I
nte
rna
t
io
na
l J
o
urna
l o
f
E
lect
rica
l a
nd
Co
m
pu
t
er
E
ng
ineering
(
I
J
E
CE
)
Vo
l.
15
,
No
.
5
,
Octo
b
er
20
25
,
p
p
.
4
8
2
9
~
4
8
3
6
I
SS
N:
2088
-
8
7
0
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijece.
v
15
i
5
.
pp
4
8
2
9
-
4
8
3
6
4829
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ec
e.
ia
esco
r
e.
co
m
La
ng
ua
g
e mo
del optimiza
tion for
menta
l healt
h que
stio
n
a
nswering
appli
c
a
tion
F
a
rda
n Z
a
m
a
k
hs
y
a
ri
1
,2
,
Ag
u
ng
F
a
t
wa
nto
2
1
I
n
f
o
r
mat
i
o
n
T
e
c
h
n
o
l
o
g
y
,
S
e
k
o
l
a
h
T
i
n
g
g
i
Te
k
n
o
l
o
g
i
C
a
h
a
y
a
S
u
r
y
a
,
K
e
d
i
r
i
,
I
n
d
o
n
e
s
i
a
2
I
n
f
o
r
mat
i
c
s De
p
a
r
t
me
n
t
,
F
a
c
u
l
t
y
o
f
S
c
i
e
n
c
e
a
n
d
T
e
c
h
n
o
l
o
g
y
,
U
n
i
v
e
r
s
i
t
a
s I
sl
a
m N
e
g
e
r
i
S
u
n
a
n
K
a
l
i
j
a
g
a
,
Y
o
g
y
a
k
a
r
t
a
,
I
n
d
o
n
e
s
i
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Au
g
1
8
,
2
0
2
4
R
ev
is
ed
J
u
n
5
,
2
0
2
5
Acc
ep
ted
J
u
n
3
0
,
2
0
2
5
Q
u
e
s
t
io
n
a
n
s
w
e
ri
n
g
(
Q
A)
i
s
a
ta
s
k
in
n
a
tu
r
a
l
l
a
n
g
u
a
g
e
p
ro
c
e
s
s
in
g
(
N
LP
)
w
h
e
r
e
th
e
b
id
i
r
e
c
t
io
n
a
l
e
n
c
o
d
e
r
r
e
p
r
e
s
e
n
t
a
t
io
n
s
f
ro
m
t
r
a
n
s
f
o
r
m
e
r
s
(
B
E
R
T
)
l
a
n
g
u
a
g
e
m
o
d
e
l
h
a
s
sh
o
wn
r
e
m
a
r
k
a
b
l
e
r
e
su
lt
s
.
T
h
i
s
r
e
s
e
a
r
c
h
f
o
c
u
s
e
s
o
n
o
p
ti
m
i
z
in
g
t
h
e
I
n
d
o
B
E
R
T
a
n
d
M
B
E
RT
m
o
d
e
l
s
f
o
r
t
h
e
Q
A
t
a
s
k
in
t
h
e
m
e
n
t
a
l
h
e
a
l
th
d
o
m
a
i
n
,
u
s
in
g
a
t
r
a
n
s
l
a
t
e
d
v
e
r
sio
n
o
f
th
e
Am
o
d
/me
n
tal_
h
e
a
lt
h
_
c
o
u
n
se
li
n
g
_
c
o
n
v
e
rsa
ti
o
n
s
d
a
tas
e
t
o
n
H
u
g
g
in
g
F
a
c
e
.
Th
e
o
p
t
imiz
a
ti
o
n
p
ro
c
e
ss
in
v
o
lv
e
s
fin
e
-
tu
n
i
n
g
I
n
d
o
BERT
a
n
d
M
BERT
t
o
e
n
h
a
n
c
e
th
e
ir
p
e
rfo
rm
a
n
c
e
,
e
v
a
lu
a
ted
u
si
n
g
BERT
S
c
o
re
c
o
m
p
o
n
e
n
ts:
F
1
,
re
c
a
ll
,
a
n
d
p
re
c
isio
n
.
T
h
e
re
su
lt
s i
n
d
ica
te t
h
a
t
f
in
e
-
tu
n
i
n
g
si
g
n
if
ica
n
tl
y
b
o
o
sts
In
d
o
BERT
’s p
e
rfo
rm
a
n
c
e
,
a
c
h
iev
in
g
a
n
F
1
-
BERT
S
c
o
re
o
f
9
1
.
8
%
,
a
re
c
a
ll
o
f
8
9
.
9
%
,
a
n
d
p
re
c
isio
n
o
f
9
3
.
9
%
,
m
a
rk
in
g
a
2
8
%
im
p
ro
v
e
m
e
n
t.
F
o
r
t
h
e
m
o
d
e
l,
M
-
BERT
’s
fin
e
-
tu
n
i
n
g
re
su
lt
s
i
n
c
lu
d
e
a
n
F
1
-
BERT
S
c
o
re
o
f
7
9
.
2
%
,
re
c
a
ll
o
f
7
3
.
4
%
,
a
n
d
p
re
c
isio
n
o
f
8
6
.
2
%
,
with
o
n
l
y
a
5
%
imp
ro
v
e
m
e
n
t.
Th
e
se
fin
d
i
n
g
s
u
n
d
e
rsc
o
re
th
e
imp
o
rt
a
n
c
e
o
f
fin
e
-
tu
n
in
g
a
n
d
u
sin
g
lan
g
u
a
g
e
-
sp
e
c
ifi
c
m
o
d
e
ls
li
k
e
I
n
d
o
BERT
f
o
r
sp
e
c
ialize
d
NL
P
tas
k
s,
d
e
m
o
n
s
tratin
g
th
e
p
o
ten
ti
a
l
to
c
re
a
te
m
o
re
a
c
c
u
ra
te
a
n
d
c
o
n
tex
tu
a
ll
y
re
lev
a
n
t
q
u
e
stio
n
-
a
n
sw
e
rin
g
sy
ste
m
s in
th
e
m
e
n
tal
h
e
a
lt
h
d
o
m
a
in
.
K
ey
w
o
r
d
s
:
B
id
ir
ec
tio
n
al
en
co
d
er
r
ep
r
esen
tatio
n
s
f
r
o
m
tr
an
s
f
o
r
m
er
s
I
n
d
o
B
E
R
T
MBERT
Natu
r
al
lan
g
u
ag
e
p
r
o
ce
s
s
in
g
Qu
esti
o
n
a
n
s
wer
T
h
is i
s
a
n
o
p
e
n
a
c
c
e
ss
a
rticle
u
n
d
e
r th
e
CC B
Y
-
SA
li
c
e
n
se
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
Ag
u
n
g
Fatwan
to
I
n
f
o
r
m
atics De
p
ar
tm
en
t,
Facu
lty
o
f
Scien
ce
an
d
T
ec
h
n
o
lo
g
y
,
Un
iv
er
s
itas
I
s
lam
Neg
er
i Su
n
an
Kalijag
a
J
l.
Ma
r
s
d
a
Ad
is
u
cip
to
,
Yo
g
y
a
k
ar
ta
5
5
2
8
1
,
Yo
g
y
a
k
ar
ta
E
m
ail: a
g
u
n
g
.
f
atwa
n
to
@
u
in
-
s
u
k
a.
ac
.
id
1.
I
NT
RO
D
UCT
I
O
N
R
e
c
e
n
t
im
p
r
o
v
em
en
ts
to
a
f
am
ily
o
f
n
eu
r
al
m
o
d
els,
k
n
o
w
n
as
b
id
ir
ec
tio
n
al
en
co
d
er
r
ep
r
esen
tatio
n
s
f
r
o
m
tr
an
s
f
o
r
m
er
s
(
B
E
R
T
)
,
h
ea
v
ily
r
ely
o
n
p
r
e
-
tr
ai
n
in
g
an
d
h
av
e
s
h
o
w
n
p
r
o
m
is
e
in
a
r
an
g
e
o
f
n
at
u
r
al
lan
g
u
ag
e
p
r
o
ce
s
s
in
g
(
NL
P)
ta
s
k
s
,
wh
ich
in
clu
d
e
tex
t
ca
teg
o
r
izatio
n
a
n
d
d
ataset
q
u
esti
o
n
-
an
s
wer
in
g
[
1
]
.
T
o
lear
n
wo
r
d
r
ep
r
esen
tatio
n
s
,
B
E
R
T
u
s
es
th
e
t
r
an
s
f
o
r
m
er
ar
ch
itectu
r
e,
wh
ich
c
o
n
s
is
ts
o
f
s
ev
er
al
en
co
d
e
r
lay
er
s
.
B
ec
au
s
e
it
p
r
o
ce
s
s
es
s
eq
u
en
tial
in
p
u
t,
lik
e
wo
r
d
s
in
tex
t,
u
s
in
g
an
atten
tio
n
m
ec
h
an
is
m
,
th
e
T
r
an
s
f
o
r
m
er
is
r
eg
ar
d
e
d
as
r
em
ar
k
ab
le
with
in
th
e
d
is
cip
lin
e
o
f
n
atu
r
al
lan
g
u
ag
e
p
r
o
ce
s
s
in
g
.
A
d
ee
p
er
c
o
m
p
r
eh
en
s
io
n
o
f
th
e
tex
t'
s
co
n
tex
t
is
m
ad
e
p
o
s
s
ib
le
b
y
th
e
atten
tio
n
m
ec
h
an
is
m
,
wh
ich
e
n
ab
les
th
e
m
o
d
el
to
a
cc
u
r
ately
r
e
p
r
esen
t
th
e
r
elatio
n
s
h
ip
b
etwe
en
wo
r
d
s
th
at
ar
e
f
ar
ap
a
r
t
[
2
]
.
T
h
e
B
E
R
T
f
r
am
ewo
r
k
c
o
n
s
is
ted
o
f
two
s
tag
es,
p
r
e
-
tr
ain
in
g
an
d
f
in
e
-
tu
n
i
n
g
.
Pre
-
t
r
ain
in
g
in
v
o
lv
es
u
s
in
g
s
ev
er
al
p
r
e
-
tr
ain
in
g
tas
k
s
to
tr
ain
th
e
m
o
d
el
o
n
u
n
la
b
eled
d
ata
[
3
]
.
Usi
n
g
lab
elled
d
ata
f
r
o
m
ea
r
lier
task
s
,
th
e
f
ir
s
t
s
tag
e
to
b
e
d
o
n
e
in
th
e
B
E
R
T
m
o
d
el
is
t
h
e
p
r
e
-
tr
ain
i
n
g
p
ar
a
m
eter
s
,
th
en
th
e
p
r
o
ce
s
s
is
f
o
llo
wed
b
y
f
i
n
e
-
tu
n
in
g
all
o
f
th
e
p
ar
am
eter
s
.
Desp
ite
b
ein
g
s
tar
ted
with
th
e
s
am
e
p
r
e
-
tr
ain
in
g
p
ar
am
eter
s
,
ev
er
y
task
h
as
a
u
n
iq
u
e
f
in
e
-
t
u
n
ed
m
o
d
el
[
4
]
.
T
h
is
s
tu
d
y
m
ak
es
u
s
e
o
f
two
B
E
R
T
m
o
d
el
v
er
s
io
n
s
th
at
wer
e
cr
ea
ted
esp
ec
ially
f
o
r
I
n
d
o
n
e
s
ian
an
d
m
u
ltil
in
g
u
al
lan
g
u
a
g
es:
th
e
B
E
R
T
m
u
ltil
in
g
u
al
b
ase
m
o
d
el
(
ca
s
ed
)
(
g
o
o
g
l
e
-
b
e
r
t
/
b
e
r
t
-
b
a
s
e
-
m
u
l
ti
l
i
n
g
u
a
l
-
c
a
s
e
d
)
a
n
d
t
h
e
I
n
d
o
B
E
R
T
b
a
s
e
u
n
c
a
s
e
d
(
R
i
f
k
y
/
I
n
d
o
b
er
t
-
Q
A
)
.
R
es
e
a
r
c
h
e
r
s
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
5
,
Octo
b
e
r
20
25
:
4
8
2
9
-
4
8
3
6
4830
u
s
e
d
t
w
o
p
r
e
-
e
x
i
s
t
i
n
g
B
E
R
T
m
o
d
e
l
s
m
o
n
o
l
i
n
g
u
a
l
B
E
R
T
f
o
r
I
n
d
o
n
e
s
i
a
n
(
c
a
l
l
e
d
“
I
n
d
o
B
E
R
T
”
)
an
d
m
u
ltil
in
g
u
al
B
E
R
T
(
“
m
B
E
R
T
”
)
.
MBE
R
T
is
tr
ain
ed
b
y
co
m
b
i
n
in
g
1
0
4
lan
g
u
ag
es
o
f
W
ik
ip
ed
ia
d
o
cu
m
en
ts
,
wh
ich
also
in
clu
d
e
I
n
d
o
n
esian
,
an
d
is
ef
f
e
ctiv
e
f
o
r
m
u
ltil
in
g
u
al
task
s
th
a
t d
o
n
o
t
r
eq
u
i
r
e
m
u
c
h
ex
p
la
n
atio
n
[
5
]
.
I
n
d
o
B
E
R
T
is
a
B
E
R
T
u
s
in
g
tr
an
s
f
o
r
m
er
-
b
ased
m
o
d
el
[
4
]
,
tr
ain
ed
with
th
e
Hu
g
g
in
g
f
ac
e
f
r
am
ewo
r
k
with
th
e
d
ef
au
lt
B
E
R
T
b
ase
(
u
n
ca
s
ed
)
c
o
n
f
ig
u
r
atio
n
[
6
]
.
I
n
c
o
n
tr
ast,
an
o
th
er
v
ar
ia
n
t
o
f
B
E
R
T
is
m
u
ltil
in
g
u
al
-
B
E
R
T
(
m
B
E
R
T
)
th
at
u
s
es
m
ask
ed
lan
g
u
ag
e
m
o
d
ellin
g
(
M
L
M)
to
tr
ain
it
o
n
th
e
la
r
g
est
W
ik
ip
ed
ia
d
ataset
ac
r
o
s
s
1
0
4
lan
g
u
ag
es,
in
clu
d
i
n
g
I
n
d
o
n
esian
.
Ov
er
s
am
p
lin
g
is
u
s
ed
f
o
r
s
m
all
lan
g
u
ag
es,
a
n
d
u
n
d
e
r
s
am
p
lin
g
is
u
s
ed
f
o
r
lar
g
e
lan
g
u
a
g
es
wh
en
d
ea
lin
g
with
u
n
b
ala
n
ce
d
d
at
a.
T
h
is
m
ak
es
it
p
o
s
s
ib
le
to
e
m
p
lo
y
th
e
m
B
E
R
T
m
o
d
el
o
n
lan
g
u
ag
e
d
atasets
with
lim
ited
r
eso
u
r
ce
s
,
s
u
ch
as
I
n
d
o
n
esian
[
7
]
.
T
h
e
B
E
R
T
co
n
f
ig
u
r
atio
n
,
1
2
atten
tio
n
h
ea
d
s
,
1
2
h
id
d
e
n
la
y
er
s
o
f
7
6
8
ea
ch
,
a
n
d
f
ee
d
-
f
o
r
war
d
h
id
d
en
la
y
er
s
o
f
3
,
0
7
2
ar
e
s
h
ar
ed
b
y
t
h
ese
two
m
o
d
els
[
8
]
.
W
e
wil
l u
s
e
th
e
d
a
tase
ts
A
m
o
d
/
m
e
n
ta
l_
h
ea
l
th
_
co
u
n
s
el
in
g
_
c
o
n
v
er
s
ati
o
n
s
t
a
k
e
n
f
r
o
m
t
h
e
H
u
g
g
in
g
f
a
ce
web
s
ite
,
t
o
e
v
a
lu
at
e
t
h
e
tw
o
d
i
f
f
er
e
n
t
m
o
d
e
ls
.
Qu
esti
o
n
s
a
n
d
r
esp
o
n
s
es
f
r
o
m
two
o
n
lin
e
co
u
n
s
ellin
g
an
d
th
er
ap
y
p
latf
o
r
m
s
ar
e
in
cl
u
d
e
d
in
th
is
d
ataset.
Qu
alif
ied
p
s
y
ch
o
lo
g
is
ts
h
av
e
an
s
wer
ed
th
e
q
u
esti
o
n
s
,
wh
ich
s
p
an
a
wid
e
r
an
g
e
o
f
m
en
tal
h
ea
lth
co
n
ce
r
n
s
.
T
h
e
aim
o
f
th
is
d
ataset
i
s
to
en
h
an
ce
th
e
q
u
ality
o
f
lan
g
u
ag
e
m
o
d
els.
T
h
e
d
ataset
is
s
till
i
n
E
n
g
lis
h
,
s
o
t
h
e
r
esear
ch
e
r
m
an
u
ally
tr
a
n
s
lated
it
in
to
I
n
d
o
n
esian
f
o
r
m
o
r
e
r
elev
an
t
test
in
g
.
T
h
e
e
v
alu
atio
n
o
f
th
e
d
ataset
will
u
s
e
B
E
R
T
-
Sco
r
e.
T
h
is
ev
alu
atio
n
m
etr
ic
is
v
er
y
u
s
ef
u
l
f
o
r
ev
alu
atin
g
th
e
p
er
f
o
r
m
an
ce
o
f
q
u
esti
o
n
-
an
s
wer
[
9
]
.
T
o
p
ics
r
elate
d
to
q
u
esti
o
n
-
a
n
s
wer
s
y
s
tem
s
with
I
n
d
o
n
esian
s
h
av
e
b
ee
n
r
esear
c
h
ed
b
ef
o
r
e,
s
u
ch
as
r
esear
ch
co
n
d
u
cted
by
Dza
k
y
et
a
l.
[
1
0
]
,
in
h
is
r
esear
ch
cr
ea
tin
g
a
ch
atb
o
t
f
o
r
m
e
n
tal
h
ea
lth
u
s
in
g
a
d
ee
p
n
eu
r
al
n
etwo
r
k
an
d
B
E
R
T
m
o
d
els
s
p
ec
if
ically
f
o
r
I
n
d
o
n
esia
n
s
.
Af
ter
test
in
g
,
th
e
r
esu
lts
o
f
th
is
s
tu
d
y
s
h
o
wed
an
ac
cu
r
ac
y
v
alu
e
o
f
7
1
.
7
3
%
[
1
0
]
.
I
n
ad
d
itio
n
,
in
r
esear
ch
co
n
d
u
cted
b
y
Hu
za
e
n
i
et
a
l.
[
1
1
]
wh
er
e
h
is
r
esear
ch
aim
ed
to
cr
ea
te
a
ch
a
tb
o
t
s
y
s
tem
to
d
iag
n
o
s
e
m
en
tal
h
ea
lth
s
y
m
p
to
m
s
u
s
in
g
th
e
B
E
R
T
m
o
d
el,
b
ased
o
n
th
e
r
esu
lts
o
f
test
s
th
at
h
a
v
e
b
ee
n
ca
r
r
ied
o
u
t,
th
e
ac
c
u
r
ac
y
v
alu
e
is
7
8
.
6
7
%
[
1
1
]
.
T
h
i
s
r
esear
ch
ass
ess
es
th
e
r
esu
lts
o
f
p
r
ev
io
u
s
s
tu
d
ies
an
d
f
o
cu
s
es
o
n
im
p
r
o
v
in
g
p
er
f
o
r
m
an
ce
f
o
r
m
e
n
tal
h
ea
lth
q
u
esti
o
n
-
an
s
wer
s
y
s
tem
s
b
y
ad
o
p
tin
g
I
n
d
o
n
esi
an
d
atasets
.
T
h
is
ex
p
er
im
e
n
tal
r
esear
ch
aim
s
to
d
em
o
n
s
tr
ate
th
e
ef
f
icien
cy
in
u
tili
zin
g
t
h
e
M
-
B
E
R
T
an
d
I
n
d
o
B
E
R
T
m
o
d
els
o
n
a
n
AI
ch
atb
o
t
f
o
cu
s
in
g
o
n
d
i
g
ital
m
en
tal
h
ea
lth
.
B
y
co
n
d
u
ctin
g
th
is
r
esear
ch
,
it
is
ex
p
ec
ted
t
o
f
i
n
d
o
u
t
th
e
r
elev
a
n
ce
o
f
B
E
R
T
m
o
d
els
t
h
at
ar
e
s
u
itab
le,
ef
f
icien
t,
an
d
s
ca
lab
le
f
o
r
u
s
e
i
n
c
h
atb
o
t
m
o
d
els.
Oth
er
co
n
tr
ib
u
tio
n
o
f
th
is
r
esear
ch
is
:
a.
O
p
t
i
m
i
z
e
d
I
n
d
o
B
E
R
T
a
n
d
M
B
E
R
T
m
o
d
e
l
f
o
r
Q
A
t
a
s
k
o
f
t
h
e
m
e
n
t
a
l
h
e
a
lt
h
d
o
m
a
i
n
i
n
I
n
d
o
n
e
s
i
a
n
la
n
g
u
a
g
e
,
b.
T
h
e
tr
an
s
latio
n
o
f
th
e
Am
o
d
/
m
en
tal_
h
ea
lth
_
c
o
u
n
s
elin
g
_
co
n
v
er
s
atio
n
s
d
ataset
to
I
n
d
o
n
es
ian
lan
g
u
a
g
e,
c.
Pro
v
id
e
a
n
o
v
er
v
iew
o
f
th
e
m
en
tal
h
ea
lth
q
u
esti
o
n
-
an
s
wer
s
y
s
tem
th
at
ca
n
b
e
a
p
p
lied
ac
co
r
d
in
g
to
th
e
in
ten
d
ed
s
p
ec
if
icatio
n
s
.
T
h
is
ty
p
e
o
f
r
esear
ch
is
p
o
t
en
tially
r
ep
ea
tab
le
to
b
e
co
n
d
u
cted
f
o
r
o
t
h
er
k
i
n
d
s
o
f
f
o
u
n
d
atio
n
al
m
o
d
els’
d
ataset.
T
h
e
r
esu
lt
o
f
th
at
k
in
d
o
f
s
tu
d
y
is
p
o
s
s
ib
le
to
b
e
a
p
p
lied
to
m
e
n
ta
l
h
ea
lth
q
u
esti
o
n
s
an
s
wer
in
g
s
y
s
tem
o
f
d
if
f
er
en
t la
n
g
u
ag
es.
I
t
ca
n
b
e
p
o
ten
tiall
y
im
p
lem
en
te
d
,
esp
ec
ially
in
c
o
u
n
tr
ies
with
lar
g
e
p
o
p
u
latio
n
s
wh
er
e
s
o
m
etim
es
s
u
f
f
er
f
r
o
m
in
s
u
f
f
icien
t m
en
ta
l h
ea
lth
p
r
o
f
ess
io
n
als
.
2.
M
E
T
H
O
D
T
h
is
s
tu
d
y
em
p
l
o
y
s
an
ex
p
la
n
ato
r
y
r
esear
ch
d
esig
n
with
a
q
u
an
titativ
e
a
p
p
r
o
ac
h
.
T
h
e
e
x
p
lan
ato
r
y
tech
n
iq
u
e
was
ch
o
s
en
s
in
ce
th
e
s
tu
d
y
'
s
g
o
al
i
s
to
d
escr
i
b
e
an
d
ass
ess
th
e
ca
u
s
e
-
an
d
-
ef
f
ec
t
r
elatio
n
s
h
ip
b
etwe
en
f
in
e
-
tu
n
in
g
an
d
p
er
f
o
r
m
an
ce
im
p
r
o
v
em
e
n
t
in
th
e
I
n
d
o
B
E
R
T
an
d
MBERT
lan
g
u
ag
e
m
o
d
els
[
1
2
]
.
T
h
e
q
u
an
titativ
e
ap
p
r
o
ac
h
was
u
tili
ze
d
b
ec
au
s
e
th
e
d
ata
ac
q
u
ir
ed
wer
e
n
u
m
e
r
ical,
in
d
icatin
g
th
e
lev
el
o
f
m
o
d
el
p
er
f
o
r
m
an
c
e
b
ef
o
r
e
an
d
af
ter
o
p
tim
izatio
n
.
T
h
e
d
ata
was
co
llected
u
s
in
g
ex
p
er
im
en
tal
tech
n
iq
u
e.
E
x
p
er
im
en
ts
wer
e
ca
r
r
ie
d
o
u
t
b
y
co
n
d
u
ctin
g
f
in
e
-
tu
n
i
n
g
t
o
t
h
e
I
n
d
o
B
E
R
T
an
d
MBERT
lan
g
u
ag
e
m
o
d
els
an
d
th
en
co
m
p
ar
in
g
t
h
eir
p
er
f
o
r
m
a
n
ce
b
ef
o
r
e
an
d
af
ter
th
is
o
p
tim
izatio
n
p
r
o
ce
s
s
[
1
3
]
.
2
.
1
.
I
ns
t
rum
ent
2
.
1
.
1
.
So
f
t
wa
re
a
nd
h
a
rdwa
re
T
h
is
r
esear
ch
u
tili
ze
d
Go
o
g
le
C
o
lab
as
th
e
m
ain
d
ev
elo
p
m
e
n
t
en
v
ir
o
n
m
en
t,
s
u
p
p
o
r
ted
b
y
a
T
esla
T
4
GPU.
T
h
e
T
4
GPU,
b
ased
o
n
NVI
DI
A'
s
T
u
r
in
g
ar
ch
itectu
r
e,
o
f
f
e
r
s
ef
f
icien
t
p
er
f
o
r
m
an
c
e
f
o
r
d
ee
p
lear
n
in
g
task
s
with
1
6
GB
GDD
R
6
m
em
o
r
y
an
d
u
p
to
8
.
1
T
FLOPS
o
f
co
m
p
u
ter
p
o
wer
.
T
h
is
s
etu
p
en
ab
les
f
aster
m
o
d
el
tr
ain
in
g
a
n
d
in
f
er
en
ce
,
esp
ec
ially
f
o
r
r
eso
u
r
ce
-
i
n
ten
s
iv
e
task
s
in
n
atu
r
al
lan
g
u
a
g
e
p
r
o
ce
s
s
in
g
.
Fo
r
o
f
f
lin
e
p
r
ep
ar
atio
n
,
we
u
s
ed
a
lap
to
p
with
an
I
n
tel
C
o
r
e
i5
p
r
o
ce
s
s
o
r
,
8
GB
R
AM
,
an
d
a
n
NVI
DI
A
GeFo
r
ce
MX
2
3
0
GPU.
Alth
o
u
g
h
l
o
ca
l h
ar
d
war
e
was
u
s
ed
f
o
r
in
itial
test
in
g
an
d
co
d
e
d
ev
elo
p
m
en
t,
all
f
in
e
-
tu
n
in
g
an
d
ev
alu
atio
n
wer
e
c
o
n
d
u
cted
in
th
e
C
o
lab
en
v
ir
o
n
m
en
t
t
o
lev
e
r
ag
e
clo
u
d
-
b
ased
ac
ce
ler
atio
n
.
T
h
is
co
m
b
in
atio
n
allo
wed
s
ea
m
less
in
teg
r
atio
n
b
etwe
en
co
d
e
p
r
o
to
t
y
p
in
g
an
d
s
ca
lab
le
m
o
d
el
tr
ain
in
g
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
La
n
g
u
a
g
e
m
o
d
el
o
p
timiz
a
tio
n
fo
r
men
ta
l h
ea
lth
q
u
esti
o
n
a
n
s
w
erin
g
…
(
F
a
r
d
a
n
Za
ma
kh
s
ya
r
i
)
4831
2
.
2
.
M
a
t
er
ia
l
2
.
2
.
1
.
F
o
un
da
t
io
na
l
m
o
del
I
n
c
o
n
d
u
c
t
i
n
g
t
h
is
r
es
e
a
r
c
h
,
w
e
u
s
e
d
t
w
o
l
a
n
g
u
a
g
e
m
o
d
e
ls
d
e
v
e
l
o
p
e
d
b
as
e
d
o
n
t
h
e
B
E
R
T
m
o
d
el,
n
a
m
e
l
y
I
n
d
o
B
E
R
T
a
n
d
MB
E
R
T
.
I
n
d
o
B
E
R
T
is
a
t
r
a
n
s
f
o
r
m
e
r
-
b
a
s
e
d
m
o
d
e
l
i
n
t
h
e
B
E
R
T
s
t
y
l
e
,
I
t
w
a
s
t
r
a
i
n
e
d
o
n
l
y
a
s
a
m
a
s
k
e
d
l
a
n
g
u
a
g
e
m
o
d
e
l
u
s
i
n
g
t
h
e
H
u
g
g
i
n
g
f
a
c
e
f
r
a
m
e
w
o
r
k
,
a
d
h
e
r
i
n
g
t
o
t
h
e
s
t
a
n
d
a
r
d
B
E
R
T
-
B
a
s
e
c
o
n
f
i
g
u
r
a
t
i
o
n
(
u
n
c
a
s
e
d
)
[
1
4
]
.
T
h
e
m
o
d
e
l
c
o
n
s
i
s
t
s
o
f
1
2
f
e
ed
-
f
o
r
w
a
r
d
h
i
d
d
e
n
l
a
y
e
r
s
w
i
t
h
3
,
0
7
2
d
,
1
2
a
t
t
e
n
t
i
o
n
h
e
a
d
s
,
a
n
d
1
2
h
i
d
d
e
n
l
a
y
e
r
s
w
it
h
7
6
8
d
e
a
c
h
.
T
h
e
m
o
d
e
l
s
e
t
t
h
e
t
r
a
i
n
i
n
g
t
o
u
s
e
5
1
2
t
o
k
e
n
s
p
er
b
a
t
c
h
a
n
d
e
x
t
e
n
d
s
t
h
e
H
u
g
g
i
n
g
f
a
c
e
f
r
a
m
e
w
o
r
k
t
o
a
n
a
l
y
z
e
d
i
s
t
i
n
c
t
s
e
ts
o
f
t
e
x
t
f
o
r
d
i
s
t
i
n
ct
d
o
c
u
m
e
n
t
b
l
o
c
k
s
.
A
v
o
c
a
b
u
l
a
r
y
o
f
3
1
,
9
2
3
I
n
d
o
n
e
s
i
a
n
W
o
r
d
P
i
e
c
es
w
a
s
u
s
e
d
t
o
t
r
ai
n
I
n
d
o
B
E
R
T
[
5
]
.
T
h
e
t
o
t
a
l
a
m
o
u
n
t
o
f
w
o
r
d
s
u
s
e
d
t
o
tr
a
i
n
t
h
e
I
n
d
o
B
E
R
T
m
o
d
e
l
w
a
s
o
v
e
r
2
2
0
m
i
l
li
o
n
,
g
a
t
h
e
r
e
d
f
r
o
m
t
h
r
e
e
p
r
e
d
o
m
i
n
a
n
t
s
o
u
r
c
e
s
:
i
)
t
h
e
7
4
m
il
l
i
o
n
w
o
r
d
s
f
o
u
n
d
o
n
t
h
e
I
n
d
o
n
e
s
i
a
n
W
i
k
i
p
e
d
i
a
;
ii
)
n
e
ws
i
t
e
m
s
s
o
u
r
c
e
d
f
r
o
m
K
o
m
p
a
s
,
L
i
p
u
t
a
n
6
,
T
e
m
p
o
(
a
t
o
t
a
l
o
f
5
5
m
i
l
l
i
o
n
w
o
r
d
s
)
;
a
n
d
iii
)
t
h
e
9
0
m
i
l
li
o
n
w
o
r
d
s
f
o
u
n
d
i
n
t
h
e
I
n
d
o
n
e
s
i
a
n
W
e
b
c
o
r
p
u
s
.
U
p
o
n
p
r
e
p
r
o
c
e
s
s
i
n
g
th
e
c
o
r
p
u
s
i
n
t
o
5
1
2
t
o
k
e
n
i
z
e
d
d
o
c
u
m
e
n
t
b
l
o
c
k
s
,
w
e
w
e
r
e
a
b
l
e
t
o
a
c
q
u
i
r
e
1
3
,
9
8
5
d
e
v
e
l
o
p
m
e
n
t
e
x
a
m
p
l
e
s
a
n
d
1
,
0
6
7
,
5
8
1
t
r
a
i
n
i
n
g
e
x
a
m
p
l
e
s
(
w
it
h
o
u
t
r
e
d
u
n
d
a
n
c
y
)
.
F
o
u
r
N
v
i
d
i
a
V
1
0
0
G
P
U
s
(
1
6
G
B
e
a
c
h
)
,
a
b
a
t
c
h
s
i
z
e
o
f
1
2
8
a
n
d
a
l
e
a
r
n
i
n
g
r
a
t
e
o
f
1
e
-
4
,
a
l
o
n
g
w
it
h
a
n
A
d
a
m
o
p
t
i
m
i
z
e
r
a
n
d
a
l
i
n
e
a
r
s
c
h
e
d
u
l
e
r
,
w
e
r
e
e
m
p
l
o
y
e
d
f
o
r
t
h
e
t
r
a
i
n
i
n
g
.
O
v
e
r
t
h
e
c
o
u
r
s
e
o
f
t
w
o
m
o
n
t
h
s
,
t
h
e
m
o
d
e
l
w
a
s
t
r
a
i
n
e
d
f
o
r
2
.
4
m
i
l
l
i
o
n
s
t
e
p
s
,
o
r
1
8
0
e
p
o
c
h
s
,
a
n
d
t
h
e
d
e
v
e
l
o
p
m
e
n
t
s
e
t'
s
u
l
t
i
m
at
e
a
c
c
u
r
a
c
y
r
a
t
e
w
as
3
.
9
7
,
o
r
c
o
m
p
a
r
a
b
l
e
t
o
B
E
R
T
-
b
as
e
i
n
E
n
g
l
is
h
[
1
5
]
.
An
I
n
d
o
n
esian
-
la
n
g
u
a
g
e
W
ik
ip
ed
ia
co
r
p
u
s
c
o
n
s
is
tin
g
o
f
1
0
4
m
o
n
o
lin
g
u
al
lan
g
u
a
g
es
w
as
u
s
ed
to
p
r
e
-
tr
ain
t
h
e
s
in
g
le
-
lan
g
u
ag
e
m
u
ltil
in
g
u
al
B
E
R
T
(
M
-
B
E
R
T
)
m
o
d
el.
R
esear
ch
er
s
ca
n
u
tili
ze
th
e
m
-
B
E
R
T
m
o
d
el
to
tack
le
task
s
in
d
if
f
er
en
t
lan
g
u
ag
es
b
ec
au
s
e
it
h
as
f
in
is
h
ed
th
e
p
r
e
-
tr
ain
in
g
p
r
o
ce
s
s
in
an
ex
ten
s
iv
e
v
ar
iety
o
f
lan
g
u
ag
es,
ex
p
an
d
i
n
g
its
u
s
ef
u
ln
ess
[
1
6
]
.
T
h
e
b
e
n
ef
it
o
f
m
u
ltil
in
g
u
al
B
E
R
T
s
i
s
th
at
th
ey
f
ac
ilit
ate
m
u
ltil
in
g
u
al
lear
n
in
g
.
T
h
ey
ar
e
ab
le
to
b
e
co
ac
h
ed
o
n
a
s
p
ec
if
ic
task
in
o
n
e
lan
g
u
ag
e
an
d
i
m
p
lem
en
t th
e
s
am
e
task
q
u
ite
well
in
an
o
th
er
la
n
g
u
ag
e,
e
v
en
if
th
e
y
ar
e
o
n
ly
tr
ain
ed
f
o
r
a
m
o
n
o
lin
g
u
al
task
.
T
h
e
m
o
d
el
u
s
es
1
2
T
r
an
s
f
o
r
m
e
r
lay
e
r
s
with
1
2
h
ea
d
s
,
7
6
8
em
b
ed
d
in
g
d
i
m
en
s
io
n
s
,
an
d
3
0
7
2
f
ee
d
-
f
o
r
war
d
h
id
d
en
lay
er
d
im
en
s
io
n
s
,
with
a
d
r
o
p
o
u
t
r
a
te
o
f
0
.
1
an
d
GE
L
U
ac
tiv
atio
n
.
Mu
ltil
in
g
u
al
B
E
R
T
was
o
p
ti
m
ized
u
s
in
g
Ad
am
Op
tim
izer
,
8
8
b
atc
h
s
izes,
an
d
m
ix
ed
-
p
r
ec
is
io
n
tr
ain
in
g
with
1
0
ep
o
c
h
s
.
E
ac
h
m
o
d
el
was
t
r
ain
ed
o
n
NVI
DI
A
R
T
X
T
itan
h
ar
d
war
e
with
2
4
GB
o
f
m
em
o
r
y
f
o
r
ap
p
r
o
x
im
at
ely
2
0
h
o
u
r
s
[
1
7
]
.
2
.
2
.
2
.
Da
t
a
s
et
T
h
i
s
r
e
s
e
a
r
c
h
w
as
c
o
n
d
u
c
t
e
d
u
s
i
n
g
t
h
e
A
m
o
d
/
m
e
n
t
a
l
_
h
e
a
lt
h
_
c
o
u
n
s
e
l
i
n
g
_
c
o
n
v
e
r
s
a
ti
o
n
s
d
a
ta
s
e
t
t
a
k
e
n
f
r
o
m
t
h
e
H
u
g
g
i
n
g
F
a
c
e
w
e
b
s
i
te
.
T
h
i
s
d
a
t
as
e
t
o
r
i
g
i
n
al
l
y
c
o
n
t
ai
n
s
c
o
n
v
e
r
s
a
ti
o
n
s
i
n
E
n
g
l
is
h
b
e
t
w
e
e
n
c
o
u
n
s
el
l
o
r
s
a
n
d
c
l
i
e
n
t
s
d
i
s
c
u
s
s
i
n
g
m
e
n
t
a
l
h
e
a
l
t
h
i
s
s
u
e
s
.
T
o
s
u
i
t
t
h
e
c
o
n
te
x
t
o
f
t
h
i
s
s
t
u
d
y
,
t
h
e
d
a
t
a
s
e
t
w
a
s
t
r
a
n
s
l
a
t
e
d
i
n
t
o
I
n
d
o
n
e
s
i
a
n
.
T
h
e
t
r
a
n
s
l
a
ti
o
n
p
r
o
c
e
s
s
wa
s
d
o
n
e
d
i
r
e
c
t
l
y
b
y
t
h
e
r
e
s
e
a
r
c
h
t
e
a
m
a
n
d
n
o
t
m
e
c
h
a
n
i
c
a
l
l
y
t
o
e
n
s
u
r
e
t
h
e
a
c
c
u
r
a
c
y
a
n
d
t
r
a
n
s
l
a
ti
o
n
q
u
a
l
i
t
y
.
T
h
e
A
m
o
d
/
m
e
n
t
al
_
h
e
a
lt
h
_
c
o
u
n
s
e
l
i
n
g
_
c
o
n
v
e
r
s
a
ti
o
n
s
d
a
t
a
s
et
c
o
n
s
i
s
t
s
o
f
c
o
n
v
e
r
s
a
t
i
o
n
s
c
o
v
e
r
i
n
g
a
r
a
n
g
e
o
f
m
e
n
t
a
l
h
e
a
lt
h
-
r
e
l
a
te
d
t
o
p
i
cs,
s
u
c
h
a
s
t
r
a
u
m
a
,
a
n
x
i
et
y
,
r
e
l
ati
o
n
s
h
i
p
is
s
u
es
,
a
n
d
d
e
p
r
e
s
s
i
o
n
.
E
a
c
h
c
o
n
v
e
r
s
at
i
o
n
c
o
n
s
i
s
ts
o
f
m
u
l
t
i
p
le
e
x
c
h
a
n
g
e
s
b
e
t
w
ee
n
t
h
e
c
o
u
n
s
e
l
l
o
r
a
n
d
t
h
e
c
l
i
e
n
t
,
w
h
i
c
h
a
l
l
o
ws
r
e
s
ea
r
c
h
e
r
s
t
o
a
n
al
y
z
e
th
e
d
y
n
a
m
i
c
s
o
f
t
h
e
c
o
n
v
e
r
s
at
i
o
n
a
n
d
e
f
f
e
c
t
i
v
e
c
o
u
n
s
e
ll
i
n
g
s
t
r
at
e
g
i
e
s
.
T
h
e
u
s
e
o
f
th
ese
d
atasets
in
r
esear
ch
p
r
o
v
id
es
s
ev
er
al
ad
v
an
tag
es.
Firstl
y
,
it
p
r
o
v
i
d
es
a
r
ich
an
d
r
ea
lis
tic
r
ep
r
esen
tatio
n
o
f
m
e
n
tal
h
ea
lth
co
u
n
s
ellin
g
s
itu
atio
n
s
,
allo
win
g
r
esear
ch
er
s
to
e
x
p
lo
r
e
th
e
is
s
u
e
in
a
r
elev
an
t
co
n
tex
t.
Seco
n
d
,
th
e
tr
an
s
latio
n
in
to
I
n
d
o
n
esian
a
llo
ws
th
is
s
tu
d
y
to
c
o
n
tr
ib
u
te
to
th
e
g
r
o
wth
o
f
r
eso
u
r
ce
s
an
d
tec
h
n
o
lo
g
ies
in
I
n
d
o
n
esian
,
wh
ic
h
is
s
till
r
elativ
ely
lack
in
g
c
o
m
p
a
r
ed
to
o
th
er
la
n
g
u
a
g
es.
T
h
ir
d
,
th
is
d
ataset
ca
n
b
e
u
s
ed
to
tr
ain
an
d
ev
alu
ate
lan
g
u
ag
e
m
o
d
els
an
d
co
n
v
er
s
atio
n
al
s
y
s
tem
s
d
esig
n
ed
to
ass
is
t in
th
e
co
n
tex
t o
f
m
en
tal
h
ea
lth
co
u
n
s
ellin
g
.
T
h
e
d
ataset
tr
an
s
latio
n
p
r
o
ce
s
s
was
d
o
n
e
m
a
n
u
ally
.
I
n
th
e
p
r
o
ce
s
s
,
we
wo
r
k
e
d
with
a
p
s
y
ch
o
th
er
ap
y
ass
is
tan
t
an
d
a
lectu
r
er
in
th
e
I
s
lam
ic
p
s
y
ch
o
lo
g
y
s
tu
d
y
p
r
o
g
r
am
.
I
n
ad
d
itio
n
,
th
e
tr
an
s
lato
r
is
a
g
r
ad
u
ate
o
f
th
e
Ma
s
ter
o
f
I
s
lam
ic
Ps
y
ch
o
lo
g
y
E
d
u
ca
tio
n
at
Su
n
an
Kal
ijag
a
State
I
s
lam
ic
Un
iv
e
r
s
ity
Yo
g
y
ak
ar
ta.
T
h
e
tr
an
s
lato
r
o
f
th
e
d
ataset
h
as g
o
o
d
I
n
d
o
n
esian
lan
g
u
ag
e
s
k
ills
an
d
u
n
d
er
s
tan
d
s
E
n
g
lis
h
as
th
e
m
ain
la
n
g
u
a
g
e
o
f
th
e
tr
an
s
lated
d
ataset.
2
.
3
.
P
r
o
ce
du
re
I
n
th
is
r
esear
c
h
,
th
e
r
e
ar
e
s
ev
er
al
p
r
o
ce
d
u
r
es
th
at
n
ee
d
t
o
b
e
d
o
n
e,
s
tar
tin
g
f
r
o
m
p
r
e
-
p
r
o
ce
s
s
in
g
d
ata,
an
d
m
o
d
el
tr
ain
in
g
to
m
o
d
el
ev
alu
atio
n
.
as
f
o
r
th
e
d
et
ails
th
at
will
b
e
ex
p
lain
ed
in
Fig
u
r
e
1
.
Fig
u
r
e
1
s
h
o
ws
th
e
p
r
o
ce
s
s
f
lo
w
in
t
h
is
r
esear
ch
,
w
h
er
e
t
h
e
f
ir
s
t
p
r
o
ce
s
s
is
d
ata
co
llectio
n
wh
ich
will
b
e
th
e
m
ain
m
ater
ial
in
th
e
q
u
esti
o
n
-
an
s
wer
s
y
s
tem
[
1
8
]
.
At
th
is
s
tag
e,
th
e
r
esear
ch
er
u
s
es
a
d
ataset
f
r
o
m
Am
o
d
/m
en
tal_
h
ea
lth
_
co
u
n
s
elin
g
_
co
n
v
er
s
atio
n
s
wh
ich
will
b
e
tr
an
s
lated
in
to
I
n
d
o
n
esian
.
Fu
r
th
er
m
o
r
e,
af
te
r
th
e
d
ata
is
s
u
cc
ess
f
u
lly
o
b
tain
ed
,
d
ata
p
r
e
-
p
r
o
ce
s
s
in
g
will
b
e
ca
r
r
ied
o
u
t,
wh
er
e
later
th
e
d
ata
will
b
e
m
ad
e
in
to
to
k
e
n
s
,
an
d
c
o
lu
m
n
s
will
b
e
ad
d
ed
to
th
e
d
ataset
[
1
9
]
.
Af
ter
p
r
e
-
p
r
o
ce
s
s
in
g
is
co
m
p
lete,
we
will
co
n
tin
u
e
with
m
o
d
el
tr
ain
in
g
f
o
r
q
u
esti
o
n
-
an
s
wer
n
ee
d
s
,
wh
er
e
o
p
ti
m
izatio
n
will
b
e
ca
r
r
ied
o
u
t
b
y
f
in
e
-
tu
n
in
g
th
e
m
o
d
el
to
im
p
r
o
v
e
th
e
q
u
ality
o
f
th
e
m
o
d
el.
Fin
ally
,
th
e
m
o
d
el
will
b
e
tes
ted
u
s
in
g
B
E
R
T
Sco
r
e
to
ass
es
s
th
e
o
p
tim
izatio
n
r
esu
lts
[
2
0
]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
5
,
Octo
b
e
r
20
25
:
4
8
2
9
-
4
8
3
6
4832
Fig
u
r
e
1
.
R
esear
ch
f
lo
w
2
.
3
.
1
.
Da
t
a
pre
-
pro
ce
s
s
ing
Data
n
ee
d
s
to
p
ass
th
e
im
p
o
r
t
an
t
s
tag
e
o
f
p
r
e
-
p
r
o
ce
s
s
in
g
as
p
r
ep
a
r
atio
n
b
ef
o
r
e
it
is
u
s
ed
t
o
tr
ain
t
h
e
q
u
esti
o
n
-
an
s
wer
in
g
m
o
d
el.
T
h
is
s
tag
e
aim
s
to
tr
an
s
f
o
r
m
u
n
p
r
o
ce
s
s
ed
d
ata
in
to
a
s
u
itab
le
f
o
r
m
at
th
at
ca
n
b
e
p
r
o
ce
s
s
ed
p
r
o
p
er
ly
b
y
d
ee
p
lear
n
in
g
m
o
d
els.
So
m
e
o
f
t
h
e
d
ata
p
r
e
-
p
r
o
ce
s
s
in
g
s
tep
s
th
at
will
b
e
ca
r
r
ied
o
u
t
ar
e
[
2
1
]
.
a.
T
o
k
en
izatio
n
o
f
q
u
esti
o
n
s
an
d
an
s
wer
s
:
T
h
e
q
u
esti
o
n
an
d
a
n
s
wer
tex
t
is
s
ep
ar
ated
in
t
o
in
d
iv
id
u
al
to
k
en
s
u
s
in
g
a
s
p
ec
ialized
to
k
en
izer
,
s
u
ch
as
th
e
B
E
R
T
to
k
en
ize
[
2
2
]
.
T
h
is
to
k
en
izatio
n
is
im
p
o
r
t
an
t
to
b
r
ea
k
th
e
tex
t
in
to
u
n
its
th
at
ca
n
b
e
u
n
d
er
s
to
o
d
b
y
th
e
m
o
d
el,
s
u
ch
as
wo
r
d
s
o
r
s
u
b
-
wo
r
d
s
.
T
h
is
s
tag
e
en
s
u
r
es
th
at
th
e
in
p
u
t
d
ata
co
n
f
o
r
m
s
to
th
e
f
o
r
m
at
r
e
q
u
ir
ed
b
y
t
h
e
B
E
R
T
m
o
d
el
[
2
3
]
.
b.
Sear
ch
f
o
r
th
e
s
tar
t
an
d
en
d
p
o
s
itio
n
o
f
th
e
an
s
wer
:
T
h
e
b
eg
in
n
in
g
an
d
co
n
clu
s
io
n
o
f
th
e
an
s
wer
in
th
e
tex
t
f
ac
to
r
s
ar
e
id
en
tifie
d
an
d
m
ar
k
ed
[
2
4
]
.
T
h
is
s
tep
h
elp
s
th
e
m
o
d
el
lear
n
th
e
co
r
r
ela
tio
n
am
o
n
g
th
e
q
u
esti
o
n
,
th
e
tex
t c
o
n
tex
t,
an
d
th
e
ac
tu
al
p
o
s
itio
n
o
f
th
e
a
n
s
wer
.
c.
A
d
d
i
t
i
o
n
o
f
s
t
a
r
t
_
p
o
s
i
ti
o
n
s
a
n
d
e
n
d
_
p
o
s
i
t
i
o
n
s
c
o
l
u
m
n
s
t
o
t
h
e
d
a
t
a
s
et
:
T
h
e
s
t
a
r
t
a
n
d
e
n
d
p
o
s
iti
o
n
i
n
f
o
r
m
a
t
i
o
n
o
b
t
a
i
n
e
d
f
r
o
m
t
h
e
p
r
e
v
i
o
u
s
s
t
e
p
i
s
a
d
d
e
d
t
o
t
h
e
d
a
t
a
s
e
t
as
n
e
w
c
o
l
u
m
n
s
,
s
u
c
h
as
s
t
a
r
t
_
p
o
s
i
t
i
o
n
s
a
n
d
e
n
d
_
p
o
s
i
t
i
o
n
s
[
2
5
]
.
T
h
e
s
e
c
o
l
u
m
n
s
w
i
ll
b
e
u
s
e
d
as
t
a
r
g
et
s
o
r
la
b
e
l
s
i
n
t
h
e
m
o
d
e
l
t
r
a
i
n
i
n
g
p
r
o
c
e
s
s
.
2
.
3
.
2
.
M
o
del
t
ra
ini
ng
T
h
e
m
o
d
el
tr
ain
in
g
s
tag
e
is
an
im
p
o
r
tan
t
p
a
r
t
o
f
d
ev
el
o
p
in
g
a
d
ee
p
lear
n
in
g
-
b
ased
s
y
s
tem
.
T
h
is
p
r
o
ce
s
s
in
v
o
lv
es
tr
ain
in
g
th
e
t
r
an
s
f
o
r
m
er
m
o
d
el
u
s
in
g
a
p
r
e
-
p
r
o
ce
s
s
ed
d
ataset.
in
th
is
s
ta
g
e,
h
y
p
e
r
p
ar
am
ete
r
d
eter
m
in
atio
n
a
n
d
f
i
n
e
-
tu
n
i
n
g
p
r
o
ce
s
s
will b
e
ex
ec
u
ted
o
n
th
e
m
o
d
el
th
at
h
as b
ee
n
d
ete
r
m
i
n
ed
[
2
6
]
.
a.
Def
in
in
g
lan
g
u
ag
e
m
o
d
el
o
p
ti
m
izatio
n
:
T
o
en
s
u
r
e
a
f
air
c
o
m
p
ar
is
o
n
,
b
o
th
I
n
d
o
B
E
R
T
an
d
MBERT
wer
e
f
in
e
-
tu
n
e
d
u
s
in
g
id
e
n
tical
h
y
p
er
p
ar
am
eter
s
:
lear
n
in
g
r
ate
o
f
2
e
-
5
,
b
atch
s
ize
o
f
8
,
3
tr
ain
in
g
ep
o
ch
s
,
a
n
d
a
d
r
o
p
o
u
t
r
ate
o
f
0
.
1
.
B
o
th
m
o
d
els
s
h
ar
e
th
e
s
am
e
ar
ch
itect
u
r
e
(
1
2
h
id
d
e
n
lay
e
r
s
,
1
2
atte
n
tio
n
h
e
ad
s
,
an
d
h
id
d
en
s
ize
o
f
7
6
8
)
.
Op
ti
m
izatio
n
s
tr
ateg
ies
in
clu
d
e
p
ar
am
eter
in
itializatio
n
,
t
o
k
en
izatio
n
,
an
d
r
eg
u
lar
izatio
n
[
2
7
]
.
b.
Fin
e
-
tu
n
in
g
:
Usi
n
g
lab
elled
d
ata
f
r
o
m
ea
r
lier
jo
b
s
,
th
e
B
E
R
T
m
o
d
el
th
en
in
itialized
b
y
u
s
in
g
p
r
e
-
tr
ain
e
d
p
ar
am
eter
s
.
T
h
e
p
r
o
ce
s
s
th
en
co
n
tin
u
es
to
ad
ju
s
tm
en
t o
f
all
p
ar
am
eter
s
.
E
v
e
n
wh
ile
all
task
s
s
tar
t
with
th
e
s
am
e
p
r
e
-
tr
ain
in
g
s
ettin
g
s
,
ea
ch
o
n
e
h
as a
u
n
iq
u
e
r
ef
i
n
ed
m
o
d
el
[
2
8
]
.
I
n
p
e
r
f
o
r
m
in
g
q
u
esti
o
n
-
a
n
s
wer
in
g
task
s
o
n
I
n
d
o
n
esian
d
at
a,
two
tr
an
s
f
o
r
m
er
m
o
d
els
ca
n
b
e
u
s
ed
,
n
am
ely
I
n
d
o
B
E
R
T
an
d
m
u
ltil
in
g
u
al
-
B
E
R
T
(
MBERT)
.
B
o
th
m
o
d
els
ar
e
b
ased
o
n
th
e
B
E
R
T
ar
ch
itectu
r
e
t
h
at
h
as p
r
o
v
e
n
r
eliab
le
in
s
ev
e
r
al
n
atu
r
al
lan
g
u
ag
e
p
r
o
ce
s
s
in
g
ta
s
k
s
[
2
9
]
.
2
.
3
.
3
.
M
o
del
e
v
a
lua
t
io
n
I
n
th
is
r
esear
ch
,
th
e
m
o
d
el
p
e
r
f
o
r
m
a
n
ce
ev
alu
atio
n
u
s
es
B
E
R
T
Sco
r
e.
B
E
R
T
Sco
r
e
is
an
ev
alu
atio
n
m
etr
ic
f
o
r
lan
g
u
ag
e
g
e
n
er
atio
n
b
ased
o
n
p
r
e
-
tr
ain
e
d
co
n
tex
t
u
al
em
b
ed
d
i
n
g
B
E
R
T
.
B
E
R
T
Sco
r
e
ca
lcu
lates
th
e
s
im
ilar
ity
b
etwe
en
a
co
u
p
le
o
f
s
en
ten
ce
s
as
th
e
to
tal
o
f
th
e
cu
m
u
lativ
e
s
im
ilar
ity
b
etwe
e
n
b
o
th
s
en
ten
ce
s
’
em
b
ed
d
in
g
to
k
e
n
s
.
B
E
R
T
s
co
r
e
ass
ig
n
s
a
tr
u
e
an
s
wer
s
en
ten
ce
(
ẋ
)
a
n
d
a
p
r
e
d
ictio
n
s
e
n
ten
ce
(
ẍ
)
,
W
e
u
s
e
co
n
tex
tu
al
e
m
b
ed
d
in
g
to
r
ep
r
esen
t
th
e
to
k
en
s
an
d
ca
lc
u
late
th
e
r
esu
lt
u
tili
zin
g
co
s
in
e
s
i
m
ilar
ity
,
o
p
tio
n
ally
im
p
o
r
tan
ce
r
ev
iewe
d
with
in
v
er
s
e
d
o
cu
m
e
n
t
f
r
e
q
u
en
c
y
(
I
D
F)
s
co
r
e
[
2
0
]
.
B
E
R
T
s
co
r
e
will
m
atch
ea
ch
to
k
e
n
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
La
n
g
u
a
g
e
m
o
d
el
o
p
timiz
a
tio
n
fo
r
men
ta
l h
ea
lth
q
u
esti
o
n
a
n
s
w
erin
g
…
(
F
a
r
d
a
n
Za
ma
kh
s
ya
r
i
)
4833
in
ẋ
with
a
to
k
en
in
ẍ
to
ca
l
cu
late
r
ec
all
B
E
R
T
Sco
r
e,
an
d
ea
ch
to
k
en
in
ẍ
with
a
to
k
en
in
ẋ
to
ca
lcu
late
p
r
ec
is
io
n
B
E
R
T
Sco
r
e.
Af
ter
th
at,
th
e
p
r
ec
is
io
n
a
n
d
r
ec
a
ll
v
alu
es
will
b
e
co
m
b
in
ed
to
ca
lcu
late
th
e
F1
B
E
R
T
Sco
r
e
s
ize.
T
h
e
f
o
llo
win
g
is
th
e
f
o
r
m
u
la
u
s
ed
to
ca
l
cu
late
r
ec
all
B
E
R
T
Sco
r
e,
p
r
e
cisi
o
n
B
E
R
T
Sco
r
e,
an
d
F1
B
E
R
T
Sco
r
e
[
9
]
.
Fo
r
th
e
ev
alu
atio
n
o
f
th
is
q
u
esti
o
n
-
a
n
s
wer
s
y
s
tem
,
we
will test u
s
i
n
g
B
E
R
T
Sco
r
e
an
d
co
m
p
ar
e
th
e
r
esu
lts
b
ef
o
r
e
an
d
af
ter
o
p
tim
izin
g
th
e
m
o
d
el,
in
o
r
d
er
t
o
g
et
clea
r
a
n
d
c
r
ed
ib
le
r
esu
lts
.
3.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
3
.
1
.
Da
t
a
pre
-
pro
ce
s
s
ing
Af
ter
th
e
d
ataset
is
p
r
ep
ar
ed
,
th
e
d
ata
g
o
es
to
t
h
e
p
r
e
-
p
r
o
ce
s
s
in
g
p
h
ase.
T
h
e
d
ataset
was
r
ea
d
ju
s
ted
b
ef
o
r
e
th
e
q
u
esti
o
n
-
a
n
s
wer
in
g
m
o
d
el
is
tr
ain
ed
.
T
h
e
g
o
al
o
f
th
is
s
tep
is
to
alter
u
n
p
r
o
ce
s
s
ed
in
p
u
t
o
n
a
f
o
r
m
th
at
d
ee
p
lear
n
in
g
m
o
d
els
ca
n
h
an
d
le
c
o
r
r
ec
tly
.
Ad
d
itio
n
all
y
,
th
e
d
ata
was
s
p
litt
ed
in
th
is
p
r
o
ce
s
s
,
with
8
0
%
o
f
th
e
d
ata
g
o
in
g
to
war
d
tr
ain
i
n
g
an
d
2
0
% g
o
in
g
t
o
war
d
v
ali
d
atio
n
.
3
.
1
.
1
.
T
o
k
eniza
t
io
n o
f
qu
estio
ns
a
nd
a
ns
wer
s
At
th
is
s
tag
e,
ea
ch
q
u
esti
o
n
an
d
an
s
wer
wer
e
to
k
en
ized
,
t
o
d
iv
id
e
th
e
tex
t
in
p
u
t
in
th
e
d
ata
in
to
s
m
aller
p
ar
ts
f
o
r
ea
s
y
p
r
o
ce
s
s
in
g
.
I
n
th
e
p
r
o
ce
s
s
,
b
ec
au
s
e
t
h
er
e
wer
e
two
d
if
f
er
en
t
m
o
d
e
ls
,
th
e
to
k
en
s
u
s
ed
wer
e
also
d
if
f
er
en
t
b
etwe
en
th
e
I
n
d
o
B
E
R
T
an
d
MBERT
m
o
d
els.
T
h
e
I
n
d
o
B
E
R
T
m
o
d
el
u
s
ed
th
e
B
E
R
T
T
o
k
en
izer
f
r
o
m
“
R
if
k
y
/
I
n
d
o
b
er
t
-
QA
”
wh
ile
MBERT
u
s
ed
“
b
er
t
-
b
ase
-
m
u
ltil
in
g
u
al
-
ca
s
ed
”
.
I
n
g
e
n
er
al,
th
er
e
was
n
o
d
if
f
er
en
ce
in
ter
m
s
o
f
th
e
p
r
o
ce
s
s
,
b
u
t
th
e
“
R
if
k
y
/I
n
d
o
b
e
r
t
-
QA
”
to
k
e
n
izatio
n
u
s
ed
an
I
n
d
o
n
esian
d
ictio
n
ar
y
,
wh
ile
“
b
er
t
-
b
ase
-
m
u
ltil
in
g
u
al
-
ca
s
ed
”
u
s
ed
a
d
ictio
n
ar
y
o
f
1
0
0
lan
g
u
ag
es
in
g
en
e
r
al,
s
o
f
o
r
I
n
d
o
n
esian
s
p
ec
if
ically
,
“
R
if
k
y
/I
n
d
o
b
er
t
-
QA
”
is
s
u
p
er
io
r
.
Af
ter
d
eter
m
i
n
in
g
th
e
B
E
R
T
t
o
k
en
izer
,
iter
ate
o
n
ea
c
h
r
o
w
an
d
r
et
r
iev
e
q
u
esti
o
n
s
an
d
a
n
s
wer
s
f
r
o
m
th
e
d
ataset.
Qu
esti
o
n
s
an
d
an
s
wer
s
wer
e
to
k
en
ized
u
s
in
g
th
e
p
r
ed
ef
in
e
d
to
k
en
ize
r
.
T
h
e
s
tar
t a
n
d
en
d
p
o
s
itio
n
s
o
f
th
e
an
s
wer
in
th
e
in
p
u
t
to
k
en
s
wer
e
d
eter
m
in
ed
b
y
f
in
d
in
g
th
e
to
k
en
[
SEP]
th
at
s
ep
ar
ates
th
e
q
u
esti
o
n
an
d
an
s
wer
.
I
f
th
e
[
SEP]
to
k
e
n
is
n
o
t
f
o
u
n
d
,
t
h
e
s
tar
t
an
d
en
d
p
o
s
itio
n
s
wer
e
s
et
to
0
.
T
h
e
s
tar
t
an
d
en
d
p
o
s
itio
n
s
wer
e
s
to
r
ed
in
th
e
s
tar
t_
p
o
s
itio
n
s
an
d
en
d
_
p
o
s
itio
n
s
lis
ts
.
F
o
r
d
eter
m
in
atio
n
,
th
e
[
C
L
S]
to
k
en
was
alwa
y
s
at
p
o
s
itio
n
0
,
s
o
a
f
ter
th
e
[
C
L
S]
t
o
k
en
was star
t_
p
o
s
itio
n
s
.
3
.
1
.
2
.
Addi
t
io
n o
f
s
t
a
rt
_
po
s
it
io
ns
a
nd
end_
po
s
it
io
ns
co
lu
m
ns
t
o
t
he
da
t
a
s
et
Af
ter
k
n
o
win
g
th
e
s
tar
t_
p
o
s
itio
n
s
an
d
en
d
_
p
o
s
itio
n
s
,
th
e
`
s
tar
t_
p
o
s
itio
n
s
`
an
d
`
en
d
_
p
o
s
itio
n
s
`
co
lu
m
n
s
wer
e
ad
d
ed
to
th
e
d
at
aset.
T
h
is
was
to
s
h
ar
e
th
e
lab
els
o
r
tar
g
ets
th
at
th
e
m
o
d
el
will
lear
n
d
u
r
in
g
th
e
tr
ain
in
g
p
r
o
ce
s
s
.
I
n
t
h
e
q
u
esti
o
n
an
s
wer
in
g
(
QA)
task
,
th
e
m
o
d
el
was
r
eq
u
ir
ed
to
d
eter
m
in
e
th
e
s
tar
t
an
d
en
d
p
o
s
itio
n
s
o
f
th
e
a
n
s
wer
f
r
o
m
t
h
e
co
n
tex
t.
B
y
ad
d
in
g
th
e
`
s
ta
r
t_
p
o
s
itio
n
s
`
a
n
d
`
en
d
_
p
o
s
itio
n
s
`
f
ield
s
wh
ich
p
u
t
th
e
s
tar
tin
g
an
d
en
d
in
g
p
o
s
itio
n
s
o
f
th
e
ac
tu
al
a
n
s
wer
,
th
e
m
o
d
el
c
o
u
ld
b
e
tr
ain
ed
t
o
s
tick
to
p
atter
n
s
ass
o
ciate
d
with
th
e
co
r
r
ec
t
an
s
wer
.
Du
r
in
g
tr
ain
in
g
,
th
e
m
o
d
el
was
r
ec
eiv
in
g
in
p
u
t
in
th
e
f
o
r
m
o
f
q
u
esti
o
n
s
an
d
tex
t
co
n
te
x
t,
an
d
tar
g
ets
in
th
e
f
o
r
m
o
f
`
s
tar
t_
p
o
s
itio
n
s
`
an
d
`
en
d
_
p
o
s
itio
n
s
`
o
f
th
e
a
ctu
al
an
s
wer
s
.
T
h
e
m
o
d
el
was
lear
n
ed
to
m
ap
th
e
in
p
u
t
to
th
e
co
r
r
ec
t
tar
g
ets
s
o
th
at
d
u
r
in
g
in
f
e
r
en
ce
(
ju
d
g
m
en
t
o
r
u
s
ag
e)
,
it
co
u
ld
f
o
r
esee
th
e
s
tar
t
an
d
en
d
p
o
s
itio
n
s
o
f
th
e
co
r
r
ec
t
an
s
wer
f
o
r
t
h
e
g
iv
e
n
q
u
esti
o
n
an
d
tex
t
co
n
tex
t.
T
h
u
s
,
th
e
ad
d
itio
n
o
f
th
e
`
s
tar
t_
p
o
s
i
tio
n
s
`
an
d
`
en
d
_
p
o
s
itio
n
s
`
f
iel
d
s
to
th
e
tr
ain
in
g
an
d
v
alid
ati
o
n
d
atasets
was
an
im
p
o
r
tan
t step
in
p
r
e
p
ar
in
g
in
f
o
r
m
atio
n
f
o
r
th
e
Qu
esti
o
n
An
s
wer
in
g
task
s
o
th
at
th
e
m
o
d
el
co
u
ld
b
e
tr
ain
e
d
to
lear
n
p
atter
n
s
th
at
wer
e
ass
o
ci
ated
with
co
r
r
ec
t a
n
s
wer
s
.
3
.
2
.
M
o
del
t
ra
ini
ng
Af
ter
d
eter
m
in
i
n
g
th
e
to
k
e
n
s
in
th
e
d
ataset
an
d
ad
d
in
g
th
e
s
tar
t_
p
o
s
itio
n
s
an
d
e
n
d
_
p
o
s
itio
n
s
co
lu
m
n
s
,
th
e
n
ex
t
s
tep
was
tr
ain
in
g
th
e
m
o
d
el.
I
n
th
e
m
o
d
el
tr
ain
i
n
g
s
tag
e,
th
e
h
y
p
er
p
ar
am
eter
s
wer
e
d
eter
m
in
ed
an
d
th
e
f
in
e
-
tu
n
i
n
g
p
r
o
ce
s
s
wer
e
p
er
f
o
r
m
e
d
o
n
ea
ch
m
o
d
el.
T
h
is
s
tag
e
was
th
e
co
r
e
p
r
o
ce
s
s
o
f
th
is
r
esear
ch
.
3
.
2
.
1
.
Def
ini
ng
la
ng
ua
g
e
mo
del o
ptim
iza
t
io
n
Dete
r
m
in
in
g
th
e
f
o
u
n
d
atio
n
al
m
o
d
el,
s
tr
ateg
y
an
d
o
p
tim
al
h
y
p
er
p
ar
am
eter
v
alu
es,
s
u
ch
as
n
u
m
b
e
r
o
f
ep
o
ch
s
,
b
atch
s
ize,
an
d
lea
r
n
in
g
r
ate,
was
v
er
y
im
p
o
r
tan
t
to
g
et
g
o
o
d
im
p
lem
en
tatio
n
f
r
o
m
th
e
m
o
d
el.
I
n
th
is
r
esear
ch
,
we
s
et
th
e
s
am
e
v
alu
e
b
etwe
en
th
e
I
n
d
o
B
E
R
T
an
d
MBERT
m
o
d
els
s
o
th
at
t
h
e
co
m
p
ar
is
o
n
g
ets
b
alan
ce
d
r
esu
lts
.
T
h
e
h
y
p
er
p
ar
am
eter
s
u
s
ed
in
b
o
th
m
o
d
e
ls
ar
e
Hid
d
en
Dr
o
p
o
u
t
Pro
b
a
b
ilit
y
o
f
0
.
1
,
u
s
in
g
Ad
am
as
th
e
o
p
tim
izer
,
lear
n
i
n
g
r
ate
o
f
2
.
0
0
E
-
0
5
as
well
as
u
s
in
g
3
e
p
o
ch
s
an
d
b
atch
s
ize
o
f
8
.
Fo
u
n
d
atio
n
al
m
o
d
els
o
f
I
n
d
o
B
E
R
T
an
d
M
B
E
R
T
h
av
e
th
e
s
am
e
p
ar
am
et
er
s
,
n
am
ely
1
2
h
i
d
d
en
lay
er
s
,
1
2
atten
tio
n
h
ea
d
s
,
an
d
Hid
d
e
n
Size
7
6
8
.
Fo
r
b
etter
m
o
d
el
o
p
tim
izatio
n
,
s
tr
ateg
ies
s
u
ch
as
p
ar
am
eter
in
itializatio
n
,
to
k
e
n
izatio
n
,
m
o
d
el
o
p
tim
izatio
n
,
an
d
r
eg
u
l
ar
izatio
n
ar
e
n
ee
d
ed
.
T
ab
le
1
s
u
m
m
a
r
izes
th
e
o
p
tim
izatio
n
s
ettin
g
s
.
T
h
e
p
ar
am
eter
s
f
o
llo
w
th
e
s
tan
d
ar
d
B
E
R
T
co
n
f
ig
u
r
atio
n
,
an
d
id
en
tical
h
y
p
er
p
ar
am
eter
s
wer
e
ap
p
lie
d
to
b
o
th
m
o
d
els
f
o
r
a
f
air
co
m
p
ar
is
o
n
.
T
h
is
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
5
,
Octo
b
e
r
20
25
:
4
8
2
9
-
4
8
3
6
4834
h
y
p
er
p
ar
am
eter
p
r
o
v
is
io
n
ad
a
p
ts
to
Dev
lin
'
s
r
esear
ch
en
tit
led
“
B
E
R
T:
P
r
e
-
tr
a
in
in
g
o
f
Dee
p
B
id
ir
ec
tio
n
a
l
Tr
a
n
s
fo
r
mer
s
fo
r
La
n
g
u
a
g
e
U
n
d
ers
ta
n
d
in
g
”
,
wh
er
e
h
is
r
ese
ar
ch
m
e
n
tio
n
ed
r
elate
d
h
y
p
er
p
ar
am
eter
s
th
at
a
r
e
o
p
tim
al
en
o
u
g
h
to
b
e
u
s
ed
in
th
e
f
in
e
-
tu
n
in
g
p
r
o
ce
s
s
.
T
h
e
s
tr
ateg
y
u
s
ed
in
t
h
is
r
esear
ch
in
clu
d
es
p
ar
a
m
eter
in
itializatio
n
,
to
k
en
izatio
n
,
m
o
d
el
o
p
tim
izatio
n
,
an
d
r
e
g
u
lar
izatio
n
.
Fo
r
p
ar
a
m
eter
in
itializatio
n
an
d
to
k
en
izatio
n
,
we
u
s
e
th
e
f
o
u
n
d
atio
n
o
f
I
n
d
o
B
E
R
T
an
d
MBERT,
wh
ile
th
e
r
eg
u
lar
iz
atio
n
tech
n
iq
u
e
u
s
es
d
r
o
p
o
u
t p
r
o
b
ab
ilit
y
,
wh
ich
wa
s
in
ten
d
ed
to
r
ed
u
ce
d
ata
o
v
e
r
f
itti
n
g
d
u
r
i
n
g
th
e
f
in
e
-
tu
n
in
g
p
r
o
ce
s
s
.
T
ab
le
1
.
Dete
r
m
in
i
n
g
lan
g
u
ag
e
m
o
d
el
o
p
tim
izatio
n
V
a
r
i
a
b
l
e
I
t
e
m
I
n
d
o
B
ER
T
M
B
ER
T
P
a
r
a
me
t
e
r
s
H
i
d
d
e
n
l
a
y
e
r
s
12
12
A
t
t
e
n
t
i
o
n
h
e
a
d
s
12
12
H
i
d
d
e
n
s
i
z
e
7
6
8
7
6
8
H
y
p
e
r
p
a
r
a
me
t
e
r
s
M
a
x
l
e
n
g
t
h
5
1
2
5
1
2
Le
a
r
n
i
n
g
r
a
t
e
2
.
0
0
E
-
05
2
.
0
0
E
-
05
O
p
t
i
mi
z
e
r
A
d
a
m
A
d
a
m
Ep
o
c
h
3
3
B
a
t
c
h
s
i
z
e
8
8
S
t
r
a
t
e
g
y
P
a
r
a
me
t
e
r
I
n
i
t
i
a
l
i
z
a
t
i
o
n
R
i
f
k
y
/
I
n
d
o
b
e
r
t
-
QA
b
e
r
t
-
b
a
s
e
-
mu
l
t
i
l
i
n
g
u
a
l
-
c
a
s
e
d
To
k
e
n
i
z
a
t
i
o
n
R
i
f
k
y
/
I
n
d
o
b
e
r
t
-
QA
b
e
r
t
-
b
a
s
e
-
mu
l
t
i
l
i
n
g
u
a
l
-
c
a
s
e
d
M
o
d
e
l
o
p
t
i
mi
z
a
t
i
o
n
U
si
n
g
H
y
p
e
r
p
a
r
a
me
t
e
r
U
si
n
g
H
y
p
e
r
p
a
r
a
me
t
e
r
R
e
g
u
l
a
r
i
z
a
t
i
o
n
D
r
o
p
o
u
t
P
r
o
b
a
b
i
l
i
t
y
(
0
.
1
)
D
r
o
p
o
u
t
P
r
o
b
a
b
i
l
i
t
y
(
0
.
1
)
3
.
2
.
2
.
F
ine
-
t
un
ing
Af
ter
s
ettin
g
in
itial
h
y
p
er
p
ar
a
m
eter
s
,
f
in
e
-
tu
n
in
g
was
p
er
f
o
r
m
ed
o
n
b
o
th
I
n
d
o
B
E
R
T
an
d
MBERT
.
T
h
e
p
r
o
ce
s
s
s
tar
ts
with
to
k
en
i
za
tio
n
,
wh
ich
d
if
f
e
r
s
f
r
o
m
th
e
ea
r
lier
p
r
e
p
r
o
ce
s
s
in
g
s
tep
.
I
t
i
n
v
o
lv
es
in
itializin
g
th
e
d
ata
f
r
am
e,
to
k
e
n
izer
,
an
d
tex
t
len
g
th
,
co
u
n
tin
g
d
atas
et
s
ize,
an
d
s
am
p
lin
g
f
r
o
m
t
h
e
“
Qu
esti
o
n
”
an
d
“
An
s
wer
”
co
lu
m
n
s
.
Sp
ec
ial
t
o
k
en
s
lik
e
[
C
L
S]
an
d
[
SEP]
wer
e
ad
d
e
d
,
te
x
ts
wer
e
tr
u
n
c
ated
o
r
p
a
d
d
ed
to
a
m
ax
im
u
m
len
g
th
,
an
d
in
p
u
ts
ar
e
co
n
v
e
r
ted
to
Py
T
o
r
ch
ten
s
o
r
s
.
Nex
t,
th
e
m
o
d
el
was
b
u
ilt
b
y
ad
ap
tin
g
it
to
th
e
q
u
esti
o
n
-
an
s
wer
task
with
th
e
s
p
ec
if
ied
d
ataset
an
d
p
ar
am
eter
s
.
T
r
ain
in
g
r
u
n
s
f
o
r
3
ep
o
ch
s
o
n
b
o
th
tr
ain
in
g
a
n
d
v
alid
atio
n
d
ata
t
o
o
p
tim
ize
th
e
m
o
d
el
weig
h
ts
f
o
r
b
etter
p
r
ed
ictio
n
s
.
Af
ter
tr
ain
in
g
,
th
e
m
o
d
el
an
d
to
k
en
ize
r
wer
e
s
av
ed
u
s
in
g
m
o
d
el.
s
av
e
_
p
r
etr
ai
n
ed
an
d
to
k
en
izer
.
s
av
e
_
p
r
etr
ain
e
d
f
o
r
later
u
s
e
with
o
u
t
r
etr
ain
in
g
.
3
.
3
.
M
o
del
e
v
a
lua
t
i
o
n
At
th
is
s
tag
e,
th
e
s
av
ed
r
es
u
lts
o
f
th
e
f
in
e
-
tu
n
in
g
p
r
o
c
ess
(
th
e
m
o
d
el)
wer
e
e
v
alu
ated
u
s
in
g
B
E
R
T
Sco
r
e.
B
E
R
T
Sco
r
e
is
a
lan
g
u
ag
e
g
en
er
atio
n
ev
alu
a
tio
n
m
etr
ic
b
ased
o
n
th
e
p
r
e
-
tr
ain
ed
co
n
tex
tu
al
em
b
ed
d
in
g
B
E
R
T
.
B
E
R
T
Sco
r
e
ca
lcu
lates
th
e
s
im
ilar
ity
o
f
two
s
en
ten
ce
s
as
th
e
to
tal
o
f
th
e
cu
m
u
lativ
e
s
im
ilar
ity
b
etwe
en
th
e
em
b
ed
d
in
g
to
k
en
s
.
As
illu
s
tr
ated
in
T
ab
le
2
,
th
e
I
n
d
o
B
E
R
T
m
o
d
el
d
em
o
n
s
tr
a
ted
a
n
o
tab
le
p
er
f
o
r
m
a
n
ce
in
ter
m
s
o
f
m
o
d
el
ev
al
u
atio
n
u
s
in
g
B
E
R
T
Sco
r
e.
T
h
e
m
o
d
el
attain
ed
an
F1
-
B
E
R
T
Sco
r
e
o
f
9
1
.
8
%,
ac
co
m
p
a
n
ied
b
y
a
r
ec
all
B
E
R
T
Sco
r
e
o
f
8
9
.
9
%
an
d
a
p
r
ec
is
io
n
B
E
R
T
Sco
r
e
o
f
9
3
.
9
%.
I
n
c
o
m
p
ar
is
o
n
,
t
h
e
MBERT
m
o
d
el
ex
h
ib
ited
a
co
m
p
ar
ativ
ely
lo
wer
p
er
f
o
r
m
an
ce
,
with
an
F
1
-
B
E
R
T
Sco
r
e
o
f
7
9
.
2
%,
a
r
e
ca
ll
B
E
R
T
Sco
r
e
o
f
7
3
.
4
%,
an
d
a
p
r
ec
is
io
n
B
E
R
T
Sco
r
e
o
f
8
6
.
2
%.
I
n
ad
d
itio
n
,
a
co
m
p
ar
is
o
n
was
m
ad
e
with
th
e
GPT
-
2
m
o
d
el,
an
d
th
e
r
esu
lts
o
b
tain
e
d
wer
e
F1
-
B
E
R
T
Sco
r
e
6
6
.
2
%,
r
ec
al
l
B
E
R
T
Sco
r
e
6
8
.
0
%,
a
n
d
p
r
e
cisi
o
n
B
E
R
T
Sco
r
e
6
5
.
5
%.
B
ased
o
n
t
h
e
v
alu
es
o
b
tain
e
d
,
it
ca
n
b
e
c
o
n
clu
d
ed
th
at
t
h
e
I
n
d
o
B
E
R
T
m
o
d
e
l
is
s
u
p
er
io
r
to
t
h
e
MBERT
an
d
GPT
-
2
m
o
d
els in
q
u
esti
o
n
an
d
an
s
wer
task
s
u
s
in
g
I
n
d
o
n
esian
-
la
n
g
u
a
g
e
m
en
ta
l h
ea
lth
d
atasets
.
T
ab
le
2
.
Dete
r
m
in
i
n
g
lan
g
u
ag
e
m
o
d
el
o
p
tim
izatio
n
M
o
d
e
l
I
t
e
m
F
1
B
ER
TSc
o
r
e
P
r
e
c
i
s
i
o
n
B
E
R
TSc
o
r
e
R
e
c
a
l
l
B
ER
TSc
o
r
e
I
n
d
o
B
ER
T
B
e
f
o
r
e
F
i
n
e
-
T
u
n
i
n
g
6
5
.
3
%
7
3
.
8
%
6
0
.
8
%
A
f
t
e
r
F
i
n
e
-
T
u
n
i
n
g
9
1
.
8
%
9
3
.
9
%
8
9
.
9
%
M
B
ER
T
B
e
f
o
r
e
F
i
n
e
-
T
u
n
i
n
g
7
4
.
2
%
7
8
.
8
%
7
2
.
2
%
A
f
t
e
r
F
i
n
e
-
T
u
n
i
n
g
7
9
.
2
%
8
6
.
2
%
7
3
.
4
%
G
P
T
-
2
B
e
f
o
r
e
F
i
n
e
-
T
u
n
i
n
g
5
8
.
4
%
5
6
.
4
%
6
0
.
0
%
A
f
t
e
r
F
i
n
e
-
T
u
n
i
n
g
6
6
.
2
%
6
8
.
0
%
6
5
.
5
%
3
.
4
.
Dis
cus
s
io
n
T
h
e
r
esu
lts
o
f
th
is
s
tu
d
y
e
m
p
h
asize
th
e
im
p
o
r
tan
ce
o
f
u
s
in
g
lan
g
u
ag
e
-
s
p
ec
if
ic
m
o
d
els
in
NL
P
task
s
.
I
n
d
o
B
E
R
T
,
wh
ich
was
p
r
e
-
tr
ain
ed
o
n
I
n
d
o
n
esian
tex
t,
s
ig
n
if
ican
tly
o
u
tp
er
f
o
r
m
ed
th
e
m
u
ltil
in
g
u
al
MBERT
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
La
n
g
u
a
g
e
m
o
d
el
o
p
timiz
a
tio
n
fo
r
men
ta
l h
ea
lth
q
u
esti
o
n
a
n
s
w
erin
g
…
(
F
a
r
d
a
n
Za
ma
kh
s
ya
r
i
)
4835
af
ter
f
in
e
-
t
u
n
in
g
.
I
n
d
o
B
E
R
T
ac
h
iev
ed
a
n
F1
-
B
E
R
T
Sco
r
e
o
f
9
1
.
8
%,
p
r
ec
is
io
n
o
f
9
3
.
9
%,
an
d
r
ec
all
o
f
8
9
.
9
%
.
I
n
co
m
p
ar
is
o
n
,
MBERT
o
n
ly
r
ea
ch
e
d
7
9
.
2
%
F1
,
with
8
6
.
2
%
p
r
ec
is
io
n
an
d
7
3
.
4
%
r
ec
all.
T
h
is
2
8
%
im
p
r
o
v
em
e
n
t
in
I
n
d
o
B
E
R
T
's
F1
s
co
r
e
co
m
p
ar
ed
t
o
its
p
er
f
o
r
m
an
ce
b
e
f
o
r
e
f
in
e
-
tu
n
in
g
,
v
er
s
u
s
o
n
ly
a
5
% g
ain
in
MBERT,
s
h
o
ws
th
at
f
in
e
-
t
u
n
in
g
is
h
ig
h
ly
ef
f
ec
tiv
e
esp
e
cially
wh
en
th
e
m
o
d
el
ar
ch
it
ec
tu
r
e
an
d
tr
ain
in
g
d
ata
ar
e
clo
s
ely
alig
n
ed
with
t
h
e
tar
g
et
lan
g
u
ag
e.
T
h
e
lo
wer
p
er
f
o
r
m
an
ce
o
f
MBERT
ca
n
b
e
attr
ib
u
ted
to
its
m
u
ltil
in
g
u
al
tr
ain
in
g
o
b
jectiv
e
,
wh
ich
r
ed
u
ce
s
its
s
p
ec
ializati
o
n
in
a
n
y
s
in
g
le
lan
g
u
ag
e,
in
cl
u
d
in
g
I
n
d
o
n
esian
.
T
h
ese
f
in
d
in
g
s
s
u
p
p
o
r
t
th
e
id
ea
th
at
f
o
r
s
p
ec
ialized
d
o
m
ain
s
lik
e
m
en
tal
h
ea
lth
in
a
s
p
e
cif
ic
lan
g
u
ag
e
,
m
o
n
o
lin
g
u
al
m
o
d
els
ar
e
m
o
r
e
s
u
itab
le.
Ad
d
itio
n
ally
,
th
e
o
p
tim
ized
I
n
d
o
B
E
R
T
m
o
d
el
h
as
p
o
ten
tial
f
o
r
r
ea
l
-
wo
r
ld
d
e
p
lo
y
m
e
n
t
in
m
en
tal
h
ea
lth
ch
atb
o
t
ap
p
licatio
n
s
.
Ho
wev
er
,
eth
ical
an
d
leg
al
c
o
n
s
id
er
atio
n
s
r
em
ai
n
ess
en
tial,
p
ar
ticu
lar
ly
r
eg
ar
d
in
g
d
ata
p
r
iv
ac
y
,
co
n
ten
t
m
o
d
er
atio
n
,
a
n
d
ap
p
r
o
p
r
iate
r
esp
o
n
s
e
h
an
d
lin
g
.
E
n
s
u
r
in
g
r
esp
o
n
s
ib
le
AI
d
esig
n
is
cr
itical
b
ef
o
r
e
s
u
ch
s
y
s
tem
s
ar
e
p
u
b
licly
im
p
lem
en
ted
.
4.
CO
NCLU
SI
O
N
T
h
e
in
ten
tio
n
o
f
th
is
wo
r
k
wa
s
to
eq
u
alize
t
h
e
p
e
r
f
o
r
m
an
ce
o
f
B
E
R
T
m
o
d
els
in
q
u
esti
o
n
-
an
s
wer
in
g
task
s
,
s
p
ec
if
ically
I
n
d
o
B
E
R
T
an
d
M
-
B
E
R
T
m
o
d
els,
u
tili
zin
g
I
n
d
o
n
esian
lan
g
u
ag
e
d
atasets
f
o
cu
s
ed
o
n
m
en
tal
h
ea
lth
d
o
m
ain
.
T
h
e
p
er
f
o
r
m
an
ce
o
f
I
n
d
o
B
E
R
T
s
u
r
p
ass
ed
M
-
B
E
R
T
,
with
an
F1
-
B
E
R
T
Sco
r
e
o
f
9
1
.
8
%,
r
ec
all
B
E
R
T
Sco
r
e
o
f
8
9
.
9
%,
an
d
p
r
ec
is
io
n
B
E
R
T
Sco
r
e
o
f
9
3
.
9
%.
Me
an
wh
ile,
th
e
MBERT
m
o
d
el
h
as
a
lo
we
r
p
er
f
o
r
m
an
ce
,
with
a
v
alu
e
o
f
F1
-
B
E
R
T
Sco
r
e
o
f
7
9
.
2
%,
r
ec
all
B
E
R
T
Sco
r
e
o
f
7
3
.
4
%,
an
d
p
r
ec
is
io
n
B
E
R
T
Sco
r
e
o
f
8
6
.
2
%.
T
h
ese
f
in
d
in
g
s
h
ig
h
lig
h
t
th
e
n
ee
d
t
o
u
s
e
lan
g
u
ag
e
-
s
p
ec
if
ic
m
o
d
els,
s
u
ch
as
I
n
d
o
B
E
R
T
f
o
r
I
n
d
o
n
esian
,
to
im
p
r
o
v
e
t
h
e
p
er
f
o
r
m
a
n
ce
an
d
r
ele
v
an
c
e
o
f
r
esp
o
n
s
es
in
q
u
esti
o
n
-
a
n
s
wer
s
y
s
tem
s
.
I
n
ad
d
itio
n
,
th
is
s
tu
d
y
d
em
o
n
s
tr
ates
th
e
ef
f
ec
tiv
en
ess
o
f
f
in
e
-
tu
n
in
g
m
eth
o
d
s
in
im
p
r
o
v
in
g
m
o
d
el
p
er
f
o
r
m
an
ce
wh
ich
in
th
is
ca
s
e
th
e
I
n
d
o
B
E
R
T
im
p
r
o
v
e
b
y
2
8
%
wh
ile
MBERT
im
p
r
o
v
e
b
y
ab
o
u
t
5
%.
T
h
e
h
ig
h
er
im
p
r
o
v
em
e
n
t
o
f
I
n
d
o
B
E
R
T
s
h
o
ws
th
at
m
o
d
els
tr
ain
ed
f
o
r
a
s
p
ec
if
ic
ty
p
e
o
f
lan
g
u
a
g
e
(
in
th
is
ca
s
e
th
e
I
n
d
o
n
esian
lan
g
u
a
g
e)
ca
n
im
p
r
o
v
e
s
ig
n
if
ica
n
tly
wh
en
o
p
tim
ize
f
o
r
NL
P
task
s
in
th
at
s
p
ec
i
f
ic
lan
g
u
a
g
e.
Oth
e
r
co
n
tr
ib
u
tio
n
o
f
t
h
is
r
esear
ch
w
er
e;
o
p
tim
ized
I
n
d
o
B
E
R
T
an
d
MBERT
m
o
d
el
f
o
r
QA
task
o
f
th
e
m
en
tal
h
ea
lth
d
o
m
ain
in
I
n
d
o
n
esian
lan
g
u
ag
e,
th
e
tr
an
s
latio
n
o
f
t
h
e
Am
o
d
/m
en
tal_
h
ea
lth
_
co
u
n
s
elin
g
_
c
o
n
v
er
s
atio
n
s
d
ataset
to
I
n
d
o
n
esian
lan
g
u
ag
e,
a
n
d
p
r
o
v
id
e
an
o
v
er
v
iew
o
f
th
e
m
en
tal
h
ea
lth
q
u
esti
o
n
-
an
s
wer
s
y
s
tem
th
at
ca
n
b
e
ap
p
lied
ac
co
r
d
in
g
to
th
e
in
ten
d
ed
s
p
ec
if
icatio
n
s
.
T
o
s
u
m
m
a
r
ize,
th
is
s
tu
d
y
n
o
t
o
n
ly
d
e
m
o
n
s
tr
ate
I
n
d
o
B
E
R
T
'
s
s
u
p
er
io
r
ity
in
I
n
d
o
n
esian
q
u
esti
o
n
-
an
s
wer
in
g
task
s
co
m
p
a
r
e
d
to
MBERT
b
u
t
also
em
p
h
as
izes
th
e
im
p
o
r
tan
ce
o
f
in
v
esti
n
g
in
th
e
d
ev
elo
p
m
en
t
an
d
o
p
tim
izatio
n
o
f
lan
g
u
ag
e
-
s
p
ec
if
ic
m
o
d
els
to
im
p
r
o
v
e
th
e
ac
ce
s
s
ib
ilit
y
an
d
q
u
ality
o
f
d
ig
ital
s
er
v
ices,
p
ar
ticu
lar
ly
in
s
en
s
itiv
e
ar
ea
s
s
u
ch
as
m
en
tal
h
ea
lth
.
T
h
is
s
tu
d
y
also
s
h
o
ws
a
h
ig
h
er
ac
cu
r
ac
y
v
alu
e
co
m
p
ar
ed
to
s
ev
er
al
p
r
ev
io
u
s
s
tu
d
ies
with
an
F1
-
B
E
R
T
Sco
r
e
o
f
9
1
.
8
%,
r
ec
all
B
E
R
T
Sco
r
e
o
f
8
9
.
9
%,
an
d
p
r
ec
is
io
n
B
E
R
T
Sco
r
e
o
f
9
3
.
9
%.
T
h
ese
f
in
d
in
g
s
p
av
e
th
e
wa
y
f
o
r
f
u
t
u
r
e
s
tu
d
ies
in
to
th
e
ad
a
p
tatio
n
o
f
NL
P m
o
d
els to
d
if
f
er
e
n
t c
u
ltu
r
al
c
o
n
te
x
ts
an
d
ap
p
licatio
n
d
o
m
ai
n
s
.
RE
F
E
R
E
NC
E
S
[
1
]
J.
A
.
A
l
z
u
b
i
,
R
.
Ja
i
n
,
A
.
S
i
n
g
h
,
P
.
P
a
r
w
e
k
a
r
,
a
n
d
M
.
G
u
p
t
a
,
“
C
O
B
ER
T:
C
O
V
I
D
-
1
9
q
u
e
s
t
i
o
n
a
n
sw
e
r
i
n
g
s
y
st
e
m
u
s
i
n
g
B
E
R
T,
”
Ara
b
i
a
n
J
o
u
rn
a
l
f
o
r
S
c
i
e
n
c
e
a
n
d
En
g
i
n
e
e
ri
n
g
,
v
o
l
.
4
8
,
n
o
.
8
,
p
p
.
1
1
0
0
3
–
1
1
0
1
3
,
2
0
2
3
,
d
o
i
:
1
0
.
1
0
0
7
/
s1
3
3
6
9
-
0
2
1
-
0
5
8
1
0
-
5.
[
2
]
A
.
K
e
sarw
a
n
i
,
S
.
D
a
s,
D
.
R
.
K
i
sk
u
,
a
n
d
M
.
D
a
l
u
i
,
“
M
u
l
t
i
-
sc
a
l
e
v
i
si
o
n
t
r
a
n
sf
o
r
m
e
r
t
o
w
a
r
d
i
mp
r
o
v
e
d
n
o
n
-
i
n
v
a
si
v
e
a
n
a
e
mi
a
d
e
t
e
c
t
i
o
n
u
s
i
n
g
p
a
l
m
v
i
d
e
o
,
”
M
u
l
t
i
m
e
d
i
a
T
o
o
l
s
a
n
d
Ap
p
l
i
c
a
t
i
o
n
s
,
v
o
l
.
8
3
,
n
o
.
3
8
,
p
p
.
8
5
8
2
5
–
8
5
8
4
8
,
2
0
2
4
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
1
0
4
2
-
0
2
4
-
2
0
1
1
8
-
w.
[
3
]
M
.
B
i
l
a
l
a
n
d
A
.
A
.
A
l
maz
r
o
i
,
“
Ef
f
e
c
t
i
v
e
n
e
ss
o
f
f
i
n
e
-
t
u
n
e
d
B
E
R
T
mo
d
e
l
i
n
c
l
a
ss
i
f
i
c
a
t
i
o
n
o
f
h
e
l
p
f
u
l
a
n
d
u
n
h
e
l
p
f
u
l
o
n
l
i
n
e
c
u
s
t
o
m
e
r
r
e
v
i
e
w
s,
”
El
e
c
t
ro
n
i
c
C
o
m
m
e
rc
e
R
e
se
a
rc
h
,
v
o
l
.
2
3
,
n
o
.
4
,
p
p
.
2
7
3
7
–
2
7
5
7
,
2
0
2
3
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
0
6
6
0
-
0
2
2
-
0
9
5
6
0
-
w.
[
4
]
J.
D
e
v
l
i
n
,
M
.
W
.
C
h
a
n
g
,
K
.
Le
e
,
a
n
d
K
.
To
u
t
a
n
o
v
a
,
“
B
ER
T
:
p
r
e
-
t
r
a
i
n
i
n
g
o
f
d
e
e
p
b
i
d
i
r
e
c
t
i
o
n
a
l
t
r
a
n
sf
o
r
me
r
s
f
o
r
l
a
n
g
u
a
g
e
u
n
d
e
r
s
t
a
n
d
i
n
g
,
”
in
N
AAC
L
H
L
T
2
0
1
9
-
2
0
1
9
C
o
n
f
e
r
e
n
c
e
o
f
t
h
e
N
o
rt
h
A
m
e
ri
c
a
n
C
h
a
p
t
e
r
o
f
t
h
e
Ass
o
c
i
a
t
i
o
n
f
o
r
C
o
m
p
u
t
a
t
i
o
n
a
l
L
i
n
g
u
i
st
i
c
s:
H
u
m
a
n
L
a
n
g
u
a
g
e
T
e
c
h
n
o
l
o
g
i
e
s
-
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
C
o
n
f
e
re
n
c
e
,
v
o
l
.
1
,
p
p
.
4
1
7
1
–
4
1
8
6
,
O
c
t
.
2
0
1
9
.
[
5
]
F
.
K
o
t
o
,
A
.
R
a
h
i
mi
,
J.
H
.
La
u
,
a
n
d
T.
B
a
l
d
w
i
n
,
“
I
n
d
o
LE
M
a
n
d
I
n
d
o
B
ER
T:
A
b
e
n
c
h
mark
d
a
t
a
se
t
a
n
d
p
r
e
-
t
r
a
i
n
e
d
l
a
n
g
u
a
g
e
m
o
d
e
l
f
o
r
I
n
d
o
n
e
s
i
a
n
N
LP,
”
i
n
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
2
8
th
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
C
o
m
p
u
t
a
t
i
o
n
a
l
L
i
n
g
u
i
st
i
c
s
,
2
0
2
0
,
p
p
.
7
5
7
–
7
7
0
,
d
o
i
:
1
0
.
1
8
6
5
3
/
v
1
/
2
0
2
0
.
c
o
l
i
n
g
-
mai
n
.
6
6
.
[
6
]
F
.
K
o
t
o
,
J.
H
.
La
u
,
a
n
d
T.
B
a
l
d
w
i
n
,
“
I
N
D
O
B
ER
TWEET
:
a
p
r
e
t
r
a
i
n
e
d
l
a
n
g
u
a
g
e
m
o
d
e
l
f
o
r
I
n
d
o
n
e
s
i
a
n
Tw
i
t
t
e
r
w
i
t
h
e
f
f
e
c
t
i
v
e
d
o
m
a
i
n
-
s
p
e
c
i
f
i
c
v
o
c
a
b
u
l
a
r
y
i
n
i
t
i
a
l
i
z
a
t
i
o
n
,
”
in
E
MN
L
P
2
0
2
1
-
2
0
2
1
C
o
n
f
e
re
n
c
e
o
n
Em
p
i
ri
c
a
l
M
e
t
h
o
d
s
i
n
N
a
t
u
r
a
l
L
a
n
g
u
a
g
e
Pro
c
e
ssi
n
g
,
Pro
c
e
e
d
i
n
g
s
,
p
p
.
1
0
6
6
0
–
1
0
6
6
8
,
2
0
2
1
,
d
o
i
:
1
0
.
1
8
6
5
3
/
v
1
/
2
0
2
1
.
e
mn
l
p
-
ma
i
n
.
8
3
3
.
[
7
]
Q
.
Li
a
n
d
Y
.
Zh
a
n
g
,
“
I
mp
r
o
v
e
d
t
e
x
t
mat
c
h
i
n
g
m
o
d
e
l
b
a
se
d
o
n
B
E
R
T,
”
Fr
o
n
t
i
e
rs i
n
C
o
m
p
u
t
i
n
g
a
n
d
I
n
t
e
l
l
i
g
e
n
t
S
y
s
t
e
m
s
,
v
o
l
.
2
,
n
o
.
3
,
p
p
.
4
0
–
4
3
,
2
0
2
3
,
d
o
i
:
1
0
.
5
4
0
9
7
/
f
c
i
s
.
v
2
i
3
.
5
2
0
9
.
[
8
]
T.
M
.
L
u
u
,
H
.
T.
Le
,
a
n
d
T.
M
.
H
o
a
n
g
,
“
A
h
y
b
r
i
d
m
o
d
e
l
u
s
i
n
g
t
h
e
p
r
e
t
r
a
i
n
e
d
B
ER
T
a
n
d
d
e
e
p
n
e
u
r
a
l
n
e
t
w
o
r
k
s
w
i
t
h
r
i
c
h
f
e
a
t
u
r
e
f
o
r
e
x
t
r
a
c
t
i
v
e
t
e
x
t
su
mm
a
r
i
z
a
t
i
o
n
,
”
J
o
u
rn
a
l
o
f
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
a
n
d
C
y
b
e
r
n
e
t
i
c
s
,
v
o
l
.
3
7
,
n
o
.
2
,
p
p
.
1
2
3
–
1
4
3
,
2
0
2
1
,
d
o
i
:
1
0
.
1
5
6
2
5
/
1
8
1
3
-
9
6
6
3
/
3
7
/
2
/
1
5
9
8
0
.
[
9
]
T.
Zh
a
n
g
,
V
.
K
i
sh
o
r
e
,
F
.
W
u
,
K
.
Q
.
W
e
i
n
b
e
r
g
e
r
,
a
n
d
Y
.
A
r
t
z
i
,
“
B
E
R
TS
C
O
R
E
:
e
v
a
l
u
a
t
i
n
g
t
e
x
t
g
e
n
e
r
a
t
i
o
n
w
i
t
h
B
ER
T,
”
in
8
t
h
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
o
n
L
e
a
r
n
i
n
g
Re
p
res
e
n
t
a
t
i
o
n
s,
I
C
L
R
2
0
2
0
,
p
p
.
1
–
4
3
,
2
0
2
0
.
[
1
0
]
A
.
A
.
D
z
a
k
y
e
t
a
l
.
,
“
O
p
t
i
m
i
z
a
t
i
o
n
C
h
a
t
b
o
t
s
e
r
v
i
c
e
s
b
a
s
e
d
o
n
D
N
N
-
B
ER
T
f
o
r
men
t
a
l
h
e
a
l
t
h
o
f
u
n
i
v
e
r
si
t
y
s
t
u
d
e
n
t
s
,
”
J
o
u
rn
a
l
o
f
Ap
p
l
i
e
d
I
n
f
o
rm
a
t
i
c
s
a
n
d
C
o
m
p
u
t
i
n
g
,
v
o
l
.
8
,
n
o
.
1
,
p
p
.
1
3
–
2
1
,
2
0
2
4
,
d
o
i
:
1
0
.
3
0
8
7
1
/
j
a
i
c
.
v
8
i
1
.
7
4
0
3
.
[
1
1
]
H
.
H
u
z
a
e
n
i
,
Z.
K
.
S
i
m
b
o
l
o
n
,
a
n
d
M
.
A
.
F
i
r
d
a
u
s,
“
M
e
n
t
a
l
h
e
a
l
t
h
d
i
so
r
d
e
r
c
h
a
t
b
o
t
u
si
n
g
N
LP
a
n
d
f
o
r
w
a
r
d
c
h
a
i
n
i
n
g
me
t
h
o
d
s,
c
a
se
st
u
d
y
o
f
C
u
t
M
e
u
t
i
a
H
o
sp
i
t
a
l
,
”
J
o
u
r
n
a
l
o
f
I
n
f
o
rm
a
t
i
o
n
T
e
c
h
n
o
l
o
g
y
a
n
d
C
o
m
p
u
t
e
rs
(
J
I
T
C
)
,
v
o
l
.
4
,
n
o
.
2
,
p
p
.
2
3
1
–
2
3
7
,
2
0
2
4
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
5
,
Octo
b
e
r
20
25
:
4
8
2
9
-
4
8
3
6
4836
[
1
2
]
M
.
S
a
u
n
d
e
r
s,
P
.
L
e
w
i
s
,
a
n
d
A
.
Th
o
r
n
h
i
l
l
,
Re
s
e
a
rc
h
m
e
t
h
o
d
s
f
o
r
b
u
n
i
e
ss s
t
u
d
e
n
t
s
.
P
e
a
r
so
n
,
2
0
0
7
.
[
1
3
]
J.
W
.
C
r
e
ssw
e
l
l
,
Ed
u
c
a
t
i
o
n
a
l
rese
a
r
c
h
p
l
a
n
n
i
n
g
,
c
o
n
d
u
c
t
i
n
g
,
a
n
d
e
v
a
l
u
a
t
i
n
g
q
u
a
n
t
i
t
a
t
i
v
e
a
n
d
q
u
a
l
i
t
a
t
i
v
e
r
e
se
a
rc
h
, 4
th
e
d
.
B
o
s
t
o
n
:
P
h
o
e
n
i
x
C
o
l
o
r
C
o
r
p
,
2
0
1
2
.
[
1
4
]
R
.
S
u
t
o
y
o
,
H
.
L.
H
.
S
.
W
a
r
n
a
r
s,
S
.
M
.
I
sa,
a
n
d
W
.
B
u
d
i
h
a
r
t
o
,
“
Emo
t
i
o
n
a
l
l
y
a
w
a
r
e
C
h
a
t
b
o
t
f
o
r
r
e
s
p
o
n
d
i
n
g
t
o
I
n
d
o
n
e
s
i
a
n
p
r
o
d
u
c
t
r
e
v
i
e
w
s,
”
I
C
I
C
Ex
p
r
e
ss Le
t
t
e
rs
,
v
o
l
.
1
9
,
n
o
.
3
,
p
p
.
8
6
1
–
8
7
6
,
2
0
2
3
,
d
o
i
:
1
0
.
2
4
5
0
7
/
i
j
i
c
i
c
.
1
9
.
0
3
.
8
6
1
.
[
1
5
]
F
.
K
o
t
o
,
J.
H
.
La
u
,
a
n
d
T.
B
a
l
d
w
i
n
,
“
L
i
p
u
t
a
n
6
:
a
l
a
r
g
e
-
s
c
a
l
e
i
n
d
o
n
e
si
a
n
d
a
t
a
s
e
t
f
o
r
t
e
x
t
su
mm
a
r
i
z
a
t
i
o
n
,
”
a
rX
i
v
p
r
e
p
r
i
n
t
a
rXi
v
:
2
0
1
1
.
0
0
6
7
9
,
n
o
.
1
,
2
0
2
0
.
[
1
6
]
N
.
K
.
N
i
ss
a
a
n
d
E.
Y
u
l
i
a
n
t
i
,
“
M
u
l
t
i
-
l
a
b
e
l
t
e
x
t
c
l
a
s
si
f
i
c
a
t
i
o
n
o
f
I
n
d
o
n
e
si
a
n
c
u
st
o
m
e
r
r
e
v
i
e
w
s
u
s
i
n
g
b
i
d
i
r
e
c
t
i
o
n
a
l
e
n
c
o
d
e
r
r
e
p
r
e
s
e
n
t
a
t
i
o
n
s fr
o
m t
r
a
n
sf
o
r
mers
l
a
n
g
u
a
g
e
m
o
d
e
l
,
”
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
rn
a
l
o
f
E
l
e
c
t
r
i
c
a
l
a
n
d
C
o
m
p
u
t
e
r E
n
g
i
n
e
e
r
i
n
g
,
v
o
l
.
1
3
,
n
o
.
5
,
p
p
.
5
6
4
1
–
5
6
5
2
,
2
0
2
3
,
d
o
i
:
1
0
.
1
1
5
9
1
/
i
j
e
c
e
.
v
1
3
i
5
.
p
p
5
6
4
1
-
5
6
5
2
.
[
1
7
]
S
.
W
u
a
n
d
M
.
D
r
e
d
z
e
,
“
A
r
e
a
l
l
l
a
n
g
u
a
g
e
s
c
r
e
a
t
e
d
e
q
u
a
l
i
n
m
u
l
t
i
l
i
n
g
u
a
l
B
ER
T
?
,
”
P
ro
c
e
e
d
i
n
g
s
o
f
t
h
e
A
n
n
u
a
l
Me
e
t
i
n
g
o
f
t
h
e
Asso
c
i
a
t
i
o
n
f
o
r
C
o
m
p
u
t
a
t
i
o
n
a
l
L
i
n
g
u
i
st
i
c
s
,
n
o
.
J
u
l
y
,
p
p
.
1
2
0
–
1
3
0
,
2
0
2
0
,
d
o
i
:
1
0
.
1
8
6
5
3
/
v
1
/
2
0
2
0
.
r
e
p
l
4
n
l
p
-
1
.
1
6
.
[
1
8
]
A
.
K
a
z
e
m
i
,
J.
M
o
z
a
f
a
r
i
,
a
n
d
M
.
A
.
N
e
m
a
t
b
a
k
h
sh
,
“
P
e
r
si
a
n
Q
u
A
D
:
t
h
e
n
a
t
i
v
e
q
u
e
s
t
i
o
n
a
n
sw
e
r
i
n
g
d
a
t
a
s
e
t
f
o
r
t
h
e
P
e
r
si
a
n
l
a
n
g
u
a
g
e
,
”
I
EEE
Ac
c
e
ss
,
v
o
l
.
1
0
,
p
p
.
2
6
0
4
5
–
2
6
0
5
7
,
2
0
2
2
,
d
o
i
:
1
0
.
1
1
0
9
/
A
C
C
ESS
.
2
0
2
2
.
3
1
5
7
2
8
9
.
[
1
9
]
A
.
K
e
s
a
r
w
a
n
i
,
S
.
D
a
s
,
D
.
R
.
K
i
sk
u
,
a
n
d
M
.
D
a
l
u
i
,
“
D
u
a
l
mo
d
e
i
n
f
o
r
ma
t
i
o
n
f
u
si
o
n
w
i
t
h
p
r
e
-
t
r
a
i
n
e
d
C
N
N
m
o
d
e
l
s
a
n
d
t
r
a
n
sf
o
r
m
e
r
f
o
r
v
i
d
e
o
-
b
a
se
d
n
o
n
-
i
n
v
a
s
i
v
e
a
n
a
e
m
i
a
d
e
t
e
c
t
i
o
n
,
”
B
i
o
m
e
d
i
c
a
l
S
i
g
n
a
l
Pr
o
c
e
ssi
n
g
a
n
d
C
o
n
t
r
o
l
,
v
o
l
.
8
8
,
p
.
1
0
5
5
9
2
,
2
0
2
4
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
b
s
p
c
.
2
0
2
3
.
1
0
5
5
9
2
.
[
2
0
]
A
.
C
h
e
n
,
G
.
S
t
a
n
o
v
s
k
y
,
S
.
S
i
n
g
h
,
a
n
d
M
.
G
a
r
d
n
e
r
,
“
E
v
a
l
u
a
t
i
n
g
q
u
e
st
i
o
n
a
n
sw
e
r
i
n
g
e
v
a
l
u
a
t
i
o
n
,
”
M
RQ
A
@
EM
N
L
P
2
0
1
9
-
Pro
c
e
e
d
i
n
g
s
o
f
t
h
e
2
n
d
W
o
rks
h
o
p
o
n
Ma
c
h
i
n
e
R
e
a
d
i
n
g
f
o
r
Q
u
e
st
i
o
n
An
s
w
e
ri
n
g
,
p
p
.
1
1
9
–
1
2
4
,
2
0
1
9
,
d
o
i
:
1
0
.
1
8
6
5
3
/
v
1
/
d
1
9
-
5
8
1
7
.
[
2
1
]
F
.
B
a
h
a
r
u
d
d
i
n
a
n
d
M
.
F
.
N
a
u
f
a
l
,
“
F
i
n
e
-
t
u
n
i
n
g
I
n
d
o
B
E
R
T
f
o
r
I
n
d
o
n
e
si
a
n
e
x
a
m
q
u
e
st
i
o
n
c
l
a
ss
i
f
i
c
a
t
i
o
n
b
a
se
d
o
n
b
l
o
o
m’
s
t
a
x
o
n
o
my
,
”
J
o
u
rn
a
l
o
f
I
n
f
o
rm
a
t
i
o
n
S
y
s
t
e
m
s
E
n
g
i
n
e
e
r
i
n
g
a
n
d
B
u
si
n
e
ss
I
n
t
e
l
l
i
g
e
n
c
e
,
v
o
l
.
9
,
n
o
.
2
,
p
p
.
2
5
3
–
2
6
3
,
2
0
2
3
,
d
o
i
:
1
0
.
2
0
4
7
3
/
j
i
s
e
b
i
.
9
.
2
.
2
5
3
-
2
6
3
.
[
2
2
]
X
.
P
.
Q
i
u
,
T.
X
.
S
u
n
,
Y
.
G
.
X
u
,
Y
.
F
.
S
h
a
o
,
N
.
D
a
i
,
a
n
d
X
.
J
.
H
u
a
n
g
,
“
Pre
-
t
r
a
i
n
e
d
mo
d
e
l
s
f
o
r
n
a
t
u
r
a
l
l
a
n
g
u
a
g
e
p
r
o
c
e
ssi
n
g
:
A
su
r
v
e
y
,
”
S
c
i
e
n
c
e
C
h
i
n
a
T
e
c
h
n
o
l
o
g
i
c
a
l
S
c
i
e
n
c
e
s
,
v
o
l
.
6
3
,
n
o
.
1
0
,
p
p
.
1
8
7
2
–
1
8
9
7
,
2
0
2
0
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
1
4
3
1
-
020
-
1
6
4
7
-
3.
[
2
3
]
M
.
R
.
R
i
z
q
u
l
l
a
h
,
A
.
P
u
r
w
a
r
i
a
n
t
i
,
a
n
d
A
.
F
.
A
j
i
,
“
Q
A
S
i
N
a
:
r
e
l
i
g
i
o
u
s
d
o
ma
i
n
q
u
e
s
t
i
o
n
a
n
sw
e
r
i
n
g
u
si
n
g
S
i
r
a
h
N
a
b
a
w
i
y
a
h
,
”
i
n
2
0
2
3
10
t
h
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
A
d
v
a
n
c
e
d
I
n
f
o
rm
a
t
i
c
s:
C
o
n
c
e
p
t
,
T
h
e
o
ry
a
n
d
A
p
p
l
i
c
a
t
i
o
n
,
I
C
AI
C
T
A
2
0
2
3
,
2
0
2
3
,
p
p
.
1
–
6
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
A
I
C
TA
5
9
2
9
1
.
2
0
2
3
.
1
0
3
9
0
1
2
3
.
[
2
4
]
M
.
S
e
o
,
A
.
K
e
mb
h
a
v
i
,
A
.
F
a
r
h
a
d
i
,
a
n
d
H
.
H
a
j
i
s
h
i
r
z
i
,
“
Bi
-
d
i
r
e
c
t
i
o
n
a
l
a
t
t
e
n
t
i
o
n
f
l
o
w
f
o
r
ma
c
h
i
n
e
c
o
m
p
r
e
h
e
n
si
o
n
,
”
in
5
th
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
L
e
a
r
n
i
n
g
R
e
p
r
e
se
n
t
a
t
i
o
n
s
,
I
C
L
R
2
0
1
7
-
C
o
n
f
e
re
n
c
e
T
r
a
c
k
Pro
c
e
e
d
i
n
g
s
,
p
p
.
1
–
1
3
,
2
0
1
7
,
d
o
i
:
a
b
s
/
1
6
1
1
.
0
1
6
0
3
.
[
2
5
]
B
.
V
a
n
A
k
e
n
,
A
.
L
ö
s
e
r
,
B
.
W
i
n
t
e
r
,
a
n
d
F
.
A
.
G
e
r
s,
“
H
o
w
d
o
e
s
B
E
R
T
a
n
sw
e
r
q
u
e
s
t
i
o
n
s?
A
l
a
y
e
r
-
w
i
s
e
a
n
a
l
y
si
s
o
f
t
r
a
n
sf
o
r
mer
r
e
p
r
e
s
e
n
t
a
t
i
o
n
s
,
”
in
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
I
n
f
o
rm
a
t
i
o
n
a
n
d
K
n
o
w
l
e
d
g
e
M
a
n
a
g
e
m
e
n
t
,
Pro
c
e
e
d
i
n
g
s
,
p
p
.
1
8
2
3
–
1
8
3
2
,
2
0
1
9
,
d
o
i
:
1
0
.
1
1
4
5
/
3
3
5
7
3
8
4
.
3
3
5
8
0
2
8
.
[
2
6
]
A
.
R
a
h
a
l
i
a
n
d
M
.
A
.
A
k
h
l
o
u
f
i
,
“
E
n
d
-
to
-
e
n
d
t
r
a
n
sf
o
r
mer
-
b
a
se
d
m
o
d
e
l
s
i
n
t
e
x
t
u
a
l
-
b
a
se
d
N
LP
,
”
AI
(
S
w
i
t
z
e
rl
a
n
d
)
,
v
o
l
.
4
,
n
o
.
1
,
p
p
.
54
–
1
1
0
,
2
0
2
3
,
d
o
i
:
1
0
.
3
3
9
0
/
a
i
4
0
1
0
0
0
4
.
[
2
7
]
A
.
K
e
sarw
a
n
i
,
S
.
D
a
s,
D
.
R
.
K
i
s
k
u
,
a
n
d
M
.
D
a
l
u
i
,
“
N
o
n
-
i
n
v
a
si
v
e
a
n
a
e
mi
a
d
e
t
e
c
t
i
o
n
b
a
s
e
d
o
n
p
a
l
m
p
a
l
l
o
r
v
i
d
e
o
u
si
n
g
t
r
e
e
-
st
r
u
c
t
u
r
e
d
3
D
C
N
N
a
n
d
v
i
s
i
o
n
t
r
a
n
sf
o
r
mer
m
o
d
e
l
s
,
”
J
o
u
r
n
a
l
o
f
Ex
p
e
ri
m
e
n
t
a
l
&
T
h
e
o
re
t
i
c
a
l
A
rt
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
,
p
p
.
1
–
2
9
,
d
o
i
:
1
0
.
1
0
8
0
/
0
9
5
2
8
1
3
X
.
2
0
2
3
.
2
3
0
1
4
0
1
.
[
2
8
]
R
.
C
a
l
i
z
z
a
n
o
,
M
.
O
st
e
n
d
o
r
f
f
,
Q
.
R
u
a
n
,
a
n
d
G
.
R
e
h
m,
“
G
e
n
e
r
a
t
i
n
g
e
x
t
e
n
d
e
d
a
n
d
m
u
l
t
i
l
i
n
g
u
a
l
s
u
mm
a
r
i
e
s
w
i
t
h
p
r
e
-
t
r
a
i
n
e
d
t
r
a
n
sf
o
r
mers
,
”
in
2
0
2
2
L
a
n
g
u
a
g
e
Re
s
o
u
r
c
e
s
a
n
d
E
v
a
l
u
a
t
i
o
n
C
o
n
f
e
r
e
n
c
e
,
L
REC
2
0
2
2
,
n
o
.
Ju
n
e
,
p
p
.
1
6
4
0
–
1
6
5
0
,
2
0
2
2
.
[
2
9
]
A
.
F
a
t
w
a
n
t
o
,
F
.
Za
m
a
k
h
s
y
a
r
i
,
R
.
N
d
u
n
g
i
,
a
n
d
L.
F
i
t
r
i
y
a
n
i
,
“
S
y
st
e
ma
t
i
c
l
i
t
e
r
a
t
u
r
e
r
e
v
i
e
w
o
f
B
ER
T
-
b
a
s
e
d
m
o
d
e
l
s
f
o
r
n
a
t
u
r
a
l
l
a
n
g
u
a
g
e
p
r
o
c
e
ssi
n
g
t
a
sk
s,
”
I
n
f
o
t
e
l
,
p
p
.
7
1
3
–
7
2
8
,
2
0
2
4
,
d
o
i
:
1
0
.
2
0
8
9
5
/
I
N
F
O
TEL.
V
1
6
I
3
.
1
2
0
6
.
B
I
O
G
RAP
H
I
E
S O
F
AUTH
O
RS
Fa
r
d
a
n
Z
a
m
a
k
h
sy
a
r
i
s
tu
d
y
i
n
g
t
o
g
e
t
a
b
a
c
h
e
l
o
r
o
f
c
o
m
p
u
ter
sc
ien
c
e
(S
.
Ko
m
.
)
d
e
g
re
e
in
in
f
o
rm
a
ti
o
n
sy
ste
m
s
,
a
t
S
u
n
a
n
Am
p
e
l
S
tate
Isla
m
ic
Un
iv
e
rsity
S
u
ra
b
a
y
a
in
2
0
1
7
,
c
o
n
ti
n
u
i
n
g
h
is
M
a
ste
r
o
f
C
o
m
p
u
ter
S
c
ien
c
e
(M
.
K
o
m
.
)
e
d
u
c
a
t
io
n
i
n
i
n
fo
rm
a
ti
c
s,
S
u
n
a
n
Ka
li
jag
a
S
tate
Isla
m
ic
Un
iv
e
rsity
Yo
g
y
a
k
a
rta
i
n
2
0
2
4
.
C
u
rre
n
tl
y
wo
r
k
in
g
a
s
a
lec
tu
re
r
a
t
th
e
De
p
a
rtme
n
t
o
f
In
f
o
rm
a
ti
c
s,
F
a
c
u
lt
y
o
f
E
n
g
in
e
e
rin
g
,
S
e
k
o
lah
T
in
g
g
i
Ca
h
a
y
a
S
u
ry
a
,
Ke
d
iri
,
I
n
d
o
n
e
sia
.
His
re
se
a
rc
h
in
tere
sts
i
n
c
lu
d
e
we
b
d
e
v
e
l
o
p
m
e
n
t
,
d
a
ta
sc
ien
c
e
,
n
a
tu
ra
l
lan
g
u
a
g
e
p
r
o
c
e
ss
in
g
,
a
n
d
a
rt
ifi
c
ial
in
telli
g
e
n
c
e
.
He
c
a
n
b
e
c
o
n
tac
ted
v
ia
e
m
a
il
:
m
a
sfa
rd
a
n
9
9
@g
m
a
il
.
c
o
m
.
A
g
u
n
g
F
a
twa
n
t
o
o
b
tai
n
e
d
a
b
a
c
h
e
lo
r
o
f
sc
ien
c
e
(S
.
S
i.
)
in
c
o
m
p
u
ter
sc
ien
c
e
a
n
d
m
a
ste
r’s
de
g
re
e
(M
.
Ko
m
.
)
i
n
C
o
m
p
u
ter
S
c
ien
c
e
fro
m
G
a
d
jah
M
a
d
a
Un
i
v
e
rsity
.
He
late
r
p
u
rsu
e
d
h
is
d
o
c
t
o
ra
l
e
d
u
c
a
ti
o
n
a
t
th
e
De
p
a
rtme
n
t
o
f
Co
m
p
u
ter
S
c
ien
c
e
,
Au
stra
li
a
n
Na
ti
o
n
a
l
Un
iv
e
rsity
.
He
is
c
u
rre
n
tl
y
a
lec
tu
re
r
a
t
th
e
In
f
o
rm
a
ti
c
s
De
p
a
rtm
e
n
t,
F
a
c
u
lt
y
o
f
S
c
ien
c
e
a
n
d
Tec
h
n
o
l
o
g
y
,
Un
i
v
e
rsitas
Isla
m
Ne
g
e
ri
S
u
n
a
n
Ka
li
jag
a
,
Yo
g
y
a
k
a
rta,
In
d
o
n
e
sia
.
His
re
se
a
r
c
h
in
tere
sts
in
c
lu
d
e
so
ftwa
re
e
n
g
in
e
e
rin
g
,
n
a
tu
ra
l
lan
g
u
a
g
e
p
ro
c
e
ss
in
g
,
a
rti
ficia
l
in
telli
g
e
n
c
e
a
n
d
d
a
ta sc
ien
c
e
.
He
c
a
n
b
e
c
o
n
tac
ted
v
ia em
a
il
:
a
g
u
n
g
.
fa
twa
n
to
@
u
in
-
s
u
k
a
.
a
c
.
id
.
Evaluation Warning : The document was created with Spire.PDF for Python.