
ti
on f
r
o
m
th
e
la
be
le
d
da
ta
s
e
t,
e
n
a
bl
in
g
th
e
a
ppl
ic
a
ti
on
of
va
r
io
us
tr
a
di
ti
ona
l
m
a
c
hi
ne
-
le
a
r
ni
ng
a
lg
or
it
hm
s
.
T
hi
s
a
ppr
oa
c
h be
gi
ns
w
it
h
lo
gi
s
ti
c
r
e
gr
e
s
s
io
n
[
14]
, a
m
e
th
od t
ha
t
m
o
de
ls
t
he
pr
oba
bi
li
ti
e
s
of
bi
na
r
y outc
om
e
s
.
N
e
xt
,
w
e
ut
il
iz
e
th
e
s
uppor
t
ve
c
to
r
c
la
s
s
if
ie
r
[
15]
,
w
hi
c
h
e
f
f
e
c
ti
ve
ly
e
m
pl
oys
hype
r
pl
a
ne
s
to
s
e
pa
r
a
te
c
la
s
s
e
s
in
hi
gh
-
di
m
e
ns
io
na
l
s
pa
c
e
.
li
ght
gr
a
di
e
nt
boos
ti
ng
m
a
c
hi
ne
[
16]
is
a
ls
o
a
ppl
ie
d,
a
nd
it
is
known
f
or
it
s
hi
gh
e
f
f
ic
ie
nc
y
w
he
n
ha
ndl
in
g
la
r
ge
-
s
c
a
le
da
ta
s
e
ts
.
A
dd
it
io
na
ll
y,
r
a
ndom
f
or
e
s
t
[
17
]
is
e
m
pl
oye
d
a
s
a
n
e
ns
e
m
bl
e
m
e
th
od t
ha
t
c
ons
tr
uc
t
s
m
ul
ti
pl
e
de
c
i
s
io
n t
r
e
e
s
t
o i
m
pr
ove
m
ode
l
a
c
c
ur
a
c
y.
X
G
B
oos
t
[
18]
,
opt
im
iz
e
d
f
or
bot
h
pe
r
f
or
m
a
nc
e
a
nd
e
f
f
ic
ie
nc
y,
is
in
c
lu
de
d
in
our
a
na
ly
s
is
.
W
e
f
ur
th
e
r
in
c
or
por
a
te
A
da
B
oos
t
[
19]
, w
hi
c
h i
te
r
a
ti
ve
ly
c
om
bi
ne
s
w
e
a
k
c
la
s
s
if
ie
r
s
t
o f
or
m
a
r
obus
t
c
la
s
s
if
ie
r
. F
in
a
ll
y, a
de
c
is
io
n
t
r
e
e
[
20]
,
w
it
h
it
s
tr
e
e
-
li
ke
s
tr
uc
tu
r
e
,
is
u
s
e
d
f
or
bot
h
c
l
a
s
s
if
ic
a
ti
on
a
nd
r
e
gr
e
s
s
io
n
ta
s
k
s
.
T
hi
s
c
om
pr
e
he
ns
iv
e
e
va
lu
a
ti
on
a
ll
ow
s
f
or
a
de
ta
il
e
d
c
om
pa
r
is
on
o
f
how
th
e
s
e
tr
a
di
ti
ona
l
m
a
c
hi
ne
le
a
r
ni
ng
m
ode
ls
pe
r
f
or
m
in
th
e
c
ont
e
xt
of
s
e
nt
im
e
nt
a
na
ly
s
is
w
h
e
n
us
in
g
T
F
-
I
D
F
a
s
th
e
f
e
a
tu
r
e
r
e
pr
e
s
e
nt
a
ti
on
te
c
hni
que
.
‒
B
E
R
T
e
m
be
ddi
ng
-
ba
s
e
d
tr
a
di
ti
ona
l
m
ode
ls
:
B
E
R
T
e
m
b
e
ddi
ngs
pr
ov
id
e
a
s
ig
ni
f
ic
a
nt
a
d
va
nt
a
ge
ove
r
TF
-
I
D
F
ve
c
to
r
s
by
c
a
pt
ur
in
g
th
e
c
ont
e
xt
in
w
hi
c
h w
or
d
s
a
ppe
a
r
,
c
r
e
a
ti
n
g a
m
or
e
hol
i
s
ti
c
r
e
pr
e
s
e
nt
a
ti
on
of
te
xt
. T
F
-
I
D
F
,
a
s
ta
ti
s
ti
c
a
l
m
e
tr
i
c
f
or
e
va
lu
a
ti
ng
a
w
or
d'
s
im
por
t
a
nc
e
w
it
hi
n a
c
or
pu
s
,
e
f
f
e
c
ti
v
e
ly
id
e
nt
if
ie
s
ke
yw
or
d
r
e
le
v
a
nc
e
but
tr
e
a
t
s
e
a
c
h
te
r
m
i
nde
p
e
nd
e
nt
ly
,
la
c
ki
ng
c
ont
e
x
tu
a
l
in
s
ig
ht
.
I
n
c
on
tr
a
s
t,
B
E
R
T
,
or
bi
di
r
e
c
t
io
na
l
e
n
c
od
e
r
r
e
pr
e
s
e
nt
a
ti
on
s
f
r
om
tr
a
ns
f
or
m
e
r
s
,
g
e
ne
r
a
te
s
c
ont
e
x
tu
a
li
z
e
d
w
or
d
e
m
b
e
ddi
n
gs
th
a
t
c
a
pt
ur
e
c
om
pl
e
x
s
e
m
a
nt
i
c
a
nd
s
ynt
a
c
ti
c
nu
a
nc
e
s
.
T
h
is
c
a
p
a
bi
li
ty
is
p
a
r
ti
c
ul
a
r
ly
v
a
lu
a
bl
e
in
s
e
n
ti
m
e
nt
a
na
ly
s
i
s
, w
h
e
r
e
a
w
or
d’
s
s
e
nt
im
e
nt
c
a
n v
a
r
y s
u
bs
t
a
nt
i
a
ll
y b
a
s
e
d
on ne
a
r
by t
e
r
m
s
.
R
e
s
e
a
r
c
h h
a
s
s
ho
w
n t
ha
t
us
in
g
B
E
R
T
e
m
be
dd
in
g
s
a
s
in
put
f
e
a
tu
r
e
s
f
or
tr
a
di
ti
on
a
l,
n
on
-
de
e
p
-
l
e
a
r
ni
n
g
c
l
a
s
s
if
ie
r
s
-
s
uc
h
a
s
s
upp
or
t
ve
c
to
r
m
a
c
hi
n
e
s
,
r
a
ndo
m
f
or
e
s
t
s
,
or
bo
os
ti
ng
a
lg
or
it
hm
s
-
e
na
bl
e
s
th
e
s
e
m
ode
l
s
to
h
a
r
ne
s
s
th
e
e
nr
i
c
he
d,
c
ont
e
xt
ua
l
in
f
or
m
a
ti
on
e
m
b
e
dd
e
d
i
n
th
e
t
e
xt
,
le
a
di
ng
to
im
pr
o
ve
d
a
c
c
ur
a
c
y
a
nd
r
obu
s
tn
e
s
s
in
s
e
nt
im
e
nt
c
la
s
s
if
i
c
a
ti
on
ta
s
ks
c
om
p
a
r
e
d
to
m
od
e
ls
r
e
ly
in
g s
ol
e
l
y
on
T
F
-
I
D
F
ve
c
to
r
s
.
E
m
pl
oyi
ng
B
E
R
T
e
m
be
ddi
ngs
in
t
hi
s
w
a
y of
f
e
r
s
a
pr
om
i
s
in
g
a
ppr
o
a
c
h
t
o
e
nha
nc
in
g s
e
nt
im
e
nt
a
na
ly
s
i
s
m
od
e
l
s
[
21]
.
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
M
ode
li
ng s
e
nt
ime
nt
analy
s
is
of
I
ndone
s
ia
n bi
odi
v
e
r
s
it
y
pol
ic
y
T
w
e
e
ts
…
(
M
ohamm
ad T
e
duh Ulini
ans
y
ah
)
2393
2.4.
M
od
e
l
p
e
r
f
or
m
an
c
e
e
val
u
at
io
n
T
hi
s
s
tu
dy
a
ddr
e
s
s
e
s
th
e
c
ha
ll
e
ng
e
of
a
na
ly
z
in
g
publ
ic
s
e
nt
i
m
e
nt
f
r
om
th
e
in
f
or
m
a
l
a
nd
di
ve
r
s
e
la
ngua
ge
us
e
d
on
I
ndone
s
ia
n
T
w
it
te
r
,
pa
r
ti
c
ul
a
r
ly
in
th
e
c
ont
e
xt
of
bi
odi
ve
r
s
it
y
pol
ic
ie
s
,
by
e
va
lu
a
ti
ng
th
e
e
f
f
e
c
ti
ve
ne
s
s
of
I
ndoB
E
R
T
w
e
e
t
c
om
p
a
r
e
d
to
s
e
ve
n
tr
a
d
it
io
na
l
c
la
s
s
if
ie
r
s
.
U
s
in
g
a
c
ur
a
te
d
da
ta
s
e
t
of
13,435
twe
e
ts
,
w
e
e
m
pl
oy
bot
h
T
F
-
I
D
F
a
nd
B
E
R
T
e
m
be
ddi
ngs
f
or
f
e
a
tu
r
e
e
xt
r
a
c
ti
on
in
th
e
tr
a
di
ti
ona
l
c
la
s
s
if
ie
r
s
,
w
hi
le
I
ndoB
E
R
T
w
e
e
t
le
ve
r
a
ge
s
it
s
pr
e
-
tr
a
in
e
d
e
m
b
e
ddi
ngs
.
K
e
y
m
e
tr
ic
s
s
uc
h
a
s
m
e
a
n
a
c
c
ur
a
c
y,
m
e
a
n
F1
-
s
c
or
e
,
a
nd
s
t
a
ti
s
ti
c
a
l
s
ig
ni
f
ic
a
nc
e
(
p
-
va
lu
e
s
)
a
r
e
us
e
d
to
c
om
pa
r
e
th
e
m
ode
l
s
to
e
ns
ur
e
a
r
obus
t
pe
r
f
or
m
a
nc
e
a
s
s
e
s
s
m
e
nt
.
F
ur
th
e
r
e
va
lu
a
ti
on
m
e
th
ods
in
c
lu
de
10
-
f
ol
d
c
r
os
s
-
va
li
da
ti
on
to
e
ns
ur
e
r
obus
tn
e
s
s
,
a
c
c
ur
a
c
y,
a
nd
F
1
-
s
c
or
e
to
m
e
a
s
ur
e
pr
e
di
c
ti
ve
c
or
r
e
c
tn
e
s
s
,
S
ta
ti
s
ti
c
a
l
S
ig
ni
f
ic
a
nc
e
te
s
t
s
to
va
li
da
te
pe
r
f
or
m
a
nc
e
di
f
f
e
r
e
nc
e
s
,
a
nd
W
or
d
A
tt
r
ib
ut
io
n
to
e
nha
nc
e
in
te
r
pr
e
ta
bi
li
ty
by
id
e
nt
if
yi
ng
ke
y
te
r
m
s
in
f
lu
e
nc
in
g
s
e
nt
im
e
nt
pr
e
di
c
ti
ons
.
T
he
s
e
e
va
lu
a
ti
ons
pr
ovi
de
c
r
it
ic
a
l
in
s
ig
ht
s
in
to
th
e
s
tr
e
ngt
hs
a
nd
li
m
it
a
ti
ons
of
e
a
c
h a
ppr
oa
c
h
f
or
a
na
ly
z
in
g s
e
nt
im
e
nt
f
r
om
I
ndone
s
ia
n T
w
it
te
r
da
ta
.
2.4.1.
10
-
f
ol
d
c
r
os
s
-
val
id
at
io
n
10
-
f
ol
d c
r
os
s
-
va
li
da
ti
on i
s
a
w
id
e
ly
r
e
c
ogni
z
e
d m
e
th
od f
or
e
va
l
ua
ti
ng ma
c
hi
ne
l
e
a
r
ni
ng mode
ls
[
22]
.
T
hi
s
t
e
c
hni
que
di
vi
de
s
t
he
da
ta
s
e
t
in
to
t
e
n e
qua
l
s
ubs
e
ts
, known a
s
“
f
ol
ds
.”
I
n e
a
c
h i
te
r
a
ti
on, one
-
f
ol
d
is
s
e
t
a
s
id
e
a
s
th
e
te
s
ti
ng
s
e
t,
w
hi
le
th
e
r
e
m
a
in
in
g
ni
ne
f
ol
ds
(
90
%
of
th
e
da
ta
)
a
r
e
us
e
d
f
o
r
t
r
a
in
in
g
th
e
m
ode
l
(
10%
f
or
t
e
s
ti
ng)
. T
hi
s
pr
oc
e
s
s
i
s
r
e
pe
a
te
d t
e
n t
im
e
s
, w
it
h e
a
c
h
s
ubs
e
t
s
e
r
vi
ng a
s
t
he
t
e
s
t
s
e
t
e
xa
c
tl
y on
c
e
. T
h
e
pe
r
f
or
m
a
nc
e
m
e
tr
ic
is
c
a
lc
ul
a
te
d
f
or
e
a
c
h
f
ol
d,
a
nd
th
e
ov
e
r
a
ll
pe
r
f
or
m
a
nc
e
is
de
te
r
m
in
e
d
by
a
ve
r
a
gi
ng
th
e
s
e
m
e
tr
ic
s
a
c
r
os
s
a
ll
t
e
n f
ol
ds
(
1)
.
=
1
10
∑
10
1
(
1)
T
hi
s
a
ve
r
a
g
e
of
f
e
r
s
a
c
om
pr
e
he
ns
iv
e
a
s
s
e
s
s
m
e
nt
of
t
he
m
ode
l’
s
c
a
pa
bi
li
ty
t
o ge
ne
r
a
li
z
e
t
o uns
e
e
n da
ta
.
10
-
f
ol
d c
r
os
s
-
va
li
da
ti
on e
nha
nc
e
s
e
va
lu
a
ti
on r
e
li
a
bi
li
ty
by
r
e
du
c
in
g va
r
ia
nc
e
f
r
om
r
a
ndom da
ta
s
pl
it
s
a
nd
e
ns
ur
in
g
e
a
c
h
in
s
ta
nc
e
ha
s
e
qu
a
l
c
ha
nc
e
s
in
tr
a
in
in
g
a
nd
te
s
ti
ng
s
e
ts
,
r
e
duc
in
g
ove
r
f
it
ti
ng
r
is
ks
.
S
c
ik
it
-
le
a
r
n’
s
c
r
os
s
_va
l_
s
c
or
e
f
unc
ti
on
a
ut
om
a
te
s
da
ta
di
vi
s
io
n,
m
ode
l
tr
a
in
in
g,
a
nd
pe
r
f
or
m
a
nc
e
e
va
lu
a
ti
on.
P
e
r
f
or
m
a
nc
e
m
e
tr
ic
s
a
r
e
a
ve
r
a
ge
d
a
c
r
os
s
a
ll
f
ol
ds
f
or
a
r
obus
t
e
s
t
im
a
te
,
m
a
ki
ng
t
hi
s
m
e
th
od
e
s
pe
c
ia
ll
y
e
f
f
e
c
ti
ve
w
he
n da
ta
i
s
l
im
it
e
d a
nd a
s
in
gl
e
t
r
a
in
-
te
s
t
s
pl
it
m
a
y not a
c
c
ur
a
t
e
ly
r
e
f
le
c
t
m
ode
l
ge
ne
r
a
li
z
a
ti
on.
2.4.2.
A
c
c
u
r
ac
y an
d
F
1
-
s
c
or
e
B
ui
ld
in
g
on
th
e
10
-
f
ol
d
c
r
os
s
-
va
li
da
ti
on
pr
oc
e
s
s
,
w
e
now
a
s
s
e
s
s
m
ode
l
pe
r
f
or
m
a
nc
e
us
in
g
a
c
c
ur
a
c
y
a
nd
F
1
-
s
c
or
e
,
w
it
h
r
e
s
ul
ts
a
ve
r
a
ge
d
a
c
r
os
s
a
ll
te
n
f
ol
ds
to
e
ns
ur
e
a
c
om
pr
e
he
ns
iv
e
e
v
a
lu
a
ti
on.
A
c
c
ur
a
c
y,
c
a
lc
ul
a
te
d
a
s
(
2)
,
pr
ovi
de
s
a
n
ove
r
a
ll
vi
e
w
of
m
ode
l
p
e
r
f
or
m
a
nc
e
,
w
he
r
e
tr
ue
pos
it
iv
e
s
(
T
P
)
a
nd
tr
ue
ne
ga
ti
ve
s
(
T
N
)
r
e
pr
e
s
e
nt
c
or
r
e
c
tl
y
pr
e
di
c
te
d
pos
it
iv
e
a
nd
ne
ga
ti
ve
in
s
ta
nc
e
s
,
a
nd
f
a
ls
e
pos
it
iv
e
s
(
F
P
)
a
nd
f
a
ls
e
ne
ga
ti
ve
s
(
F
N
)
a
c
c
ount
f
or
i
nc
or
r
e
c
t
pr
e
di
c
ti
ons
.
=
+
+
+
+
(
2)
W
hi
le
a
c
c
ur
a
c
y
gi
ve
s
a
br
oa
d
ove
r
vi
e
w
,
th
e
F
1
-
s
c
or
e
of
f
e
r
s
a
m
or
e
nua
nc
e
d
e
va
lu
a
ti
on,
pa
r
ti
c
ul
a
r
ly
f
or
im
ba
la
nc
e
d
da
ta
s
e
ts
.
P
r
e
c
is
io
n,
th
e
pr
opor
ti
on
of
T
P
out
of
a
l
l
pos
it
iv
e
pr
e
di
c
ti
ons
(
T
P
+
F
P
)
,
r
e
f
le
c
ts
th
e
m
ode
l’
s
a
bi
li
ty
to
a
voi
d
f
a
ls
e
pos
it
iv
e
s
,
w
h
il
e
r
e
c
a
ll
,
th
e
pr
opor
t
io
n
of
T
P
out
of
a
ll
a
c
tu
a
l
pos
it
iv
e
s
(
T
P
+
F
N
)
,
in
di
c
a
te
s
th
e
m
ode
l’
s
e
f
f
e
c
ti
ve
ne
s
s
in
id
e
nt
if
yi
ng
tr
ue
pos
it
iv
e
s
.
T
he
F
1
-
s
c
or
e
,
de
f
in
e
d
a
s
th
e
ha
r
m
oni
c
m
e
a
n
of
pr
e
c
is
io
n a
nd r
e
c
a
ll
, i
s
gi
ve
n by (
3)
.
1
_
=
2
×
×
+
(
3)
T
hi
s
m
e
tr
ic
ba
la
nc
e
s
t
he
t
r
a
de
-
of
f
be
twe
e
n
pr
e
c
is
io
n a
nd r
e
c
a
ll
,
m
a
ki
ng i
t
e
s
pe
c
ia
ll
y us
e
f
ul
w
he
n po
s
it
iv
e
a
nd
ne
ga
ti
ve
c
la
s
s
e
s
a
r
e
not
e
qua
ll
y
r
e
pr
e
s
e
nt
e
d.
C
a
lc
ul
a
ti
ng
bot
h
a
c
c
ur
a
c
y
a
nd
F
1
-
s
c
or
e
f
or
e
a
c
h
f
ol
d
dur
in
g
c
r
os
s
-
va
li
da
ti
on
e
ns
ur
e
s
th
a
t
our
e
va
lu
a
ti
on
r
e
f
le
c
ts
th
e
m
ode
l’
s
pe
r
f
or
m
a
nc
e
a
c
r
os
s
di
f
f
e
r
e
nt
da
ta
s
u
bs
e
ts
,
e
nha
n
c
in
g
th
e
r
obus
tn
e
s
s
of
our
r
e
s
ul
ts
.
2.4.3.
S
t
at
is
t
ic
al
s
ig
n
if
i
c
an
c
e
T
o a
s
s
e
s
s
t
he
r
e
li
a
bi
li
ty
of
pe
r
f
or
m
a
nc
e
di
f
f
e
r
e
nc
e
s
a
m
ong the
s
e
nt
im
e
nt
a
na
ly
s
is
m
ode
ls
, w
e
us
e
t
he
pa
ir
e
d
t
-
te
s
t
[
23]
.
T
hi
s
s
ta
ti
s
ti
c
a
l
te
s
t
c
om
pa
r
e
s
th
e
m
e
a
n a
c
c
ur
a
c
y
a
nd
F
1
-
s
c
or
e
of
th
e
m
ode
ls
,
w
hi
c
h
in
c
lu
d
e
B
E
R
T
-
ba
s
e
d
de
e
p
le
a
r
ni
ng
a
nd
tr
a
di
ti
ona
l
m
a
c
hi
ne
le
a
r
ni
ng
m
ode
ls
e
m
pl
oyi
ng
e
it
he
r
T
F
-
I
D
F
or
B
E
R
T
e
m
be
ddi
ngs
.
T
he
p
a
ir
e
d
t
-
te
s
t
e
v
a
lu
a
te
s
w
he
th
e
r
th
e
obs
e
r
ve
d
di
f
f
e
r
e
nc
e
s
in
pe
r
f
or
m
a
nc
e
m
e
tr
ic
s
a
c
r
o
s
s
th
e
10
-
f
ol
d c
r
os
s
-
va
li
da
ti
on a
r
e
s
ta
ti
s
ti
c
a
ll
y s
ig
ni
f
ic
a
nt
or
i
f
t
he
y c
o
ul
d be
a
tt
r
ib
ut
e
d t
o r
a
ndom va
r
ia
ti
on
[
24]
.
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
, N
o.
3
,
J
une
20
25
:
2389
-
2401
2394
T
he
t
-
va
lu
e
f
or
th
e
p
a
ir
e
d
t
-
te
s
t
is
c
a
lc
ul
a
te
d
a
s
(
4)
,
w
he
r
e
d
̅
is
t
he
m
e
a
n
of
th
e
di
f
f
e
r
e
nc
e
s
be
tw
e
e
n
pa
ir
e
d
obs
e
r
va
ti
ons
,
s
d
is
th
e
s
ta
nda
r
d
de
vi
a
ti
on
of
th
e
s
e
di
f
f
e
r
e
nc
e
s
,
a
nd
n
is
th
e
num
be
r
of
f
ol
ds
.
T
hi
s
t
-
va
lu
e
m
e
a
s
ur
e
s
ho
w
s
ig
ni
f
ic
a
nt
ly
th
e
m
e
a
n
di
f
f
e
r
e
nc
e
de
vi
a
te
s
f
r
om
z
e
r
o,
w
it
h
th
e
c
or
r
e
s
ponding
p
-
va
lu
e
de
r
iv
e
d
f
r
om
t
he
t
-
di
s
tr
ib
ut
io
n w
it
h n−
1 de
gr
e
e
s
of
f
r
e
e
dom
.
t
=
d
̅
s
d
√
n
⁄
(
4)
A
p
-
va
lu
e
le
s
s
th
a
n
0.05
in
di
c
a
te
s
th
a
t
th
e
p
e
r
f
or
m
a
nc
e
di
f
f
e
r
e
nc
e
s
a
r
e
s
ta
ti
s
ti
c
a
ll
y
s
ig
ni
f
ic
a
nt
[
25]
,
[
26]
,
s
ugge
s
ti
ng
th
a
t
one
m
ode
l
pe
r
f
or
m
s
be
tt
e
r
th
a
n
a
not
he
r
m
e
a
ni
ngf
ul
ly
.
T
hi
s
a
na
ly
s
i
s
e
ns
ur
e
s
th
a
t
our
c
onc
lu
s
io
ns
a
bout
m
ode
l
e
f
f
e
c
ti
ve
ne
s
s
a
r
e
s
uppor
te
d
by
s
ta
ti
s
ti
c
a
l
e
vi
de
nc
e
,
pr
ovi
di
ng
a
r
e
li
a
bl
e
ba
s
is
f
o
r
e
va
lu
a
ti
ng t
he
a
dva
nt
a
ge
s
of
di
f
f
e
r
e
nt
s
e
nt
im
e
nt
a
na
ly
s
is
t
e
c
hni
que
s
[
27]
.
2.4.4.
Wor
d
at
t
r
ib
u
t
io
n
I
n
a
ddi
ti
on
to
a
s
s
e
s
s
in
g
m
ode
l
pe
r
f
or
m
a
nc
e
w
it
h
a
c
c
ur
a
c
y
a
n
d
F
1
-
s
c
or
e
,
th
is
s
tu
dy
e
m
pl
oys
w
or
d
a
tt
r
ib
ut
io
n
te
c
hni
que
s
to
de
e
pe
n
in
s
ig
ht
s
in
to
s
e
nt
im
e
nt
a
na
l
ys
is
m
ode
ls
,
pa
r
ti
c
ul
a
r
ly
th
os
e
us
in
g
B
E
R
T
e
m
be
ddi
ngs
[
28]
,
[
29]
.
W
or
d
a
tt
r
ib
ut
io
n
is
c
r
uc
ia
l
f
or
unde
r
s
ta
n
di
ng
th
e
in
f
lu
e
nc
e
of
in
di
vi
dua
l
w
or
ds
or
to
ke
ns
on
m
ode
l
pr
e
di
c
ti
ons
,
e
ns
ur
in
g
th
a
t
th
e
de
r
iv
e
d
in
s
ig
ht
s
a
r
e
a
c
c
ur
a
te
a
nd
in
te
r
pr
e
ta
bl
e
.
T
o
qua
nt
if
y
e
a
c
h
w
or
d’
s
c
ont
r
ib
ut
io
n
to
s
e
nt
im
e
nt
pr
e
di
c
ti
on,
w
e
us
e
th
e
in
te
gr
a
te
d
gr
a
di
e
nt
s
m
e
th
od,
w
hi
c
h
c
om
put
e
s
th
e
a
ve
r
a
ge
gr
a
di
e
nt
of
th
e
m
ode
l’
s
out
put
w
it
h
r
e
s
pe
c
t
to
w
or
d
e
m
be
ddi
ngs
in
te
gr
a
te
d
f
r
om
a
ba
s
e
li
ne
(
e
.g.,
a
z
e
r
o
ve
c
to
r
)
to
t
he
a
c
tu
a
l
in
put
. T
he
f
or
m
ul
a
f
or
I
nt
e
gr
a
te
d G
r
a
di
e
nt
s
i
s
de
f
i
ne
d i
n (
5)
.
(
)
=
(
−
′
)
×
∫
(
′
+
×
(
−
′
)
)
1
0
(
5)
w
he
r
e
x
is
t
he
i
nput
f
e
a
tu
r
e
,
′
is
t
he
ba
s
e
li
ne
i
nput
,
F
is
t
he
m
ode
l
f
unc
ti
on, a
nd
α
is
a
s
c
a
li
ng f
a
c
to
r
[
29]
.
T
o
va
li
da
te
a
nd
vi
s
ua
li
z
e
r
e
s
ul
ts
,
w
e
us
e
w
or
d
c
lo
uds
to
di
s
pl
a
y
w
or
d
pr
om
in
e
nc
e
ba
s
e
d
on
a
tt
r
ib
ut
io
n
s
c
or
e
s
[
30]
,
[
31]
.
H
ig
hl
ig
ht
in
g
w
or
ds
th
a
t
s
ig
ni
f
ic
a
nt
ly
im
pa
c
t
s
e
nt
im
e
nt
pr
e
di
c
ti
ons
.
T
hi
s
c
om
bi
na
ti
on
o
f
qua
nt
it
a
ti
ve
a
tt
r
ib
ut
io
n
a
nd
qua
li
ta
ti
ve
v
is
ua
li
z
a
ti
on
e
nha
nc
e
s
t
he
in
te
r
pr
e
ta
bi
li
ty
a
nd
r
e
li
a
bi
li
ty
of
s
e
nt
im
e
nt
a
na
ly
s
is
m
ode
l
s
.
3.
R
E
S
U
L
T
S
A
N
D
D
I
S
C
U
S
S
I
O
N
T
hi
s
s
e
c
ti
on
pr
e
s
e
nt
s
th
e
r
e
s
ul
ts
of
c
om
pa
r
in
g
I
ndoB
E
R
T
w
e
e
t
w
it
h
tr
a
di
ti
ona
l
m
a
c
hi
ne
le
a
r
ni
ng
m
ode
ls
to
th
e
c
ur
a
te
d
da
ta
s
e
t,
w
it
h
pe
r
f
or
m
a
nc
e
m
e
a
s
ur
e
d
by
a
c
c
ur
a
c
y
a
nd
F
1
-
s
c
or
e
.
W
e
a
l
s
o
a
na
ly
z
e
s
e
nt
im
e
nt
di
s
tr
ib
ut
io
n
a
c
r
os
s
bi
odi
ve
r
s
it
y
pol
ic
y
s
ubdom
a
in
s
a
nd
e
v
a
lu
a
te
m
ode
l
p
e
r
f
or
m
a
nc
e
th
r
ough
s
ta
ti
s
ti
c
a
l
s
ig
ni
f
ic
a
nc
e
a
nd
w
or
d
-
a
tt
r
ib
ut
io
n
a
na
ly
s
is
.
T
he
s
e
f
in
di
ngs
pr
ovi
de
in
s
ig
ht
s
in
to
th
e
e
f
f
e
c
ti
ve
n
e
s
s
of
th
e
m
ode
ls
a
nd of
f
e
r
r
e
c
om
m
e
nda
ti
ons
f
or
f
ut
ur
e
i
m
pr
ove
m
e
nt
s
.
3.1.
L
ab
e
li
n
g
r
e
s
u
lt
s
T
he
a
n
a
ly
s
is
of
s
e
nt
im
e
nt
di
s
tr
ib
ut
io
n
w
it
hi
n
our
la
be
le
d
d
a
ta
s
e
t
of
13,435
tw
e
e
ts
r
e
v
e
a
ls
di
s
ti
nc
t
pa
tt
e
r
ns
a
c
r
os
s
va
r
io
us
s
ubdoma
in
s
of
bi
odi
ve
r
s
it
y
pol
ic
ie
s
in
I
ndone
s
ia
.
T
h
e
he
a
lt
h
s
ubdoma
in
e
xhi
bi
ts
th
e
hi
ghe
s
t
pr
opor
ti
on
of
pos
it
iv
e
twe
e
ts
,
w
hi
le
f
oo
d
s
e
c
ur
it
y
is
pr
e
d
om
in
a
nt
ly
a
s
s
oc
ia
t
e
d
w
it
h
ne
ga
ti
ve
s
e
nt
im
e
nt
.
T
he
s
e
f
in
di
ngs
unde
r
s
c
or
e
th
e
h
e
te
r
oge
ne
ous
na
tu
r
e
of
pu
bl
ic
s
e
nt
im
e
nt
,
w
it
h
F
ig
ur
e
3
pr
ovi
di
ng
a
c
om
pr
e
he
ns
iv
e
ove
r
vi
e
w
of
pos
it
iv
e
, ne
g
a
ti
ve
, a
nd ne
ut
r
a
l
s
e
nt
i
m
e
nt
s
a
c
r
os
s
t
h
e
s
ubdoma
in
s
.
F
ig
ur
e
3. D
is
tr
ib
ut
io
n of
s
e
nt
im
e
nt
l
a
be
ls
a
c
r
os
s
I
ndone
s
i
a
n T
w
it
te
r
s
ubdoma
in
s
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
M
ode
li
ng s
e
nt
ime
nt
analy
s
is
of
I
ndone
s
ia
n bi
odi
v
e
r
s
it
y
pol
ic
y
T
w
e
e
ts
…
(
M
ohamm
ad T
e
duh Ulini
ans
y
ah
)
2395
3.2.
E
val
u
at
io
n
r
e
s
u
lt
s
T
hi
s
e
xpe
r
im
e
nt
e
va
lu
a
te
d
s
e
ve
n
non
-
de
e
p
-
le
a
r
ni
ng
m
e
th
od
s
-
lo
gi
s
ti
c
r
e
gr
e
s
s
io
n,
s
uppor
t
v
e
c
to
r
c
la
s
s
if
ie
r
,
r
a
ndom
f
or
e
s
t,
L
G
B
M
C
la
s
s
if
ie
r
,
X
G
B
oos
t,
A
da
B
o
os
t,
a
nd
de
c
is
io
n
tr
e
e
-
us
in
g
bot
h
T
F
-
I
D
F
a
nd
B
E
R
T
e
m
be
ddi
ng
s
a
s
in
put
f
e
a
tu
r
e
s
.
A
ddi
ti
ona
ll
y,
th
e
I
ndoB
E
R
T
w
e
e
t
pr
e
-
tr
a
in
e
d
m
ode
l
w
a
s
te
s
te
d
w
it
h
th
r
e
e
va
r
ia
ti
ons
of
a
f
ul
ly
c
onne
c
te
d
la
y
e
r
:
no
a
c
ti
va
ti
on
f
unc
ti
on,
G
E
L
U
,
a
nd
T
a
nh.
I
n
to
ta
l,
17
s
e
nt
im
e
nt
a
na
ly
s
i
s
m
ode
ls
w
e
r
e
e
va
lu
a
te
d
us
in
g
10
-
f
ol
d
c
r
os
s
-
va
li
da
ti
on,
w
it
h
a
c
c
ur
a
c
y
a
nd
F
1
-
s
c
or
e
m
e
tr
ic
s
c
om
pa
r
e
d
a
c
r
os
s
m
ode
ls
. S
ta
ti
s
ti
c
a
l
s
ig
ni
f
ic
a
nc
e
t
e
s
ts
w
e
r
e
c
ondu
c
te
d t
o c
om
pa
r
e
m
ode
l
pa
ir
s
, a
nd t
he
be
s
t
-
pe
r
f
or
m
in
g B
E
R
T
-
ba
s
e
d
m
o
de
l
unde
r
w
e
nt
w
or
d
-
a
tt
r
ib
ut
io
n
a
na
ly
s
is
to
unde
r
s
ta
nd
th
e
im
pa
c
t
of
s
pe
c
if
ic
w
or
ds
on
s
e
nt
im
e
nt
pr
e
di
c
ti
on.
3.2.1. P
e
r
f
or
m
an
c
e
m
e
t
r
ic
T
a
bl
e
1 pr
e
s
e
nt
s
t
he
e
va
lu
a
ti
on me
tr
ic
s
f
or
e
a
c
h m
ode
l,
i
nc
lu
di
ng me
a
n a
c
c
ur
a
c
y a
nd me
a
n F
1 s
c
or
e
,
a
ve
r
a
ge
d
ove
r
te
n
f
ol
ds
a
nd
hi
ghl
ig
ht
in
g
th
e
be
s
t
pe
r
f
or
m
a
n
c
e
f
r
om
th
e
f
ol
ds
.
T
he
B
E
R
T
-
ba
s
e
d
m
ode
ls
de
m
ons
tr
a
te
d
s
up
e
r
io
r
pe
r
f
or
m
a
nc
e
to
tr
a
di
ti
ona
l
m
a
c
hi
ne
le
a
r
ni
ng
m
ode
ls
,
a
s
il
lu
s
tr
a
te
d
in
F
ig
ur
e
4,
w
hi
c
h
s
how
s
e
a
c
h
c
la
s
s
if
ie
r
’
s
m
e
a
n
a
c
c
ur
a
c
y
a
nd
m
e
a
n F
1
s
c
or
e
.
T
he
hi
ghe
s
t
m
e
a
n
F
1 s
c
or
e
of
0.7633
w
a
s
a
c
hi
e
v
e
d
us
in
g
th
e
T
a
nh a
c
ti
va
ti
on
f
unc
ti
on,
w
hi
le
th
e
G
E
L
U
a
c
ti
va
ti
on f
unc
ti
on
yi
e
ld
e
d
th
e
hi
ghe
s
t
m
e
a
n
a
c
c
ur
a
c
y
of
0.7899.
T
he
s
e
r
e
s
ul
ts
unde
r
s
c
or
e
B
E
R
T
’
s
e
x
c
e
pt
io
na
l
c
a
pa
bi
li
t
y
to
c
a
pt
ur
e
th
e
nua
nc
e
d
s
e
nt
im
e
nt
e
xpr
e
s
s
e
d
w
it
hi
n
th
e
da
ta
s
e
t.
F
ur
th
e
r
m
or
e
,
th
e
B
E
R
T
-
ba
s
e
d
m
ode
ls
s
how
e
d
r
e
m
a
r
ka
bl
e
c
ons
is
te
nc
y
in
th
e
ir
pe
r
f
or
m
a
nc
e
,
w
it
h s
li
ght
va
r
ia
ti
ons
i
n a
c
c
ur
a
c
y
a
nd F
1 s
c
or
e
s
a
c
r
os
s
di
f
f
e
r
e
nt
a
c
ti
va
ti
on f
unc
ti
ons
.
T
a
bl
e
1.
T
he
a
c
c
ur
a
c
y a
nd F
1
-
s
c
or
e
f
or
e
a
c
h m
ode
l
V
e
c
t
or
i
z
a
t
i
on
C
l
a
s
s
i
f
i
e
r
M
e
a
n
a
c
c
ur
a
c
y
S
t
d.
D
e
v
a
c
c
ur
a
c
y
M
e
a
n
F
1
s
c
or
e
S
t
d.
D
e
v
F
1
s
c
or
e
B
e
s
t
a
c
c
ur
a
c
y
B
e
s
t
F
1
s
c
or
e
TF
-
I
D
F
L
ogi
s
t
i
c
r
e
gr
e
s
s
i
on
0.7144
0.0135
0.6688
0.0094
0.7337
0.6819
S
uppor
t
ve
c
t
or
c
l
a
s
s
i
f
i
e
r
0.7101
0.0142
0.6556
0.0097
0.7307
0.6689
R
a
ndom
f
or
e
s
t
0.6670
0.0137
0.6022
0.0114
0.6884
0.6212
L
G
B
M
C
l
a
s
s
i
f
i
e
r
0.6915
0.0118
0.6475
0.0104
0.7149
0.6700
X
G
B
oos
t
0.6762
0.0151
0.6281
0.0151
0.7119
0.6637
A
da
B
oos
t
0.6153
0.0144
0.5688
0.0152
0.6399
0.5891
D
e
c
i
s
i
on t
r
e
e
0.5596
0.0100
0.5263
0.0124
0.5781
0.5479
B
E
R
T
e
m
be
ddi
ng
L
ogi
s
t
i
c
r
e
gr
e
s
s
i
on
0.7409
0.0087
0.7057
0.0058
0.7559
0.7145
S
uppor
t
ve
c
t
or
c
l
a
s
s
i
f
i
e
r
0.7616
0.0096
0.7198
0.0047
0.7792
0.7277
R
a
ndom
f
or
e
s
t
0.6828
0.0155
0.6075
0.0121
0.7041
0.6192
L
G
B
M
C
l
a
s
s
i
f
i
e
r
0.7362
0.0124
0.6952
0.0086
0.7511
0.7102
X
G
B
oos
t
0.7315
0.0122
0.6878
0.0118
0.7443
0.7040
A
da
B
oos
t
0.6689
0.0125
0.6144
0.0120
0.6920
0.6287
D
e
c
i
s
i
on t
r
e
e
0.5396
0.0121
0.5053
0.0108
0.5615
0.5201
B
E
R
T
I
ndoB
E
R
T
w
e
e
t
-
G
E
L
U
0.7899
0.0123
0.7632
0.0140
0.8065
0.7810
I
ndoB
E
R
T
w
e
e
t
-
T
A
N
H
0.7891
0.0129
0.7633
0.0156
0.8088
0.7907
I
ndoB
E
R
T
w
e
e
t
-
NONE
0.7868
0.0103
0.7582
0.0153
0.8080
0.7917
U
s
in
g
B
E
R
T
w
or
d
e
m
be
ddi
ngs
a
s
f
e
a
tu
r
e
s
f
or
tr
a
di
ti
ona
l
m
a
c
hi
ne
le
a
r
ni
ng
m
ode
ls
,
ot
he
r
th
a
n
D
e
c
is
io
n
T
r
e
e
,
in
c
r
e
a
s
e
d
th
e
pe
r
f
or
m
a
nc
e
of
th
e
r
e
s
ul
ti
ng
m
ode
ls
.
H
ow
e
ve
r
,
th
e
y
w
e
r
e
s
ti
ll
lo
w
e
r
th
a
n
th
e
B
E
R
T
-
ba
s
e
d
m
ode
ls
.
T
he
pe
r
f
or
m
a
nc
e
de
c
r
e
a
s
e
of
th
e
D
e
c
i
s
io
n
T
r
e
e
w
h
e
n
us
in
g
B
E
R
T
e
m
be
ddi
ngs
c
a
n
be
a
tt
r
ib
ut
e
d
to
th
e
de
ns
e
na
tu
r
e
o
f
B
E
R
T
e
m
be
ddi
ng
s
,
w
hi
c
h
m
a
ke
s
it
ha
r
de
r
to
f
in
d
opt
im
a
l
s
pl
it
s
f
or
th
e
D
e
c
is
io
n
T
r
e
e
.
I
n
c
ont
r
a
s
t,
T
F
-
I
D
F
f
e
a
tu
r
e
s
a
r
e
s
pa
r
s
e
,
w
hi
c
h
is
m
or
e
s
ui
ta
bl
e
f
or
D
e
c
is
io
n T
r
e
e
s
a
s
th
e
y c
a
n
e
f
f
e
c
ti
ve
ly
l
e
ve
r
a
ge
t
he
s
pa
r
s
it
y f
or
be
tt
e
r
s
pl
it
s
.
P
r
io
r
s
e
nt
im
e
nt
a
na
ly
s
is
s
tu
di
e
s
in
th
is
f
ie
ld
h
a
ve
us
e
d
ge
ne
r
a
l
m
a
c
hi
ne
le
a
r
ni
ng
a
nd
N
L
P
m
ode
ls
,
but
th
e
y
ha
ve
not
f
ul
ly
a
ddr
e
s
s
e
d
th
e
uni
que
li
ngui
s
ti
c
a
nd
c
ul
tu
r
a
l
c
ha
r
a
c
te
r
is
ti
c
s
of
I
ndone
s
ia
n
s
oc
ia
l
m
e
di
a
di
s
c
our
s
e
.
I
ndoB
E
R
T
w
e
e
t
e
f
f
e
c
ti
ve
ly
f
il
ls
th
i
s
r
e
s
e
a
r
c
h
ga
p,
de
m
ons
tr
a
ti
ng
hi
ghe
r
a
c
c
ur
a
c
y
a
nd
F
1
s
c
or
e
s
th
a
n
tr
a
di
ti
ona
l
m
ode
ls
li
ke
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
s
,
a
nd
r
a
ndom
f
or
e
s
t
.
S
pe
c
if
ic
a
ll
y,
I
ndoB
E
R
T
w
e
e
t
a
c
hi
e
ve
d
a
m
e
a
n
a
c
c
ur
a
c
y
of
78.99%
a
nd
a
n
F
1
s
c
or
e
of
0.7633,
s
ur
pa
s
s
in
g
th
e
hi
gh
e
s
t
a
c
c
ur
a
c
y
a
nd
F
1
s
c
or
e
of
71.44%
a
nd
0.6688,
r
e
s
pe
c
ti
ve
ly
,
a
c
hi
e
ve
d
by
tr
a
di
ti
ona
l
m
ode
l
s
.
I
ts
pr
e
-
tr
a
in
e
d
c
a
pa
bi
li
ti
e
s
e
na
bl
e
a
m
or
e
nua
nc
e
d
unde
r
s
ta
ndi
ng
of
c
om
pl
e
x
s
e
nt
im
e
nt
e
xpr
e
s
s
io
ns
s
pe
c
if
ic
to
I
ndone
s
ia
n
bi
odi
ve
r
s
it
y poli
c
y. W
hi
le
t
r
a
di
ti
on
a
l
m
ode
ls
pe
r
f
or
m
e
d r
e
a
s
on
a
bl
y w
it
h T
F
-
I
D
F
ve
c
to
r
iz
a
ti
on, t
he
y
s
tr
uggl
e
d
w
it
h
th
e
in
f
or
m
a
l
la
ngua
ge
pr
e
va
le
nt
on
s
oc
ia
l
m
e
di
a
.
D
e
s
pi
te
I
ndoB
E
R
T
w
e
e
t'
s
a
dva
nt
a
ge
s
,
tr
a
di
ti
ona
l
m
ode
ls
r
e
m
a
in
va
lu
a
bl
e
in
r
e
s
our
c
e
-
li
m
it
e
d
s
e
tt
in
gs
f
or
th
e
ir
s
im
pl
e
r
im
pl
e
m
e
nt
a
ti
on
a
nd
lo
w
e
r
c
om
put
a
ti
ona
l
de
m
a
nds
.
T
he
s
e
f
in
di
ngs
unde
r
s
c
or
e
th
e
im
por
ta
nc
e
of
s
e
le
c
ti
n
g
m
ode
ls
s
ui
te
d
to
th
e
uni
que
r
e
qui
r
e
m
e
nt
s
of
a
gi
ve
n
ta
s
k
a
nd
c
ont
e
xt
,
r
e
ve
a
li
ng
I
ndoB
E
R
T
w
e
e
t’
s
pot
e
nt
ia
l
to
a
dva
nc
e
m
or
e
e
qui
ta
bl
e
a
nd
pr
e
c
is
e
s
e
nt
im
e
nt
a
na
ly
s
is
i
n I
ndone
s
i
a
n bi
odi
ve
r
s
it
y di
s
c
our
s
e
.
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
, N
o.
3
,
J
une
20
25
:
2389
-
2401
2396
H
ow
e
ve
r
, t
hi
s
s
tu
dy i
s
l
im
it
e
d by it
s
r
e
li
a
nc
e
on a
s
in
gl
e
da
ta
s
e
t
f
r
om
T
w
it
te
r
, w
hi
c
h, w
hi
le
e
f
f
e
c
ti
ve
in
e
va
lu
a
ti
ng
I
ndoB
E
R
T
w
e
e
t’
s
s
tr
e
ngt
hs
,
m
a
y
not
f
ul
ly
c
a
pt
ur
e
s
e
nt
im
e
nt
nua
nc
e
s
a
c
r
os
s
ot
he
r
s
oc
ia
l
m
e
di
a
pl
a
tf
or
m
s
.
T
hi
s
li
m
it
a
ti
on
c
oul
d
a
f
f
e
c
t
th
e
m
o
de
l'
s
a
ppl
ic
a
bi
li
ty
in
br
oa
de
r
c
ont
e
xt
s
,
a
s
s
e
nt
im
e
nt
e
xpr
e
s
s
io
ns
m
a
y
va
r
y
s
ig
ni
f
ic
a
nt
ly
a
c
r
os
s
di
f
f
e
r
e
nt
pl
a
tf
or
m
s
.
A
s
a
r
e
s
ul
t,
th
e
f
in
di
ngs
s
houl
d
be
in
te
r
pr
e
te
d w
it
h
c
a
ut
io
n,
r
e
c
ogni
z
in
g t
ha
t
a
ddi
ti
ona
l
da
ta
s
e
t
s
m
a
y be
ne
e
d
e
d t
o c
onf
ir
m
t
he
m
ode
l’
s
ge
n
e
r
a
li
z
a
bi
li
ty
.
F
ig
ur
e
4. C
om
pa
r
is
on of
m
e
a
n a
c
c
ur
a
c
ie
s
a
nd F
1
-
s
c
or
e
s
f
or
e
a
c
h m
ode
l
3.2.2. S
t
at
is
t
ic
al
s
ig
n
if
ic
an
c
e
an
al
ys
is
F
ig
ur
e
5
p
r
ovi
de
s
th
e
p
-
va
lu
e
s
f
or
pa
i
r
w
is
e
c
om
pa
r
is
ons
of
th
e
F
1
-
s
c
or
e
be
twe
e
n
a
ll
17
m
ode
ls
.
A
l
l
p
-
va
lu
e
s
a
m
ong B
E
R
T
-
ba
s
e
d
a
nd t
r
a
di
ti
ona
l
m
a
c
hi
ne
l
e
a
r
ni
ng mode
ls
ha
ve
va
lu
e
s
l
e
s
s
t
ha
n
0.05, indi
c
a
ti
ng a
s
ta
ti
s
ti
c
a
ll
y
s
ig
ni
f
ic
a
nt
di
f
f
e
r
e
nc
e
.
T
he
p
-
va
lu
e
s
a
m
ong
B
E
RT
-
ba
s
e
d
m
ode
l
s
a
r
e
a
ll
gr
e
a
te
r
th
a
n
0.05
,
in
di
c
a
ti
ng
th
e
y
do
not
d
if
f
e
r
s
ig
ni
f
ic
a
nt
ly
.
T
he
ta
bl
e
a
ls
o
s
how
s
th
a
t
th
e
us
e
of
B
E
R
T
-
e
m
be
ddi
ng
in
t
r
a
di
ti
ona
l
m
a
c
hi
ne
le
a
r
ni
ng
m
e
th
ods
m
os
tl
y
di
f
f
e
r
s
s
ig
ni
f
ic
a
nt
ly
f
r
om
us
in
g
T
F
-
I
D
F
,
e
xc
e
pt
f
or
th
e
pa
ir
of
r
a
ndo
m
f
o
r
e
s
t
us
in
g B
E
R
T
e
m
be
ddi
ng
a
nd T
F
-
I
D
F
, w
hi
c
h ha
s
a
p
-
va
lu
e
of
0.
3.
F
ig
ur
e
5.
T
he
p
-
va
lu
e
s
f
or
pa
ir
w
is
e
c
om
pa
r
is
on
s
be
twe
e
n m
ode
ls
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
M
ode
li
ng s
e
nt
ime
nt
analy
s
is
of
I
ndone
s
ia
n bi
odi
v
e
r
s
it
y
pol
ic
y
T
w
e
e
ts
…
(
M
ohamm
ad T
e
duh Ulini
ans
y
ah
)
2397
3.2.3. Wor
d
-
at
t
r
ib
u
t
io
n
an
al
ys
is
T
hi
s
s
tu
dy
f
oc
us
e
s
on
s
e
nt
im
e
nt
c
la
s
s
if
ic
a
ti
on
us
in
g
w
or
d
a
tt
r
ib
ut
io
n
de
r
iv
e
d
f
r
om
th
e
in
te
gr
a
te
d
gr
a
di
e
nt
s
m
e
th
od,
s
uppl
e
m
e
nt
e
d
by
w
or
d
f
r
e
que
nc
y
a
na
ly
s
is
.
T
he
in
te
gr
a
te
d
gr
a
di
e
nt
s
m
e
th
od
e
nh
a
nc
e
s
th
e
in
te
r
pr
e
ta
bi
li
ty
of
s
e
nt
im
e
nt
pr
e
di
c
ti
ons
by
c
a
lc
ul
a
ti
ng
w
or
d
-
le
ve
l
c
ont
r
ib
ut
io
ns
f
r
om
a
ba
s
e
li
ne
to
th
e
a
c
tu
a
l
in
put
,
pa
r
ti
c
ul
a
r
ly
in
th
e
I
ndoB
E
R
T
w
e
e
t
-
G
E
L
U
m
ode
l,
w
hi
c
h
is
th
e
be
s
t
-
pe
r
f
or
m
in
g
m
ode
l
in
th
is
r
e
s
e
a
r
c
h.
A
lt
hough
ot
he
r
m
e
th
ods
li
ke
s
ha
pl
e
y
a
ddi
ti
ve
e
xpl
a
na
ti
ons
(
S
H
A
P
)
[
32]
a
nd
lo
c
a
l
in
te
r
pr
e
ta
bl
e
m
ode
l
-
a
gnos
ti
c
e
xpl
a
na
ti
ons
(
L
I
M
E
)
[
33]
c
oul
d
pr
ovi
de
a
ddi
ti
ona
l
i
nt
e
r
pr
e
ta
bi
l
it
y by a
na
ly
z
in
g w
or
d
in
te
r
a
c
ti
ons
or
bui
ld
in
g
lo
c
a
l
a
ppr
oxi
m
a
ti
ons
, t
he
y w
e
r
e
not
i
nc
lu
de
d i
n t
hi
s
s
tu
dy.
a.
W
or
d
f
r
e
que
nc
y
c
lo
uds
W
or
d
f
r
e
que
nc
y
c
lo
uds
vi
s
ua
ll
y
s
um
m
a
r
iz
e
th
e
m
os
t
f
r
e
qu
e
nt
ly
m
e
nt
io
ne
d
te
r
m
s
w
it
hi
n
e
a
c
h
s
e
nt
im
e
nt
c
a
te
gor
y
-
ne
ga
ti
ve
,
n
e
ut
r
a
l,
a
nd
pos
it
iv
e
.
T
he
s
e
c
lo
uds
he
lp
id
e
nt
if
y
dom
in
a
nt
th
e
m
e
s
in
publ
ic
di
s
c
our
s
e
.
‒
N
e
ga
ti
ve
s
e
nt
im
e
nt
:
A
s
s
how
n
in
F
ig
ur
e
6
(
a
)
,
th
e
w
or
d
f
r
e
que
nc
y
c
lo
ud
f
or
ne
ga
ti
ve
s
e
nt
im
e
nt
hi
ghl
ig
ht
s
te
r
m
s
s
uc
h
a
s
‘
k
e
bak
ar
an
hut
an
’
(
f
or
e
s
t
f
ir
e
s
)
a
nd ‘
impor
be
r
as
’
(
r
ic
e
im
po
r
ts
)
.
T
he
s
e
te
r
m
s
s
tr
ongl
y
f
oc
us
on e
nvi
r
onm
e
nt
a
l
c
r
is
e
s
a
nd f
ood s
e
c
ur
it
y, r
e
f
le
c
ti
ng w
id
e
s
pr
e
a
d publi
c
c
onc
e
r
n.
‒
N
e
ut
r
a
l
s
e
nt
im
e
nt
:
F
ig
ur
e
6
(
b)
il
lu
s
tr
a
te
s
th
a
t
ne
ut
r
a
l
s
e
nt
im
e
nt
is
dom
in
a
te
d
by
w
or
ds
li
ke
‘
m
obi
l
li
s
tr
ik
’
(
e
le
c
tr
ic
c
a
r
s
)
a
nd
‘
s
tu
nt
in
g,’
w
hi
c
h
s
ugge
s
t
ongoing
di
s
c
us
s
io
ns
a
r
ound
te
c
hnol
ogy
a
nd
publ
ic
he
a
lt
h,
ty
pi
c
a
ll
y r
e
por
te
d i
n a
f
a
c
tu
a
l
or
ne
ut
r
a
l
to
ne
.
‒
P
os
it
iv
e
s
e
nt
im
e
nt
:
T
he
w
or
d f
r
e
que
nc
y c
lo
ud
f
or
pos
it
iv
e
s
e
nt
i
m
e
nt
F
ig
ur
e
6
(
c
)
e
m
pha
s
iz
e
s
t
e
r
m
s
s
u
c
h a
s
‘
I
ndone
s
ia
’
a
nd
‘
k
e
ndar
aan
li
s
tr
ik
’
(
e
le
c
tr
ic
ve
hi
c
le
s
)
,
in
di
c
a
ti
ve
of
na
ti
ona
l
pr
id
e
a
nd
opt
im
is
m
to
w
a
r
ds
te
c
hnol
ogi
c
a
l
pr
ogr
e
s
s
a
nd e
nvi
r
onm
e
nt
a
l
s
us
ta
in
a
bi
li
ty
.
I
t
is
im
por
ta
nt
to
not
e
th
a
t
th
e
s
e
w
or
d
c
lo
uds
a
r
e
ba
s
e
d
on
r
a
w
w
or
d
f
r
e
que
nc
ie
s
a
nd
do
not
a
c
c
ount
f
o
r
th
e
s
pe
c
if
ic
pe
r
f
or
m
a
nc
e
of
s
e
nt
im
e
nt
a
na
ly
s
i
s
m
ode
ls
.
b.
W
or
d
a
tt
r
ib
ut
io
n c
lo
u
ds
W
or
d
a
tt
r
ib
ut
io
n
c
lo
uds
,
de
r
iv
e
d
f
r
om
I
n
te
gr
a
te
d
G
r
a
di
e
nt
s
,
s
pot
li
ght
th
e
w
or
ds
w
it
h
th
e
hi
ghe
s
t
m
e
a
n
a
tt
r
ib
ut
io
n va
lu
e
s
, s
ig
ni
f
ic
a
nt
ly
i
m
pa
c
ti
ng s
e
nt
im
e
nt
c
la
s
s
if
ic
a
ti
on.
‒
N
e
ga
ti
ve
s
e
nt
im
e
nt
:
A
s
de
pi
c
te
d
in
F
ig
ur
e
7
(
a
)
,
w
or
ds
li
ke
‘
pos
i
ti
f
c
ov
id
’
(
C
O
V
I
D
-
pos
it
iv
e
)
a
nd
‘
k
e
bi
ja
k
an
impor
’
(
im
por
t
pol
ic
y)
e
xhi
bi
t
hi
gh
a
tt
r
ib
ut
io
n
va
lu
e
s
.
T
he
s
e
te
r
m
s
a
r
e
c
r
it
ic
a
l
in
dr
iv
in
g
ne
g
a
ti
ve
s
e
nt
im
e
nt
,
hi
ghl
ig
ht
in
g publi
c
c
onc
e
r
ns
ove
r
he
a
lt
h c
r
is
e
s
a
nd e
c
onomi
c
p
ol
ic
ie
s
.
‒
N
e
ut
r
a
l
s
e
nt
im
e
nt
:
F
ig
ur
e
7
(
b)
s
ho
w
s
th
a
t
in
ne
ut
r
a
l
s
e
nt
im
e
nt
,
w
or
ds
s
uc
h
a
s
‘
hut
an
m
angr
o
v
e
’
(
m
a
ngr
ove
f
or
e
s
t)
a
nd ‘
obat
he
r
bal
’
(
he
r
ba
l
m
e
di
c
in
e
)
ha
ve
hi
gh
a
tt
r
ib
ut
io
n va
lu
e
s
.
T
he
s
e
w
or
d
s
a
r
e
c
e
nt
r
a
l
to
ne
ut
r
a
l
di
s
c
our
s
e
, r
e
f
le
c
ti
ng ba
la
nc
e
d, c
ont
e
xt
-
s
pe
c
if
ic
c
ont
e
nt
.
‒
P
os
it
iv
e
s
e
nt
im
e
nt
:
F
or
pos
it
iv
e
s
e
nt
im
e
nt
,
F
ig
ur
e
7
(
c
)
r
e
v
e
a
ls
th
a
t
w
or
ds
li
ke
‘
s
al
ur
k
an
bant
uan
’
(
di
s
tr
ib
ut
e
a
id
)
a
nd ‘
ja
ga k
e
ta
hanan
’
(
m
a
in
ta
in
r
e
s
il
ie
nc
e
)
hol
d
th
e
hi
ghe
s
t
a
tt
r
ib
ut
io
n va
lu
e
s
. T
he
s
e
t
e
r
m
s
unde
r
s
c
or
e
t
he
i
m
por
ta
nc
e
of
c
om
m
uni
ty
s
uppor
t
a
nd r
e
s
il
ie
nc
e
i
n pos
it
iv
e
s
e
nt
im
e
nt
e
xpr
e
s
s
io
ns
.
T
he
s
e
a
tt
r
ib
ut
io
n
c
lo
uds
a
r
e
ge
n
e
r
a
te
d
u
s
in
g
th
e
I
ndoB
E
R
T
w
e
e
t
-
G
E
L
U
m
ode
l,
w
hi
c
h
is
th
e
be
s
t
-
pe
r
f
or
m
in
g
m
ode
l
in
our
a
na
ly
s
is
, t
he
r
e
by pr
ovi
di
ng mor
e
a
c
c
ur
a
te
a
nd
s
e
n
ti
m
e
nt
-
s
pe
c
if
ic
i
ns
ig
ht
s
.
c.
C
om
pa
r
is
on of
w
or
d c
lo
uds
:
f
r
e
que
nc
y vs
. a
tt
r
ib
ut
io
n
T
a
bl
e
2
c
om
pa
r
e
s
th
e
to
p
f
our
w
or
ds
de
r
iv
e
d
f
r
om
bot
h
f
r
e
que
nc
y
a
nd
a
tt
r
ib
ut
io
n
a
na
ly
s
e
s
f
or
e
a
c
h
s
e
nt
im
e
nt
c
a
te
gor
y t
o unde
r
s
ta
nd t
he
di
f
f
e
r
e
nc
e
s
i
n w
or
d s
ig
ni
f
ic
a
nc
e
ba
s
e
d on a
na
ly
s
is
a
ppr
oa
c
h
e
s
.
‒
N
e
ga
ti
ve
s
e
nt
i
m
e
nt
:
T
h
e
f
r
e
que
nc
y
c
lo
ud
e
m
pha
s
iz
e
s
br
oa
de
r
s
oc
ie
ta
l
is
s
ue
s
li
ke
‘
k
e
bak
a
r
an
hut
an
’
(
f
or
e
s
t
f
ir
e
s
)
,
w
hi
le
th
e
a
tt
r
ib
ut
io
n
c
lo
ud
id
e
nt
if
ie
s
‘
pos
it
if
c
ov
id
’
(
C
O
V
I
D
-
pos
it
iv
e
)
a
s
a
ke
y
dr
iv
e
r
of
ne
ga
ti
ve
s
e
nt
im
e
nt
de
s
pi
te
i
ts
l
ow
e
r
f
r
e
que
nc
y. T
hi
s
r
e
s
ul
t
de
m
ons
tr
a
te
s
t
he
I
ndoB
E
R
T
w
e
e
t
-
G
E
L
U
m
ode
l’
s
a
bi
li
ty
to
di
s
c
e
r
n m
or
e
i
m
pa
c
tf
ul
w
or
ds
c
ont
r
ib
ut
in
g t
o ne
ga
ti
ve
s
e
nt
im
e
nt
.
‒
N
e
ut
r
a
l
s
e
nt
i
m
e
nt
:
C
om
m
on t
opi
c
s
l
ik
e
‘
m
obi
l
li
s
tr
ik
’
(
e
le
c
tr
ic
c
a
r
s
)
a
r
e
pr
om
in
e
nt
i
n t
he
f
r
e
que
nc
y c
lo
ud,
w
he
r
e
a
s
‘
hut
an
m
ang
r
ov
e
’
(
m
a
ngr
ove
f
or
e
s
t)
e
m
e
r
ge
s
in
th
e
a
tt
r
ib
ut
io
n
c
lo
ud,
s
how
in
g
it
s
im
por
ta
nc
e
in
ne
ut
r
a
l
di
s
c
us
s
io
n
s
de
s
pi
te
i
t
s
r
e
la
ti
ve
i
nf
r
e
que
nc
y. T
hi
s
r
e
s
ul
t
r
e
f
le
c
ts
t
he
m
ode
l’
s
nua
nc
e
d unde
r
s
ta
ndi
ng
of
ne
ut
r
a
l
s
e
nt
im
e
nt
.
‒
P
os
it
iv
e
s
e
nt
im
e
nt
:
W
hi
le
‘
I
ndone
s
ia
’
is
f
r
e
que
nt
ly
m
e
nt
io
ne
d
in
pos
it
iv
e
s
e
nt
im
e
nt
,
th
e
a
tt
r
ib
ut
io
n
c
lo
ud
e
m
pha
s
iz
e
s
a
c
ti
on
-
or
ie
nt
e
d
te
r
m
s
li
ke
‘
s
al
ur
k
an
bant
uan
’
(
di
s
tr
ib
ut
e
a
id
)
,
r
e
ve
a
li
ng
th
e
ir
de
e
pe
r
e
m
ot
io
na
l
im
pa
c
t.
T
hi
s
r
e
s
ul
t
hi
ghl
ig
ht
s
th
e
r
e
f
in
e
d
s
e
nt
im
e
nt
a
na
ly
s
is
c
a
pa
bi
li
ti
e
s
of
th
e
I
ndoB
E
R
T
w
e
e
t
-
G
E
L
U
m
ode
l.
T
he
w
or
d
-
a
tt
r
ib
ut
io
n a
na
ly
s
is
us
in
g t
he
I
ndoB
E
R
T
w
e
e
t
-
G
E
L
U
m
ode
l
s
how
s
a
m
a
r
ke
d i
m
pr
ove
m
e
nt
in
th
e
a
c
c
ur
a
c
y
a
nd
de
pt
h
of
s
e
nt
im
e
nt
a
na
ly
s
is
,
e
s
pe
c
ia
ll
y
w
it
hi
n
I
ndone
s
ia
n
T
w
it
te
r
di
s
c
us
s
io
ns
on
bi
odi
ve
r
s
it
y poli
c
y, by ove
r
c
om
in
g t
he
l
im
it
a
ti
ons
of
t
r
a
di
ti
ona
l
f
r
e
que
nc
y
-
ba
s
e
d m
e
th
ods
, w
hi
c
h of
te
n f
a
il
t
o
c
a
pt
ur
e
c
ont
e
xt
ua
l
de
pt
h.
U
nl
ik
e
th
e
s
e
c
onve
nt
io
na
l
m
e
th
od
s
,
I
ndoB
E
R
T
w
e
e
t
-
G
E
L
U
e
f
f
e
c
ti
ve
ly
gr
a
s
ps
th
e
c
ont
e
xt
ua
l
r
e
le
va
nc
e
of
w
or
ds
in
I
ndone
s
ia
n’
s
uni
que
li
ngui
s
ti
c
a
nd
c
ul
tu
r
a
l
nua
nc
e
s
,
id
e
nt
if
yi
ng
f
r
e
que
nt
l
y
us
e
d
te
r
m
s
w
hi
le
a
c
c
ur
a
te
ly
a
tt
r
ib
ut
in
g
s
e
nt
im
e
nt
,
th
us
unc
ove
r
in
g
s
ubt
le
f
a
c
to
r
s
li
ke
s
pe
c
if
ic
le
xi
c
a
l
c
hoi
c
e
s
th
a
t
s
ha
pe
e
m
ot
io
na
l
r
e
s
pons
e
s
.
T
he
s
e
f
in
di
ngs
a
r
e
c
r
uc
ia
l
f
or
f
i
e
ld
s
r
e
qui
r
in
g
pr
e
c
is
e
s
e
nt
im
e
nt
in
te
r
pr
e
ta
ti
on,
s
uc
h a
s
publi
c
opi
ni
on on e
n
vi
r
onm
e
nt
a
l
pol
ic
ie
s
, a
s
t
he
y e
n
a
bl
e
pol
ic
ym
a
ke
r
s
t
o m
a
ke
de
c
is
io
ns
gr
ounde
d i
n
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
, N
o.
3
,
J
une
20
25
:
2389
-
2401
2398
m
or
e
a
c
c
ur
a
te
,
a
c
ti
ona
bl
e
in
s
ig
ht
s
.
T
hi
s
s
tu
dy
e
m
pha
s
i
z
e
s
th
e
e
s
s
e
nt
ia
l
r
ol
e
of
a
dva
n
c
e
d
m
ode
l
s
li
ke
I
ndoB
E
R
T
w
e
e
t
-
G
E
L
U
in
de
li
ve
r
in
g
hi
gh
-
qua
li
ty
a
na
ly
ti
c
a
l
r
e
s
ul
ts
a
nd
s
tr
e
s
s
e
s
th
e
ne
e
d
f
or
c
ont
in
uous
m
ode
l
r
e
f
in
e
m
e
nt
t
o m
e
e
t
th
e
c
om
pl
e
xi
ti
e
s
of
s
e
nt
im
e
nt
a
na
ly
s
is
a
c
r
o
s
s
di
ve
r
s
e
l
in
gui
s
ti
c
s
e
tt
in
gs
.
(
a
)
(
b)
(
c
)
F
ig
ur
e
6
.
W
or
d f
r
e
que
nc
y c
lo
uds
f
or
e
a
c
h s
e
nt
im
e
nt
c
la
s
s
:
(
a
)
ne
ga
ti
ve
, (
b)
ne
ut
r
a
l,
(
c
)
pos
it
iv
e
(
a
)
(
b)
(
c
)
F
ig
ur
e
7
. W
or
d A
tt
r
ib
ut
io
n
c
lo
uds
f
or
e
a
c
h s
e
nt
im
e
nt
c
la
s
s
:
(
a
)
ne
ga
ti
ve
, (
b)
ne
ut
r
a
l,
(
c
)
pos
it
iv
e
T
a
bl
e
2
. C
om
pa
r
is
on
of
t
op w
or
ds
by f
r
e
que
nc
y
a
nd a
tt
r
ib
ut
io
n
S
e
nt
i
m
e
nt
T
op
w
or
ds
(
f
r
e
que
nc
y
)
T
op
w
or
ds
(
a
t
t
r
i
but
i
on
)
N
e
ga
t
i
ve
'
k
e
bak
ar
an
hut
an'
(
f
or
e
s
t
f
i
r
e
s
)
,
'
i
m
por
be
r
as
'
(
r
i
c
e
i
m
por
t
s
)
, '
s
t
unt
i
ng'
, '
pe
t
ani
'
(
f
a
r
m
e
r
s
)
'
pos
i
t
i
f
c
ov
i
d'
(
C
O
V
I
D
-
pos
i
t
i
ve
)
,
'
k
e
bi
j
ak
an
i
m
por
'
(
i
m
por
t
pol
i
c
y)
,
'
k
e
na
s
t
unt
i
ng'
(
a
f
f
e
c
t
e
d
by
s
t
unt
i
ng)
,
'
r
ak
y
at
k
e
c
i
l
'
(
c
om
m
on pe
opl
e
)
N
e
ut
r
a
l
'
m
obi
l
l
i
s
t
r
i
k
'
(
e
l
e
c
t
r
i
c
c
a
r
s
)
,
'
s
t
unt
i
ng'
,
'
m
i
ny
ak
gor
e
ng'
(
c
ooki
ng oi
l
)
, '
I
ndone
s
i
a'
'
hut
an
m
angr
ov
e
'
(
m
a
ngr
ove
f
or
e
s
t
)
,
'
m
at
e
r
i
'
(
m
a
t
e
r
i
a
l
)
,
'
obat
he
r
bal
'
(
he
r
ba
l
m
e
di
c
i
ne
)
, '
be
r
al
i
h'
(
s
w
i
t
c
h)
P
os
i
t
i
ve
'
i
ndone
s
i
a'
,
'
k
e
ndar
aan
l
i
s
t
r
i
k
'
(
e
l
e
c
t
r
i
c
ve
hi
c
l
e
s
)
,
'
m
obi
l
l
i
s
t
r
i
k
(
e
l
e
c
t
r
i
c
c
a
r
s
)
'
,
'
angk
a
s
t
unt
i
ng'
(
s
t
unt
i
ng f
i
gur
e
s
)
'
s
al
ur
k
an
bant
uan'
(
di
s
t
r
i
but
e
a
i
d)
,
'
j
aga
k
e
t
ahanan'
(
m
a
i
nt
a
i
n
r
e
s
i
l
i
e
nc
e
)
,
'
m
ar
i
be
r
s
a
m
a'
(
l
e
t
'
s
be
t
oge
t
he
r
)
,
'
be
r
s
am
a c
e
gah'
(
t
oge
t
he
r
pr
e
ve
nt
)
4.
C
O
N
C
L
U
S
I
O
N
T
hi
s
s
tu
dy
c
onf
ir
m
s
th
a
t
I
ndoB
E
R
T
w
e
e
t
w
it
h
th
e
G
E
L
U
a
c
ti
va
ti
on
e
nha
nc
e
s
bot
h
a
c
c
ur
a
c
y
a
nd
in
te
r
pr
e
ta
bi
li
ty
i
n s
e
nt
im
e
nt
a
na
ly
s
is
f
or
I
ndone
s
ia
n bi
odi
ve
r
s
it
y poli
c
y di
s
c
our
s
e
, e
s
ta
bl
is
hi
ng i
t
a
s
a
v
a
lu
a
bl
e
to
ol
f
or
publ
ic
s
e
c
to
r
e
va
lu
a
ti
ons
.
A
ppl
ie
d
to
13,4
35
twe
e
ts
,
I
nd
oB
E
R
T
w
e
e
t
-
G
E
L
U
c
ons
is
te
nt
ly
out
pe
r
f
or
m
e
d
tr
a
di
ti
ona
l
c
la
s
s
if
ie
r
s
, w
it
h a
m
e
a
n a
c
c
ur
a
c
y of
78.99%
, i
m
pr
ovi
ng by 7.55%
ove
r
t
he
be
s
t
T
F
-
I
D
F
m
ode
l
a
n
d
by
2.83%
ove
r
ot
he
r
B
E
R
T
-
ba
s
e
d
m
ode
ls
,
w
hi
le
W
or
d
-
A
tt
r
ib
ut
io
n
A
na
ly
s
is
de
m
on
s
tr
a
te
d
I
ndoB
E
R
T
w
e
et
-
G
E
L
U
’
s
c
a
pa
c
it
y
f
or
c
ont
e
xt
ua
l
r
e
le
va
nc
e
,
pr
ovi
di
ng
nua
nc
e
d s
e
nt
im
e
nt
in
s
ig
ht
s
w
it
hi
n
I
ndone
s
ia
n
T
w
it
te
r
’
s
uni
que
li
ngui
s
ti
c
c
ont
e
xt
.
H
ow
e
ve
r
,
pot
e
nt
ia
l
bi
a
s
e
s
r
e
qui
r
e
a
tt
e
nt
io
n,
a
s
I
ndoB
E
R
T
w
e
e
t
r
e
f
le
c
ts
de
m
ogr
a
phi
c
Evaluation Warning : The document was created with Spire.PDF for Python.