I
AE
S In
t
er
na
t
io
na
l J
o
urna
l o
f
Art
if
icia
l In
t
ellig
ence
(
I
J
-
AI
)
Vo
l.
15
,
No
.
1
,
Feb
r
u
ar
y
20
26
,
p
p
.
909
~
918
I
SS
N:
2
2
5
2
-
8
9
3
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijai.v
15
.i
1
.
p
p
9
0
9
-
9
1
8
909
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
i
.
ia
esco
r
e.
co
m
Deep
lea
rni
ng
-
ba
sed spa
m det
ectio
n
for W
ha
tsApp
c
ha
tbot
fallba
ck redu
ctio
n
Sa
t
rio
Sa
dewo
,
Am
a
lia
Z
a
hra
D
e
p
a
r
t
me
n
t
o
f
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
,
B
I
N
U
S
G
r
a
d
u
a
t
e
P
r
o
g
r
a
m
-
M
a
s
t
e
r
o
f
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
,
B
i
n
a
N
u
s
a
n
t
a
r
a
U
n
i
v
e
r
si
t
y
,
J
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
Sep
22
,
2
0
2
5
R
ev
is
ed
J
an
2
,
2
0
2
6
Acc
ep
ted
J
an
22
,
2
0
2
6
Ch
a
tb
o
ts
o
n
Wh
a
tsAp
p
a
re
wi
d
e
ly
u
se
d
fo
r
c
u
sto
m
e
r
se
rv
ice
,
b
u
t
th
e
i
r
e
ffe
c
ti
v
e
n
e
ss
is
o
ften
u
n
d
e
rm
in
e
d
b
y
fa
ll
b
a
c
k
re
sp
o
n
se
s
wh
e
n
u
se
r
in
p
u
t
c
a
n
n
o
t
b
e
u
n
d
e
rsto
o
d
.
A
m
a
jo
r
c
a
u
se
o
f
th
e
se
fa
ll
b
a
c
k
s
is
u
n
so
li
c
it
e
d
sp
a
m
,
wh
ich
d
isr
u
p
ts
i
n
tera
c
ti
o
n
s
a
n
d
r
e
d
u
c
e
s
se
rv
ice
q
u
a
li
ty
.
Th
is
stu
d
y
d
e
v
e
l
o
p
s
a
n
d
e
v
a
lu
a
tes
a
sp
a
m
d
e
tec
ti
o
n
sy
ste
m
a
ime
d
a
t
re
d
u
c
in
g
fa
ll
b
a
c
k
ra
tes
a
n
d
e
n
h
a
n
c
in
g
u
se
r
e
x
p
e
rien
c
e
.
A
c
o
m
p
a
ra
ti
v
e
a
n
a
ly
sis
wa
s
c
o
n
d
u
c
te
d
b
e
twe
e
n
trad
it
io
n
a
l
m
a
c
h
in
e
lea
rn
i
n
g
m
o
d
e
ls
(
su
p
p
o
rt
v
e
c
to
r
m
a
c
h
in
e
(S
VM)
a
n
d
d
e
c
isio
n
tree
(DT)
)
a
n
d
a
d
v
a
n
c
e
d
d
e
e
p
lea
rn
in
g
a
rc
h
it
e
c
tu
re
s,
i
n
c
lu
d
in
g
lo
n
g
sh
o
rt
-
term
m
e
m
o
ry
(LS
TM
)
v
a
rian
ts
(
v
a
n
il
la,
b
id
irec
ti
o
n
a
l,
sta
c
k
e
d
,
c
o
n
v
o
lu
ti
o
n
a
l
n
e
u
ra
l
n
e
two
r
k
(
CNN
)
-
LS
TM
,
a
n
d
e
n
c
o
d
e
r
-
d
e
c
o
d
e
r
)
a
n
d
tran
sfo
rm
e
r
-
b
a
se
d
m
o
d
e
ls
(
b
i
d
irec
ti
o
n
a
l
e
n
c
o
d
e
r
re
p
re
se
n
tatio
n
s
fr
o
m
tran
sfo
rm
e
rs
(
BERT
)
-
b
a
se
,
Distil
BERT
,
a
n
d
c
ro
ss
-
li
n
g
u
a
l
lan
g
u
a
g
e
m
o
d
e
l
-
ro
b
u
stly
o
p
ti
m
ize
d
BERT
p
re
tra
in
in
g
a
p
p
r
o
a
c
h
(
XLM
-
ROBERT
a
)
).
Us
in
g
1
7
0
,
0
0
0
m
e
ss
a
g
e
s
sa
m
p
led
fro
m
1
8
m
il
li
o
n
in
tera
c
ti
o
n
s
c
o
ll
e
c
ted
b
e
twe
e
n
Ju
ly
2
0
2
2
a
n
d
De
c
e
m
b
e
r
2
0
2
3
,
th
e
m
o
d
e
ls
we
re
a
ss
e
s
se
d
with
sta
n
d
a
rd
e
v
a
lu
a
ti
o
n
m
e
tri
c
s.
Re
su
lt
s
sh
o
w
th
a
t
CNN
-
LS
T
M
a
n
d
Disti
lBE
RT
a
c
h
iev
e
d
th
e
m
o
st
ro
b
u
st
p
e
rfo
rm
a
n
c
e
.
CNN
-
LS
TM
a
tt
a
in
e
d
a
p
re
c
isio
n
o
f
0
.
9
2
,
re
c
a
ll
o
f
0
.
9
1
,
F
1
-
sc
o
re
o
f
0
.
9
1
,
a
n
d
a
c
c
u
ra
c
y
o
f
0
.
9
4
,
wh
il
e
Di
stil
BERT
a
c
h
iev
e
d
p
re
c
isio
n
o
f
0
.
9
2
,
re
c
a
ll
o
f
0
.
8
9
,
F
1
-
sc
o
re
o
f
0
.
9
0
,
a
n
d
a
c
c
u
ra
c
y
o
f
0
.
9
3
.
T
h
e
se
fin
d
in
g
s
h
i
g
h
l
ig
h
t
t
h
e
ir
su
p
e
ri
o
r
a
b
i
li
ty
to
c
a
p
t
u
re
c
o
n
tex
t
u
a
l
p
a
tt
e
rn
s
i
n
sp
a
m
m
e
ss
a
g
e
s.
Im
p
lem
e
n
ti
n
g
su
c
h
m
o
d
e
ls
is
e
x
p
e
c
ted
to
sig
n
ifi
c
a
n
t
ly
lo
we
r
fa
ll
b
a
c
k
ra
tes
,
t
h
e
re
b
y
imp
ro
v
i
n
g
c
h
a
tb
o
t
re
li
a
b
il
it
y
a
n
d
u
se
r
sa
ti
sfa
c
ti
o
n
.
K
ey
w
o
r
d
s
:
Dee
p
lear
n
in
g
Fallb
ac
k
r
esp
o
n
s
e
Ma
ch
in
e
lear
n
in
g
Natu
r
al
lan
g
u
ag
e
p
r
o
ce
s
s
in
g
Sp
am
d
etec
tio
n
T
r
an
s
f
o
r
m
e
r
m
o
d
els
W
h
atsAp
p
ch
atb
o
t
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
:
Satr
io
Sad
ewo
Dep
ar
tm
en
t o
f
C
o
m
p
u
ter
Scie
n
ce
,
B
I
NUS
Gr
ad
u
ate
Pro
g
r
a
m
-
Ma
s
ter
o
f
C
o
m
p
u
te
r
Scien
c
e
B
in
a
Nu
s
an
tar
a
Un
iv
er
s
ity
J
ak
ar
ta,
I
n
d
o
n
esia
E
m
ail:
s
atr
io
.
s
ad
ewo
@
b
in
u
s
.
ac
.
id
1.
I
NT
RO
D
UCT
I
O
N
I
n
t
h
e
co
n
t
em
p
o
r
ar
y
d
ig
i
ta
l
l
an
d
s
c
ap
e
,
W
h
a
t
s
A
p
p
h
a
s
e
s
t
ab
l
i
s
h
ed
i
t
s
e
lf
a
s
a
c
r
i
t
ic
a
l
c
o
m
m
u
n
ic
a
t
i
o
n
m
ed
i
u
m
,
s
er
v
i
n
g
a
g
lo
b
a
l
u
s
e
r
b
a
s
e
ex
c
e
ed
i
n
g
t
wo
b
i
l
l
io
n
[
1
]
.
T
h
e
in
t
eg
r
a
t
i
o
n
o
f
ch
a
t
b
o
t
s
o
n
t
o
th
i
s
p
l
a
t
f
o
r
m
h
a
s
p
r
e
c
ip
i
t
a
t
e
d
a
p
ar
ad
i
g
m
s
h
if
t
in
cu
s
t
o
m
er
s
er
v
i
c
e,
m
a
r
k
e
t
in
g
s
t
r
a
t
e
g
i
e
s
,
an
d
u
s
er
e
n
g
ag
e
m
e
n
t
p
r
o
to
c
o
l
s
[
2
]
.
A
lt
h
o
u
g
h
ch
a
tb
o
t
s
p
r
e
s
e
n
t
co
n
s
i
d
e
r
ab
l
e
a
d
v
an
t
a
g
e
s
,
c
o
n
t
r
i
b
u
t
i
n
g
t
o
an
a
v
er
ag
e
r
e
v
en
u
e
in
cr
e
a
s
e
o
f
6
7
%
a
n
d
a
c
c
o
u
n
t
in
g
f
o
r
2
6
%
o
f
to
t
a
l
s
a
l
e
s
i
n
t
e
r
ac
t
i
o
n
s
[
3
]
,
th
e
i
r
o
p
er
a
t
i
o
n
a
l
e
f
f
i
c
a
cy
i
s
f
r
e
q
u
en
t
l
y
i
m
p
ed
e
d
b
y
t
e
ch
n
i
c
a
l
l
i
m
i
t
a
t
io
n
s
[
4
]
.
A
p
r
ed
o
m
in
an
t
i
s
s
u
e
i
s
t
h
e
h
i
g
h
r
a
te
o
f
"f
a
l
l
b
a
c
k
"
r
e
s
p
o
n
s
e
s
,
wh
e
r
e
in
th
e
ch
a
t
b
o
t
i
s
u
n
a
b
le
t
o
i
n
t
er
p
r
e
t
a
u
s
er
's
q
u
e
r
y
,
co
n
s
eq
u
en
t
ly
d
e
l
iv
e
r
i
n
g
a
g
e
n
e
r
i
c
a
n
d
u
n
h
e
l
p
f
u
l
r
ep
l
y
.
T
h
i
s
p
h
en
o
m
e
n
o
n
n
o
t
o
n
ly
d
i
m
i
n
i
s
h
e
s
th
e
u
s
e
r
e
x
p
e
r
ie
n
c
e
b
u
t
a
l
s
o
s
i
g
n
i
f
i
e
s
u
n
d
er
l
y
in
g
a
r
c
h
i
t
ec
t
u
r
a
l
o
r
t
r
a
in
i
n
g
i
n
ef
f
i
c
i
e
n
c
ie
s
[
5
]
.
A
p
r
i
n
c
ip
a
l
f
a
c
t
o
r
co
n
t
r
ib
u
t
in
g
to
th
e
h
i
g
h
r
a
t
e
o
f
f
a
l
l
b
a
ck
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2252
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
1
,
Feb
r
u
ar
y
20
26
:
909
-
9
1
8
910
r
e
s
p
o
n
s
e
s
i
s
t
h
e
p
e
r
v
a
s
i
v
en
e
s
s
o
f
u
n
s
o
l
i
c
i
t
ed
s
p
a
m
m
e
s
s
a
g
e
s
.
S
u
ch
m
e
s
s
a
g
e
s
d
i
s
r
u
p
t
th
e
in
t
e
n
d
e
d
c
o
n
v
e
r
s
a
t
io
n
a
l
tr
a
j
e
c
to
r
y
,
tr
i
g
g
e
r
n
o
n
-
p
r
o
d
u
c
t
iv
e
f
a
l
l
b
a
c
k
r
e
s
p
o
n
s
e
s
,
a
n
d
m
ay
in
t
r
o
d
u
c
e
s
e
c
u
r
i
t
y
v
u
l
n
e
r
ab
i
l
i
t
i
e
s
,
i
n
c
lu
d
in
g
p
h
is
h
i
n
g
a
t
t
a
ck
s
[
6
]
,
[
7
]
.
T
h
i
s
r
es
e
a
r
c
h
c
o
n
f
r
o
n
t
s
th
e
ch
a
l
l
en
g
e
o
f
s
p
a
m
-
in
d
u
c
ed
f
a
l
l
b
a
ck
r
e
s
p
o
n
s
e
s
w
i
t
h
in
a
W
h
a
t
s
A
p
p
ch
a
tb
o
t
ec
o
s
y
s
t
em
.
T
h
e
i
n
v
e
s
t
i
g
a
t
io
n
l
e
v
e
r
ag
e
s
a
s
u
b
s
t
an
t
i
a
l
d
a
t
a
s
e
t
o
f
1
8
m
i
l
l
i
o
n
m
e
s
s
ag
e
s
,
co
l
l
e
c
t
e
d
f
r
o
m
an
o
p
er
a
t
i
o
n
a
l
c
h
a
tb
o
t
s
er
v
i
c
e,
o
f
wh
i
c
h
6
m
i
l
l
io
n
(
3
3
%
)
c
u
l
m
in
a
t
e
d
in
a
f
a
l
l
b
a
c
k
.
T
h
e
c
en
t
r
a
l
o
b
j
e
c
t
iv
e
i
s
to
e
n
g
in
e
e
r
a
r
o
b
u
s
t
s
p
a
m
d
e
te
c
t
i
o
n
m
o
d
e
l
c
ap
ab
l
e
o
f
r
e
d
u
c
i
n
g
th
e
f
a
l
l
b
ac
k
r
a
t
e
t
o
a
t
a
r
g
e
t
t
h
r
e
s
h
o
l
d
o
f
1
5
%
o
r
l
es
s
.
Prio
r
r
esear
ch
h
as
af
f
ir
m
ed
th
e
u
tili
ty
o
f
m
ac
h
in
e
lear
n
in
g
alg
o
r
ith
m
s
s
u
c
h
as
d
ec
is
io
n
tr
ee
s
(
DT
)
an
d
s
u
p
p
o
r
t
v
ec
to
r
m
ac
h
in
e
(
SVM)
f
o
r
s
p
am
d
etec
tio
n
ac
r
o
s
s
v
ar
io
u
s
p
latf
o
r
m
s
,
i
n
clu
d
in
g
e
m
ail
an
d
SMS
[
8
]
,
[
9
]
.
Mo
r
e
r
ec
en
tl
y
,
d
ee
p
lear
n
in
g
m
o
d
els,
n
o
tab
ly
lo
n
g
s
h
o
r
t
-
ter
m
m
e
m
o
r
y
(
L
STM
)
a
n
d
b
id
ir
ec
tio
n
al
e
n
co
d
e
r
r
e
p
r
esen
tatio
n
s
f
r
o
m
tr
an
s
f
o
r
m
er
s
(
B
E
R
T
)
,
h
av
e
d
em
o
n
s
tr
ated
s
u
p
e
r
io
r
p
e
r
f
o
r
m
an
ce
i
n
m
an
ag
in
g
th
e
c
o
m
p
lex
ity
a
n
d
co
n
tex
tu
al
n
u
an
ce
s
o
f
n
at
u
r
al
lan
g
u
ag
e
,
attain
in
g
h
ig
h
ac
cu
r
ac
y
in
s
p
am
class
if
icatio
n
task
s
[
1
0
]
,
[
1
1
]
.
T
o
p
r
o
v
id
e
a
clea
r
b
en
ch
m
a
r
k
o
f
th
e
cu
r
r
en
t
s
tate
-
of
-
t
h
e
-
ar
t a
n
d
h
i
g
h
lig
h
t
t
h
ese
co
m
p
ar
ativ
e
f
in
d
i
n
g
s
,
T
ab
le
1
p
r
esen
ts
a
co
m
p
a
r
ativ
e
tax
o
n
o
m
y
o
f
r
ec
e
n
t
s
p
am
d
etec
tio
n
s
tu
d
ies
ac
r
o
s
s
v
ar
io
u
s
p
latf
o
r
m
s
an
d
m
eth
o
d
o
lo
g
ies.
T
h
is
s
tu
d
y
ad
d
r
ess
es
a
cr
itica
l
g
ap
b
y
ev
alu
atin
g
s
p
am
d
ete
ctio
n
m
o
d
els
s
p
ec
if
ically
o
n
W
h
atsAp
p
ch
atb
o
t
d
ata,
wh
ich
p
r
esen
ts
a
u
n
iq
u
e
an
d
n
o
n
-
tr
iv
ial
c
h
allen
g
e
co
m
p
ar
ed
to
p
r
ev
io
u
s
ly
s
tu
d
ied
d
o
m
ain
s
lik
e
em
ail
an
d
SMS.
Un
lik
e
tr
ad
iti
o
n
al
s
p
am
,
W
h
atsAp
p
s
p
am
i
s
ch
ar
ac
ter
ized
b
y
ex
tr
em
e
b
r
ev
ity
,
h
ea
v
y
u
s
e
o
f
in
f
o
r
m
al
lan
g
u
ag
e
a
n
d
s
lan
g
,
an
d
co
n
te
x
tu
al
m
im
icr
y
t
h
at
o
f
ten
im
itates
leg
itima
te
u
s
er
q
u
er
ies
to
ev
a
d
e
d
etec
tio
n
[
1
2
]
.
T
h
ese
ch
a
r
ac
ter
is
tics
r
en
d
er
m
an
y
f
r
eq
u
e
n
c
y
-
b
ased
m
ac
h
in
e
lear
n
in
g
f
ea
tu
r
es
less
ef
f
ec
tiv
e
an
d
n
ec
ess
itate
ad
v
an
ce
d
d
e
ep
lear
n
in
g
m
o
d
els
ca
p
a
b
le
o
f
u
n
d
er
s
tan
d
in
g
n
u
an
ce
d
,
co
n
tex
t
-
d
ep
en
d
en
t
p
atter
n
s
.
T
h
e
p
r
im
ar
y
c
o
n
tr
ib
u
tio
n
o
f
th
is
wo
r
k
is
a
r
ig
o
r
o
u
s
,
d
ir
ec
t
c
o
m
p
ar
is
o
n
o
f
tr
ad
i
tio
n
al
an
d
ad
v
a
n
ce
d
d
ee
p
lear
n
in
g
m
o
d
els
o
n
a
la
r
g
e
-
s
ca
le,
r
ea
l
-
wo
r
ld
d
ataset
to
id
en
tify
th
e
m
o
s
t
r
o
b
u
s
t
a
r
ch
itectu
r
e
f
o
r
t
h
is
s
p
ec
if
ic,
ch
allen
g
in
g
e
n
v
ir
o
n
m
en
t.
T
o
p
r
o
v
i
d
e
a
clea
r
s
tr
u
ctu
r
e
f
o
r
th
is
in
v
esti
g
atio
n
,
th
is
p
ap
er
aim
s
to
an
s
wer
two
p
r
im
ar
y
r
esear
ch
q
u
esti
o
n
s
:
f
i
r
s
t,
(
R
Q1
)
h
o
w
ca
n
s
p
am
m
ess
ag
es
th
at
ca
u
s
e
f
allb
ac
k
s
in
a
W
h
atsAp
p
ch
atb
o
t
b
e
ef
f
ec
tiv
ely
d
etec
ted
u
s
in
g
tr
ad
itio
n
a
l
m
ac
h
in
e
lear
n
in
g
(
SVM
an
d
DT
)
v
er
s
u
s
d
ee
p
lear
n
in
g
m
o
d
els
(
L
STM
v
ar
ia
n
ts
,
an
d
B
E
R
T
v
ar
ia
n
ts
)
?;
an
d
s
ec
o
n
d
,
(
R
Q2
)
t
o
wh
at
e
x
te
n
t
d
o
d
ee
p
lear
n
in
g
m
o
d
els
o
u
tp
er
f
o
r
m
t
r
ad
itio
n
a
l
m
ac
h
in
e
lear
n
in
g
m
o
d
els
in
m
in
im
izin
g
th
e
f
allb
ac
k
r
a
te
th
r
o
u
g
h
s
u
p
er
io
r
s
p
am
class
if
icatio
n
o
n
th
is
u
n
i
q
u
e
d
ataset?
T
ab
le
1
.
C
o
m
p
a
r
ativ
e
tax
o
n
o
m
y
o
f
s
tate
-
of
-
th
e
-
ar
t sp
a
m
d
e
tectio
n
m
o
d
els
Ti
t
l
e
/
a
u
t
h
o
r
(
s)
P
l
a
t
f
o
r
m
M
e
t
h
o
d
(
s)
Ev
a
l
u
a
t
i
o
n
me
t
r
i
c
s
K
e
y
r
e
su
l
t
(
s)
Emai
l
s
p
a
m
d
e
t
e
c
t
i
o
n
u
s
i
n
g
d
e
e
p
l
e
a
r
n
i
n
g
a
p
p
r
o
a
c
h
[
6
]
Emai
l
LSTM
,
B
i
LST
M
,
B
ER
T
A
c
c
u
r
a
c
y
,
p
r
e
c
i
s
i
o
n
,
r
e
c
a
l
l
,
F1
-
sc
o
r
e
B
ER
T:
9
9
.
1
4
%
(
h
i
g
h
e
st
a
c
c
u
r
a
c
y
)
S
M
S
s
p
a
m
d
e
t
e
c
t
i
o
n
u
s
i
n
g
m
a
c
h
i
n
e
l
e
a
r
n
i
n
g
a
n
d
d
e
e
p
l
e
a
r
n
i
n
g
t
e
c
h
n
i
q
u
e
s
[
1
0
]
S
M
S
n
a
i
v
e
B
a
y
e
s,
LR
,
R
F
,
S
V
M
,
K
N
N
,
D
T
,
LSTM
P
r
e
c
i
s
i
o
n
,
r
e
c
a
l
l
,
a
c
c
u
r
a
c
y
LSTM
:
9
8
.
5
%
(
h
i
g
h
e
st
a
c
c
u
r
a
c
y
)
E
-
mai
l
s
p
a
m
d
e
t
e
c
t
i
o
n
u
s
i
n
g
mac
h
i
n
e
l
e
a
r
n
i
n
g
[
1
2
]
Emai
l
D
T,
R
F
,
n
a
i
v
e
B
a
y
e
s,
S
V
M
,
L
R
,
M
LP
A
c
c
u
r
a
c
y
M
LP:
9
8
%
(
h
i
g
h
e
st
a
c
c
u
r
a
c
y
)
Emai
l
s
p
a
m
c
l
a
ss
i
f
i
c
a
t
i
o
n
u
si
n
g
D
i
st
i
l
B
ER
T
[
1
3
]
Emai
l
D
i
st
i
l
B
ER
T
A
c
c
u
r
a
c
y
D
i
st
i
l
B
ER
T
:
9
7
.
8
4
%
(
v
a
l
i
d
a
t
i
o
n
a
c
c
u
r
a
c
y
)
LSTM
n
e
t
w
o
r
k
s f
o
r
e
mai
l
s
p
a
m
c
l
a
ss
i
f
i
c
a
t
i
o
n
[
1
4
]
Emai
l
LSTM
A
c
c
u
r
a
c
y
LSTM
:
9
7
.
4
%
a
c
c
u
r
a
c
y
A
c
o
m
p
r
e
h
e
n
s
i
v
e
r
e
v
i
e
w
o
n
e
ma
i
l
sp
a
m
c
l
a
ssi
f
i
c
a
t
i
o
n
u
si
n
g
m
a
c
h
i
n
e
l
e
a
r
n
i
n
g
a
l
g
o
r
i
t
h
ms
[
9
]
Emai
l
S
V
M
,
n
a
i
v
e
B
a
y
e
s
,
D
T,
R
F
,
n
e
u
r
a
l
n
e
t
w
o
r
k
s
A
c
c
u
r
a
c
y
,
p
r
e
c
i
s
i
o
n
,
r
e
c
a
l
l
S
V
M
:
9
8
.
3
2
%
(
h
i
g
h
e
st
a
c
c
u
r
a
c
y
)
S
M
S
s
p
a
m
c
l
a
ssi
f
i
c
a
t
i
o
n
u
si
n
g
mac
h
i
n
e
l
e
a
r
n
i
n
g
t
e
c
h
n
i
q
u
e
s
[
8
]
S
M
S
n
a
i
v
e
B
a
y
e
s,
LR
,
S
V
M
,
RF
A
c
c
u
r
a
c
y
S
V
M
:
9
8
.
7
9
%
(
h
i
g
h
e
st
a
c
c
u
r
a
c
y
)
2.
M
E
T
H
O
D
T
h
e
r
esear
ch
m
eth
o
d
o
lo
g
y
was sy
s
tem
atica
lly
d
esig
n
ed
to
f
ac
ilit
ate
th
e
d
ev
elo
p
m
en
t a
n
d
s
u
b
s
eq
u
en
t
ev
alu
atio
n
o
f
a
s
p
am
d
etec
tio
n
m
o
d
el.
As
d
etailed
i
n
th
e
f
o
llo
win
g
s
ec
tio
n
s
,
th
is
p
r
o
ce
s
s
b
eg
an
with
d
ata
co
llectio
n
an
d
lab
elin
g
,
f
o
llo
wed
b
y
p
r
ep
r
o
ce
s
s
in
g
,
m
o
d
el
b
u
ild
in
g
,
h
y
p
er
p
ar
a
m
eter
tu
n
in
g
,
an
d
f
in
ally
,
a
co
m
p
r
eh
e
n
s
iv
e
m
o
d
el
ev
alu
at
io
n
.
T
h
e
p
r
o
ce
d
u
r
al
f
r
am
ew
o
r
k
,
d
e
p
icted
in
Fig
u
r
e
1
,
co
m
m
en
ce
d
with
d
ata
ac
q
u
is
itio
n
an
d
co
n
clu
d
ed
wit
h
th
is
co
m
p
r
eh
en
s
iv
e
e
v
alu
ati
o
n
o
f
th
e
r
esu
ltan
t
m
o
d
els.
Fi
g
u
r
e
1
illu
s
tr
atin
g
th
e
s
eq
u
en
tial step
s
f
r
o
m
d
ata
co
llectio
n
an
d
a
n
n
o
tatio
n
to
m
o
d
el
ev
alu
atio
n
.
2
.
1
.
Da
t
a
c
o
llect
io
n a
nd
prepa
ra
t
io
n
T
h
e
d
a
t
a
s
e
t
w
a
s
p
r
o
c
u
r
e
d
f
r
o
m
t
h
e
i
n
t
e
r
a
c
ti
o
n
l
o
g
s
o
f
a
W
h
a
t
s
A
p
p
c
h
a
t
b
o
t
,
c
o
v
e
r
i
n
g
t
h
e
p
e
r
i
o
d
f
r
o
m
J
u
l
y
2
0
2
2
t
o
D
e
c
e
m
b
e
r
2
0
2
3
a
n
d
e
n
c
o
m
p
a
s
s
i
n
g
1
8
m
i
l
li
o
n
m
e
s
s
a
g
es
.
F
o
r
t
h
e
p
u
r
p
o
s
e
o
f
t
h
is
s
t
u
d
y
,
a
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Dee
p
lea
r
n
in
g
-
b
a
s
ed
s
p
a
m
d
et
ec
tio
n
fo
r
Wh
a
ts
A
p
p
ch
a
tb
o
t f
a
llb
a
ck
r
ed
u
ctio
n
(
S
a
tr
io
S
a
d
e
w
o
)
911
r
e
p
r
e
s
e
n
t
a
ti
v
e
s
a
m
p
l
e
o
f
1
7
0
,
0
0
0
m
e
s
s
a
g
e
s
t
h
a
t h
a
d
t
r
i
g
g
e
r
e
d
a
f
a
ll
b
a
c
k
r
es
p
o
n
s
e
w
as
is
o
l
ate
d
.
T
h
i
s
d
a
t
as
e
t w
as
s
u
b
s
e
q
u
e
n
t
l
y
p
a
r
ti
t
i
o
n
e
d
i
n
t
o
t
r
a
i
n
i
n
g
(
7
0
%
)
,
v
a
l
i
d
a
ti
o
n
(
1
5
%
)
,
a
n
d
t
e
s
t
i
n
g
(
1
5
%
)
s
u
b
s
e
ts
[
1
5
]
.
T
h
e
p
a
r
t
i
t
i
o
n
i
n
g
w
a
s
p
e
r
f
o
r
m
e
d
c
h
r
o
n
o
l
o
g
i
c
a
l
l
y
t
o
r
i
g
o
r
o
u
s
l
y
a
s
s
es
s
t
h
e
m
o
d
el'
s
c
a
p
a
c
it
y
f
o
r
t
e
m
p
o
r
a
l
g
e
n
e
r
a
li
z
a
t
i
o
n
.
A
r
ig
o
r
o
u
s
an
n
o
tatio
n
p
r
o
ce
s
s
was
th
en
co
n
d
u
cte
d
b
y
a
p
an
el
o
f
th
r
ee
d
o
m
ain
ex
p
er
ts
(
co
m
p
o
s
ed
o
f
o
n
e
d
ata
s
cien
tis
t
an
d
two
p
r
o
d
u
ct
o
wn
er
r
e
p
r
esen
tativ
es)
to
class
if
y
ea
ch
m
ess
ag
e
a
s
eith
er
'
s
p
am
'
o
r
'
n
o
n
-
s
p
am
'
[
1
6
]
.
T
h
is
p
r
o
ce
s
s
was
g
o
v
er
n
ed
b
y
a
s
tr
ict
s
et
o
f
class
if
icatio
n
r
u
les
d
ef
in
ed
in
co
llab
o
r
atio
n
wit
h
th
e
p
r
o
d
u
ct
o
wn
er
t
o
en
s
u
r
e
d
o
m
ain
r
elev
a
n
ce
,
as
o
u
tlin
ed
i
n
T
ab
le
2
.
T
o
en
s
u
r
e
c
o
n
s
is
ten
cy
an
d
m
in
im
ize
s
u
b
jectiv
e
b
ias,
an
in
itial
v
alid
atio
n
was
p
er
f
o
r
m
e
d
wh
er
e
all
th
r
ee
ex
p
er
ts
i
n
d
ep
e
n
d
en
tly
l
ab
eled
a
s
am
p
le
o
f
1
5
,
0
0
0
m
ess
ag
es;
th
e
f
in
al
l
ab
el
f
o
r
t
h
is
s
et
was
d
eter
m
in
ed
b
y
a
m
ajo
r
ity
v
o
te.
Fo
llo
win
g
th
is
m
an
u
al
v
alid
atio
n
,
a
s
em
i
-
s
u
p
er
v
is
ed
ap
p
r
o
ac
h
was
u
s
ed
to
an
n
o
tat
e
th
e
f
u
ll
d
ataset:
a
p
r
elim
in
ar
y
m
o
d
el
tr
ai
n
ed
o
n
th
e
1
5
,
0
0
0
-
m
ess
ag
e
s
et
was
u
s
ed
to
p
r
o
v
id
e
in
itial
p
r
ed
ictio
n
s
,
wh
ich
wer
e
th
e
n
v
alid
ated
an
d
co
r
r
ec
ted
b
y
th
e
ex
p
er
t
p
an
el
t
o
en
s
u
r
e
h
ig
h
-
q
u
ality
la
b
els
f
o
r
t
h
e
f
in
al
tr
ain
in
g
d
ata.
T
h
is
lab
elin
g
p
r
o
ce
s
s
r
esu
lted
in
th
e
d
ata
d
is
tr
ib
u
tio
n
ac
r
o
s
s
th
e
tr
ain
in
g
,
v
alid
atio
n
,
an
d
test
in
g
s
u
b
s
ets
d
etailed
in
T
ab
le
3
,
wh
ich
n
o
tab
ly
in
clu
d
es tem
p
o
r
al
v
ar
iatio
n
i
n
s
p
am
p
r
ev
alen
ce
.
Fig
u
r
e
1
.
R
esear
ch
m
eth
o
d
o
l
o
g
y
f
lo
wch
a
r
t
T
ab
le
2
.
Sp
am
class
if
icatio
n
c
r
iter
ia
in
d
ata
lab
elin
g
S
p
a
m r
u
l
e
Ex
a
m
p
l
e
messa
g
e
I
r
r
e
l
e
v
a
n
t
m
e
ssa
g
e
s
"
S
h
e
w
a
s
c
r
y
i
n
g
i
n
t
h
e
l
a
k
e
s"
,
"
J
u
a
l
b
e
l
i
t
a
n
a
h
l
e
n
g
k
a
p
d
i
si
n
i
"
P
r
o
mo
t
i
o
n
a
l
mess
a
g
e
s (
n
o
n
-
c
o
m
p
a
n
y
)
"
S
i
t
u
s
p
e
rt
a
ru
h
a
n
o
n
l
i
n
e
,
si
n
i
k
a
k
a
k
j
o
i
n
"
,
"
S
a
y
a
m
e
n
j
u
a
l
l
a
m
p
u
b
a
c
a
,
b
i
s
a
k
e
l
i
n
k
i
n
i
y
a
"
M
e
ss
a
g
e
s wi
t
h
o
n
l
y
a
s
i
n
g
l
e
c
h
a
r
a
c
t
e
r
(
u
n
l
e
ss
sp
e
c
i
f
i
c
mea
n
i
n
g
)
"
A
B
C
D
E
F
G
Z
,
?
/
’
;
"
(
e
x
c
e
p
t
i
o
n
:
Y
/
N
)
M
e
ss
a
g
e
s wi
t
h
me
a
n
i
n
g
l
e
ss
c
h
a
r
a
c
t
e
r
c
o
m
b
i
n
a
t
i
o
n
s
"
B
N
V
M
M
D
F
K
D
sf
sd
"
,
"
D
sa
d
a
d
k
h
e
u
r
3
4
3
2
4
2
3
5
"
P
r
o
f
a
n
e
o
r
v
u
l
g
a
r
w
o
r
d
s wi
t
h
o
u
t
c
o
n
t
e
x
t
(
S
p
e
c
i
f
i
c
v
u
l
g
a
r
e
x
a
m
p
l
e
s)
C
o
p
i
e
d
/
p
a
st
e
d
c
h
a
t
(
e
x
c
e
p
t
r
e
p
o
r
t
f
o
r
mat
s)
"
Ma
a
f
,
p
a
s
t
i
k
a
n
t
i
d
a
k
a
d
a
p
e
n
u
l
i
s
a
n
d
a
l
a
m
c
h
a
t
"
,
"
Ap
a
b
i
sa
sa
y
a
b
a
n
t
u
?
?
"
M
e
ss
a
g
e
s wi
t
h
l
i
n
k
s
(
e
x
c
e
p
t
c
o
m
p
a
n
y
d
o
ma
i
n
s)
"
w
w
w
.
i
n
st
a
g
r
a
m.
c
o
m
/
a
k
u
j
u
a
l
n
a
si
"
,
"
w
w
w
.
h
o
t
sp
o
t
.
c
o
m"
T
ab
le
3
.
L
a
b
el
d
is
tr
ib
u
tio
n
p
er
d
ata
s
u
b
s
et
D
a
t
a
s
e
t
na
me
La
b
e
l
To
t
a
l
N
o
n
-
s
p
a
m
S
p
a
m
D
a
t
a
t
r
a
i
n
i
n
g
0
7
2
0
2
2
t
o
0
7
2
0
2
3
4
8
,
0
2
0
1
6
,
9
8
0
6
5
,
0
0
0
D
a
t
a
v
a
l
i
d
a
t
i
o
n
0
8
2
0
2
3
1
5
,
1
5
3
9
,
8
3
1
2
4
,
9
8
4
D
a
t
a
t
e
s
t
i
n
g
0
9
2
0
2
3
1
4
,
9
2
3
1
0
,
0
5
8
2
4
,
9
8
1
D
a
t
a
t
e
s
t
i
n
g
1
0
2
0
2
3
1
6
,
9
7
1
8
,
0
2
7
2
4
,
9
9
8
D
a
t
a
t
e
s
t
i
n
g
1
1
2
0
2
3
1
2
,
5
8
2
2
,
4
1
5
1
4
,
9
9
7
D
a
t
a
t
e
s
t
i
n
g
1
2
2
0
2
3
1
2
,
3
5
9
2
,
6
4
1
1
5
,
0
0
0
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2252
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
1
,
Feb
r
u
ar
y
20
26
:
909
-
9
1
8
912
2
.
2
.
Da
t
a
prepro
ce
s
s
ing
Prio
r
to
m
o
d
el
tr
ain
in
g
,
t
h
e
te
x
tu
al
d
ata
was
s
u
b
jecte
d
to
a
co
m
p
r
eh
e
n
s
iv
e
p
r
ep
r
o
ce
s
s
in
g
p
ip
elin
e
to
en
s
u
r
e
its
clea
n
lin
ess
an
d
n
o
r
m
aliza
tio
n
.
A
cr
itical
s
tep
f
o
r
im
p
r
o
v
in
g
m
o
d
el
p
e
r
f
o
r
m
an
ce
[
1
7
]
.
T
h
is
p
i
p
eli
n
e
in
v
o
l
v
e
d
s
e
v
e
r
a
l
s
ta
g
es:
c
o
n
v
er
s
i
o
n
o
f
all
te
x
t
t
o
l
o
w
er
ca
s
e
;
r
e
m
o
v
al
o
f
p
u
n
c
tu
ati
o
n
a
n
d
s
p
e
cia
l
c
h
a
r
a
cte
r
s
;
n
o
r
m
al
iz
ati
o
n
o
f
s
l
a
n
g
a
n
d
in
f
o
r
m
a
l
te
r
m
i
n
o
l
o
g
y
t
o
t
h
ei
r
s
ta
n
d
a
r
d
l
ex
ica
l
f
o
r
m
s
;
el
im
in
ati
o
n
o
f
s
to
p
w
o
r
d
s
[
1
8
]
;
to
k
en
izatio
n
to
s
eg
m
en
t
t
h
e
te
x
t
in
to
d
is
cr
ete
wo
r
d
s
[
1
9
]
;
an
d
th
e
ap
p
licatio
n
o
f
s
tem
m
in
g
to
r
ed
u
ce
wo
r
d
s
to
th
eir
m
o
r
p
h
o
lo
g
ical
r
o
o
ts
[
2
0
]
.
2
.
3
.
F
e
a
t
ure
ex
t
r
a
ct
io
n a
nd
m
o
delin
g
Fo
r
th
e
co
n
v
en
tio
n
al
m
ac
h
in
e
lear
n
in
g
m
o
d
els
(
DT
an
d
SVM)
,
th
e
p
r
ep
r
o
ce
s
s
ed
tex
t
d
ata
was
tr
an
s
f
o
r
m
ed
in
to
n
u
m
er
ical
f
e
atu
r
e
v
ec
t
o
r
s
.
Var
io
u
s
f
ea
tu
r
e
ex
tr
ac
tio
n
tech
n
i
q
u
es
wer
e
e
v
alu
ated
,
i
n
clu
d
in
g
C
o
u
n
tVec
to
r
izer
,
Hash
in
g
Vec
to
r
izer
,
ter
m
f
r
e
q
u
en
c
y
-
in
v
e
r
s
e
d
o
cu
m
en
t
f
r
eq
u
e
n
cy
(TF
-
I
D
F),
W
o
r
d
2
V
ec
,
an
d
Fas
tTe
x
t.
Fo
llo
win
g
p
r
elim
in
ar
y
ev
alu
atio
n
s
,
T
F
-
I
DF
was
d
eter
m
in
ed
to
b
e
th
e
o
p
tim
al
m
eth
o
d
f
o
r
f
ea
t
u
r
e
r
ep
r
esen
tatio
n
,
c
o
n
s
is
ten
t w
ith
its
wid
esp
r
ea
d
s
u
cc
ess
f
u
l a
p
p
licatio
n
in
tex
t c
lass
if
icatio
n
liter
atu
r
e
[
2
1
]
.
Fo
r
th
e
d
ee
p
lear
n
in
g
p
ar
a
d
ig
m
,
two
p
r
in
ci
p
al
ar
ch
itectu
r
al
ca
teg
o
r
ies we
r
e
in
v
esti
g
ated
:
i)
L
STM
:
a
r
an
g
e
o
f
L
STM
v
ar
ian
ts
wer
e
im
p
lem
en
ted
an
d
test
ed
,
in
clu
d
in
g
v
a
n
illa
L
STM
,
s
tack
ed
L
STM
,
b
id
ir
ec
tio
n
al
L
STM
,
an
d
a
h
y
b
r
id
C
NN
-
L
STM
ar
ch
itectu
r
e.
T
h
ese
m
o
d
els
ar
e
r
ec
o
g
n
ized
f
o
r
th
eir
p
r
o
f
icien
cy
in
c
ap
tu
r
in
g
s
eq
u
en
tial d
ep
en
d
en
cies w
ith
in
tex
tu
al
d
ata
[
2
2
]
–
[
2
4
]
.
ii)
B
E
R
T
:
p
re
-
tr
ain
ed
B
E
R
T
m
o
d
els,
s
p
ec
if
ically
B
E
R
T
-
b
as
e
,
Dis
til
B
E
R
T
,
an
d
cr
o
s
s
-
lin
g
u
al
lan
g
u
ag
e
m
o
d
el
-
r
o
b
u
s
tly
o
p
tim
ized
B
E
R
T
p
r
etr
ain
in
g
ap
p
r
o
ac
h
(
XL
M
-
R
OB
E
R
T
a
)
,
wer
e
f
in
e
-
tu
n
ed
f
o
r
th
e
b
in
ar
y
s
p
am
class
if
icatio
n
tas
k
.
T
h
e
ef
f
icac
y
o
f
th
ese
m
o
d
e
ls
is
d
er
iv
ed
f
r
o
m
th
eir
ad
v
a
n
ce
d
ab
ilit
y
to
p
r
o
ce
s
s
th
e
b
id
ir
ec
tio
n
al
co
n
te
x
t o
f
wo
r
d
s
with
in
a
s
en
ten
ce
[
1
1
]
,
[
1
3
]
,
[
2
5
]
.
T
o
p
r
o
v
id
e
th
e
r
e
q
u
ested
m
eth
o
d
o
lo
g
ical
clar
ity
,
T
ab
le
4
d
etails
th
e
ar
c
h
itectu
r
es
f
o
r
th
e
tw
o
to
p
-
p
er
f
o
r
m
i
n
g
m
o
d
els.
T
h
e
C
NN
-
L
STM
ar
ch
itectu
r
e
was
b
u
ilt
f
r
o
m
s
cr
atch
,
wh
ile
th
e
Dis
tilB
E
R
T
ar
ch
itectu
r
e
s
h
o
ws
th
e
class
if
icatio
n
h
ea
d
a
d
d
ed
o
n
t
o
p
o
f
th
e
p
r
e
-
t
r
ain
ed
b
ase
m
o
d
el
f
o
r
f
in
e
-
t
u
n
in
g
.
T
h
e
m
o
d
els
wer
e
im
p
lem
en
ted
u
tili
zin
g
th
e
T
en
s
o
r
Flo
w
a
n
d
Ker
as
f
r
am
ewo
r
k
s
.
A
cr
i
tical
p
h
ase
o
f
th
e
m
eth
o
d
o
l
o
g
y
was
h
y
p
er
p
a
r
a
m
eter
tu
n
in
g
,
wh
er
ein
tec
h
n
iq
u
es
s
u
ch
as
g
r
id
s
ea
r
ch
wer
e
em
p
lo
y
e
d
to
s
y
s
tem
atica
lly
id
en
tify
th
e
o
p
tim
al
co
n
f
ig
u
r
atio
n
o
f
ea
ch
m
o
d
el'
s
p
ar
am
eter
s
(
e.
g
.
,
lear
n
i
n
g
r
ate,
n
u
m
b
er
o
f
lay
er
s
,
an
d
d
r
o
p
o
u
t
r
ate)
to
m
ax
im
ize
p
r
ed
ictiv
e
p
er
f
o
r
m
an
ce
[
2
6
]
.
T
h
e
f
in
al
o
p
tim
al
h
y
p
er
p
ar
am
eter
s
f
o
r
th
e
to
p
-
p
er
f
o
r
m
i
n
g
m
o
d
els
wer
e
i
d
en
tifie
d
f
r
o
m
th
is
p
r
o
ce
s
s
.
F
o
r
th
e
SVM
m
o
d
el,
a
p
o
ly
n
o
m
ial
k
er
n
el
with
a
d
eg
r
ee
o
f
2
an
d
a
C
v
alu
e
o
f
1
was
u
s
ed
.
Fo
r
th
e
Dis
til
B
E
R
T
m
o
d
el,
th
e
f
in
e
-
tu
n
e
d
ar
ch
itectu
r
e
co
n
s
is
ted
o
f
th
r
ee
h
id
d
en
d
en
s
e
lay
er
s
wit
h
5
1
2
,
2
5
6
,
an
d
1
2
8
n
eu
r
o
n
s
,
r
esp
ec
tiv
ely
,
ea
ch
f
o
llo
we
d
b
y
a
d
r
o
p
o
u
t
lay
e
r
with
r
ates o
f
0
.
5
,
0
.
4
,
an
d
0
.
3
.
An
Ad
am
o
p
tim
izer
with
a
lea
r
n
in
g
r
ate
o
f
0
.
0
0
0
1
was u
tili
z
ed
f
o
r
t
r
ain
in
g
.
T
ab
le
4
.
Ar
c
h
itectu
r
al
s
ch
em
a
tic
o
f
C
NN
-
L
STM
an
d
Dis
tilB
E
R
T
(
f
in
e
-
tu
n
in
g
)
La
y
e
r
A
r
c
h
i
t
e
c
t
u
r
e
o
f
C
N
N
-
LST
M
A
r
c
h
i
t
e
c
t
u
r
e
o
f
f
i
n
e
-
t
u
n
i
n
g
D
i
s
t
i
l
B
ER
T
I
n
p
u
t
Emb
e
d
d
i
n
g
L
a
y
e
r
I
n
p
u
t
f
r
o
m
D
i
st
i
l
B
ER
T
B
a
se
(
7
6
8
u
n
i
t
)
1
C
o
n
v
1
D
(
6
4
f
i
l
t
e
r
,
k
e
r
n
e
l
=
5
)
D
e
n
se
(
5
1
2
u
n
i
t
)
2
M
a
x
P
o
o
l
i
n
g
1
D
(
p
o
o
l
si
z
e
=
2
)
D
r
o
p
o
u
t
(
R
a
t
e
=
0
.
5
)
3
LSTM
(
6
4
u
n
i
t
)
D
e
n
se
(
2
5
6
u
n
i
t
)
4
-
D
r
o
p
o
u
t
(
R
a
t
e
=
0
.
4
)
5
-
D
e
n
se
(
1
2
8
u
n
i
t
)
6
-
D
r
o
p
o
u
t
(
R
a
t
e
=
0
.
3
)
O
u
t
p
u
t
D
e
n
se
(
1
u
n
i
t
,
's
i
g
mo
i
d
')
D
e
n
se
(
1
u
n
i
t
,
's
i
g
mo
i
d
')
3.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
T
h
e
p
er
f
o
r
m
an
ce
o
f
th
e
tr
ain
ed
m
o
d
els
was
r
ig
o
r
o
u
s
ly
ev
alu
ated
u
s
in
g
p
r
ec
is
io
n
,
r
ec
all
,
F1
-
s
co
r
e,
an
d
ac
cu
r
ac
y
[
2
7
]
.
T
h
is
ev
al
u
atio
n
was
co
n
d
u
cted
o
n
d
is
tin
ct
test
d
atasets
f
r
o
m
f
o
u
r
c
o
n
s
ec
u
tiv
e
m
o
n
th
s
(
Sep
tem
b
er
to
Dec
em
b
er
2
0
2
3
)
to
n
o
t
o
n
l
y
m
ea
s
u
r
e
p
er
f
o
r
m
an
ce
b
u
t
also
to
ass
ess
th
e
m
o
d
els'
s
tab
il
ity
an
d
r
o
b
u
s
tn
ess
ag
ain
s
t
tem
p
o
r
al
s
h
if
ts
in
d
ata
p
atter
n
s
.
T
h
is
lo
n
g
itu
d
in
al
ap
p
r
o
ac
h
is
cr
itical,
as
th
e
n
atu
r
e
o
f
s
p
am
ca
n
ev
o
lv
e
o
v
er
tim
e
[
2
8
]
.
All
r
esu
lts
p
r
esen
ted
in
th
e
s
u
m
m
ar
y
tab
les
r
ef
lect
th
e
av
er
ag
e
p
er
f
o
r
m
a
n
ce
ac
r
o
s
s
th
ese
f
o
u
r
test
in
g
p
e
r
io
d
s
u
n
less
o
th
er
wis
e
s
p
ec
if
ied
.
3
.
1
.
P
re
li
m
ina
ry
f
e
a
t
ure
ex
t
ra
ct
io
n e
v
a
lua
t
i
o
n
B
ased
o
n
th
e
p
r
elim
in
ar
y
r
esu
l
ts
s
h
o
wn
in
T
ab
le
5
,
T
F
-
I
DF
was selec
ted
as th
e
o
p
tim
al
v
ec
to
r
izatio
n
tech
n
iq
u
e
f
o
r
t
h
e
f
in
al
m
ac
h
in
e
lear
n
in
g
m
o
d
el
co
m
p
a
r
is
o
n
s
.
W
h
ile
Hash
in
g
Vec
to
r
izer
an
d
C
o
u
n
tVec
to
r
ize
r
g
av
e
th
e
DT
m
o
d
el
h
ig
h
ac
cu
r
ac
y
(
0
.
9
2
)
,
th
eir
p
er
f
o
r
m
a
n
c
e
with
SVM
was
s
ig
n
if
ican
tl
y
lo
wer
,
with
r
ec
all
s
co
r
es
o
f
0
.
7
3
a
n
d
0
.
6
8
,
r
esp
ec
tiv
ely
.
T
h
e
r
ef
o
r
e
,
T
F
-
I
DF
c
o
m
b
in
ed
with
SVM,
w
h
ich
y
i
eld
ed
a
s
tr
o
n
g
an
d
m
o
r
e
b
alan
ce
d
p
er
f
o
r
m
a
n
ce
(
Acc
u
r
ac
y
0
.
8
8
,
F1
-
s
co
r
e
0
.
8
7
)
,
was
id
e
n
tifie
d
as
t
h
e
m
o
s
t
r
o
b
u
s
t
ch
o
ice
f
o
r
b
o
th
alg
o
r
ith
m
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Dee
p
lea
r
n
in
g
-
b
a
s
ed
s
p
a
m
d
et
ec
tio
n
fo
r
Wh
a
ts
A
p
p
ch
a
tb
o
t f
a
llb
a
ck
r
ed
u
ctio
n
(
S
a
tr
io
S
a
d
e
w
o
)
913
T
ab
le
5
.
Su
m
m
a
r
y
o
f
ev
alu
ati
o
n
r
esu
lts
o
f
m
et
h
o
d
s
with
f
ea
tu
r
e
ex
tr
ac
tio
n
F
e
a
t
u
r
e
e
x
t
r
a
c
t
i
o
n
M
e
t
h
o
d
P
r
e
c
i
s
i
o
n
R
e
c
a
l
l
F1
-
s
c
o
r
e
A
c
c
u
r
a
c
y
C
o
u
n
t
V
e
c
t
o
r
i
z
e
r
DT
0
.
9
1
0
.
9
2
0
.
9
2
0
.
9
2
S
V
M
0
.
8
5
0
.
6
8
0
.
6
7
0
.
7
4
W
o
r
d
2
V
ec
DT
0
.
8
7
0
.
8
7
0
.
8
7
0
.
8
7
S
V
M
0
.
8
5
0
.
8
6
0
.
8
5
0
.
8
5
F
a
st
T
e
x
t
DT
0
.
8
8
0
.
8
4
0
.
8
5
0
.
8
6
S
V
M
0
.
8
9
0
.
8
5
0
.
8
7
0
.
8
8
H
a
sh
i
n
g
V
e
c
t
o
r
DT
0
.
9
2
0
.
9
3
0
.
9
2
0
.
9
2
S
V
M
0
.
8
6
0
.
7
3
0
.
7
4
0
.
7
8
Tf
i
d
f
V
e
c
t
o
r
i
z
e
r
DT
0
.
9
1
0
.
9
2
0
.
9
2
0
.
9
2
S
V
M
0
.
8
9
0
.
8
7
0
.
8
7
0
.
8
8
3
.
2
.
M
o
del per
f
o
rma
nce
Fo
llo
win
g
an
ex
ten
s
iv
e
h
y
p
er
p
ar
am
eter
tu
n
i
n
g
p
r
o
ce
s
s
,
b
o
t
h
th
e
tr
ad
itio
n
al
m
ac
h
in
e
lear
n
in
g
an
d
ad
v
an
ce
d
d
ee
p
lear
n
in
g
m
o
d
el
s
d
em
o
n
s
tr
ated
s
tr
o
n
g
p
r
ed
icti
v
e
ca
p
ab
ilit
ies.
Ho
wev
er
,
a
d
is
tin
ct
p
er
f
o
r
m
an
ce
h
ier
ar
ch
y
was
o
b
s
er
v
e
d
,
wi
th
th
e
d
ee
p
lear
n
in
g
m
o
d
e
ls
co
n
s
is
ten
tly
o
u
tp
er
f
o
r
m
in
g
th
eir
t
r
ad
itio
n
al
co
u
n
ter
p
a
r
ts
.
T
h
is
a
d
v
an
tag
e
i
s
attr
ib
u
tab
le
to
th
eir
in
h
e
r
en
t
ca
p
ac
ity
to
lear
n
c
o
m
p
lex
,
h
i
er
ar
ch
ical
f
ea
t
u
r
es
an
d
u
n
d
er
s
tan
d
co
n
te
x
tu
al
r
el
atio
n
s
h
ip
s
with
in
th
e
tex
t,
wh
ich
is
a
lim
itatio
n
f
o
r
alg
o
r
i
th
m
s
lik
e
DT
an
d
SVM
th
at
r
ely
o
n
m
o
r
e
s
u
p
er
f
icial,
f
r
eq
u
en
cy
-
b
ased
f
ea
t
u
r
es.
T
ab
le
6
p
r
esen
ts
th
e
a
v
er
ag
e
p
er
f
o
r
m
an
ce
m
etr
ics o
f
th
e
m
o
s
t n
o
tab
le
m
o
d
els o
v
er
t
h
e
f
o
u
r
-
m
o
n
th
test
in
g
p
er
io
d
.
T
ab
le
6
.
Av
e
r
ag
e
p
e
r
f
o
r
m
an
ce
co
m
p
ar
is
o
n
o
f
all
m
o
d
els ac
r
o
s
s
th
e
f
o
u
r
-
m
o
n
th
test
p
er
io
d
(
Sep
tem
b
er
to
Dec
em
b
er
2
0
2
3
)
M
o
d
e
l
s
P
r
e
c
i
s
i
o
n
R
e
c
a
l
l
F1
-
sc
o
r
e
A
c
c
u
r
a
c
y
DT
0
.
9
0
0
.
9
2
0
.
9
1
0
.
9
3
S
V
M
0
.
9
0
0
.
9
2
0
.
9
1
0
.
9
3
S
t
a
c
k
e
d
LSTM
0
.
9
1
0
.
9
1
0
.
9
1
0
.
9
4
C
N
N
-
LSTM
0
.
9
2
0
.
9
1
0
.
9
1
0
.
9
4
En
c
o
d
e
r
d
e
c
o
d
e
r
LST
M
0
.
9
1
0
.
9
0
0
.
9
1
0
.
9
4
B
ER
T
-
B
A
S
E
0
.
9
1
0
.
8
9
0
.
9
0
0
.
9
2
D
i
st
i
l
B
ER
T
0
.
9
2
0
.
8
9
0
.
9
0
0
.
9
3
X
LM
-
R
o
b
e
r
t
a
0
.
9
2
0
.
7
9
0
.
8
3
0
.
8
9
T
h
e
em
p
ir
ical
r
esu
lts
id
en
ti
f
ied
th
e
C
NN
-
L
STM
an
d
Dis
til
B
E
R
T
m
o
d
els
as
th
e
p
r
ee
m
in
en
t
ar
ch
itectu
r
es
in
ter
m
s
o
f
p
er
f
o
r
m
an
ce
.
T
h
e
h
y
b
r
id
C
NN
-
L
STM
m
o
d
el
ac
h
iev
ed
th
e
h
i
g
h
e
s
t
o
v
er
all
ac
cu
r
ac
y
(
0
.
9
4
)
,
s
h
o
wca
s
in
g
th
e
p
o
wer
o
f
its
ar
ch
itectu
r
e.
T
h
is
s
y
n
e
r
g
y
is
p
a
r
ticu
lar
ly
ef
f
ec
tiv
e
f
o
r
id
en
tify
i
n
g
s
p
am
;
th
e
C
NN
lay
er
s
ac
t
a
s
a
p
o
w
er
f
u
l
f
ea
tu
r
e
ex
tr
ac
to
r
,
id
en
tif
y
in
g
lo
ca
l
n
-
g
r
a
m
p
atter
n
s
an
d
k
ey
wo
r
d
s
o
f
ten
ass
o
ciate
d
with
s
p
am
(
e.
g
.
,
'
p
r
o
mo
'
or
'lin
k
i
n
i'
)
,
wh
ile
th
e
L
STM
lay
er
s
an
aly
ze
th
e
s
eq
u
e
n
ce
in
wh
ich
th
ese
p
atter
n
s
ap
p
ea
r
,
ca
p
tu
r
i
n
g
th
e
co
n
tex
tu
al
s
tr
u
ctu
r
e
o
f
t
h
e
m
e
s
s
ag
e.
Dis
til
B
E
R
T
,
a
l
ig
h
ter
an
d
f
aster
v
er
s
io
n
o
f
B
E
R
T
,
also
d
em
o
n
s
tr
ated
ex
ce
llen
t,
wel
l
-
r
o
u
n
d
ed
p
er
f
o
r
m
an
ce
with
h
ig
h
p
r
ec
is
io
n
(
0
.
9
2
)
a
n
d
ac
c
u
r
ac
y
(
0
.
9
3
)
.
I
ts
s
tr
en
g
th
lies
in
its
ab
ilit
y
t
o
r
etain
a
p
o
wer
f
u
l
g
r
asp
o
f
b
i
d
ir
ec
tio
n
al
c
o
n
tex
t,
ev
en
as
a
d
is
till
ed
m
o
d
el.
T
h
i
s
is
cr
u
cial
f
o
r
d
is
tin
g
u
is
h
in
g
s
o
p
h
is
ticated
s
p
am
th
at
m
im
ics
co
n
v
er
s
atio
n
.
Fo
r
ex
am
p
le,
in
a
m
ess
ag
e
lik
e
"
K
a
ka
k
d
m
",
a
u
n
id
ir
ec
tio
n
a
l
m
o
d
el
m
ig
h
t
b
e
co
n
f
u
s
ed
,
b
u
t
a
b
id
ir
ec
tio
n
al
m
o
d
el
ca
n
an
aly
ze
th
e
co
n
tex
t
f
r
o
m
b
o
t
h
d
ir
ec
tio
n
s
to
b
ett
er
ass
ess
th
e
in
ten
t,
r
elatin
g
it to
p
atter
n
s
s
ee
n
in
o
th
er
p
r
o
m
o
tio
n
al
s
p
am
.
I
n
co
n
tr
ast,
th
e
m
u
ltil
in
g
u
al
m
o
d
el
XL
M
-
R
OB
E
R
T
a,
wh
ile
ac
h
iev
in
g
h
ig
h
p
r
ec
is
io
n
,
e
x
h
ib
ited
a
n
o
tab
le
d
ef
icien
c
y
in
r
ec
all
(
0
.
7
9
)
.
T
h
is
s
u
g
g
ests
th
at
its
g
en
er
alize
d
,
cr
o
s
s
-
lin
g
u
al
tr
ain
in
g
was
les
s
ef
f
ec
tiv
e
f
o
r
th
is
d
o
m
ain
-
s
p
ec
if
ic
task
.
T
h
e
m
o
d
el,
tr
ain
ed
o
n
1
0
0
la
n
g
u
ag
es,
m
ay
lack
th
e
s
p
ec
if
i
c
to
k
en
izatio
n
an
d
em
b
ed
d
in
g
r
ep
r
esen
tatio
n
s
n
ee
d
ed
to
u
n
d
er
s
tan
d
th
e
n
u
an
ce
s
o
f
I
n
d
o
n
esian
s
lan
g
,
in
f
o
r
m
alities
,
an
d
co
n
tex
t
-
s
p
ec
if
ic
s
p
am
(
e.
g
.
,
"
S
itu
s
p
erta
r
u
h
a
n
o
n
lin
e
")
as
ef
f
ec
tiv
ely
as
th
e
m
o
d
els
f
in
e
-
tu
n
ed
s
p
ec
if
ically
o
n
th
is
d
ataset.
3
.
3
.
Ana
ly
s
is
o
f
m
is
cla
s
s
if
ica
t
io
ns
A
m
o
r
e
g
r
an
u
la
r
an
aly
s
is
o
f
th
e
m
is
class
if
icat
io
n
s
h
ig
h
lig
h
ts
th
e
p
er
s
is
ten
t
ch
allen
g
es
in
au
to
m
ated
ch
atb
o
t
s
p
am
d
etec
tio
n
.
Fig
u
r
es
2
an
d
3
p
r
esen
t
th
e
c
o
n
f
u
s
io
n
m
atr
ices
f
o
r
th
e
to
p
-
p
er
f
o
r
m
in
g
C
NN
-
L
STM
an
d
Dis
tilB
E
R
T
m
o
d
els,
r
esp
ec
tiv
ely
.
T
h
ese
f
ig
u
r
es
d
etail
th
eir
p
er
f
o
r
m
an
ce
ac
r
o
s
s
th
e
f
o
u
r
-
m
o
n
th
test
in
g
p
er
io
d
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2252
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
1
,
Feb
r
u
ar
y
20
26
:
909
-
9
1
8
914
Fig
u
r
e
2
.
Mo
n
th
ly
c
o
n
f
u
s
io
n
m
atr
ices f
o
r
th
e
C
NN
-
L
STM
m
o
d
el,
s
h
o
win
g
p
r
e
d
ictiv
e
p
er
f
o
r
m
an
ce
o
n
t
h
e
test
s
ets f
r
o
m
Sep
tem
b
er
to
Dec
e
m
b
er
2
0
2
3
Fig
u
r
e
3
.
Mo
n
th
ly
c
o
n
f
u
s
io
n
m
atr
ices f
o
r
th
e
Dis
tilB
E
R
T
m
o
d
el,
s
h
o
win
g
p
r
e
d
ictiv
e
p
er
f
o
r
m
an
ce
o
n
t
h
e
test
s
ets f
r
o
m
Sep
tem
b
er
to
Dec
e
m
b
er
2
0
2
3
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Dee
p
lea
r
n
in
g
-
b
a
s
ed
s
p
a
m
d
et
ec
tio
n
fo
r
Wh
a
ts
A
p
p
ch
a
tb
o
t f
a
llb
a
ck
r
ed
u
ctio
n
(
S
a
tr
io
S
a
d
e
w
o
)
915
As
s
u
m
m
ar
ized
i
n
T
a
b
le
7
,
th
e
m
is
class
if
icatio
n
er
r
o
r
s
o
b
s
er
v
ed
in
t
h
e
to
p
-
p
e
r
f
o
r
m
in
g
m
o
d
els
ca
n
b
e
g
r
o
u
p
ed
in
to
two
m
ai
n
ca
te
g
o
r
ies b
ased
o
n
th
e
d
i
r
ec
tio
n
o
f
th
e
p
r
e
d
ictio
n
er
r
o
r
:
i)
Fals
e
p
o
s
itiv
es
(
non
-
s
p
am
class
if
ied
as
s
p
am
)
:
th
e
m
o
d
els
o
cc
asio
n
ally
m
is
class
if
ied
m
ess
ag
es
ch
ar
ac
ter
ized
b
y
ex
tr
em
e
b
r
e
v
ity
,
in
f
o
r
m
ality
,
o
r
a
m
b
ig
u
it
y
(
e.
g
.
,
"
A
ku
g
ita
"
[
I
am
Git
a]
,
"
Ha
lo
b
u
k
"
[
Hello
m
a'
am
]
,
o
r
s
in
g
le
-
wo
r
d
r
ep
lies
)
as
s
p
am
.
T
h
is
er
r
o
r
is
lik
ely
b
ec
au
s
e
th
ese
m
ess
ag
es
lack
s
u
f
f
icien
t
co
n
tex
t
f
o
r
a
h
i
g
h
-
c
o
n
f
id
en
ce
p
r
ed
ictio
n
.
Fo
r
ar
ch
itectu
r
es
lik
e
C
N
N
-
L
STM
,
b
r
ief
,
leg
itima
te
m
ess
ag
es
s
u
ch
as
'A
ku
g
ita
'
m
ay
tr
ig
g
er
lo
ca
l
p
atter
n
d
ete
cto
r
s
(
C
NNs)
th
at
h
av
e
lear
n
ed
to
ass
o
ciate
s
h
o
r
t,
n
o
n
-
g
r
am
m
atica
l in
p
u
ts
with
lo
w
-
ef
f
o
r
t sp
am
,
f
ailin
g
to
ca
p
tu
r
e
th
e
b
en
ig
n
s
eq
u
e
n
tial c
o
n
tex
t th
at
th
e
L
STM
lay
er
wo
u
l
d
ty
p
ical
ly
an
aly
ze
.
ii)
Fals
e
n
eg
ativ
es
(
s
p
am
class
if
ied
as
n
o
n
-
s
p
am
)
:
m
o
r
e
im
p
o
r
tan
tly
,
th
e
m
o
d
els
s
tr
u
g
g
l
ed
to
class
if
y
s
o
p
h
is
ticated
s
p
am
th
at
ad
e
p
tly
m
im
ics
h
u
m
an
co
n
v
er
s
atio
n
o
r
leg
itima
te
s
y
s
tem
n
o
tific
atio
n
s
.
I
n
s
tan
ce
s
s
u
ch
as
"
K
a
ka
k
d
m
"
[
Sis
ter
,
DM
m
e]
em
p
lo
y
in
f
o
r
m
al,
p
er
s
o
n
a
b
le
lan
g
u
ag
e
t
o
cir
cu
m
v
en
t
f
ilter
s
,
wh
er
ea
s
m
ess
ag
es
lik
e
"Welco
m
e
to
Gb
o
ar
d
clip
b
o
ar
d
.
.
.
"
m
asq
u
e
r
ad
e
as
b
en
ig
n
s
y
s
tem
aler
ts
.
T
h
is
ev
id
en
ce
d
em
o
n
s
tr
ates
a
clea
r
ev
o
lu
tio
n
in
s
p
am
m
i
n
g
tech
n
iq
u
es
t
o
war
d
s
o
cial
e
n
g
in
ee
r
in
g
an
d
co
n
tex
tu
al
ca
m
o
u
f
lag
e
t
o
e
v
ad
e
d
etec
tio
n
,
wh
ich
p
r
esen
ts
a
n
o
n
g
o
in
g
ch
allen
g
e
t
h
at
n
ec
ess
itates
m
o
d
els
ca
p
ab
le
o
f
d
is
ce
r
n
in
g
n
o
t ju
s
t
lex
ical
co
n
ten
t,
b
u
t a
ls
o
u
n
d
er
ly
in
g
in
ten
t.
T
h
is
an
aly
s
is
in
d
icate
s
th
at
w
h
ile
d
ee
p
lear
n
in
g
m
o
d
els
p
o
s
s
ess
co
n
s
id
er
ab
le
p
r
e
d
ictiv
e
p
o
wer
,
th
eir
ac
cu
r
ac
y
ca
n
b
e
co
m
p
r
o
m
i
s
ed
wh
en
co
n
f
r
o
n
ted
with
m
ess
ag
es
th
at
ar
e
ei
th
er
co
n
tex
t
-
d
e
f
icien
t
o
r
in
ten
tio
n
ally
d
ec
e
p
tiv
e.
I
n
th
is
r
esp
ec
t,
th
e
Dis
tilB
E
R
T
m
o
d
el
d
em
o
n
s
tr
ated
a
m
ar
g
i
n
al
ad
v
an
ta
g
e
o
v
e
r
C
NN
-
L
STM
b
y
ex
h
ib
itin
g
a
lo
wer
in
cid
en
ce
o
f
f
alse
n
e
g
ativ
es.
T
h
is
is
a
cr
u
cial
d
is
tin
ctio
n
f
o
r
p
r
ac
tical
im
p
lem
en
tatio
n
.
A
f
alse
n
eg
a
tiv
e
(
lettin
g
s
p
am
th
r
o
u
g
h
)
d
i
r
ec
tly
d
eg
r
a
d
es
th
e
u
s
er
ex
p
e
r
ien
ce
an
d
f
ails
to
s
o
lv
e
th
e
co
r
e
p
r
o
b
lem
.
I
n
c
o
n
tr
ast,
a
f
alse
p
o
s
itiv
e
(
b
lo
ck
in
g
a
leg
itima
te
m
ess
ag
e)
is
al
s
o
u
n
d
esira
b
le
b
u
t
ca
n
b
e
m
itig
ated
with
a
u
s
er
f
ee
d
b
ac
k
m
ec
h
an
is
m
.
Dis
tilB
E
R
T
's
b
id
ir
ec
tio
n
al
co
n
tex
t
-
a
war
en
ess
ap
p
ea
r
s
to
g
iv
e
it
a
s
lig
h
t
e
d
g
e
in
id
en
ti
f
y
in
g
t
h
ese
ca
m
o
u
f
lag
ed
s
p
a
m
m
ess
ag
es,
m
ak
in
g
it
a
m
o
r
e
r
o
b
u
s
t
ch
o
ice
f
o
r
en
s
u
r
in
g
a
h
ig
h
-
q
u
ality
,
s
ec
u
r
e
u
s
er
ex
p
er
ie
n
ce
,
wh
ich
is
o
f
ten
th
e
p
ar
a
m
o
u
n
t o
b
jectiv
e.
T
ab
le
7
.
R
ep
r
esen
tativ
e
ex
a
m
p
les o
f
f
alse p
o
s
itiv
e
an
d
f
alse n
eg
ativ
e
m
is
class
if
icatio
n
s
b
y
th
e
to
p
m
o
d
els
M
i
s
c
l
a
ssi
f
i
c
a
t
i
o
n
t
y
p
e
M
o
d
e
l
Ex
a
m
p
l
e
messa
g
e
A
n
a
l
y
s
i
s/
r
e
a
so
n
f
o
r
e
r
r
o
r
F
a
l
se
p
o
si
t
i
v
e
(
n
o
n
-
sp
a
m
a
s s
p
a
m)
C
N
N
-
LSTM
/
D
i
s
t
i
l
B
E
R
T
"
Ak
u
g
i
t
a
"
,
"
cs
"
,
"
H
a
l
o
b
u
k
"
Le
g
i
t
i
ma
t
e
m
e
ssa
g
e
s a
r
e
t
o
o
b
r
i
e
f
,
i
n
f
o
r
mal
,
o
r
l
a
c
k
c
o
n
t
e
x
t
,
mi
m
i
c
k
i
n
g
l
o
w
-
e
f
f
o
r
t
sp
a
m
.
F
a
l
se
n
e
g
a
t
i
v
e
(
s
p
a
m
a
s
n
o
n
-
s
p
a
m)
C
N
N
-
LSTM
/
D
i
s
t
i
l
B
E
R
T
"
Ak
u
ri
o
"
,
"
K
a
k
a
k
d
m
"
,
"
S
e
l
a
m
a
t
d
a
t
a
n
g
d
i
p
a
p
a
n
k
l
i
p
G
b
o
a
r
d
.
.
."
S
p
a
m
i
s
c
a
m
o
u
f
l
a
g
e
d
,
m
i
mi
c
k
i
n
g
f
r
i
e
n
d
l
y
c
o
n
v
e
r
sa
t
i
o
n
o
r
b
e
n
i
g
n
s
y
st
e
m
n
o
t
i
f
i
c
a
t
i
o
n
s
t
o
e
v
a
d
e
d
e
t
e
c
t
i
o
n
.
3
.
4
.
E
s
t
im
a
t
ed
im
pa
ct
o
n f
a
l
lba
ck
ra
t
e
A
p
r
im
ar
y
o
b
jectiv
e
o
f
t
h
is
r
e
s
ea
r
ch
was
to
r
ed
u
ce
th
e
ch
at
b
o
t'
s
f
allb
ac
k
r
ate,
wh
ich
s
to
o
d
at
3
3
%
(
6
m
illi
o
n
o
u
t
o
f
1
8
m
illi
o
n
m
ess
ag
es),
to
a
tar
g
et
o
f
1
5
%
o
r
less
.
T
h
e
an
aly
s
is
in
s
u
b
-
s
ec
tio
n
3
.
2
id
e
n
tifie
d
th
e
C
NN
-
L
STM
m
o
d
el
as
th
e
to
p
p
er
f
o
r
m
e
r
in
ter
m
s
o
f
o
v
er
all
ac
c
u
r
ac
y
an
d
F1
-
s
co
r
e
.
T
h
e
r
ef
o
r
e
,
we
ca
n
n
o
w
s
im
u
late
th
e
im
p
lem
en
tatio
n
o
f
th
is
s
p
ec
if
ic
m
o
d
el
to
esti
m
ate
its
d
ir
ec
t
im
p
ac
t
o
n
th
e
o
v
er
all
f
allb
ac
k
r
ate.
B
ased
o
n
th
e
m
an
u
al
an
n
o
t
atio
n
o
f
t
h
e
1
7
0
,
0
0
0
-
m
ess
ag
e
s
am
p
le,
an
esti
m
ated
2
9
.
4
%
o
f
th
e
f
allb
ac
k
-
in
d
u
cin
g
m
ess
ag
es
wer
e
id
en
tifie
d
as
s
p
am
(
4
9
,
9
5
2
s
p
am
m
ess
ag
es
o
u
t
o
f
1
6
9
,
9
6
0
to
tal
s
am
p
les).
E
x
tr
ap
o
latin
g
th
is
to
th
e
en
tir
e
p
o
p
u
latio
n
,
ap
p
r
o
x
im
ately
1
.
7
6
4
m
illi
o
n
o
f
th
e
6
m
illi
o
n
f
allb
ac
k
m
ess
ag
es
ar
e
s
p
am
,
wh
ile
th
e
r
em
ain
in
g
4
.
2
3
6
m
illi
o
n
ar
e
leg
itim
ate,
n
o
n
-
s
p
am
q
u
er
ies
th
at
th
e
ch
atb
o
t
f
ailed
to
co
m
p
r
eh
e
n
d
.
B
y
im
p
lem
en
ti
n
g
th
e
C
NN
-
L
STM
m
o
d
el
(
r
ec
all
:
0
.
9
1
,
p
r
ec
is
io
n
:
0
.
9
2
)
,
we
ca
n
p
r
o
ject
th
e
f
o
llo
win
g
:
i)
Sp
am
f
ilter
ed
(
tr
u
e
p
o
s
itiv
es):
th
e
m
o
d
el
wo
u
l
d
c
o
r
r
ec
tl
y
id
en
tif
y
a
n
d
f
ilter
1
.
6
0
5
m
illi
o
n
s
p
am
m
ess
ag
es (
1
.
7
6
4
m
illi
o
n
*
0
.
9
1
r
ec
all)
.
T
h
ese
m
ess
ag
es wo
u
ld
b
e
p
r
e
v
en
ted
f
r
o
m
t
r
ig
g
er
in
g
a
f
allb
ac
k
.
ii)
Sp
am
m
is
s
ed
(
f
alse
n
eg
ativ
es
)
:
ap
p
r
o
x
im
ately
0
.
1
5
9
m
illi
o
n
s
p
am
m
ess
ag
es
(
1
.
7
6
4
m
ill
io
n
*
(
1
-
0
.
9
1
)
)
wo
u
ld
b
e
m
is
s
ed
b
y
th
e
f
ilter
an
d
wo
u
ld
s
till
r
esu
lt in
a
f
allb
ac
k
.
iii)
L
eg
itima
te
f
allb
ac
k
s
(
tr
u
e
n
e
g
ativ
es):
th
e
m
o
d
el
m
u
s
t
also
p
r
o
ce
s
s
th
e
4
.
2
3
6
m
illi
o
n
leg
itima
te
f
allb
ac
k
q
u
er
ies.
B
ased
o
n
th
e
m
o
d
el'
s
p
er
f
o
r
m
an
ce
o
n
th
e
test
s
ets,
it
h
as
a
f
alse
p
o
s
itiv
e
r
ate
(
FP
R
)
o
f
4
.
5
2
%
.
T
h
is
m
ea
n
s
it
w
o
u
ld
in
co
r
r
ec
tly
f
ilter
0
.
1
9
1
m
illi
o
n
leg
iti
m
ate
m
ess
ag
es
(
4
.
2
3
6
m
illi
o
n
*
0
.
0
4
5
2
)
.
T
h
e
r
em
ain
in
g
4
.
0
4
5
m
illi
o
n
leg
iti
m
ate
q
u
er
ies
(
tr
u
e
n
eg
ativ
es
)
wo
u
ld
co
r
r
ec
tly
p
ass
th
r
o
u
g
h
th
e
f
ilter
an
d
wo
u
ld
s
till
ca
u
s
e
a
f
allb
ac
k
,
a
s
th
e
m
o
d
el
is
d
esig
n
ed
to
d
et
ec
t
s
p
am
,
n
o
t
to
f
ix
th
e
c
h
atb
o
t'
s
u
n
d
er
ly
in
g
co
m
p
r
eh
e
n
s
io
n
is
s
u
e.
T
h
e
n
ew
esti
m
ated
to
tal
f
allb
ac
k
co
u
n
t
wo
u
ld
th
e
r
ef
o
r
e
b
e
th
e
s
u
m
o
f
m
is
s
ed
s
p
am
(
0
.
1
5
9
m
illi
o
n
)
an
d
th
e
leg
itima
te
f
ailu
r
es
th
at
p
ass
ed
th
e
f
ilter
(
4
.
0
4
5
m
illi
o
n
)
,
r
esu
ltin
g
in
4
.
2
0
4
m
ill
io
n
f
allb
ac
k
s
.
T
h
is
in
ter
v
en
tio
n
wo
u
ld
r
e
d
u
ce
t
h
e
o
v
er
all
f
allb
ac
k
r
ate
f
r
o
m
3
3
%
(
6
m
illi
o
n
/
1
8
m
illi
o
n
)
to
a
n
esti
m
ated
2
3
.
3
5
%
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2252
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
1
,
Feb
r
u
ar
y
20
26
:
909
-
9
1
8
916
(
4
.
2
0
4
m
illi
o
n
/
1
8
m
illi
o
n
)
.
W
h
ile
th
is
r
ep
r
esen
ts
a
s
ig
n
if
ica
n
t
r
ed
u
ctio
n
i
n
f
allb
ac
k
i
n
cid
e
n
ts
,
it
d
o
es n
o
t m
ee
t
th
e
am
b
itio
u
s
1
5
%
tar
g
et.
T
h
i
s
f
in
d
in
g
is
cr
itical,
as
it
d
em
o
n
s
tr
ates
th
at
s
p
am
m
itig
atio
n
is
o
n
ly
o
n
e
p
a
r
t
o
f
th
e
s
o
lu
tio
n
;
a
s
u
b
s
tan
tial
p
o
r
tio
n
o
f
t
h
e
f
allb
ac
k
p
r
o
b
lem
(
4
.
0
4
5
m
illi
o
n
m
ess
ag
es)
is
d
u
e
to
le
g
itima
te
u
s
er
q
u
er
ies
th
at
th
e
ch
atb
o
t
f
ails
t
o
u
n
d
er
s
tan
d
.
T
o
ad
d
r
ess
th
e
r
ig
o
r
o
f
th
is
esti
m
atio
n
,
a
s
en
s
i
tiv
ity
an
aly
s
is
wa
s
p
er
f
o
r
m
ed
to
u
n
d
er
s
tan
d
h
o
w
th
e
f
in
al
f
allb
a
ck
r
ate
w
o
u
ld
ch
an
g
e
b
ased
o
n
s
m
all
v
ar
iatio
n
s
in
th
e
m
o
d
el'
s
p
er
f
o
r
m
an
ce
an
d
d
ata
ass
u
m
p
t
io
n
s
.
T
h
is
a
n
aly
s
is
test
s
th
e
s
tab
ilit
y
o
f
th
e
2
3
.
3
5
%
esti
m
ate
b
y
v
ar
y
in
g
th
e
two
m
o
s
t
cr
itical
m
etr
ics:
th
e
m
o
d
el'
s
r
ec
all
(
it
s
ab
ilit
y
t
o
ca
tch
s
p
am
)
an
d
its
FP
R
(
its
er
r
o
r
r
ate
o
n
leg
itima
te
m
ess
ag
es).
T
h
e
s
e
n
s
itiv
ity
an
aly
s
is
in
T
ab
le
8
d
em
o
n
s
tr
ates
th
at
e
v
en
with
a
1
0
%
f
l
u
ctu
atio
n
in
m
o
d
el
p
er
f
o
r
m
a
n
ce
,
th
e
f
in
al
f
allb
ac
k
r
ate
r
em
ain
s
r
elativ
ely
s
tab
le
(
b
etwe
en
2
2
.
6
7
%
an
d
2
4
.
1
3
%).
T
h
is
f
in
d
in
g
r
ein
f
o
r
ce
s
th
e
p
ap
er
'
s
m
ain
co
n
cl
u
s
io
n
f
o
r
th
is
s
ec
ti
o
n
:
wh
ile
th
e
s
p
am
f
ilter
is
h
i
g
h
ly
ef
f
ec
tiv
e,
t
h
e
m
ajo
r
ity
o
f
th
e
f
allb
ac
k
p
r
o
b
le
m
(
o
v
e
r
4
m
illi
o
n
m
ess
ag
es)
is
ca
u
s
ed
b
y
leg
itima
te,
n
o
n
-
s
p
am
q
u
e
r
ies
th
at
th
e
ch
atb
o
t’
s
co
r
e
i
n
ten
t r
ec
o
g
n
iti
o
n
m
o
d
u
le
ca
n
n
o
t u
n
d
er
s
tan
d
.
T
ab
le
8
.
Sen
s
itiv
ity
an
aly
s
is
o
f
esti
m
ated
f
allb
ac
k
r
ate
S
c
e
n
a
r
i
o
A
ssu
me
d
r
e
c
a
l
l
A
ssu
me
d
F
P
R
Est
i
m
a
t
e
d
n
e
w
f
a
l
l
b
a
c
k
r
a
t
e
P
e
ssi
m
i
st
i
c
(
1
0
%
w
o
r
se
p
e
r
f
o
r
ma
n
c
e
)
0
.
8
2
(
f
r
o
m
0
.
9
1
)
4
.
9
7
%
(
f
r
o
m
4
.
5
2
%)
2
4
.
1
3
%
B
a
se
l
i
n
e
(
a
s
c
a
l
c
u
l
a
t
e
d
)
0
.
9
1
4
.
5
2
%
2
3
.
3
5
%
O
p
t
i
mi
s
t
i
c
(
1
0
%
b
e
t
t
e
r
p
e
r
f
o
r
ma
n
c
e
)
0
.
9
9
(
f
r
o
m
0
.
9
1
)
4
.
0
7
%
(
f
r
o
m
4
.
5
2
%)
2
2
.
6
7
%
4.
CO
NCLU
SI
O
N
T
h
is
r
esear
ch
s
u
cc
ess
f
u
lly
en
g
in
ee
r
ed
a
n
d
e
m
p
ir
ically
ev
a
lu
ated
a
s
u
ite
o
f
m
ac
h
in
e
lea
r
n
in
g
a
n
d
d
ee
p
lear
n
in
g
m
o
d
els
f
o
r
th
e
p
u
r
p
o
s
e
o
f
s
p
am
d
etec
tio
n
with
in
W
h
atsAp
p
ch
atb
o
t
co
m
m
u
n
icatio
n
s
.
T
h
e
f
in
d
in
g
s
f
r
o
m
th
is
r
esear
ch
c
o
n
clu
s
iv
ely
d
em
o
n
s
tr
ate
th
at
d
e
ep
lear
n
in
g
m
o
d
els,
p
ar
ticu
la
r
ly
th
e
C
NN
-
L
STM
an
d
Dis
tilB
E
R
T
ar
ch
itectu
r
es,
ar
e
s
ig
n
if
ican
tly
m
o
r
e
ef
f
ec
ti
v
e
th
an
tr
ad
itio
n
al
m
ac
h
in
e
le
ar
n
in
g
alg
o
r
ith
m
s
f
o
r
th
is
class
if
icat
io
n
task
.
T
h
ese
m
o
d
els
ex
h
ib
ited
h
ig
h
lev
els
o
f
ac
cu
r
ac
y
an
d
a
s
o
p
h
is
ticated
ca
p
ac
ity
f
o
r
u
n
d
er
s
tan
d
i
n
g
th
e
c
o
n
tex
tu
al
n
u
an
ce
s
in
h
er
e
n
t
in
ch
at
b
o
t
m
ess
ag
es.
I
n
ter
m
s
o
f
p
r
ac
tical
i
m
p
ac
t,
a
s
im
u
latio
n
b
ased
o
n
th
e
to
p
-
p
er
f
o
r
m
in
g
C
NN
-
L
STM
m
o
d
el
(
ac
c
u
r
ac
y
0
.
9
4
,
r
ec
all
0
.
9
1
)
p
r
o
jects
th
a
t
its
im
p
lem
en
tatio
n
wo
u
ld
r
ed
u
ce
th
e
ch
atb
o
t'
s
o
v
er
all
f
allb
ac
k
r
ate
f
r
o
m
3
3
%
d
o
wn
to
an
esti
m
ated
2
3
.
3
5
%.
T
h
is
d
ir
ec
tly
ad
d
r
ess
es
th
e
r
esear
ch
p
r
o
b
l
em
b
y
d
em
o
n
s
tr
atin
g
a
s
ig
n
if
ican
t
m
itig
atio
n
o
f
s
p
a
m
-
in
d
u
ce
d
f
allb
ac
k
s
.
Ho
wev
er
,
th
is
an
al
y
s
is
also
r
ev
ea
ls
th
at
a
lar
g
e
v
o
lu
m
e
o
f
f
allb
ac
k
s
is
ca
u
s
ed
b
y
le
g
itima
te,
n
o
n
-
s
p
am
q
u
er
ies,
in
d
icatin
g
th
at
ac
h
i
ev
in
g
th
e
1
5
%
tar
g
et
r
eq
u
ir
es
n
o
t
o
n
ly
s
p
am
f
ilter
in
g
b
u
t
also
s
u
b
s
tan
tial
im
p
r
o
v
em
e
n
ts
to
th
e
ch
atb
o
t's
co
r
e
in
ten
t
r
ec
o
g
n
itio
n
m
o
d
u
le.
Fro
m
a
n
in
f
o
r
m
atics
an
d
en
g
in
ee
r
in
g
p
er
s
p
ec
tiv
e,
th
e
s
tr
o
n
g
p
er
f
o
r
m
an
ce
o
f
th
e
m
o
r
e
co
m
p
u
t
atio
n
ally
ef
f
icien
t
C
NN
-
L
STM
an
d
Dis
tilB
E
R
T
m
o
d
els
is
h
ig
h
ly
s
ig
n
if
ican
t.
T
h
is
ef
f
icien
cy
is
cr
itica
l
f
o
r
r
ea
l
-
tim
e
p
r
o
ce
s
s
in
g
in
h
ig
h
-
th
r
o
u
g
h
p
u
t
s
m
ar
t
cu
s
to
m
er
s
er
v
ice
s
y
s
tem
s
,
as
lo
w
laten
cy
is
ess
en
tial
f
o
r
u
s
er
s
atis
f
ac
tio
n
.
Fu
r
t
h
er
m
o
r
e
,
th
is
h
as
d
ir
ec
t
s
y
s
tem
-
lev
el
i
n
teg
r
ati
o
n
im
p
licatio
n
s
r
elev
a
n
t
to
elec
tr
ical
en
g
in
ee
r
in
g
.
U
n
lik
e
lar
g
e
r
m
o
d
els
(
lik
e
B
E
R
T
-
b
ase
o
r
XL
M
-
R
OB
E
R
T
a)
th
at
o
f
ten
r
eq
u
i
r
e
h
e
av
y
clo
u
d
r
eso
u
r
ce
s
,
t
h
ese
ef
f
icien
t
ar
ch
itectu
r
es
ar
e
id
ea
l
ca
n
d
id
ates
f
o
r
ed
g
e
d
ep
lo
y
m
e
n
t.
T
h
ey
ar
e
v
iab
le
f
o
r
im
p
lem
en
tatio
n
in
em
b
ed
d
ed
n
atu
r
al
lan
g
u
ag
e
p
r
o
ce
s
s
in
g
(
NL
P
)
m
o
d
u
les
o
n
lo
ca
l
s
er
v
er
s
,
wh
ich
wo
u
ld
r
e
d
u
ce
n
etwo
r
k
d
e
p
en
d
e
n
cy
,
m
i
n
im
ize
ap
p
licatio
n
p
r
o
g
r
a
m
m
in
g
in
ter
f
ac
e
(
API
)
laten
cy
,
an
d
lo
wer
o
p
er
atio
n
al
co
s
ts
.
B
u
ild
in
g
o
n
th
ese
f
i
n
d
in
g
s
,
av
e
n
u
es
f
o
r
f
u
tu
r
e
r
esear
ch
in
clu
d
e
th
e
ex
p
lo
r
atio
n
an
d
f
in
e
-
t
u
n
in
g
o
f
o
t
h
er
a
d
v
an
ce
d
tr
a
n
s
f
o
r
m
er
-
b
ased
m
o
d
els
s
p
ec
if
ically
tr
ain
ed
o
n
I
n
d
o
n
e
s
ian
,
s
u
ch
as
I
n
d
o
n
esian
B
E
R
T
(
I
ndo
B
E
R
T
)
an
d
a
lite
B
E
R
T
(
AL
B
E
R
T
)
,
o
r
th
e
d
ev
elo
p
m
en
t
o
f
n
o
v
el
h
y
b
r
id
ar
c
h
itectu
r
es
d
esig
n
ed
t
o
m
o
r
e
ef
f
ec
tiv
ely
co
u
n
ter
th
e
ev
o
lv
in
g
tactics
o
f
s
p
am
m
er
s
.
Fu
r
th
er
m
o
r
e,
th
e
e
x
p
an
s
io
n
o
f
th
e
tr
ain
in
g
d
atas
et
an
d
th
e
ex
ec
u
tio
n
o
f
r
ea
l
-
ti
m
e
A/B
te
s
tin
g
to
v
alid
ate
th
e
s
im
u
lated
f
allb
ac
k
r
ed
u
ctio
n
wo
u
ld
c
o
n
s
titu
te
v
a
lu
ab
le
n
ex
t step
s
f
o
r
th
is
r
esear
ch
.
F
UNDING
I
NF
O
R
M
A
T
I
O
N
Au
th
o
r
s
s
tate
n
o
f
u
n
d
in
g
in
v
o
lv
ed
.
AUTHO
R
CO
NT
RI
B
UT
I
O
NS ST
A
T
E
M
E
N
T
T
h
is
jo
u
r
n
al
u
s
es
th
e
C
o
n
tr
ib
u
to
r
R
o
les
T
ax
o
n
o
m
y
(
C
R
ed
iT
)
to
r
ec
o
g
n
ize
in
d
iv
id
u
al
au
th
o
r
co
n
tr
ib
u
tio
n
s
,
r
ed
u
ce
au
th
o
r
s
h
ip
d
is
p
u
tes,
an
d
f
ac
ilit
ate
co
llab
o
r
atio
n
.
Na
m
e
o
f
Aut
ho
r
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
Satr
io
Sad
ewo
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
Am
alia
Z
ah
r
a
✓
✓
✓
✓
✓
✓
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Dee
p
lea
r
n
in
g
-
b
a
s
ed
s
p
a
m
d
et
ec
tio
n
fo
r
Wh
a
ts
A
p
p
ch
a
tb
o
t f
a
llb
a
ck
r
ed
u
ctio
n
(
S
a
tr
io
S
a
d
e
w
o
)
917
C
:
C
o
n
c
e
p
t
u
a
l
i
z
a
t
i
o
n
M
:
M
e
t
h
o
d
o
l
o
g
y
So
:
So
f
t
w
a
r
e
Va
:
Va
l
i
d
a
t
i
o
n
Fo
:
Fo
r
mal
a
n
a
l
y
s
i
s
I
:
I
n
v
e
s
t
i
g
a
t
i
o
n
R
:
R
e
so
u
r
c
e
s
D
:
D
a
t
a
C
u
r
a
t
i
o
n
O
:
W
r
i
t
i
n
g
-
O
r
i
g
i
n
a
l
D
r
a
f
t
E
:
W
r
i
t
i
n
g
-
R
e
v
i
e
w
&
E
d
i
t
i
n
g
Vi
:
Vi
su
a
l
i
z
a
t
i
o
n
Su
:
Su
p
e
r
v
i
s
i
o
n
P
:
P
r
o
j
e
c
t
a
d
mi
n
i
st
r
a
t
i
o
n
Fu
:
Fu
n
d
i
n
g
a
c
q
u
i
si
t
i
o
n
CO
NF
L
I
C
T
O
F
I
N
T
E
R
E
S
T
ST
A
T
E
M
E
NT
Au
th
o
r
s
s
tate
n
o
co
n
f
lict o
f
in
t
er
est.
DATA AV
AI
L
AB
I
L
I
T
Y
T
h
e
d
ata
t
h
at
s
u
p
p
o
r
t
th
e
f
in
d
in
g
s
o
f
t
h
is
s
tu
d
y
a
r
e
p
r
o
p
r
iet
ar
y
an
d
ar
e
n
o
t
p
u
b
licly
av
ail
ab
le
d
u
e
to
p
r
iv
ac
y
a
n
d
co
m
m
er
cial
s
en
s
itiv
ity
.
RE
F
E
R
E
NC
E
S
[
1
]
S
.
K
e
m
p
,
“
D
i
g
i
t
a
l
2
0
2
3
:
g
l
o
b
a
l
o
v
e
r
v
i
e
w
r
e
p
o
r
t
,
”
d
a
t
a
r
e
p
o
r
t
a
l
.
c
o
m
.
A
c
c
e
ss
e
d
:
N
o
v
.
0
4
,
2
0
2
3
.
[
O
n
l
i
n
e
]
.
A
v
a
i
l
a
b
l
e
:
h
t
t
p
s
:
/
/
d
a
t
a
r
e
p
o
r
t
a
l
.
c
o
m/
r
e
p
o
r
t
s
/
d
i
g
i
t
a
l
-
2
0
2
3
-
g
l
o
b
a
l
-
o
v
e
r
v
i
e
w
-
r
e
p
o
r
t
[
2
]
C
.
V
.
M
i
si
s
c
h
i
a
,
F
.
P
o
e
c
z
e
,
a
n
d
C
.
S
t
r
a
u
ss,
“
C
h
a
t
b
o
t
s
i
n
c
u
s
t
o
mer
ser
v
i
c
e
:
t
h
e
i
r
r
e
l
e
v
a
n
c
e
a
n
d
i
m
p
a
c
t
o
n
ser
v
i
c
e
q
u
a
l
i
t
y
,
”
Pro
c
e
d
i
a
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
,
v
o
l
.
2
0
1
,
p
p
.
4
2
1
–
4
2
8
,
2
0
2
2
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
p
r
o
c
s.
2
0
2
2
.
0
3
.
0
5
5
.
[
3
]
V
.
K
a
u
s
h
a
l
a
n
d
R
.
Y
a
d
a
v
,
“
L
e
a
r
n
i
n
g
s
u
c
c
e
ssf
u
l
i
m
p
l
e
me
n
t
a
t
i
o
n
o
f
c
h
a
t
b
o
t
s
i
n
b
u
s
i
n
e
sses
f
r
o
m
B
2
B
c
u
st
o
mer
e
x
p
e
r
i
e
n
c
e
p
e
r
s
p
e
c
t
i
v
e
,
”
C
o
n
c
u
r
re
n
c
y
a
n
d
C
o
m
p
u
t
a
t
i
o
n
:
P
ra
c
t
i
c
e
a
n
d
E
x
p
e
ri
e
n
c
e
,
v
o
l
.
3
5
,
n
o
.
1
,
Ja
n
.
2
0
2
3
,
d
o
i
:
1
0
.
1
0
0
2
/
c
p
e
.
7
4
5
0
.
[
4
]
L.
A
n
a
y
a
,
A
.
B
r
a
i
z
a
t
,
a
n
d
R
.
A
l
-
A
n
i
,
“
I
mp
l
e
me
n
t
i
n
g
A
I
-
b
a
se
d
c
h
a
t
b
o
t
:
b
e
n
e
f
i
t
s
a
n
d
c
h
a
l
l
e
n
g
e
s
,
”
P
ro
c
e
d
i
a
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
,
v
o
l
.
2
3
9
,
p
p
.
1
1
7
3
–
1
1
7
9
,
2
0
2
4
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
p
r
o
c
s.
2
0
2
4
.
0
6
.
2
8
4
.
[
5
]
A
.
K
.
S
h
r
i
v
a
s,
A
.
K
.
D
e
w
a
n
g
a
n
,
S
.
M
.
G
h
o
sh
,
a
n
d
D
.
S
i
n
g
h
,
“
D
e
v
e
l
o
p
me
n
t
o
f
p
r
o
p
o
se
d
e
n
sem
b
l
e
m
o
d
e
l
f
o
r
sp
a
m
e
-
ma
i
l
c
l
a
ss
i
f
i
c
a
t
i
o
n
,
”
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
n
t
r
o
l
,
v
o
l
.
5
0
,
n
o
.
3
,
S
e
p
.
2
0
2
1
,
d
o
i
:
1
0
.
5
7
5
5
/
j
0
1
.
i
t
c
.
5
0
.
3
.
2
7
3
4
9
.
[
6
]
K
.
D
e
b
n
a
t
h
a
n
d
N
.
K
a
r
,
“
Em
a
i
l
s
p
a
m
d
e
t
e
c
t
i
o
n
u
s
i
n
g
d
e
e
p
l
e
a
r
n
i
n
g
a
p
p
r
o
a
c
h
,
”
i
n
2
0
2
2
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
M
a
c
h
i
n
e
L
e
a
r
n
i
n
g
,
Bi
g
D
a
t
a
,
C
l
o
u
d
a
n
d
Pa
r
a
l
l
e
l
C
o
m
p
u
t
i
n
g
(
C
O
M
-
IT
-
C
O
N
)
,
M
a
y
2
0
2
2
,
p
p
.
3
7
–
41
,
d
o
i
:
1
0
.
1
1
0
9
/
C
O
M
-
IT
-
C
O
N
5
4
6
0
1
.
2
0
2
2
.
9
8
5
0
5
8
8
.
[
7
]
S
.
K
a
d
d
o
u
r
a
,
G
.
C
h
a
n
d
r
a
se
k
a
r
a
n
,
D
.
E
.
P
o
p
e
sc
u
,
a
n
d
J
.
H
.
D
u
r
a
i
sam
y
,
“
A
sy
st
e
ma
t
i
c
l
i
t
e
r
a
t
u
r
e
r
e
v
i
e
w
o
n
s
p
a
m
c
o
n
t
e
n
t
d
e
t
e
c
t
i
o
n
a
n
d
c
l
a
ss
i
f
i
c
a
t
i
o
n
,
”
Pe
e
rJ
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
,
v
o
l
.
8
,
J
a
n
.
2
0
2
2
,
d
o
i
:
1
0
.
7
7
1
7
/
p
e
e
r
j
-
c
s
.
8
3
0
.
[
8
]
T.
Jai
n
,
P
.
G
a
r
g
,
N
.
C
h
a
l
i
l
,
A
.
S
i
n
h
a
,
V
.
K
.
V
e
r
m
a
,
a
n
d
R
.
G
u
p
t
a
,
“
S
M
S
sp
a
m
c
l
a
ssi
f
i
c
a
t
i
o
n
u
s
i
n
g
m
a
c
h
i
n
e
l
e
a
r
n
i
n
g
t
e
c
h
n
i
q
u
e
s,
”
i
n
2
0
2
2
1
2
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
C
l
o
u
d
C
o
m
p
u
t
i
n
g
,
D
a
t
a
S
c
i
e
n
c
e
&
E
n
g
i
n
e
e
ri
n
g
(
C
o
n
f
l
u
e
n
c
e
)
,
2
0
2
2
,
p
p
.
2
7
3
–
2
7
9
,
d
o
i
:
1
0
.
1
1
0
9
/
C
o
n
f
l
u
e
n
c
e
5
2
9
8
9
.
2
0
2
2
.
9
7
3
4
1
2
8
.
[
9
]
M
.
R
a
z
a
,
N
.
D
.
J
a
y
a
si
n
g
h
e
,
a
n
d
M
.
M
.
A
.
M
u
s
l
a
m,
“
A
c
o
mp
r
e
h
e
n
s
i
v
e
r
e
v
i
e
w
o
n
e
ma
i
l
sp
a
m
c
l
a
ssi
f
i
c
a
t
i
o
n
u
si
n
g
m
a
c
h
i
n
e
l
e
a
r
n
i
n
g
a
l
g
o
r
i
t
h
ms,
”
i
n
2
0
2
1
I
n
t
e
rn
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
o
n
I
n
f
o
rm
a
t
i
o
n
N
e
t
w
o
rk
i
n
g
(
I
C
O
I
N
)
,
Ja
n
.
2
0
2
1
,
p
p
.
3
2
7
–
332
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
O
I
N
5
0
8
8
4
.
2
0
2
1
.
9
3
3
4
0
2
0
.
[
1
0
]
S
.
G
a
d
d
e
,
A
.
L
a
k
s
h
m
a
n
a
r
a
o
,
a
n
d
S
.
S
a
t
y
a
n
a
r
a
y
a
n
a
,
“
S
M
S
s
p
a
m
d
e
t
e
c
t
i
o
n
u
si
n
g
ma
c
h
i
n
e
l
e
a
r
n
i
n
g
a
n
d
d
e
e
p
l
e
a
r
n
i
n
g
t
e
c
h
n
i
q
u
e
s
,
”
i
n
2
0
2
1
7
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
C
o
m
p
u
t
i
n
g
a
n
d
C
o
m
m
u
n
i
c
a
t
i
o
n
S
y
st
e
m
s
(
I
C
AC
C
S
)
,
M
a
r
.
2
0
2
1
,
p
p
.
3
5
8
–
3
6
2
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
A
C
C
S
5
1
4
3
0
.
2
0
2
1
.
9
4
4
1
7
8
3
.
[
1
1
]
J.
D
e
v
l
i
n
,
M
.
W
.
C
h
a
n
g
,
K
.
L
e
e
,
a
n
d
K
.
T
o
u
t
a
n
o
v
a
,
“
B
E
R
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
,
”
i
n
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
2
0
1
9
C
o
n
f
e
r
e
n
c
e
o
f
t
h
e
N
o
rt
h
,
2
0
1
9
,
p
p
.
4
1
7
1
–
4
1
8
6
,
d
o
i
:
1
0
.
1
8
6
5
3
/
v
1
/
N
1
9
-
1
4
2
3
.
[
1
2
]
B
.
S
o
n
a
r
e
,
G
.
J.
D
h
a
r
mal
e
,
A
.
R
e
n
a
p
u
r
e
,
H
.
K
h
a
n
d
e
l
w
a
l
,
a
n
d
S
.
N
a
r
h
a
r
sh
e
t
t
i
w
a
r
,
“
E
-
mai
l
sp
a
m
d
e
t
e
c
t
i
o
n
u
s
i
n
g
m
a
c
h
i
n
e
l
e
a
r
n
i
n
g
,
”
i
n
2
0
2
3
4
t
h
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
f
o
r
Em
e
r
g
i
n
g
T
e
c
h
n
o
l
o
g
y
(
I
N
C
ET)
,
M
a
y
2
0
2
3
,
p
p
.
1
–
5
,
d
o
i
:
1
0
.
1
1
0
9
/
I
N
C
ET
5
7
9
7
2
.
2
0
2
3
.
1
0
1
7
0
1
8
7
.
[
1
3
]
V
.
I
.
D
.
R
o
sar
i
o
,
B
.
D
.
P
.
F
e
r
n
a
n
d
e
z
,
a
n
d
D
.
A
.
P
a
d
i
l
l
a
,
“
Ema
i
l
sp
a
m
c
l
a
ss
i
f
i
c
a
t
i
o
n
u
s
i
n
g
D
i
st
i
l
B
E
R
T,
”
i
n
2
0
2
3
I
EE
E
1
5
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
H
u
m
a
n
o
i
d
,
N
a
n
o
t
e
c
h
n
o
l
o
g
y
,
I
n
f
o
rm
a
t
i
o
n
T
e
c
h
n
o
l
o
g
y
,
C
o
m
m
u
n
i
c
a
t
i
o
n
a
n
d
C
o
n
t
r
o
l
,
En
v
i
r
o
n
m
e
n
t
,
a
n
d
Ma
n
a
g
e
m
e
n
t
(
H
N
I
C
E
M)
,
2
0
2
3
,
p
p
.
1
–
6
,
d
o
i
:
1
0
.
1
1
0
9
/
H
N
I
C
EM
6
0
6
7
4
.
2
0
2
3
.
1
0
5
8
9
2
1
1
.
[
1
4
]
V
.
S
.
V
i
n
i
t
h
a
,
D
.
K
.
R
e
n
u
k
a
,
a
n
d
L
.
A
.
K
u
mar
,
“
L
o
n
g
s
h
o
r
t
-
t
e
r
m
m
e
m
o
r
y
n
e
t
w
o
r
k
s
f
o
r
e
m
a
i
l
sp
a
m
c
l
a
ss
i
f
i
c
a
t
i
o
n
,
”
i
n
2
0
2
3
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
I
n
t
e
l
l
i
g
e
n
t
S
y
st
e
m
s
f
o
r
C
o
m
m
u
n
i
c
a
t
i
o
n
,
I
o
T
a
n
d
S
e
c
u
r
i
t
y
(
I
C
I
S
C
o
I
S
)
,
2
0
2
3
,
p
p
.
1
7
6
–
1
8
0
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
I
S
C
o
I
S
5
6
5
4
1
.
2
0
2
3
.
1
0
1
0
0
4
4
5
.
[
1
5
]
G
.
Jam
e
s,
D
.
W
i
t
t
e
n
,
T.
H
a
s
t
i
e
,
a
n
d
R
.
T
i
b
s
h
i
r
a
n
i
,
A
n
i
n
t
ro
d
u
c
t
i
o
n
t
o
st
a
t
i
st
i
c
a
l
l
e
a
r
n
i
n
g
,
v
o
l
.
1
0
3
.
i
n
S
p
r
i
n
g
e
r
T
e
x
t
s
i
n
S
t
a
t
i
st
i
c
s,
v
o
l
.
1
0
3
.
N
e
w
Y
o
r
k
,
U
n
i
t
e
d
S
t
a
t
e
s
:
S
p
r
i
n
g
e
r
N
e
w
Y
o
r
k
,
2
0
1
3
,
d
o
i
:
1
0
.
1
0
0
7
/
9
7
8
-
1
-
4
6
1
4
-
7
1
3
8
-
7.
[
1
6
]
R
.
E
.
S
c
h
a
p
i
r
e
,
“
T
h
e
b
o
o
st
i
n
g
a
p
p
r
o
a
c
h
t
o
m
a
c
h
i
n
e
l
e
a
r
n
i
n
g
:
a
n
o
v
e
r
v
i
e
w
,
”
i
n
N
o
n
l
i
n
e
a
r
Es
t
i
m
a
t
i
o
n
a
n
d
C
l
a
s
si
f
i
c
a
t
i
o
n
,
N
e
w
Y
o
r
k
,
U
n
i
t
e
d
S
t
a
t
e
s
:
S
p
r
i
n
g
e
r
,
2
0
0
3
,
p
p
.
1
4
9
–
1
7
1
,
d
o
i
:
1
0
.
1
0
0
7
/
9
7
8
-
0
-
3
8
7
-
2
1
5
7
9
-
2
_
9
.
[
1
7
]
K
.
P
o
t
d
a
r
,
T
.
S
.
P
a
r
d
a
w
a
l
a
,
a
n
d
C
.
D.
P
a
i
,
“
A
c
o
m
p
a
r
a
t
i
v
e
st
u
d
y
o
f
c
a
t
e
g
o
r
i
c
a
l
v
a
r
i
a
b
l
e
e
n
c
o
d
i
n
g
t
e
c
h
n
i
q
u
e
s
f
o
r
n
e
u
r
a
l
n
e
t
w
o
r
k
c
l
a
ss
i
f
i
e
r
s,”
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
rn
a
l
o
f
C
o
m
p
u
t
e
r
A
p
p
l
i
c
a
t
i
o
n
s
,
v
o
l
.
1
7
5
,
n
o
.
4
,
p
p
.
7
–
9
,
O
c
t
.
2
0
1
7
,
d
o
i
:
1
0
.
5
1
2
0
/
i
j
c
a
2
0
1
7
9
1
5
4
9
5
.
[
1
8
]
S
.
B
i
r
d
,
E.
K
l
e
i
n
,
a
n
d
E.
L
o
p
e
r
,
N
a
t
u
r
a
l
l
a
n
g
u
a
g
e
p
r
o
c
e
ssi
n
g
w
i
t
h
Py
t
h
o
n
.
S
e
b
a
s
t
o
p
o
l
,
U
n
i
t
e
d
S
t
a
t
e
s:
O
’
R
e
i
l
l
y
M
e
d
i
a
,
I
n
c
.
,
2
0
0
9
.
[
1
9
]
P
.
W
a
n
g
,
B
.
X
u
,
J.
X
u
,
G
.
T
i
a
n
,
C
.
-
L.
Li
u
,
a
n
d
H
.
H
a
o
,
“
S
e
ma
n
t
i
c
e
x
p
a
n
si
o
n
u
s
i
n
g
w
o
r
d
e
mb
e
d
d
i
n
g
c
l
u
s
t
e
r
i
n
g
a
n
d
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
f
o
r
i
mp
r
o
v
i
n
g
sh
o
r
t
t
e
x
t
c
l
a
ssi
f
i
c
a
t
i
o
n
,
”
N
e
u
r
o
c
o
m
p
u
t
i
n
g
,
v
o
l
.
1
7
4
,
p
p
.
8
0
6
–
8
1
4
,
Ja
n
.
2
0
1
6
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
n
e
u
c
o
m.
2
0
1
5
.
0
9
.
0
9
6
.
[
2
0
]
M
.
H
a
r
o
o
n
,
“
C
o
m
p
a
r
a
t
i
v
e
a
n
a
l
y
si
s
o
f
st
e
mm
i
n
g
a
l
g
o
r
i
t
h
ms
f
o
r
w
e
b
t
e
x
t
m
i
n
i
n
g
,
”
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
r
n
a
l
o
f
Mo
d
e
rn
Ed
u
c
a
t
i
o
n
a
n
d
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
,
v
o
l
.
1
0
,
n
o
.
9
,
p
p
.
2
0
–
2
5
,
S
e
p
.
2
0
1
8
,
d
o
i
:
1
0
.
5
8
1
5
/
i
j
m
e
c
s.2
0
1
8
.
0
9
.
0
3
.
[
2
1
]
N
.
N
.
A
.
S
j
a
r
i
f
,
N
.
F
.
M
.
A
z
m
i
,
S
.
C
h
u
p
r
a
t
,
H
.
M
.
S
a
r
k
a
n
,
Y
.
Y
a
h
y
a
,
a
n
d
S
.
M
.
S
a
m,
“
S
M
S
s
p
a
m
m
e
ssa
g
e
d
e
t
e
c
t
i
o
n
u
si
n
g
t
e
r
m
f
r
e
q
u
e
n
c
y
-
i
n
v
e
r
s
e
d
o
c
u
me
n
t
f
r
e
q
u
e
n
c
y
a
n
d
r
a
n
d
o
m
f
o
r
e
st
a
l
g
o
r
i
t
h
m,”
Pr
o
c
e
d
i
a
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
,
v
o
l
.
1
6
1
,
p
p
.
5
0
9
–
5
1
5
,
2
0
1
9
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
p
r
o
c
s.
2
0
1
9
.
1
1
.
1
5
0
.
[
2
2
]
Y
.
W
u
,
M
.
Y
u
a
n
,
S
.
D
o
n
g
,
L.
Li
n
,
a
n
d
Y
.
L
i
u
,
“
R
e
m
a
i
n
i
n
g
u
s
e
f
u
l
l
i
f
e
e
st
i
mat
i
o
n
o
f
e
n
g
i
n
e
e
r
e
d
s
y
s
t
e
ms
u
si
n
g
v
a
n
i
l
l
a
LST
M
n
e
u
r
a
l
n
e
t
w
o
r
k
s
,
”
N
e
u
r
o
c
o
m
p
u
t
i
n
g
,
v
o
l
.
2
7
5
,
p
p
.
1
6
7
–
1
7
9
,
Ja
n
.
2
0
1
8
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
n
e
u
c
o
m.
2
0
1
7
.
0
5
.
0
6
3
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2252
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
1
,
Feb
r
u
ar
y
20
26
:
909
-
9
1
8
918
[
2
3
]
J.
Li
,
“
A
c
o
mp
a
r
a
t
i
v
e
st
u
d
y
o
f
LST
M
v
a
r
i
a
n
t
s
i
n
p
r
e
d
i
c
t
i
o
n
f
o
r
T
e
sl
a
’
s
st
o
c
k
p
r
i
c
e
,
”
BC
P
Bu
s
i
n
e
ss
&
M
a
n
a
g
e
m
e
n
t
,
v
o
l
.
3
4
,
p
p
.
3
0
–
3
8
,
D
e
c
.
2
0
2
2
,
d
o
i
:
1
0
.
5
4
6
9
1
/
b
c
p
b
m.
v
3
4
i
.
2
8
6
1
.
[
2
4
]
N
.
C
h
e
n
,
“
E
x
p
l
o
r
i
n
g
t
h
e
d
e
v
e
l
o
p
me
n
t
a
n
d
a
p
p
l
i
c
a
t
i
o
n
o
f
LST
M
v
a
r
i
a
n
t
s
,
”
Ap
p
l
i
e
d
a
n
d
C
o
m
p
u
t
a
t
i
o
n
a
l
En
g
i
n
e
e
ri
n
g
,
v
o
l
.
5
3
,
n
o
.
1
,
p
p
.
1
0
3
–
1
0
7
,
M
a
r
.
2
0
2
4
,
d
o
i
:
1
0
.
5
4
2
5
4
/
2
7
5
5
-
2
7
2
1
/
5
3
/
2
0
2
4
1
2
8
8
.
[
2
5
]
S
.
K
.
A
k
p
a
t
sa
e
t
a
l
.
,
“
O
n
l
i
n
e
n
e
w
s
s
e
n
t
i
m
e
n
t
c
l
a
ssi
f
i
c
a
t
i
o
n
u
si
n
g
D
i
s
t
i
l
B
ER
T,
”
J
o
u
r
n
a
l
o
f
Q
u
a
n
t
u
m
C
o
m
p
u
t
i
n
g
,
v
o
l
.
4
,
n
o
.
1
,
p
p
.
1
–
1
1
,
2
0
2
2
,
d
o
i
:
1
0
.
3
2
6
0
4
/
j
q
c
.
2
0
2
2
.
0
2
6
6
5
8
.
[
2
6
]
L.
Y
a
n
g
a
n
d
A
.
S
h
a
mi
,
“
O
n
h
y
p
e
r
p
a
r
a
met
e
r
o
p
t
i
mi
z
a
t
i
o
n
o
f
ma
c
h
i
n
e
l
e
a
r
n
i
n
g
a
l
g
o
r
i
t
h
ms
:
t
h
e
o
r
y
a
n
d
p
r
a
c
t
i
c
e
,
”
N
e
u
ro
c
o
m
p
u
t
i
n
g
,
v
o
l
.
4
1
5
,
p
p
.
2
9
5
–
3
1
6
,
N
o
v
.
2
0
2
0
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
n
e
u
c
o
m
.
2
0
2
0
.
0
7
.
0
6
1
.
[
2
7
]
M
.
H
o
ss
i
n
a
n
d
M
.
N
.
S
u
l
a
i
m
a
n
,
“
A
r
e
v
i
e
w
o
n
e
v
a
l
u
a
t
i
o
n
me
t
r
i
c
s
f
o
r
d
a
t
a
c
l
a
ss
i
f
i
c
a
t
i
o
n
e
v
a
l
u
a
t
i
o
n
s,
”
I
n
t
e
rn
a
t
i
o
n
a
l
J
o
u
r
n
a
l
o
f
D
a
t
a
M
i
n
i
n
g
&
K
n
o
w
l
e
d
g
e
M
a
n
a
g
e
m
e
n
t
Pr
o
c
e
ss
,
v
o
l
.
5
,
n
o
.
2
,
p
p
.
1
–
1
1
,
M
a
r
.
2
0
1
5
,
d
o
i
:
1
0
.
5
1
2
1
/
i
j
d
k
p
.
2
0
1
5
.
5
2
0
1
.
[
2
8
]
S
.
La
q
t
i
b
,
K
.
E
.
Y
a
ss
i
n
i
,
a
n
d
M
.
L
.
H
a
sn
a
o
u
i
,
“
A
t
e
c
h
n
i
c
a
l
r
e
v
i
e
w
a
n
d
c
o
mp
a
r
a
t
i
v
e
a
n
a
l
y
si
s
o
f
mac
h
i
n
e
l
e
a
r
n
i
n
g
t
e
c
h
n
i
q
u
e
s
f
o
r
i
n
t
r
u
s
i
o
n
d
e
t
e
c
t
i
o
n
s
y
st
e
ms
i
n
M
A
N
ET,
”
I
n
t
e
rn
a
t
i
o
n
a
l
J
o
u
rn
a
l
o
f
E
l
e
c
t
ri
c
a
l
a
n
d
C
o
m
p
u
t
e
r
E
n
g
i
n
e
e
ri
n
g
,
v
o
l
.
1
0
,
n
o
.
3
,
p
p
.
2
7
0
1
–
2
7
0
9
,
Ju
n
.
2
0
2
0
,
d
o
i
:
1
0
.
1
1
5
9
1
/
i
j
e
c
e
.
v
1
0
i
3
.
p
p
2
7
0
1
-
2
7
0
9
.
B
I
O
G
RAP
H
I
E
S O
F
AUTH
O
RS
S
a
tr
io
S
a
d
e
wo
is
c
u
rre
n
t
ly
p
u
rsu
in
g
a
m
a
ste
r
o
f
Co
m
p
u
ter
S
c
ie
n
c
e
d
e
g
re
e
a
t
th
e
BINU
S
G
r
a
d
u
a
te
P
ro
g
ra
m
,
Bin
a
Nu
sa
n
tara
Un
iv
e
rsity
,
Ja
k
a
rta,
In
d
o
n
e
sia
.
He
re
c
e
iv
e
d
h
is
Ba
c
h
e
lo
r
o
f
En
g
i
n
e
e
rin
g
(S
.
Ko
m
.
)
in
In
f
o
rm
a
ti
c
s
fro
m
Un
iv
e
rsita
s
Isla
m
In
d
o
n
e
sia
(UII)
in
2
0
1
5
.
Alo
n
g
si
d
e
h
is
st
u
d
ies
,
h
e
c
u
rre
n
t
ly
wo
r
k
s
a
s
a
p
r
o
f
e
ss
io
n
a
l
a
t
a
n
I
n
d
o
n
e
sia
n
G
o
v
e
rn
m
e
n
t
Ag
e
n
c
y
.
His
p
rim
a
ry
re
se
a
rc
h
in
tere
sts
i
n
c
lu
d
e
a
rti
ficia
l
in
tell
ig
e
n
c
e
(AI),
m
a
c
h
in
e
lea
rn
in
g
,
a
n
d
d
e
e
p
lea
rn
in
g
.
Th
is
p
a
p
e
r
is
su
b
m
it
ted
a
s
a
p
a
rti
a
l
fu
lfi
ll
m
e
n
t
o
f
t
h
e
re
q
u
irem
e
n
ts
fo
r
h
is
m
a
ste
r'
s
d
e
g
re
e
g
ra
d
u
a
ti
o
n
.
He
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
sa
tri
o
.
sa
d
e
wo
@b
i
n
u
s.a
c
.
i
d
.
Am
a
li
a
Za
h
r
a
h
o
l
d
s
a
P
h
.
D.
i
n
Co
m
p
u
ter
S
c
ien
c
e
fr
o
m
th
e
S
c
h
o
o
l
o
f
Co
m
p
u
ter
S
c
ien
c
e
a
n
d
In
f
o
rm
a
ti
c
s,
Un
iv
e
r
sity
Co
l
leg
e
Du
b
li
n
(UCD
),
Ire
l
a
n
d
,
o
b
tai
n
e
d
i
n
2
0
1
4
.
S
h
e
re
c
e
iv
e
d
h
e
r
b
a
c
h
e
lo
r'
s
d
e
g
re
e
i
n
C
o
m
p
u
ter
S
c
ien
c
e
f
ro
m
th
e
U
n
iv
e
rsit
y
o
f
In
d
o
n
e
sia
(UI),
In
d
o
n
e
sia
,
in
2
0
0
8
.
S
h
e
is
c
u
rr
e
n
tl
y
a
lec
tu
re
r
a
n
d
q
u
a
li
ty
c
o
o
rd
i
n
a
to
r
a
t
t
h
e
m
a
ste
r
o
f
Co
m
p
u
ter
S
c
ien
c
e
P
ro
g
ra
m
,
Bi
n
a
Nu
sa
n
tara
Un
iv
e
rsit
y
,
a
p
o
siti
o
n
sh
e
h
a
s
h
e
l
d
si
n
c
e
2
0
1
7
.
P
rio
r
t
o
t
h
is,
sh
e
se
rv
e
d
a
s
a
lec
tu
re
r
a
n
d
re
se
a
rc
h
e
r
a
t
th
e
F
a
c
u
lt
y
o
f
C
o
m
p
u
ter
S
c
ien
c
e
,
Un
iv
e
rsity
o
f
I
n
d
o
n
e
sia
.
He
r
re
se
a
rc
h
in
tere
sts
in
c
lu
d
e
sp
e
e
c
h
p
r
o
c
e
ss
in
g
,
sp
e
e
c
h
re
c
o
g
n
it
i
o
n
,
sp
e
a
k
e
r
re
c
o
g
n
it
i
o
n
,
sp
o
k
e
n
la
n
g
u
a
g
e
i
d
e
n
ti
f
ica
ti
o
n
,
m
a
c
h
in
e
lea
rn
in
g
,
d
e
e
p
lea
rn
i
n
g
,
b
ig
d
a
t
a
a
n
a
ly
ti
c
s,
a
n
d
c
o
m
p
u
tati
o
n
a
l
in
telli
g
e
n
c
e
.
S
h
e
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
a
m
a
li
a
.
z
a
h
ra
@b
in
u
s.a
c
.
id
.
Evaluation Warning : The document was created with Spire.PDF for Python.