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.
1
5
,
No
.
1
,
Feb
r
u
ar
y
2
0
2
6
,
p
p
.
962
~
971
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
6
2
-
9
7
1
962
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
i
.
ia
esco
r
e.
co
m
Enha
nced f
ra
mewo
rk f
o
r d
etec
tin
g
Vietnam
ese
ha
t
e and
o
ff
ensiv
e spa
ns
Dinh
-
H
o
ng
Vu
1
,
T
uo
ng
L
e
2
1
N
a
t
u
r
a
l
La
n
g
u
a
g
e
P
r
o
c
e
ss
i
n
g
a
n
d
K
n
o
w
l
e
d
g
e
D
i
sc
o
v
e
r
y
R
e
se
a
r
c
h
G
r
o
u
p
,
F
a
c
u
l
t
y
o
f
I
n
f
o
r
ma
t
i
o
n
T
e
c
h
n
o
l
o
g
y
,
To
n
D
u
c
T
h
a
n
g
U
n
i
v
e
r
s
i
t
y
,
H
o
C
h
i
M
i
n
h
C
i
t
y
,
V
i
e
t
n
a
m
2
F
a
c
u
l
t
y
o
f
I
n
f
o
r
m
a
t
i
o
n
T
e
c
h
n
o
l
o
g
y
,
H
U
TEC
H
U
n
i
v
e
r
si
t
y
,
H
o
C
h
i
M
i
n
h
C
i
t
y
,
V
i
e
t
n
a
m
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
J
u
n
1
3
,
2
0
2
5
R
ev
is
ed
J
an
1
,
2
0
2
6
Acc
ep
ted
J
an
2
2
,
2
0
2
6
Th
e
rise
o
f
h
a
te
a
n
d
o
ffe
n
si
v
e
c
o
n
ten
t
o
n
so
c
ial
m
e
d
ia
p
latfo
rm
s,
su
c
h
a
s
F
a
c
e
b
o
o
k
a
n
d
Twit
ter,
h
a
s
e
m
e
r
g
e
d
a
s
a
n
e
sc
a
latin
g
c
o
n
c
e
rn
,
e
sp
e
c
ially
in
Vie
tn
a
m
.
Co
n
se
q
u
e
n
tl
y
,
d
e
tec
ti
n
g
h
a
te
a
n
d
o
ffe
n
si
v
e
sp
a
n
s
i
n
V
ietn
a
m
e
se
tex
t
is
a
n
e
ss
e
n
ti
a
l
a
re
a
o
f
re
se
a
rc
h
.
Th
is
st
u
d
y
in
tr
o
d
u
c
e
s
ViHa
teO
ff,
a
n
a
d
v
a
n
c
e
d
fra
m
e
wo
rk
th
a
t
c
o
m
b
in
e
s
a
h
a
ted
sp
e
e
c
h
d
icti
o
n
a
ry
(HSD)
a
u
to
m
a
ti
c
a
ll
y
c
o
n
stru
c
te
d
fro
m
th
e
Vie
tn
a
m
e
se
h
a
te
a
n
d
o
ffe
n
siv
e
sp
a
n
s
(
ViHO
S
)
d
a
tas
e
t
with
t
h
e
p
re
-
train
e
d
lan
g
u
a
g
e
m
o
d
e
ls
fo
r
Vie
tn
a
m
e
se
(
P
h
o
BER
T
)
-
l
a
rg
e
lan
g
u
a
g
e
m
o
d
e
l
to
e
n
h
a
n
c
e
th
e
d
e
tec
ti
o
n
o
f
o
ffe
n
siv
e
e
x
p
re
ss
io
n
s.
Th
e
fra
m
e
wo
rk
f
u
n
c
ti
o
n
s t
h
r
o
u
g
h
tw
o
p
r
ima
ry
m
o
d
u
l
e
s.
F
irst,
it
c
o
n
stru
c
ts
an
HSD
fro
m
th
e
ViH
OS
d
a
tas
e
t,
w
h
ich
se
rv
e
s
a
s
a
re
f
e
re
n
c
e
fo
r
id
e
n
ti
f
y
in
g
h
a
te
a
n
d
o
ffe
n
si
v
e
l
a
n
g
u
a
g
e
in
Vie
t
n
a
m
e
se
tex
t.
S
e
c
o
n
d
,
th
e
fra
m
e
wo
rk
in
teg
ra
tes
t
h
e
P
h
o
BERT
-
l
a
rg
e
lan
g
u
a
g
e
m
o
d
e
l
with
HSD,
e
n
h
a
n
c
in
g
th
e
d
e
tec
ti
o
n
o
f
h
a
rm
fu
l
wo
r
d
s
in
t
h
e
in
p
u
t
tex
t.
E
x
p
e
rime
n
tal
re
su
lt
s
d
e
m
o
n
stra
te
t
h
a
t
th
e
p
ro
p
o
se
d
fra
m
e
wo
rk
sig
n
ifi
c
a
n
t
ly
o
u
tp
e
rfo
rm
s
e
x
isti
n
g
s
tate
-
of
-
th
e
-
a
rt
(S
OTA)
,
a
c
h
iev
in
g
a
n
F
1
-
sc
o
re
o
f
0
.
8
6
9
3
o
n
t
h
e
a
ll
-
sp
a
n
s
su
b
se
t
a
n
d
0
.
8
7
0
9
o
n
th
e
m
u
lt
ip
le
-
sp
a
n
s
su
b
se
t
re
p
re
se
n
ti
n
g
re
lativ
e
imp
ro
v
e
m
e
n
ts
o
f
o
v
e
r
1
0
%
c
o
m
p
a
re
d
to
t
h
e
stro
n
g
e
st
b
a
se
li
n
e
.
K
ey
w
o
r
d
s
:
Hate
s
p
ee
ch
d
etec
tio
n
Hate
d
s
p
ee
ch
d
ictio
n
ar
y
Natu
r
al
lan
g
u
ag
e
p
r
o
ce
s
s
in
g
Of
f
en
s
iv
e
lan
g
u
a
g
e
So
cial
m
ed
ia
Vietn
am
ese
tex
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
:
T
u
o
n
g
L
e
Facu
lty
o
f
I
n
f
o
r
m
atio
n
T
ec
h
n
o
lo
g
y
,
HUT
E
C
H
Un
iv
er
s
ity
4
7
5
A
Dien
B
ien
Ph
u
,
W
ar
d
2
5
,
B
in
h
T
h
a
n
h
Dis
tr
ict,
Ho
C
h
i M
in
h
C
ity
,
Vietn
am
E
m
ail: lc
.
tu
o
n
g
@
h
u
tech
.
e
d
u
.
v
n
1.
I
NT
RO
D
UCT
I
O
N
T
h
e
r
is
e
o
f
in
ter
n
et
ac
ce
s
s
an
d
s
o
cial
m
ed
ia
h
as
tr
an
s
f
o
r
m
ed
co
m
m
u
n
icatio
n
,
en
ab
led
r
ea
l
-
tim
e
in
f
o
r
m
atio
n
ex
c
h
an
g
e
,
a
n
d
f
o
s
ter
ed
g
l
o
b
al
c
o
n
n
ec
tiv
ity
.
Ho
wev
er
,
it h
as
also
f
ac
ilit
ated
th
e
s
p
r
ea
d
o
f
h
ar
m
f
u
l
co
n
ten
t
s
u
c
h
as
h
ate
s
p
ee
ch
a
n
d
o
f
f
en
s
iv
e
lan
g
u
ag
e.
Hate
s
p
ee
ch
,
as
d
ef
in
e
d
b
y
th
e
U
n
ited
Natio
n
s
,
ta
r
g
ets
in
d
iv
id
u
als
o
r
g
r
o
u
p
s
b
ased
o
n
id
en
tity
f
ac
to
r
s
lik
e
r
elig
io
n
,
eth
n
icity
,
o
r
g
en
d
er
,
wh
ile
o
f
f
en
s
iv
e
lan
g
u
ag
e
en
co
m
p
ass
es
v
u
lg
ar
,
ab
u
s
iv
e,
o
r
d
er
o
g
ato
r
y
e
x
p
r
ess
io
n
s
th
at,
alth
o
u
g
h
n
o
t
alwa
y
s
class
i
f
ied
as
h
ate
s
p
ee
ch
,
co
n
tr
ib
u
te
s
ig
n
if
ican
tly
to
a
to
x
ic
an
d
h
o
s
tile
o
n
lin
e
en
v
ir
o
n
m
en
t.
Su
ch
co
n
ten
t
n
o
t
o
n
ly
u
n
d
er
m
in
es
s
o
cial
co
h
esio
n
a
n
d
m
u
tu
al
r
esp
ec
t
b
u
t
also
n
eg
ativ
ely
af
f
ec
ts
u
s
er
s
’
m
en
tal
h
ea
lth
,
in
cr
ea
s
in
g
s
tr
ess
,
an
x
iety
,
an
d
d
ep
r
ess
io
n
.
Ad
d
itio
n
ally
,
it
ca
n
d
am
ag
e
t
h
e
r
ep
u
tatio
n
an
d
u
s
er
b
ase
o
f
o
n
lin
e
p
la
tf
o
r
m
s
,
p
r
o
m
p
tin
g
g
o
v
er
n
m
en
ts
wo
r
ld
wid
e
to
e
n
f
o
r
ce
s
tr
icter
r
eg
u
latio
n
s
o
n
co
n
ten
t
m
o
d
er
atio
n
.
I
n
th
e
Vietn
am
ese
co
n
tex
t,
d
etec
tin
g
h
ate
an
d
o
f
f
en
s
iv
e
lan
g
u
ag
e
p
o
s
es
ad
d
itio
n
al
ch
allen
g
es
d
u
e
to
lin
g
u
is
tic
n
u
a
n
ce
s
s
u
ch
as
to
n
al
v
ar
iatio
n
,
r
eg
i
o
n
al
d
ialec
ts
,
lim
ited
an
n
o
tated
d
ata,
an
d
f
r
e
q
u
en
t
u
s
e
o
f
ir
o
n
y
,
s
lan
g
,
o
r
s
ar
ca
s
m
.
Ad
d
r
ess
in
g
th
ese
ch
allen
g
es
r
e
q
u
ir
es
n
o
t
o
n
ly
a
d
v
an
ce
d
m
ac
h
in
e
lea
r
n
i
n
g
tech
n
iq
u
es
b
u
t
also
a
d
ee
p
u
n
d
er
s
tan
d
in
g
o
f
Vietn
am
ese
cu
ltu
r
al
an
d
lin
g
u
is
tic
s
u
b
tletie
s
to
en
s
u
r
e
ac
cu
r
ate
an
d
co
n
tex
t
-
awa
r
e
d
etec
tio
n
.
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
E
n
h
a
n
ce
d
fr
a
mewo
r
k
fo
r
d
etec
tin
g
V
ietn
a
mese
h
a
te
a
n
d
o
ffen
s
ive
s
p
a
n
s
(
Din
h
-
Ho
n
g
V
u
)
963
D
e
e
p
l
e
a
r
n
i
n
g
h
a
s
b
r
o
u
g
h
t
s
i
g
n
i
f
i
c
a
n
t
a
d
v
a
n
c
e
s
i
n
t
h
e
f
i
e
ld
o
f
n
a
t
u
r
a
l
l
a
n
g
u
a
g
e
p
r
o
c
e
s
s
i
n
g
(
N
L
P
)
,
i
m
p
r
o
v
i
n
g
c
o
m
p
u
t
e
r
s
’
a
b
i
l
it
y
t
o
u
n
d
e
r
s
t
a
n
d
a
n
d
u
s
e
n
at
u
r
a
l
l
a
n
g
u
a
g
e
.
D
e
e
p
l
e
a
r
n
i
n
g
m
o
d
e
l
s
,
p
a
r
t
ic
u
l
a
r
l
y
r
e
c
u
r
r
e
n
t
n
e
u
r
a
l
n
e
t
w
o
r
k
s
(
R
N
Ns
)
,
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
al
n
e
t
w
o
r
k
s
(
C
N
Ns
)
,
a
n
d
r
ec
e
n
t
l
y
,
t
r
a
n
s
f
o
r
m
e
r
a
r
c
h
i
t
e
c
t
u
r
es
,
h
a
v
e
d
e
m
o
n
s
t
r
ate
d
e
x
c
e
p
t
i
o
n
a
l
e
f
f
e
c
ti
v
e
n
e
s
s
in
v
a
r
i
o
u
s
N
L
P
t
as
k
s
,
s
u
c
h
a
s
m
a
c
h
i
n
e
t
r
a
n
s
la
t
i
o
n
[
1
]
–
[
3
]
,
t
e
x
t
c
l
a
s
s
i
f
i
ca
t
i
o
n
[
4
]
–
[
8
]
,
s
e
n
t
i
m
e
n
t
a
n
al
y
s
is
[
9
]
–
[
1
3
]
,
a
n
d
q
u
e
s
t
i
o
n
a
n
s
we
r
i
n
g
[
1
4
]
–
[
1
7
]
.
N
o
t
a
b
l
y
,
t
h
e
e
m
e
r
g
e
n
c
e
o
f
l
a
r
g
e
l
a
n
g
u
a
g
e
m
o
d
e
l
s
li
k
e
b
i
d
i
r
e
ct
i
o
n
a
l
e
n
c
o
d
e
r
r
e
p
r
e
s
e
n
t
a
ti
o
n
s
f
r
o
m
t
r
a
n
s
f
o
r
m
e
r
s
(
B
E
R
T
)
[
1
8
]
,
g
e
n
e
r
a
t
i
v
e
p
r
e
-
t
r
a
i
n
e
d
t
r
a
n
s
f
o
r
m
e
r
s
(
GP
T
)
[
1
9
]
,
p
r
e
-
t
r
a
i
n
e
d
la
n
g
u
a
g
e
m
o
d
e
l
s
f
o
r
Vi
e
t
n
a
m
es
e
(
P
h
o
B
E
R
T
)
[
2
0
]
a
n
d
a
d
v
a
n
c
e
d
v
e
r
s
i
o
n
s
s
u
c
h
as
G
P
T
-
4
h
a
s
s
i
g
n
i
f
i
c
a
n
t
l
y
i
m
p
r
o
v
e
d
t
h
e
a
b
i
l
it
y
t
o
c
a
p
t
u
r
e
c
o
n
t
ex
t
a
n
d
r
e
l
a
t
i
o
n
s
h
i
p
s
b
e
t
w
e
e
n
w
o
r
d
s
i
n
s
e
n
t
e
n
ce
s
,
e
n
a
b
l
i
n
g
d
e
e
p
e
r
u
n
d
e
r
s
t
a
n
d
in
g
o
f
s
e
m
a
n
t
ic
s
a
n
d
g
r
a
m
m
a
r
.
M
o
r
e
o
v
e
r
,
d
e
e
p
l
e
a
r
n
i
n
g
e
n
a
b
l
e
s
s
y
s
t
e
m
s
t
o
l
ea
r
n
f
r
o
m
m
a
s
s
i
v
e
d
a
t
a
s
et
s
,
a
ch
i
e
v
i
n
g
h
i
g
h
a
c
c
u
r
a
c
y
i
n
r
e
a
l
-
w
o
r
l
d
a
p
p
l
i
c
a
t
i
o
n
s
s
u
c
h
a
s
c
h
a
t
b
o
ts
,
i
n
f
o
r
m
a
t
i
o
n
r
e
t
r
i
e
v
a
l
,
a
n
d
h
at
e
s
p
a
n
s
d
e
te
c
t
io
n
.
T
h
e
c
o
n
t
i
n
u
o
u
s
a
d
v
a
n
c
e
m
en
t
o
f
d
e
e
p
l
e
a
r
n
i
n
g
i
s
u
n
l
o
c
k
i
n
g
n
ew
o
p
p
o
r
t
u
n
i
t
i
es,
p
o
s
i
t
i
o
n
i
n
g
N
L
P a
s
a
c
r
it
i
c
al
a
r
e
a
i
n
a
r
ti
f
i
c
ia
l
i
n
t
e
l
li
g
e
n
c
e
.
R
e
c
e
n
t
e
f
f
o
r
ts
i
n
h
a
te
s
p
e
e
c
h
a
n
d
o
f
f
e
n
s
i
v
e
l
a
n
g
u
a
g
e
d
e
t
e
c
t
i
o
n
h
a
v
e
f
o
c
u
s
e
d
o
n
r
e
d
u
c
i
n
g
m
i
s
c
l
a
s
s
i
f
i
ca
t
i
o
n
b
e
t
w
e
e
n
t
h
e
t
w
o
,
a
s
d
e
m
o
n
s
t
r
a
t
e
d
b
y
Y
u
a
n
e
t
a
l
.
[
2
1
]
w
i
t
h
a
n
a
d
v
e
r
s
a
r
ia
l
d
e
b
i
a
s
i
n
g
a
p
p
r
o
a
c
h
t
h
a
t
i
m
p
r
o
v
e
s
m
u
l
t
i
li
n
g
u
a
l
p
e
r
f
o
r
m
a
n
c
e
.
T
h
e
O
f
f
e
n
s
E
v
a
l
s
h
a
r
e
d
t
as
k
s
(
S
e
m
E
v
a
l
-
2
0
1
9
a
n
d
2
0
2
0
)
f
u
r
t
h
e
r
a
d
v
a
n
c
e
d
t
h
e
f
i
e
l
d
w
it
h
o
f
f
e
n
s
i
v
e
l
a
n
g
u
a
g
e
i
d
e
n
t
i
f
i
c
at
i
o
n
d
a
t
as
e
t
(
O
L
I
D
)
-
b
a
s
e
d
b
e
n
c
h
m
a
r
k
d
a
t
as
e
ts
,
w
h
e
r
e
Z
a
m
p
ie
r
i
e
t
a
l
.
[
2
2
]
s
h
o
w
e
d
t
h
a
t
la
r
g
e
l
a
n
g
u
a
g
e
m
o
d
e
l
s
,
w
h
il
e
s
t
r
o
n
g
,
s
ti
l
l
l
a
g
b
e
h
i
n
d
t
o
p
s
y
s
te
m
s
.
R
e
g
i
o
n
-
s
p
e
ci
f
i
c
w
o
r
k
,
s
u
c
h
as
T
a
m
i
l
h
a
t
e
s
p
e
e
c
h
d
e
t
e
c
t
i
o
n
u
s
i
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
m
e
n
t
f
r
e
q
u
e
n
c
y
(
T
F
-
I
DF
)
a
n
d
t
r
a
n
s
f
o
r
m
e
r
m
o
d
e
l
s
[
2
3
]
,
a
n
d
m
u
l
t
i
c
l
ass
,
m
u
l
t
il
a
b
e
l
d
e
e
p
le
a
r
n
i
n
g
f
o
r
I
n
d
o
n
e
s
i
a
n
t
w
e
e
ts
[
2
4
]
,
r
e
f
l
e
c
ts
g
r
o
w
i
n
g
s
p
e
ci
a
l
i
za
t
i
o
n
.
H
o
a
n
g
e
t
a
l
.
[
2
5
]
i
n
t
r
o
d
u
c
e
d
Vi
e
t
n
a
m
es
e
h
a
t
e
a
n
d
o
f
f
e
n
s
i
v
e
s
p
a
n
s
(
V
i
HO
S
)
,
t
h
e
f
i
r
s
t
h
u
m
a
n
-
a
n
n
o
t
a
t
e
d
d
a
t
a
s
et
f
o
r
i
d
e
n
t
i
f
y
i
n
g
h
a
t
e
a
n
d
o
f
f
e
n
s
i
v
e
s
p
a
n
s
i
n
Vi
e
t
n
a
m
es
e
c
o
m
m
e
n
t
s
.
U
n
li
k
e
e
a
r
li
e
r
d
a
t
as
e
ts
t
h
a
t
f
o
c
u
s
o
n
l
y
o
n
f
l
a
g
g
e
d
k
e
y
w
o
r
d
s
,
V
i
H
OS
c
a
p
t
u
r
e
s
t
h
e
u
n
d
e
r
l
y
i
n
g
i
d
e
a
s
a
n
d
o
p
i
n
i
o
n
s
e
x
p
r
e
s
s
e
d
w
i
t
h
i
n
c
o
m
m
e
n
ts
.
I
t
i
n
c
l
u
d
e
s
2
6
,
4
7
6
s
p
a
n
s
f
r
o
m
1
1
,
0
5
6
c
o
m
m
e
n
t
s
,
w
i
t
h
c
l
e
a
r
s
p
l
i
t
s
f
o
r
t
r
a
i
n
i
n
g
,
d
e
v
e
l
o
p
m
e
n
t
,
a
n
d
t
e
s
t
i
n
g
.
T
h
e
d
a
t
a
s
et
a
ls
o
p
r
o
v
i
d
e
s
d
e
t
a
il
e
d
a
n
n
o
t
a
t
i
o
n
g
u
i
d
e
l
i
n
es
a
n
d
s
t
r
a
te
g
i
e
s
f
o
r
h
a
n
d
l
i
n
g
c
o
m
p
l
e
x
l
an
g
u
a
g
e
p
h
e
n
o
m
e
n
a
s
u
c
h
a
s
t
e
e
n
c
o
d
e
,
m
et
a
p
h
o
r
s
,
a
n
d
p
u
n
s
.
V
iH
O
S
s
e
r
v
es
as
a
v
a
l
u
a
b
l
e
r
es
o
u
r
c
e
f
o
r
i
m
p
r
o
v
i
n
g
h
a
t
e
s
p
e
e
c
h
d
e
t
e
c
t
i
o
n
,
e
s
p
e
ci
a
l
l
y
a
t
t
h
e
s
p
a
n
l
e
v
e
l
i
n
V
i
e
t
n
a
m
e
s
e
.
H
o
we
v
e
r
,
s
p
a
n
-
l
e
v
el
d
e
te
c
t
i
o
n
a
c
cu
r
a
c
y
i
n
[
2
5
]
is
s
t
il
l
l
i
m
i
t
e
d
,
i
n
d
i
c
a
ti
n
g
a
n
e
e
d
f
o
r
f
u
r
t
h
e
r
i
m
p
r
o
v
e
m
e
n
t
.
T
h
e
r
e
f
o
r
e
,
t
h
i
s
s
t
u
d
y
i
n
t
r
o
d
u
c
es
Vi
H
a
teO
f
f
t
o
e
n
h
a
n
c
e
h
a
t
e
a
n
d
o
f
f
e
n
s
i
v
e
s
p
a
n
s
d
et
e
c
ti
o
n
i
n
V
ie
t
n
a
m
es
e
,
c
o
m
p
a
r
e
d
t
o
e
x
i
s
ti
n
g
s
ta
t
e
-
of
-
t
h
e
-
a
r
t
(
S
O
T
A
)
m
e
t
h
o
d
s
i
n
[
2
5
]
.
T
h
e
m
a
i
n
c
o
n
t
r
i
b
u
t
i
o
n
s
o
f
t
h
is
s
t
u
d
y
a
r
e
as
f
o
ll
o
w
s
:
i
)
d
a
t
a
p
r
e
p
r
o
c
e
s
s
i
n
g
i
n
t
h
e
Vi
H
OS
d
ata
s
e
t
t
o
c
r
ea
t
e
h
ated
s
p
ee
ch
d
ictio
n
ar
y
(
H
S
D
)
,
w
h
i
c
h
c
a
n
b
e
u
s
e
d
a
s
a
r
e
f
e
r
e
n
c
e
d
i
c
ti
o
n
a
r
y
f
o
r
o
t
h
e
r
h
at
e
an
d
o
f
f
e
n
s
i
v
e
s
p
a
n
s
d
e
t
e
c
t
i
o
n
t
as
k
s
,
i
i
)
c
o
m
b
i
n
i
n
g
P
h
o
B
E
R
T
-
L
a
r
g
e
,
l
a
r
g
e
l
a
n
g
u
ag
e
m
o
d
e
l
(
LLM
)
w
i
t
h
t
h
e
H
SD
h
e
l
p
s
t
h
e
m
o
d
e
l
b
e
t
t
e
r
d
e
t
e
c
t
h
a
r
m
f
u
l
w
o
r
d
s
i
n
i
n
p
u
t
t
e
x
t
,
a
n
d
i
i
i
)
e
x
p
e
r
i
m
e
n
t
a
l
r
e
s
u
lt
s
s
h
o
w
t
h
at
V
i
H
at
e
O
f
f
c
a
n
b
e
a
p
p
l
i
e
d
t
o
d
e
t
e
c
t
n
e
g
a
ti
v
e
c
o
m
m
e
n
t
s
o
n
s
o
c
i
a
l
m
e
d
i
a
o
r
s
i
m
il
a
r
p
l
a
t
f
o
r
m
s
.
T
h
e
r
em
ai
n
d
er
o
f
th
is
p
ap
e
r
is
o
r
g
an
ized
as
f
o
llo
ws:
s
ec
tio
n
2
p
r
esen
ts
th
e
p
r
o
p
o
s
ed
ViHate
Of
f
f
r
am
ewo
r
k
,
in
clu
d
i
n
g
th
e
c
o
n
s
tr
u
ctio
n
o
f
HSD,
f
ea
tu
r
e
e
n
g
i
n
ee
r
in
g
,
a
n
d
m
o
d
el
in
teg
r
atio
n
.
Sectio
n
3
r
e
p
o
r
ts
ex
p
er
im
en
tal
r
esu
lts
an
d
co
m
p
ar
es
o
u
r
ap
p
r
o
ac
h
with
b
ase
lin
e
m
eth
o
d
s
.
Sectio
n
4
co
n
cl
u
d
es
th
e
s
tu
d
y
an
d
o
u
tlin
es f
u
tu
r
e
r
esear
ch
d
ir
ec
ti
o
n
s
.
2.
M
E
T
H
O
D
2
.
1
.
H
a
t
ed
s
peec
h dict
io
na
ry
Utilizin
g
th
e
ViHOS
d
ataset,
we
d
ev
elo
p
ed
th
e
HSD,
wh
ich
co
n
tain
s
a
co
m
p
ilatio
n
o
f
s
y
llab
les
an
d
wo
r
d
s
d
esig
n
ated
as
"h
ated
"
with
in
th
is
d
ataset.
W
e
a
s
s
u
m
e
th
at
s
y
llab
les/
wo
r
d
s
lab
eled
“h
ated
”
in
th
e
in
p
u
t
s
en
ten
ce
ar
e
h
ig
h
ly
lik
ely
t
o
in
d
icate
h
ate
f
u
l
c
o
n
ten
t
i
n
t
h
e
m
o
d
el’
s
o
u
t
p
u
t.
T
h
er
ef
o
r
e
,
th
is
d
ictio
n
ar
y
is
u
tili
ze
d
to
m
ar
k
“h
ated
”
s
y
llab
les/
wo
r
d
s
in
th
e
in
p
u
t
s
en
ten
ce
,
th
er
eb
y
in
cr
e
asin
g
th
eir
weig
h
t
d
u
r
in
g
th
e
m
o
d
el’
s
p
r
o
ce
s
s
in
g
.
T
h
e
d
ict
io
n
ar
y
co
n
s
tr
u
ctio
n
p
r
o
ce
s
s
is
as
f
o
llo
ws:
f
ir
s
t,
we
tr
an
s
f
o
r
m
th
e
s
p
an
-
le
v
el
lab
els
in
th
e
ViHOS
d
ataset
in
to
s
y
llab
le
-
o
r
w
o
r
d
-
le
v
el
la
b
els.
I
f
a
s
y
llab
le
o
r
wo
r
d
f
a
lls
with
in
a
h
atef
u
l
s
p
an
,
its
lab
el
is
s
et
to
1
;
o
th
e
r
wis
e,
it
is
ass
ig
n
ed
a
la
b
el
o
f
0
.
Seco
n
d
ly
,
we
c
o
llect
all
s
y
llab
les
an
d
wo
r
d
s
lab
eled
as 1
to
co
n
s
tr
u
ct
HSD.
Ou
r
r
esu
ltin
g
d
ictio
n
ar
y
co
n
ta
in
s
2
,
8
9
9
item
s
in
s
y
llab
le
f
o
r
m
an
d
3
,
8
9
2
item
s
in
wo
r
d
f
o
r
m
.
T
h
is
d
ictio
n
ar
y
is
d
er
iv
ed
ex
clu
s
iv
ely
f
r
o
m
th
e
t
r
ain
in
g
d
ataset
to
en
s
u
r
e
its
u
s
ag
e
d
o
es
n
o
t
in
f
lu
en
ce
th
e
m
o
d
el’
s
test
r
esu
lt
s
,
th
er
eb
y
m
ain
tain
in
g
o
b
jectiv
ity
an
d
ac
cu
r
ac
y
in
ev
alu
atin
g
m
o
d
el
p
er
f
o
r
m
a
n
c
e.
T
ab
le
1
co
n
tain
s
en
tr
ies
illu
s
tr
atin
g
an
ex
am
p
le
o
f
s
y
llab
les
an
d
wo
r
d
s
in
clu
d
ed
in
HSD,
f
o
cu
s
in
g
o
n
v
u
lg
ar
lan
g
u
ag
e,
n
eg
ativ
e
em
o
tio
n
s
,
an
d
o
f
f
e
n
s
iv
e
wo
r
d
s
.
2
.
2
.
T
he
o
v
er
a
ll a
rc
hite
ct
ure
Ou
r
p
r
o
p
o
s
ed
f
r
am
ewo
r
k
co
n
t
ain
s
th
r
ee
m
o
d
u
les
,
r
elatio
n
e
x
tr
ac
to
r
,
h
ated
m
ask
er
an
d
cla
s
s
if
icatio
n
mod
u
les.
T
h
e
r
elatio
n
s
b
etwe
en
th
r
ee
m
o
d
u
les
ar
e
s
h
o
wn
in
Fig
u
r
e
1
.
T
h
e
ViHate
Of
f
f
r
am
ewo
r
k
p
r
o
ce
s
s
es
in
p
u
t
tex
t
u
s
in
g
two
d
is
tin
ct
t
o
k
en
izatio
n
m
eth
o
d
s
:
wo
r
d
to
k
en
izatio
n
an
d
s
y
llab
le
to
k
en
i
za
tio
n
.
ViHate
Of
f
u
s
es
s
y
llab
le
to
k
en
izatio
n
,
wh
ich
s
p
lits
th
e
tex
t
at
wh
ites
p
ac
e
an
d
is
d
en
o
ted
b
y
ViHate
Of
f
s
y
llab
le.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
5
,
No
.
1
,
Feb
r
u
ar
y
2
0
2
6
:
962
-
9
7
1
964
Me
an
wh
ile,
ViHate
Of
f
to
k
e
n
izes
tex
t
in
to
Vietn
am
ese
wo
r
d
s
is
d
en
o
ted
b
y
ViHate
Of
f
wo
r
d
.
Fo
r
in
s
tan
ce
,
th
e
s
en
ten
ce
“
p
h
ả
i
ch
ă
n
g
đ
â
y
là
l
ý
d
o
ô
n
g
n
à
y
kh
ô
n
g
mu
ố
n
cá
c
h
ly
”
(
tr
an
s
lated
as
“M
ay
b
e
th
is
is
th
e
r
ea
s
o
n
h
e
d
o
esn
’
t
wan
t
to
b
e
q
u
ar
a
n
tin
e
d
”)
wo
u
ld
b
e
to
k
e
n
ized
in
to
w
o
r
d
s
as
“
p
h
ả
i_
c
h
ă
n
g
đ
â
y
là
lý
_
d
o
ô
n
g
n
à
y
kh
ô
n
g
mu
ố
n
cá
c
h
_
ly
”
an
d
th
en
f
u
r
t
h
er
s
p
lit
in
to
to
k
e
n
s
b
ased
o
n
wh
ites
p
ac
e.
I
n
s
u
m
m
a
r
y
,
th
e
ViHate
Of
f
s
y
llab
le
an
d
ViHate
Of
f
wo
r
d
f
r
a
m
ewo
r
k
s
d
if
f
er
o
n
ly
i
n
th
eir
in
p
u
t
to
k
en
s
(
s
y
llab
les
v
er
s
u
s
wo
r
d
s
)
,
wh
ile
t
h
eir
u
n
d
er
ly
i
n
g
m
o
d
el
s
tr
u
ctu
r
es r
e
m
ain
id
en
tical.
T
ab
le
1
.
E
x
am
p
le
o
f
s
y
llab
les an
d
wo
r
d
s
in
HSD
S
TT
W
o
r
d
(
i
n
V
i
e
t
n
a
m
e
se)
W
o
r
d
(
i
n
E
n
g
l
i
s
h
)
N
o
t
e
Ty
p
e
1
vl
S
l
a
n
g
a
b
b
r
e
v
i
a
t
i
o
n
f
o
r
“
v
ã
i
l
*
n
”
V
u
l
g
a
r
l
a
n
g
u
a
g
e
S
y
l
l
a
b
l
e
2
vét
Li
c
k
u
p
/
scr
a
p
e
(
d
e
r
o
g
a
t
o
r
y
)
V
u
l
g
a
r
l
a
n
g
u
a
g
e
S
y
l
l
a
b
l
e
3
c
h
á
n
B
o
r
e
d
N
e
g
a
t
i
v
e
e
m
o
t
i
o
n
S
y
l
l
a
b
l
e
4
s
ủ
a
B
a
r
k
(
i
n
su
l
t
i
n
g
,
l
i
k
e
a
d
o
g
)
O
f
f
e
n
si
v
e
w
o
r
d
S
y
l
l
a
b
l
e
5
n
á
t
R
u
i
n
e
d
(
t
o
r
n
,
c
o
m
p
l
e
t
e
l
y
b
a
d
)
O
f
f
e
n
si
v
e
w
o
r
d
S
y
l
l
a
b
l
e
6
đ
c
m
f
*
*
k
V
u
l
g
a
r
l
a
n
g
u
a
g
e
S
y
l
l
a
b
l
e
7
h
ú
t
I
n
j
e
c
t
/
u
s
e
d
r
u
g
s (
e
.
g
.
,
h
e
r
o
i
n
)
O
f
f
e
n
si
v
e
w
o
r
d
S
y
l
l
a
b
l
e
8
í
c
h
_
k
ỷ
S
e
l
f
i
s
h
N
e
g
a
t
i
v
e
e
m
o
t
i
o
n
W
o
r
d
9
đ
ộ
c
_
m
ồ
m
S
p
i
t
e
f
u
l
,
sarc
a
st
i
c
O
f
f
e
n
si
v
e
w
o
r
d
W
o
r
d
10
th
ầ
n
_
k
i
n
h
N
e
r
v
o
u
s,
a
n
x
i
e
t
y
-
r
e
l
a
t
e
d
O
f
f
e
n
si
v
e
w
o
r
d
W
o
r
d
11
t
ổ
_
c
h
a
O
f
f
e
n
si
v
e
f
a
mi
l
y
-
r
e
l
a
t
e
d
i
n
s
u
l
t
O
f
f
e
n
si
v
e
w
o
r
d
W
o
r
d
12
đ
ô
_
h
ộ
C
o
l
o
n
i
z
e
,
o
p
p
r
e
ss
N
e
g
a
t
i
v
e
e
m
o
t
i
o
n
W
o
r
d
13
k
h
i
n
h
_
b
ỉ
D
i
sd
a
i
n
,
c
o
n
t
e
m
p
t
O
f
f
e
n
si
v
e
w
o
r
d
W
o
r
d
14
d
ơ
_
b
ẩ
n
D
i
r
t
y
,
f
i
l
t
h
y
O
f
f
e
n
si
v
e
w
o
r
d
W
o
r
d
Fig
u
r
e
1
.
ViHate
Of
f
ar
c
h
itectu
r
e:
d
etec
tin
g
Vietn
am
ese
h
at
e
an
d
o
f
f
en
s
iv
e
s
p
an
s
2
.
2
.
1
.
Rela
t
io
n e
x
t
r
a
ct
o
r
m
o
du
le
T
h
is
m
o
d
u
l
e
u
ti
liz
es
th
e
P
h
o
B
E
R
T
-
l
a
r
g
e
s
y
l
la
b
l
e
co
n
f
i
g
u
r
ed
t
o
ac
h
i
ev
e
th
e
o
p
ti
m
a
l
r
es
u
lts
r
e
p
o
r
te
d
in
[
2
0
]
.
T
h
e
o
u
tp
u
t
o
f
t
h
is
m
o
d
u
l
e
is
a
v
e
ct
o
r
r
e
p
r
es
en
t
ati
o
n
f
o
r
ea
ch
to
k
en
in
t
h
e
i
n
p
u
t
te
x
t
,
c
ap
tu
r
i
n
g
r
el
ati
o
n
al
in
f
o
r
m
ati
o
n
a
m
o
n
g
t
o
k
e
n
s
.
T
o
e
n
h
a
n
c
e
t
h
e
m
o
d
el’
s
lea
r
n
i
n
g
ca
p
ab
ilit
y
,
we
a
ls
o
i
n
c
o
r
p
o
r
a
te
t
h
e
[
C
L
S
]
t
o
k
en
o
u
t
p
u
t
t
o
class
if
y
th
e
i
n
p
u
t
te
x
t
as “
h
at
ed
”
o
r
n
o
t
.
T
h
e
l
ab
el
f
o
r
[
C
L
S]
is
g
e
n
e
r
a
ted
d
u
r
in
g
t
r
a
in
in
g
as
f
o
ll
o
ws:
i
f
a
n
y
to
k
en
i
n
th
e
tex
t
is
l
ab
el
ed
as
“
h
a
te
d
”
,
th
en
t
h
e
la
b
e
l
f
o
r
[
C
L
S]
is
s
et
t
o
1
;
o
th
er
wis
e
,
it
is
s
e
t
to
0
.
T
h
e
o
u
t
p
u
t
o
f
th
is
m
o
d
u
le
is
th
e
m
at
r
i
x
1
wit
h
d
im
e
n
s
i
o
n
s
(
N
+
2
)
×
1
0
2
4
,
w
h
e
r
e
N
is
t
h
e
m
a
x
i
m
u
m
n
u
m
b
e
r
o
f
t
o
k
en
s
i
n
t
h
e
i
n
p
u
t
t
ex
t
.
C
o
n
s
is
t
e
n
t
wit
h
[
2
0
]
,
we
u
s
e
N
=
6
4
,
a
lo
n
g
wit
h
t
h
e
[
C
L
S]
a
n
d
[
SEP
]
t
o
k
e
n
s
.
1
=
ℎ
(
)
(
1
)
W
h
er
e
X
is
th
e
to
k
en
I
D
v
ec
to
r
g
en
e
r
ated
with
th
e
Ph
o
B
E
R
T
-
L
ar
g
e
to
k
e
n
izer
,
y
1
is
o
u
tp
u
t
o
f
Ph
o
B
E
R
T
-
L
ar
g
e
m
o
d
el.
2
.
2
.
2
.
H
a
t
ed
m
a
s
k
ed
m
o
du
le
T
h
is
m
o
d
u
le
u
s
es
HSD
to
id
en
tify
p
o
s
itio
n
s
o
f
to
k
e
n
s
th
at
m
ay
b
e
“
h
ated
to
k
en
s
”
in
th
e
in
p
u
t
tex
t.
T
h
e
g
o
al
is
to
d
o
u
b
le
t
h
e
weig
h
t o
f
th
ese
to
k
e
n
s
in
m
o
d
el
ca
lcu
latio
n
s
.
T
h
e
m
o
d
u
le’
s
o
u
tp
u
t is a
b
in
ar
y
v
ec
to
r
2
with
(
N
+2
)
d
im
en
s
io
n
s
,
wh
e
r
e
ea
ch
elem
en
t h
as a
v
alu
e
o
f
1
if
its
in
p
u
t to
k
en
in
t
h
e
h
ated
d
ict
else 0
.
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
E
n
h
a
n
ce
d
fr
a
mewo
r
k
fo
r
d
etec
tin
g
V
ietn
a
mese
h
a
te
a
n
d
o
ffen
s
ive
s
p
a
n
s
(
Din
h
-
Ho
n
g
V
u
)
965
2
=
(
,
)
(
2
)
I
n
th
is
co
n
tex
t,
X
is
th
e
to
k
en
I
D
v
ec
to
r
g
en
er
ated
with
th
e
Ph
o
B
E
R
T
-
l
ar
g
e
to
k
en
izer
,
D
r
ep
r
esen
ts
th
e
HSD,
y
2
is
o
u
tp
u
t
v
ec
to
r
o
f
m
ask
e
d
m
o
d
u
le.
T
h
e
in
teg
r
atio
n
o
f
H
SD
en
h
an
ce
s
th
e
m
o
d
el’
s
atten
tio
n
m
ec
h
an
is
m
b
y
ass
ig
n
in
g
g
r
ea
ter
weig
h
ts
to
to
k
en
s
k
n
o
wn
to
b
e
in
d
icativ
e
o
f
h
ate
o
r
o
f
f
e
n
s
iv
e
in
ten
t.
I
n
s
tead
o
f
tr
ea
tin
g
all
to
k
en
s
eq
u
ally
,
th
e
m
o
d
el
lev
er
ag
es
p
r
i
o
r
k
n
o
wled
g
e
f
r
o
m
an
n
o
tated
tr
ain
in
g
d
ata
to
em
p
h
asize
p
o
te
n
tially
h
ar
m
f
u
l
s
p
an
s
.
T
h
is
m
ec
h
an
is
m
im
p
r
o
v
es
d
etec
tio
n
ac
cu
r
ac
y
in
c
o
m
p
lex
o
r
len
g
th
y
s
en
ten
ce
s
,
an
d
h
elp
s
th
e
m
o
d
el
g
en
er
alize
b
etter
to
r
ar
e
o
r
d
is
to
r
ted
f
o
r
m
s
o
f
o
f
f
en
s
iv
e
lan
g
u
ag
e
(
e.
g
.
,
s
lan
g
o
r
ab
b
r
ev
iated
ex
p
r
ess
io
n
s
)
,
m
an
y
o
f
w
h
ich
a
r
e
ca
p
tu
r
e
d
in
th
e
d
ictio
n
ar
y
.
2
.
2
.
3
.
Cla
s
s
if
ica
t
io
n m
o
du
le
T
h
is
m
o
d
u
le
co
m
b
i
n
es
th
e
o
u
tp
u
ts
o
f
th
e
two
p
r
ev
io
u
s
m
o
d
u
les
u
s
in
g
two
s
ep
ar
ate
d
e
n
s
e
lay
er
s
(
DL
1
an
d
DL
2
)
.
DL
1
h
an
d
les
class
if
icatio
n
b
ased
o
n
th
e
o
u
tp
u
t
o
f
th
e
r
elatio
n
ex
tr
ac
t
o
r
m
o
d
u
le,
wh
ile
DL
2
f
o
cu
s
es
o
n
class
if
icatio
n
u
s
in
g
to
k
e
n
s
in
HSD.
T
h
is
is
ac
h
iev
ed
b
y
m
u
ltip
ly
in
g
th
e
o
u
tp
u
t
m
atr
ix
f
r
o
m
t
h
e
r
elatio
n
ex
t
r
ac
to
r
with
th
e
b
i
n
ar
y
v
ec
to
r
f
r
o
m
th
e
h
ated
m
ask
ed
m
o
d
u
le
,
r
etain
in
g
o
n
ly
v
ec
to
r
s
f
o
r
to
k
en
s
f
o
u
n
d
in
HSD.
T
h
e
o
u
tp
u
ts
f
r
o
m
th
ese
two
d
en
s
e
lay
er
s
p
r
o
v
id
e
two
d
is
tin
ct
class
if
icati
o
n
p
er
s
p
ec
tiv
es
o
n
th
e
s
am
e
d
ata.
W
e
th
e
n
m
er
g
e
th
ese
p
e
r
s
p
ec
tiv
es
b
y
a
d
d
in
g
th
e
two
m
atr
ices,
r
esu
ltin
g
in
an
(
N+
2
)
×
1
0
2
4
m
atr
ix
to
cr
ea
te
th
e
f
o
llo
win
g
3
.
3
=
1
(
1
)
+
2
(
1
⨀
2
)
(
3
)
W
h
er
e
y
1
,
y
2
a
r
e
r
esu
lt
o
f
(
1
)
an
d
(
2
)
,
DL
s
tan
d
s
f
o
r
d
en
s
e
lay
er
,
⨀
is
elem
en
t
-
wis
e
m
u
ltip
licatio
n
with
b
r
o
ad
ca
s
tin
g
: e
ac
h
r
o
w
i
in
y
1
is
s
ca
led
b
y
y
2
[
i
]
.
T
h
en
th
e
r
es
u
lt ser
v
es a
s
in
p
u
t to
th
e
two
DL
an
d
o
u
tp
u
t la
y
er
with
a
s
ig
m
o
id
ac
tiv
atio
n
f
u
n
c
tio
n
(
)
to
class
if
y
wh
eth
er
a
to
k
en
is
h
ated
o
r
n
o
t.
̂
=
(
4
(
3
(
3
)
)
)
(
4
)
W
h
er
e
y
3
is
r
esu
lt
o
f
(
3
)
,
DL
s
tan
d
s
f
o
r
d
en
s
e
l
ay
er
.
T
h
e
in
p
u
t
to
k
e
n
is
class
if
ied
a
s
a
h
ated
to
k
en
if
̂
>
0
.
5
,
an
d
as
a
n
o
n
-
h
ated
to
k
e
n
if
̂
≤
0
.
5
.
Alth
o
u
g
h
we
h
a
v
e
n
o
t
co
n
d
u
cted
an
a
b
latio
n
s
tu
d
y
to
is
o
la
te
th
e
d
ictio
n
ar
y
’
s
im
p
ac
t,
th
e
s
u
p
er
io
r
p
er
f
o
r
m
an
ce
o
f
ViHate
Of
f
co
m
p
a
r
ed
to
b
aseli
n
es
u
s
in
g
th
e
s
am
e
b
ac
k
b
o
n
e
(
Ph
o
B
E
R
T
)
im
p
lies
th
at
th
e
d
ictio
n
ar
y
p
lay
s
a
s
ig
n
if
ican
t
r
o
le.
Fu
t
u
r
e
wo
r
k
will
in
clu
d
e
d
etailed
s
tatis
t
ical
v
alid
atio
n
(
e.
g
.
,
p
-
v
alu
es
an
d
co
n
f
id
en
ce
i
n
ter
v
als)
an
d
ab
latio
n
ex
p
er
im
en
t
s
to
q
u
an
tify
th
is
co
n
tr
ib
u
tio
n
m
o
r
e
p
r
ec
is
ely
.
3.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
3
.
1
.
E
x
perim
ent
a
l
s
et
t
ing
s
T
h
is
s
tu
d
y
tr
ain
e
d
th
e
ViHate
Of
f
m
o
d
el
u
s
in
g
t
h
e
Ad
am
o
p
tim
izer
with
a
lear
n
i
n
g
r
ate
o
f
6
×1
0
⁻⁵,
a
b
atch
s
ize
o
f
1
2
8
,
an
d
4
0
tr
ai
n
in
g
e
p
o
ch
s
.
T
h
e
o
p
tim
al
ep
o
ch
th
r
esh
o
l
d
s
wer
e
s
elec
ted
f
o
r
th
e
t
h
r
ee
s
u
b
s
ets:
s
in
g
le
s
p
an
,
m
u
ltip
le
s
p
an
s
,
an
d
all
s
p
a
n
s
(
d
etails
in
s
ec
tio
n
3
.
2
)
.
ViHate
Of
f
u
tili
ze
s
th
e
p
r
e
-
tr
ai
n
ed
Ph
o
B
E
R
T
-
L
ar
g
e
m
o
d
el
[
2
0
]
v
ia
th
e
h
u
g
g
in
g
f
ac
e
lib
r
a
r
y
,
with
in
p
u
t
d
ata
f
r
o
m
th
e
ViHOS
d
ata
s
et
p
r
e
-
to
k
e
n
ized
ac
co
r
d
in
g
l
y
.
T
r
ain
in
g
was p
er
f
o
r
m
ed
o
n
a
h
ig
h
-
p
er
f
o
r
m
a
n
ce
s
y
s
tem
(
1
2
6
GB
R
AM
,
8
0
C
PU
s
)
.
Fo
r
co
m
p
ar
is
o
n
,
we
ad
o
p
ted
t
h
e
b
aselin
e
m
o
d
els
an
d
r
esu
lts
f
r
o
m
[
2
5
]
.
B
o
th
s
y
llab
le
-
b
ased
an
d
wo
r
d
-
b
ased
to
k
en
izatio
n
ap
p
r
o
ac
h
es
wer
e
test
ed
(
ViHate
Of
f
syllable
an
d
ViHate
Of
f
word
)
,
an
d
ev
alu
ated
u
s
in
g
m
ac
r
o
F1
-
s
co
r
e,
p
r
ec
is
io
n
,
an
d
r
ec
all,
co
n
s
is
ten
t w
ith
[
2
5
]
.
3
.
2
.
E
x
perim
ent
s
a
nd
re
s
ults
3
.
2
.
1
.
E
x
perim
ent
1
-
det
er
m
ini
ng
t
he
o
ptim
a
l num
ber
o
f
t
ra
ini
ng
epo
chs
f
o
r
ViH
a
t
eO
f
f
syllable
T
h
is
ex
p
er
im
en
t,
co
n
d
u
cte
d
s
p
ec
if
ically
o
n
ViHate
Of
f
syllable
as
s
h
o
wn
in
Fig
u
r
e
2
,
aim
s
t
o
id
en
tify
th
e
id
ea
l
n
u
m
b
er
o
f
ep
o
c
h
s
f
o
r
tr
ain
in
g
th
e
m
o
d
el
b
y
o
b
s
er
v
in
g
tr
en
d
s
o
v
er
a
4
0
-
ep
o
ch
r
u
n
.
Fig
u
r
e
2
(
a
)
an
d
2
(
b
)
s
h
o
ws
th
e
tr
ain
in
g
an
d
v
alid
atio
n
lo
s
s
o
v
er
4
0
ep
o
ch
s
.
W
h
ile
tr
ain
in
g
lo
s
s
s
tead
ily
d
ec
lin
es,
it
p
latea
u
s
af
ter
ep
o
ch
2
5
.
Valid
atio
n
lo
s
s
,
h
o
w
ev
er
,
in
cr
ea
s
es
af
ter
ep
o
ch
1
8
an
d
s
h
ar
p
ly
r
is
es
p
o
s
t
-
ep
o
ch
3
0
,
in
d
icatin
g
o
v
e
r
f
itti
n
g
.
T
h
u
s
,
tr
ain
in
g
b
e
y
o
n
d
3
0
ep
o
ch
s
o
f
f
er
s
litt
le
b
en
ef
it
an
d
m
ay
h
ar
m
g
en
er
aliza
tio
n
.
I
n
Fig
u
r
e
s
2
(
c
)
an
d
2
(
d
)
,
v
alid
atio
n
p
r
e
cisi
o
n
p
latea
u
s
ar
o
u
n
d
e
p
o
ch
2
6
,
s
tab
ilizin
g
n
ea
r
0
.
8
4
.
T
h
e
all
-
s
p
an
s
u
b
s
et
ex
h
ib
its
t
h
e
m
o
s
t
s
tab
le
an
d
h
ig
h
est
p
e
r
f
o
r
m
a
n
ce
ac
r
o
s
s
all
m
etr
ics.
Fig
u
r
es
2
(
e)
to
2
(
h
)
(
r
ec
all
an
d
F1
-
s
co
r
e
)
f
o
llo
w
s
im
ilar
tr
en
d
s
,
co
n
f
ir
m
in
g
th
e
m
o
d
el’
s
s
tr
en
g
th
o
n
m
o
r
e
co
m
p
lex
m
u
lti
-
s
p
an
d
ata.
C
o
n
v
er
s
ely
,
f
o
r
th
e
s
in
g
l
e
s
p
an
s
u
b
s
et,
p
er
f
o
r
m
an
ce
r
e
m
ain
s
lo
w
in
th
e
f
ir
s
t
1
5
ep
o
c
h
s
,
with
F1
ar
o
u
n
d
0
.
3
3
,
an
d
o
n
ly
im
p
r
o
v
es
s
lig
h
tly
b
y
ep
o
c
h
2
6
.
Valid
atio
n
lo
s
s
in
cr
ea
s
es
af
ter
ep
o
c
h
1
8
,
i
n
d
icatin
g
o
v
er
f
itti
n
g
r
is
k
.
T
h
er
ef
o
r
e,
t
h
e
o
p
tim
al
tr
ain
in
g
r
a
n
g
e
f
o
r
t
h
is
s
u
b
s
et
is
b
etwe
en
ep
o
c
h
s
1
8
-
2
5
.
E
p
o
c
h
2
2
was
s
elec
ted
f
o
r
f
in
al
ev
alu
atio
n
,
as
it
p
r
o
v
i
d
es
th
e
b
est
p
er
f
o
r
m
an
ce
ac
r
o
s
s
th
e
“sin
g
le
s
p
an
,
”
“m
u
ltip
le
s
p
an
s
,
”
an
d
“a
ll
s
p
an
s
”
s
u
b
s
ets.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
5
,
No
.
1
,
Feb
r
u
ar
y
2
0
2
6
:
962
-
9
7
1
966
(
a)
(
b
)
(
c)
(
d
)
(
e)
(f)
(
g
)
(
h
)
Fig
u
r
e
2
.
T
r
ain
in
g
/v
alid
ati
o
n
m
etr
ics f
o
r
ViHate
Of
f
s
y
llab
le
o
f
(
a)
t
r
ain
in
g
l
o
s
s
,
(
b
)
v
alid
ati
o
n
lo
s
s
,
(
c)
tr
ain
in
g
p
r
ec
is
io
n
,
(
d
)
v
alid
atio
n
p
r
ec
is
io
n
,
(
e)
tr
ai
n
in
g
r
ec
all,
(
f
)
v
alid
atio
n
r
ec
all,
(
g
)
tr
ain
in
g
F1
-
s
co
r
e
,
an
d
(
h
)
v
alid
atio
n
F1
-
s
co
r
e
3
.
2
.
2
.
E
x
perim
ent
2
-
det
er
m
ini
ng
t
he
o
ptim
a
l num
ber
o
f
t
ra
ini
ng
epo
chs
f
o
r
ViH
a
t
eO
f
f
word
T
h
i
s
s
t
u
d
y
f
o
c
u
s
es
o
n
V
i
Ha
t
eO
f
f
word
,
a
s
s
h
o
w
n
i
n
Fi
g
u
r
e
3
,
to
d
e
t
e
r
m
i
n
e
t
h
e
b
e
s
t
n
u
m
b
e
r
o
f
e
p
o
c
h
s
f
o
r
t
r
a
i
n
i
n
g
.
B
y
a
n
al
y
z
i
n
g
t
r
e
n
d
s
o
v
e
r
t
h
e
c
o
u
r
s
e
o
f
4
0
e
p
o
c
h
s
,
w
e
ass
e
s
s
t
h
e
m
o
d
el
’
s
le
a
r
n
i
n
g
e
f
f
i
c
i
e
n
c
y
a
n
d
i
d
e
n
t
i
f
y
p
o
t
e
n
t
i
a
l
s
i
g
n
s
o
f
o
v
e
r
f
i
t
t
i
n
g
.
I
n
t
h
i
s
e
x
p
e
r
i
m
e
n
t
,
s
i
m
il
a
r
t
r
e
n
d
s
t
o
e
x
p
e
r
i
m
e
n
t
1
w
e
r
e
o
b
s
e
r
v
e
d
,
w
i
t
h
t
h
e
v
a
l
i
d
a
t
i
o
n
l
o
s
s
s
t
a
r
t
i
n
g
t
o
r
is
e
a
f
t
e
r
e
p
o
c
h
1
8
(
F
i
g
u
r
e
s
3
(
a
)
a
n
d
3
(
b)
)
w
h
i
l
e
o
t
h
e
r
m
e
t
r
i
c
s
,
s
u
c
h
a
s
p
r
e
ci
s
i
o
n
(
F
i
g
u
r
e
s
3
(
c
)
a
n
d
3
(
d)
)
,
r
e
c
a
l
l
(
F
i
g
u
r
e
s
3
(
e
)
a
n
d
3
(
f)
)
,
a
n
d
F
1
-
s
c
o
r
e
(
Fi
g
u
r
e
s
3
(
g
)
a
n
d
3
(
h)
)
s
t
a
b
i
li
z
e
d
a
r
o
u
n
d
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
E
n
h
a
n
ce
d
fr
a
mewo
r
k
fo
r
d
etec
tin
g
V
ietn
a
mese
h
a
te
a
n
d
o
ffen
s
ive
s
p
a
n
s
(
Din
h
-
Ho
n
g
V
u
)
967
e
p
o
c
h
2
6
.
T
h
i
s
i
n
d
i
c
a
t
es
t
h
a
t
th
e
m
o
d
e
l
c
o
u
l
d
s
t
a
r
t
t
o
o
v
e
r
f
i
t
i
f
t
r
ai
n
i
n
g
p
r
o
c
e
e
d
s
b
e
y
o
n
d
t
h
i
s
p
o
i
n
t
.
T
h
e
r
e
f
o
r
e
,
t
h
e
o
p
t
i
m
a
l
t
r
a
i
n
i
n
g
r
a
n
g
e
f
o
r
t
h
i
s
s
u
b
s
et
a
p
p
e
a
r
s
t
o
b
e
b
etw
e
e
n
1
8
a
n
d
2
5
e
p
o
c
h
s
,
as
i
t
s
t
r
i
k
es
a
b
al
a
n
c
e
b
e
t
w
e
e
n
e
f
f
ec
t
i
v
e
l
e
a
r
n
i
n
g
a
n
d
p
r
e
v
e
n
t
i
n
g
o
v
e
r
f
i
tt
i
n
g
.
F
o
r
c
o
n
s
i
s
t
e
n
c
y
wi
t
h
t
h
e
p
r
e
v
i
o
u
s
e
x
p
e
r
i
m
e
n
t
,
e
p
o
c
h
2
2
w
a
s
s
e
l
e
ct
e
d
,
p
r
o
v
i
d
i
n
g
t
h
e
b
e
s
t
p
e
r
f
o
r
m
a
n
c
e
a
c
r
o
s
s
t
h
e
“s
i
n
g
le
s
p
a
n
,
”
“
m
u
l
ti
p
l
e
s
p
a
n
s
,
”
a
n
d
“
a
l
l
s
p
a
n
s
”
s
u
b
s
e
ts
.
(
a)
(
b
)
(
c)
(
d
)
(
e)
(f)
(
g
)
(
h
)
Fig
u
r
e
3
.
T
r
ain
in
g
/v
alid
ati
o
n
m
etr
ics f
o
r
ViHate
Of
f
wo
r
d
o
f
(
a)
tr
ain
in
g
lo
s
s
,
(
b
)
v
alid
atio
n
lo
s
s
,
(
c)
tr
ain
in
g
p
r
ec
is
io
n
,
(
d
)
v
alid
atio
n
p
r
ec
is
io
n
,
(
e)
tr
ain
in
g
r
ec
all,
(
f
)
v
ali
d
atio
n
r
ec
all,
(
g
)
tr
ain
i
n
g
F1
-
s
co
r
e,
an
d
(
h
)
v
alid
atio
n
F1
-
s
co
r
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
5
,
No
.
1
,
Feb
r
u
ar
y
2
0
2
6
:
962
-
9
7
1
968
3
.
2
.
3
.
E
x
perim
ent
3
-
co
m
pa
ri
s
o
n wit
h sta
t
e
-
of
-
t
he
-
a
r
t
mo
dels
I
n
th
is
ex
p
er
im
e
n
t,
we
ev
alu
ate
o
u
r
f
r
am
ewo
r
k
,
tr
ain
ed
u
s
in
g
th
e
o
p
tim
ized
p
ar
am
eter
s
o
u
tlin
ed
ea
r
lier
,
ag
ain
s
t
th
e
SOTA
m
o
d
els
f
o
r
d
etec
tin
g
h
ate
an
d
o
f
f
en
s
iv
e
s
p
an
s
.
T
ab
le
2
s
h
o
ws
th
at
th
e
B
iLST
M
-
C
R
F+Ph
o
2
W
m
o
d
el
p
er
f
o
r
m
ed
p
o
o
r
l
y
,
with
th
e
lo
west
F
1
-
s
co
r
es:
0
.
3
5
9
4
(
wo
r
d
)
an
d
0
.
4
3
2
9
(
s
y
llab
le)
,
h
ig
h
lig
h
tin
g
its
lim
itatio
n
s
o
n
th
e
“
s
in
g
le
s
p
a
n
”
s
u
b
s
et.
I
n
co
n
tr
ast,
m
o
d
er
n
m
o
d
els
lik
e
Ph
o
B
E
R
T
an
d
cr
o
s
s
-
lin
g
u
al
lan
g
u
a
g
e
m
o
d
e
l
-
R
o
B
E
R
T
a
(
XL
M
-
R
)
s
h
o
wed
b
etter
p
er
f
o
r
m
an
ce
,
esp
ec
ially
th
e
p
r
o
p
o
s
ed
ViHate
Of
f
word
,
wh
ich
ac
h
iev
e
d
th
e
h
ig
h
est
F1
-
s
co
r
e
(
0
.
8
1
7
4
)
an
d
r
ec
all
(
0
.
8
4
2
7
)
,
with
a
n
av
er
a
g
e
r
an
k
o
f
1
.
3
3
.
ViHate
Of
f
syllable
also
led
in
p
r
ec
is
io
n
(
0
.
8
5
1
9
)
with
in
th
e
s
y
llab
le
-
b
ased
g
r
o
u
p
an
d
r
a
n
k
ed
s
ec
o
n
d
o
v
er
all.
T
h
ese
r
esu
lts
co
n
f
ir
m
th
e
s
u
p
er
io
r
ity
o
f
th
e
ViHate
Of
f
f
r
am
ewo
r
k
in
b
o
th
s
y
llab
le
an
d
wo
r
d
-
lev
e
l
s
p
an
d
etec
tio
n
.
T
ab
le
3
h
i
g
h
lig
h
ts
th
e
s
u
p
er
io
r
p
er
f
o
r
m
a
n
ce
o
f
ViHate
Of
f
word
o
n
t
h
e
“
m
u
ltip
le
s
p
a
n
s
”
s
u
b
s
et,
ac
h
iev
in
g
th
e
b
est
s
co
r
es
in
p
r
ec
is
io
n
(
0
.
8
5
9
7
)
,
r
ec
all
(
0
.
8
8
3
0
)
,
an
d
F1
-
s
co
r
e
(
0
.
8
7
0
9
)
,
w
ith
an
av
er
ag
e
r
a
n
k
o
f
1
.
0
0
.
ViHate
Of
f
syllable
also
p
er
f
o
r
m
ed
well,
r
an
k
in
g
s
ec
o
n
d
ac
r
o
s
s
all
m
etr
ics
with
an
a
v
er
ag
e
r
an
k
o
f
2
.
0
0
.
I
n
c
o
n
tr
ast,
B
iLST
M
-
C
R
F+P
h
o
2
W
s
h
o
wed
th
e
wea
k
est
p
er
f
o
r
m
a
n
ce
in
b
o
th
i
n
p
u
t
ty
p
es,
co
n
f
ir
m
in
g
its
lim
itatio
n
s
in
d
etec
tin
g
m
u
ltip
le
s
p
an
s
.
W
h
ile
Ph
o
B
E
R
T
an
d
XL
M
-
R
m
o
d
els
p
er
f
o
r
m
e
d
co
m
p
etitiv
ely
,
t
h
ey
r
em
ain
ed
b
eh
in
d
ViHate
Of
f
.
No
tab
ly
,
Ph
o
B
E
R
T
large
o
u
tp
er
f
o
r
m
e
d
Ph
o
B
E
R
T
base
,
an
d
XL
M
-
R
large
s
lig
h
tly
s
u
r
p
ass
ed
XL
M
-
R
base
in
all
k
ey
m
etr
ics.
T
ab
le
4
s
u
m
m
ar
izes
m
o
d
el
p
e
r
f
o
r
m
a
n
ce
o
n
th
e
“
all
s
p
an
s
”
s
u
b
s
et.
ViHate
Of
f
word
ac
h
iev
ed
th
e
b
est
r
esu
lts
ac
r
o
s
s
all
m
etr
ics
—
p
r
ec
is
io
n
(
0
.
8
8
2
3
)
,
r
ec
all
(
0
.
8
5
7
3
)
,
F1
-
s
co
r
e
(
0
.
8
6
9
3
)
—
with
th
e
to
p
av
er
a
g
e
r
an
k
o
f
1
.
3
3
.
ViHate
Of
f
syllable
also
p
er
f
o
r
m
ed
s
tr
o
n
g
ly
,
r
a
n
k
in
g
s
ec
o
n
d
o
v
e
r
all
with
th
e
h
i
g
h
est
r
ec
all
(
0
.
8
6
1
1
)
an
d
an
F1
-
s
co
r
e
o
f
0
.
8
6
7
8
.
I
n
co
n
tr
ast,
B
iLST
M
-
C
R
F+Ph
o
2
W
m
o
d
els
s
h
o
wed
th
e
we
ak
est
p
er
f
o
r
m
a
n
ce
,
esp
ec
ially
th
e
wo
r
d
-
b
ased
v
e
r
s
io
n
with
th
e
lo
west
F1
-
s
co
r
e
(
0
.
7
0
3
6
)
an
d
av
er
ag
e
r
a
n
k
o
f
8
.
0
0
.
Mo
d
er
n
m
o
d
els
lik
e
Ph
o
B
E
R
T
large
a
n
d
XL
M
-
R
large
o
u
tp
er
f
o
r
m
ed
th
eir
b
ase
co
u
n
ter
p
ar
ts
b
u
t
r
em
ain
ed
b
e
h
in
d
ViHate
Of
f
,
co
n
f
ir
m
in
g
th
e
a
d
v
an
tag
e
o
f
task
-
s
p
ec
if
ic
an
d
Vietn
am
ese
-
o
p
tim
ized
ar
c
h
itectu
r
es.
T
ab
le
2
.
Per
f
o
r
m
an
ce
co
m
p
a
r
is
o
n
o
f
ex
p
er
im
en
tal
m
et
h
o
d
s
o
n
“
s
in
g
le
s
p
an
”
s
u
b
s
ets
M
o
d
e
l
P
(
r
a
n
k
)
R
(
r
a
n
k
)
F
1
(
r
a
n
k
)
A
v
e
r
a
g
e
r
a
n
k
S
y
l
l
a
b
l
e
B
i
LST
M
-
C
R
F
+
P
h
o
2
W
s
y
l
l
a
bl
e
0
.
4
2
2
2
(
7
)
0
.
5
0
0
9
(
7
)
0
.
4
3
2
9
(
7
)
7
.
0
0
X
LM
-
R
b
a
s
e
0
.
7
6
0
4
(
3
)
0
.
7
6
5
3
(
3
)
0
.
7
2
0
3
(
4
)
3
.
3
3
X
LM
-
R
l
a
rg
e
0
.
7
5
7
7
(
4
)
0
.
7
6
7
9
(
2
)
0
.
7
2
1
4
(
3
)
3
.
0
0
V
i
H
a
t
e
O
f
f
s
yl
l
a
b
l
e
0
.
8
5
1
9
(
1
)
0
.
7
4
9
2
(
5
)
0
.
7
9
0
7
(
2
)
2
.
6
7
W
o
r
d
B
i
LST
M
-
C
R
F
+
P
h
o
2
W
w
ord
0
.
3
1
9
6
(
8
)
0
.
4
4
6
8
(
8
)
0
.
3
5
9
4
(
8
)
8
.
0
0
P
h
o
B
E
R
T
b
a
s
e
0
.
7
3
9
2
(
6
)
0
.
7
4
8
5
(
6
)
0
.
7
0
1
6
(
6
)
6
.
0
0
P
h
o
B
E
R
T
l
a
rg
e
0
.
7
4
3
5
(
5
)
0
.
7
5
6
7
(
4
)
0
.
7
0
6
7
(
5
)
4
.
6
7
V
i
H
a
t
e
O
f
f
w
ord
0
.
7
9
5
7
(
2
)
0
.
8
4
2
7
(
1
)
0
.
8
1
7
4
(
1
)
1
.
3
3
T
ab
le
3
.
Per
f
o
r
m
an
ce
co
m
p
a
r
is
o
n
o
f
ex
p
er
im
en
tal
m
et
h
o
d
s
o
n
“m
u
ltip
le
s
p
an
s
”
s
u
b
s
ets
M
o
d
e
l
P
(
r
a
n
k
)
R
(
r
a
n
k
)
F
1
(
r
a
n
k
)
A
v
e
r
a
g
e
ra
nk
S
y
l
l
a
b
l
e
B
i
LST
M
-
C
R
F
+
P
h
o
2
W
s
y
l
l
a
bl
e
0
.
5
1
3
4
(
7
)
0
.
5
7
1
2
(
7
)
0
.
5
0
6
8
(
7
)
7
.
0
0
X
LM
-
R
b
a
s
e
0
.
7
2
0
3
(
6
)
0
.
7
9
2
7
(
3
)
0
.
7
3
2
7
(
4
)
4
.
3
3
X
LM
-
R
l
a
rg
e
0
.
7
8
2
9
(
4
)
0
.
7
5
6
9
(
4
)
0
.
7
3
5
7
(
3
)
3
.
6
7
V
i
H
a
t
e
O
f
f
s
yl
l
a
b
l
e
0
.
8
4
3
3
(
2
)
0
.
8
4
7
3
(
2
)
0
.
8
4
5
3
(
2
)
2
.
0
0
W
o
r
d
B
i
LST
M
-
C
R
F
+
P
h
o
2
W
w
o
rd
0
.
3
5
3
3
(
8
)
0
.
5
0
0
1
(
8
)
0
.
4
0
1
3
(
8
)
8
.
0
0
P
h
o
B
E
R
T
b
a
s
e
0
.
7
7
6
1
(
5
)
0
.
7
3
2
9
(
6
)
0
.
7
0
9
2
(
6
)
5
.
6
7
P
h
o
B
E
R
T
l
a
rg
e
0
.
7
8
7
8
(
3
)
0
.
7
5
5
7
(
5
)
0
.
7
3
2
1
(
5
)
4
.
3
3
V
i
H
a
t
e
O
f
f
w
ord
0
.
8
5
9
7
(
1
)
0
.
8
8
3
0
(
1
)
0
.
8
7
0
9
(
1
)
1
.
0
0
T
ab
le
4
.
Per
f
o
r
m
an
ce
co
m
p
a
r
is
o
n
o
f
ex
p
er
im
en
tal
m
et
h
o
d
s
o
n
“a
ll sp
an
s
”
s
u
b
s
ets
M
o
d
e
l
P
(
r
a
n
k
)
R
(
r
a
n
k
)
F
1
(
r
a
n
k
)
A
v
e
r
a
g
e
r
a
n
k
S
y
l
l
a
b
l
e
B
i
LST
M
-
C
R
F
+
P
h
o
2
W
s
yl
l
a
b
l
e
0
.
7
4
5
2
(
7
)
0
.
7
7
6
9
(
5
)
0
.
7
4
5
3
(
7
)
6
.
3
3
X
LM
-
R
b
a
s
e
0
.
7
7
6
6
(
6
)
0
.
7
5
7
4
(
7
)
0
.
7
4
6
7
(
6
)
6
.
3
3
X
LM
-
R
l
a
rg
e
0
.
8
0
7
1
(
3
)
0
.
7
8
8
7
(
3
)
0
.
7
7
7
0
(
3
)
3
.
0
0
V
i
H
a
t
e
O
f
f
s
yl
l
a
b
l
e
0
.
8
7
4
8
(
2
)
0
.
8
6
1
1
(
1
)
0
.
8
6
7
8
(
2
)
1
.
6
7
W
o
r
d
B
i
LST
M
-
C
R
F
+
P
h
o
2
W
w
ord
0
.
6
8
2
3
(
8
)
0
.
7
4
8
9
(
8
)
0
.
7
0
3
6
(
8
)
8
.
0
0
P
h
o
B
E
R
T
b
a
s
e
0
.
7
8
7
0
(
5
)
0
.
7
6
8
0
(
6
)
0
.
7
5
6
9
(
5
)
5
.
3
3
P
h
o
B
E
R
T
l
a
rg
e
0
.
8
0
2
8
(
4
)
0
.
7
8
3
5
(
4
)
0
.
7
7
1
6
(
4
)
4
.
0
0
V
i
H
a
t
e
O
f
f
w
ord
0
.
8
8
2
3
(
1
)
0
.
8
5
7
3
(
2
)
0
.
8
6
9
3
(
1
)
1
.
3
3
T
h
e
r
esu
lts
ac
r
o
s
s
all
th
r
ee
tab
les
co
n
s
is
ten
tly
co
n
f
i
r
m
th
e
s
u
p
er
io
r
ity
o
f
th
e
ViHate
Of
f
f
r
a
m
ewo
r
k
s
,
esp
ec
ially
ViHate
Of
f
word
,
wh
ich
ac
h
iev
ed
t
h
e
h
ig
h
est
m
etr
ics
an
d
lo
west
av
er
ag
e
r
a
n
k
s
a
cr
o
s
s
all
s
p
an
ty
p
es.
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
E
n
h
a
n
ce
d
fr
a
mewo
r
k
fo
r
d
etec
tin
g
V
ietn
a
mese
h
a
te
a
n
d
o
ffen
s
ive
s
p
a
n
s
(
Din
h
-
Ho
n
g
V
u
)
969
ViHate
Of
f
syllable
also
p
er
f
o
r
m
ed
s
tr
o
n
g
ly
,
co
n
s
is
ten
tly
r
an
k
in
g
s
ec
o
n
d
a
n
d
h
ig
h
lig
h
tin
g
th
e
v
al
u
e
o
f
s
y
llab
le
-
lev
el
p
r
o
ce
s
s
in
g
.
T
h
ese
f
in
d
in
g
s
d
em
o
n
s
tr
ate
th
at
lev
er
ag
in
g
a
d
v
an
ce
d
,
Vietn
am
ese
-
s
p
ec
if
ic
d
ee
p
lear
n
in
g
tech
n
iq
u
es
as
d
o
n
e
in
ViHate
Of
f
is
cr
u
cial
f
o
r
ac
h
iev
in
g
SOTA
p
er
f
o
r
m
an
ce
i
n
d
etec
tin
g
h
ate
an
d
o
f
f
en
s
iv
e
s
p
an
s
.
B
ey
o
n
d
ev
alu
atio
n
m
etr
ics,
we
also
ass
e
s
s
ed
th
e
in
f
er
en
ce
ef
f
icien
cy
o
f
th
e
ViHate
Of
f
f
r
am
ewo
r
k
u
n
d
er
d
if
f
er
e
n
t
d
ep
lo
y
m
en
t
s
ce
n
ar
io
s
.
On
a
C
PU
-
o
n
ly
m
ac
h
in
e
(
8
0
v
C
PUs
,
1
2
6
GB
R
AM
)
,
th
e
m
o
d
el
ac
h
iev
es
ap
p
r
o
x
im
ately
1
0
-
1
5
co
m
m
en
ts
p
e
r
s
ec
o
n
d
,
s
u
itab
le
f
o
r
b
atch
p
r
o
ce
s
s
in
g
o
r
o
f
f
lin
e
m
o
d
e
r
atio
n
.
On
a
h
ig
h
-
p
e
r
f
o
r
m
an
ce
s
er
v
er
with
an
NVI
DI
A
A1
0
0
GPU
,
th
r
o
u
g
h
p
u
t
r
ea
c
h
es
u
p
to
9
0
0
-
1
,
2
0
0
co
m
m
en
ts
p
er
s
ec
o
n
d
i
n
b
atch
m
o
d
e,
o
r
1
0
0
-
2
0
0
c
o
m
m
en
ts
p
er
s
ec
o
n
d
f
o
r
r
ea
l
-
tim
e
s
ettin
g
s
.
T
h
ese
r
esu
lts
in
d
icate
th
at
ViHate
Of
f
is
f
ea
s
ib
le
f
o
r
lar
g
e
-
s
ca
le
d
ep
lo
y
m
e
n
t,
p
ar
ticu
lar
l
y
wh
en
GPU
r
eso
u
r
ce
s
ar
e
av
ailab
le.
3
.
3
.
E
rr
o
r
a
na
ly
s
is
Du
r
in
g
o
u
r
e
v
alu
atio
n
,
we
id
en
tifie
d
co
m
m
o
n
s
o
u
r
ce
s
o
f
m
is
class
if
icatio
n
in
th
e
ViHate
Of
f
f
r
am
ewo
r
k
,
p
r
im
ar
ily
d
u
e
to
lin
g
u
is
tic
co
m
p
le
x
ities
in
h
er
en
t
in
Vietn
am
ese
s
o
cial
m
ed
ia
lan
g
u
ag
e.
T
h
e
k
e
y
er
r
o
r
ca
teg
o
r
ies
in
clu
d
e:
em
er
g
in
g
o
r
u
n
lis
ted
s
lan
g
ex
p
r
ess
io
n
s
n
o
t
ca
p
tu
r
ed
in
th
e
cu
r
r
en
t
HSD
(
e.
g
.
,
“
ă
n
h
à
n
h
”,
“
g
ã
y
kè
o
”,
“
nổ
”
d
ep
e
n
d
in
g
o
n
co
n
tex
t)
;
s
ar
ca
s
tic
o
r
ir
o
n
ic
ex
p
r
ess
io
n
s
,
wh
ich
r
eq
u
ir
e
d
ee
p
co
n
tex
tu
al
u
n
d
er
s
tan
d
in
g
an
d
o
f
ten
r
ev
er
s
e
th
e
ap
p
ar
en
t
s
en
tim
en
t
.
An
o
th
er
ca
te
g
o
r
y
i
s
m
etap
h
o
r
ica
l
o
r
in
d
ir
ec
t
in
s
u
lts
,
wh
er
e
o
f
f
en
s
iv
e
m
ea
n
in
g
is
im
p
lied
r
ath
e
r
th
an
d
ir
ec
tly
s
tated
(
e.
g
.
,
s
ee
m
in
g
ly
p
o
s
itiv
e
p
h
r
ases
u
s
ed
in
m
o
c
k
in
g
t
o
n
e
lik
e
“
đ
ỉn
h
th
ậ
t
”)
.
Fo
r
in
s
tan
ce
,
in
t
h
e
s
en
ten
ce
“
ô
i
b
ố
c
á
i
lũ
t
h
a
n
h
n
iên
h
ã
m
lo
l.
đ
ẹp
mặ
t
q
u
á
”/“o
h
f
ath
er
o
f
th
o
s
e
d
is
g
u
s
tin
g
y
o
u
t
h
s
[
v
u
lg
ar
s
la
n
g
]
h
o
w
g
lo
r
io
u
s
…”,
th
e
g
o
l
d
lab
els
in
clu
d
e
“
b
ố
,
cá
i,
lũ
,
h
ã
m,
lo
l,
đ
ẹ
p
,
mặ
t,
quá
”
b
u
t
th
e
m
o
d
el
o
n
ly
i
d
e
n
tifie
d
“
lũ
,
h
ã
m,
l
o
l
”
as
h
atef
u
l
to
k
en
s
.
T
o
k
en
s
“
b
ố
,
cá
i
”
ca
r
r
y
a
co
n
te
x
tu
al
in
s
u
ltin
g
to
n
e
,
b
u
t
a
r
e
n
o
t
i
n
h
er
en
tly
o
f
f
en
s
iv
e
o
n
th
eir
o
w
n
.
T
o
k
en
s
“
đ
ẹp
,
mặ
t,
q
u
á
”
is
s
ar
ca
s
tic,
r
ev
er
s
in
g
th
e
liter
al
s
en
tim
en
t
(
liter
all
y
“
s
o
g
lo
r
io
u
s
”)
in
to
r
id
icu
l
e
ch
allen
g
in
g
f
o
r
th
e
m
o
d
el
to
d
etec
t
with
o
u
t
p
r
ag
m
atic
u
n
d
er
s
tan
d
in
g
.
T
h
ese
er
r
o
r
s
h
ig
h
lig
h
t
th
e
li
m
itatio
n
s
o
f
b
o
th
t
h
e
s
tatic
d
ictio
n
ar
y
a
p
p
r
o
ac
h
an
d
cu
r
r
en
t
m
o
d
el
ar
ch
itectu
r
e
in
h
an
d
lin
g
n
u
an
c
ed
o
r
co
n
tex
t
-
d
ep
e
n
d
en
t
lan
g
u
ag
e.
I
n
f
u
t
u
r
e
wo
r
k
,
we
p
r
o
p
o
s
e
to
:
co
n
tin
u
o
u
s
ly
u
p
d
ate
t
h
e
HSD
with
n
ew
s
lan
g
a
n
d
s
o
cial
m
ed
ia
ter
m
s
.
I
n
co
r
p
o
r
ate
s
ar
ca
s
m
-
awa
r
e
p
r
etr
ain
in
g
o
r
f
in
e
-
tu
n
e
o
n
d
atasets
s
p
ec
if
ically
a
n
n
o
tated
f
o
r
ir
o
n
y
.
I
n
te
g
r
ate
co
n
tex
t
-
s
en
s
itiv
e
s
en
tim
en
t
m
o
d
els
to
im
p
r
o
v
e
r
o
b
u
s
tn
ess
ag
ain
s
t in
d
ir
ec
t e
x
p
r
ess
io
n
s
.
4.
CO
NCLU
SI
O
N
T
h
is
s
tu
d
y
p
r
o
p
o
s
ed
an
e
n
h
an
ce
d
f
r
am
ewo
r
k
f
o
r
d
etec
tin
g
h
ate
an
d
o
f
f
e
n
s
iv
e
s
p
an
s
in
Vietn
am
ese
b
y
ad
d
r
ess
in
g
k
e
y
lin
g
u
is
tic
ch
allen
g
es
th
r
o
u
g
h
a
co
m
b
in
atio
n
o
f
d
ata
p
r
ep
r
o
ce
s
s
in
g
,
t
h
e
co
n
s
tr
u
ctio
n
o
f
a
HSD,
an
d
th
e
ap
p
licatio
n
o
f
t
h
e
Ph
o
B
E
R
T
-
L
ar
g
e
lan
g
u
a
g
e
m
o
d
el.
T
h
e
ex
p
er
im
en
tal
r
esu
lts
o
n
th
e
ViHOS
d
ataset
d
em
o
n
s
tr
ate
th
at
o
u
r
a
p
p
r
o
ac
h
co
n
s
is
ten
tly
o
u
t
p
er
f
o
r
m
s
SOTA
m
eth
o
d
s
ac
r
o
s
s
all
ev
alu
atio
n
s
u
b
s
ets:
s
in
g
le
s
p
an
,
m
u
ltip
le
s
p
an
s
,
a
n
d
all
s
p
an
s
.
T
h
ese
f
in
d
in
g
s
co
n
f
ir
m
th
e
r
o
b
u
s
tn
ess
an
d
ef
f
ec
tiv
en
ess
o
f
th
e
f
r
am
ewo
r
k
i
n
id
en
tify
in
g
h
ar
m
f
u
l
co
n
ten
t,
p
ar
ticu
lar
ly
o
n
s
o
cial
m
ed
ia
p
latf
o
r
m
s
.
Desp
ite
th
e
d
em
o
n
s
tr
ated
ef
f
ec
tiv
en
ess
o
f
th
e
HSD,
its
co
v
er
a
g
e
r
em
ain
s
in
h
er
en
tly
lim
ited
d
u
e
to
its
s
tatic
n
atu
r
e
an
d
r
elian
ce
o
n
m
an
u
ally
an
n
o
tated
d
ata.
As
s
lan
g
an
d
o
f
f
en
s
iv
e
e
x
p
r
ess
io
n
s
ev
o
lv
e
r
a
p
id
ly
o
n
s
o
cial
m
ed
ia,
m
an
y
n
ewly
em
er
g
in
g
ter
m
s
ar
e
n
o
t
ca
p
tu
r
ed
in
th
e
cu
r
r
en
t
d
ictio
n
ar
y
.
Fu
r
th
er
m
o
r
e
,
th
e
m
o
d
el
s
tr
u
g
g
les
with
s
ar
ca
s
m
,
ir
o
n
y
,
a
n
d
m
etap
h
o
r
ical
lan
g
u
ag
e,
wh
er
e
th
e
liter
al
m
ea
n
in
g
o
f
ten
d
if
f
e
r
s
f
r
o
m
th
e
in
ten
d
ed
s
en
tim
en
t.
T
o
ad
d
r
ess
th
ese
ch
allen
g
es,
we
p
lan
to
d
e
v
elo
p
an
a
u
to
m
a
ted
d
ictio
n
ar
y
u
p
d
ate
p
ip
elin
e
,
in
teg
r
ate
s
ar
ca
s
m
-
awa
r
e
p
r
etr
ain
in
g
an
d
co
n
te
x
t
-
s
en
s
itiv
e
s
en
tim
en
t
m
o
d
els,
an
d
co
llect
ad
d
itio
n
al
an
n
o
tated
d
atasets
to
en
h
an
ce
th
e
a
d
ap
tab
ilit
y
a
n
d
r
o
b
u
s
tn
ess
o
f
th
e
s
y
s
tem
in
r
ea
l
-
wo
r
ld
s
ce
n
ar
io
s
.
I
n
f
u
tu
r
e
wo
r
k
,
we
p
la
n
to
en
r
ich
th
e
d
ataset
b
y
c
o
llectin
g
m
o
r
e
lab
eled
d
ata
f
r
o
m
d
iv
er
s
e
s
o
u
r
ce
s
t
o
im
p
r
o
v
e
t
h
e
m
o
d
el'
s
g
en
er
aliza
b
ilit
y
.
Ad
d
itio
n
ally
,
in
co
r
p
o
r
atin
g
ad
v
a
n
ce
d
tec
h
n
iq
u
es
s
u
ch
as
h
y
p
er
p
ar
am
et
er
tu
n
in
g
,
en
s
em
b
le
s
tr
ateg
ies,
an
d
ex
ter
n
al
k
n
o
wled
g
e
b
ases
m
ay
f
u
r
th
e
r
en
h
a
n
ce
th
e
ad
ap
tab
ilit
y
an
d
p
er
f
o
r
m
an
ce
o
f
th
e
s
y
s
tem
in
r
ea
l
-
wo
r
ld
a
p
p
licatio
n
s
.
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
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
5
,
No
.
1
,
Feb
r
u
ar
y
2
0
2
6
:
962
-
9
7
1
970
Na
m
e
o
f
Aut
ho
r
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
Din
h
-
Ho
n
g
Vu
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
T
u
o
n
g
L
e
✓
✓
✓
✓
✓
✓
✓
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.
I
NF
O
RM
E
D
CO
NS
E
N
T
T
h
is
s
tu
d
y
d
id
n
o
t
in
v
o
l
v
e
h
u
m
an
s
u
b
jects,
p
er
s
o
n
al
i
n
f
o
r
m
atio
n
,
o
r
i
d
en
tifia
b
le
d
ata
r
eq
u
ir
i
n
g
in
f
o
r
m
e
d
co
n
s
en
t.
E
T
H
I
CAL AP
P
RO
V
AL
T
h
is
s
tu
d
y
d
id
n
o
t
i
n
v
o
l
v
e
h
u
m
an
s
u
b
jects,
an
im
als,
o
r
an
y
e
x
p
er
im
e
n
ts
r
eq
u
i
r
in
g
in
s
titu
tio
n
al
r
ev
iew
b
o
ar
d
a
p
p
r
o
v
al,
as
it
is
b
ased
s
o
lely
o
n
co
m
p
u
tatio
n
al
an
aly
s
is
u
s
in
g
p
u
b
licly
av
ailab
le
d
atasets
an
d
p
r
e
-
tr
ain
ed
lan
g
u
a
g
e
m
o
d
els.
DATA AV
AI
L
AB
I
L
I
T
Y
T
h
e
d
ataset
u
s
ed
in
th
is
s
tu
d
y
is
p
u
b
licly
av
ailab
le
i
n
Git
Hu
b
at
h
ttp
s
://g
ith
u
b
.
co
m
/p
h
u
s
r
o
y
al/
ViHOS.
RE
F
E
R
E
NC
E
S
[
1
]
S
.
R
a
n
a
t
h
u
n
g
a
,
E
.
S
.
A
.
Le
e
,
M
.
P
.
S
k
e
n
d
u
l
i
,
R
.
S
h
e
k
h
a
r
,
M
.
A
l
a
m,
a
n
d
R
.
K
a
u
r
,
“
N
e
u
r
a
l
m
a
c
h
i
n
e
t
r
a
n
s
l
a
t
i
o
n
f
o
r
l
o
w
-
r
e
s
o
u
r
c
e
l
a
n
g
u
a
g
e
s
:
A
s
u
r
v
e
y
,
”
A
C
M
C
o
m
p
u
t
i
n
g
S
u
rv
e
y
s
,
v
o
l
.
5
5
,
n
o
.
1
1
,
p
p
.
1
–
3
7
,
2
0
2
3
,
d
o
i
:
1
0
.
1
1
4
5
/
3
5
6
7
5
9
2
.
[
2
]
S
.
Z
h
u
,
C
.
M
i
,
T
.
L
i
,
Y
.
Y
a
n
g
,
a
n
d
C
.
X
u
,
“
U
n
s
u
p
e
r
v
i
s
e
d
p
a
r
a
l
l
e
l
s
e
n
t
e
n
c
e
s
o
f
m
a
c
h
i
n
e
t
r
a
n
s
l
a
t
i
o
n
f
o
r
A
s
i
a
n
l
a
n
g
u
a
g
e
p
a
i
r
s,
”
AC
M
T
ra
n
s
a
c
t
i
o
n
s
o
n
As
i
a
n
a
n
d
L
o
w
-
R
e
so
u
r
c
e
L
a
n
g
u
a
g
e
I
n
f
o
rm
a
t
i
o
n
Pro
c
e
ss
i
n
g
,
v
o
l
.
2
2
,
n
o
.
3
,
p
p
.
6
4
:
1
-
6
4
:
1
4
,
2
0
2
3
,
d
o
i
:
1
0
.
1
1
4
5
/
3
4
8
6
6
7
7
.
[
3
]
P
.
N
.
N
g
u
y
e
n
a
n
d
P
.
Tr
a
n
,
“
C
o
n
s
t
r
u
c
t
i
n
g
a
C
h
i
n
e
se
-
V
i
e
t
n
a
mes
e
b
i
l
i
n
g
u
a
l
c
o
r
p
u
s
f
r
o
m
s
u
b
t
i
t
l
e
w
e
b
s
i
t
e
s,
”
I
n
t
e
rn
a
t
i
o
n
a
l
J
o
u
rn
a
l
o
f
I
n
t
e
l
l
i
g
e
n
t
I
n
f
o
rm
a
t
i
o
n
a
n
d
D
a
t
a
b
a
s
e
S
y
s
t
e
m
s
,
v
o
l
.
1
6
,
n
o
.
4
,
p
p
.
3
8
5
–
4
0
8
,
2
0
2
4
,
d
o
i
:
1
0
.
1
5
0
4
/
I
J
I
I
D
S
.
2
0
2
4
.
1
4
1
7
4
8
.
[
4
]
D
.
H
.
V
u
,
K
.
N
g
u
y
e
n
,
K
.
T.
Tr
a
n
,
B
.
V
o
,
a
n
d
T
.
Le
,
“
I
mp
r
o
v
i
n
g
f
a
k
e
j
o
b
d
e
s
c
r
i
p
t
i
o
n
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
-
b
a
s
e
d
N
LP
t
e
c
h
n
i
q
u
e
s,
”
J
o
u
r
n
a
l
o
f
I
n
f
o
r
m
a
t
i
o
n
a
n
d
T
e
l
e
c
o
m
m
u
n
i
c
a
t
i
o
n
,
v
o
l
.
9
,
n
o
.
1
,
p
p
.
1
1
3
–
1
2
5
,
2
0
2
5
,
d
o
i
:
1
0
.
1
0
8
0
/
2
4
7
5
1
8
3
9
.
2
0
2
4
.
2
3
8
7
3
8
0
.
[
5
]
Y
.
W
u
a
n
d
J
.
W
a
n
,
“
A
s
u
r
v
e
y
o
f
t
e
x
t
c
l
a
ssi
f
i
c
a
t
i
o
n
b
a
se
d
o
n
p
r
e
-
t
r
a
i
n
e
d
l
a
n
g
u
a
g
e
m
o
d
e
l
,
”
N
e
u
ro
c
o
m
p
u
t
i
n
g
,
v
o
l
.
6
1
6
,
2
0
2
5
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
n
e
u
c
o
m.
2
0
2
4
.
1
2
8
9
2
1
.
[
6
]
H
.
A
l
l
a
m,
L
.
M
a
k
u
b
v
u
r
e
,
B
.
G
y
a
mf
i
,
K
.
N
.
G
r
a
h
a
m
,
a
n
d
K
.
A
k
i
n
w
o
l
e
r
e
,
“
Te
x
t
c
l
a
ssi
f
i
c
a
t
i
o
n
:
h
o
w
m
a
c
h
i
n
e
l
e
a
r
n
i
n
g
i
s
r
e
v
o
l
u
t
i
o
n
i
z
i
n
g
t
e
x
t
c
a
t
e
g
o
r
i
z
a
t
i
o
n
,
”
I
n
f
o
rm
a
t
i
o
n
,
v
o
l
.
1
6
,
n
o
.
2
,
2
0
2
5
,
d
o
i
:
1
0
.
3
3
9
0
/
i
n
f
o
1
6
0
2
0
1
3
0
.
[
7
]
N
.
A
.
S
a
p
u
t
r
a
,
K
.
A
e
n
i
,
a
n
d
N
.
M
.
S
a
r
a
sw
a
t
i
,
“
I
n
d
o
n
e
si
a
n
h
a
t
e
s
p
e
e
c
h
t
e
x
t
c
l
a
ss
i
f
i
c
a
t
i
o
n
u
si
n
g
i
m
p
r
o
v
e
d
k
-
n
e
a
r
e
st
n
e
i
g
h
b
o
r
w
i
t
h
TF
-
I
D
F
-
I
C
S
ρ
F
,
”
S
c
i
e
n
t
i
f
i
c
J
o
u
r
n
a
l
o
f
I
n
f
o
rm
a
t
i
c
s
,
v
o
l
.
1
1
,
n
o
.
1
,
p
p
.
2
1
–
3
0
,
2
0
2
4
,
d
o
i
:
1
0
.
1
5
2
9
4
/
s
j
i
.
v
1
1
i
1
.
4
8
0
8
5
.
[
8
]
G
.
Z.
N
a
b
i
i
l
a
h
,
I
.
N
.
A
l
a
m,
E
.
S
.
P
u
r
w
a
n
t
o
,
a
n
d
M
.
F
.
H
i
d
a
y
a
t
,
“
I
n
d
o
n
e
s
i
a
n
m
u
l
t
i
l
a
b
e
l
c
l
a
ssi
f
i
c
a
t
i
o
n
u
si
n
g
I
n
d
o
B
ER
T
e
mb
e
d
d
i
n
g
a
n
d
M
B
E
R
T
c
l
a
ssi
f
i
c
a
t
i
o
n
,
”
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
&
C
o
m
p
u
t
e
r
E
n
g
i
n
e
e
r
i
n
g
,
v
o
l
.
1
4
,
n
o
.
1
,
2
0
2
4
,
d
o
i
:
1
0
.
1
1
5
9
1
/
i
j
e
c
e
.
v
1
4
i
1
.
p
p
1
0
7
1
-
1
0
7
8
.
[
9
]
T.
K
.
Tr
a
n
,
H
.
M
.
D
i
n
h
,
a
n
d
T
.
T.
P
h
a
n
,
“
B
u
i
l
d
i
n
g
a
n
e
n
h
a
n
c
e
d
s
e
n
t
i
m
e
n
t
c
l
a
ss
i
f
i
c
a
t
i
o
n
f
r
a
mew
o
r
k
b
a
s
e
d
o
n
n
a
t
u
r
a
l
l
a
n
g
u
a
g
e
p
r
o
c
e
ss
i
n
g
,
”
J
o
u
rn
a
l
o
f
I
n
t
e
l
l
i
g
e
n
t
&
Fu
zzy
S
y
st
e
m
s
,
v
o
l
.
4
3
,
n
o
.
2
,
p
p
.
1
7
7
1
–
1
7
7
7
,
2
0
2
2
,
d
o
i
:
1
0
.
3
2
3
3
/
JI
F
S
-
2
1
9
2
7
8
.
[
1
0
]
M
.
W
a
n
k
h
a
d
e
,
A
.
C
.
S
.
R
a
o
,
a
n
d
C
.
K
u
l
k
a
r
n
i
,
“
A
su
r
v
e
y
o
n
s
e
n
t
i
me
n
t
a
n
a
l
y
si
s
me
t
h
o
d
s,
a
p
p
l
i
c
a
t
i
o
n
s
,
a
n
d
c
h
a
l
l
e
n
g
e
s,”
Ar
t
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
Re
v
i
e
w
,
v
o
l
.
5
5
,
n
o
.
7
,
p
p
.
5
7
3
1
–
5
7
8
0
,
2
0
2
2
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
0
4
6
2
-
0
2
2
-
1
0
1
4
4
-
1.
[
1
1
]
H
.
M
u
r
f
i
,
T.
G
o
w
a
n
d
i
,
G
.
A
r
d
a
n
e
s
w
a
r
i
,
a
n
d
S
.
N
u
r
r
o
h
m
a
h
,
“
B
E
R
T
-
b
a
sed
c
o
mb
i
n
a
t
i
o
n
o
f
c
o
n
v
o
l
u
t
i
o
n
a
l
a
n
d
r
e
c
u
r
r
e
n
t
n
e
u
r
a
l
n
e
t
w
o
r
k
f
o
r
I
n
d
o
n
e
s
i
a
n
s
e
n
t
i
m
e
n
t
a
n
a
l
y
s
i
s,”
A
p
p
l
i
e
d
S
o
f
t
C
o
m
p
u
t
i
n
g
,
v
o
l
.
1
5
1
,
2
0
2
4
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
a
s
o
c
.
2
0
2
3
.
1
1
1
1
1
2
.
[
1
2
]
L.
Z
h
u
,
H
.
Z
h
a
o
,
Z.
Zh
u
,
C
.
Zh
a
n
g
,
a
n
d
X
.
K
o
n
g
,
“
M
u
l
t
i
m
o
d
a
l
s
e
n
t
i
m
e
n
t
a
n
a
l
y
si
s
w
i
t
h
u
n
i
m
o
d
a
l
l
a
b
e
l
g
e
n
e
r
a
t
i
o
n
a
n
d
m
o
d
a
l
i
t
y
d
e
c
o
m
p
o
si
t
i
o
n
,
”
I
n
f
o
rm
a
t
i
o
n
Fu
s
i
o
n
,
v
o
l
.
1
1
6
,
2
0
2
5
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
i
n
f
f
u
s.
2
0
2
4
.
1
0
2
7
8
7
.
[
1
3
]
N
.
D
a
r
r
a
z
,
I
.
K
a
r
a
b
i
l
a
,
A
.
El
-
A
n
sari
,
N
.
A
l
a
mi
,
a
n
d
M
.
E
l
M
a
l
l
a
h
i
,
“
I
n
t
e
g
r
a
t
e
d
se
n
t
i
me
n
t
a
n
a
l
y
s
i
s
w
i
t
h
B
E
R
T
f
o
r
e
n
h
a
n
c
e
d
h
y
b
r
i
d
r
e
c
o
m
me
n
d
a
t
i
o
n
s
y
s
t
e
ms,
”
Ex
p
e
r
t
S
y
s
t
e
m
s
w
i
t
h
A
p
p
l
i
c
a
t
i
o
n
s
,
v
o
l
.
2
6
1
,
2
0
2
5
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
e
sw
a
.
2
0
2
4
.
1
2
5
5
3
3
.
[
1
4
]
Y
.
Zh
u
a
n
g
,
Y
.
Y
u
,
K
.
W
a
n
g
,
H
.
S
u
n
,
a
n
d
C
.
Z
h
a
n
g
,
“
To
o
l
Q
A
:
a
d
a
t
a
s
e
t
f
o
r
LL
M
q
u
e
st
i
o
n
a
n
sw
e
r
i
n
g
w
i
t
h
e
x
t
e
r
n
a
l
t
o
o
l
s
,
”
i
n
Pro
c
e
e
d
i
n
g
s
o
f
t
h
e
3
7
t
h
I
n
t
e
rn
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
N
e
u
r
a
l
I
n
f
o
rm
a
t
i
o
n
Pr
o
c
e
ssi
n
g
S
y
st
e
m
s
,
2
0
2
3
,
p
p
.
5
0
1
1
7
–
5
0
1
4
3
.
[
1
5
]
P
.
L
u
e
t
a
l
.
,
“
Le
a
r
n
t
o
e
x
p
l
a
i
n
:
mu
l
t
i
mo
d
a
l
r
e
a
s
o
n
i
n
g
v
i
a
t
h
o
u
g
h
t
c
h
a
i
n
s
f
o
r
sc
i
e
n
c
e
q
u
e
st
i
o
n
a
n
sw
e
r
i
n
g
,
”
in
P
ro
c
e
e
d
i
n
g
s
o
f
t
h
e
3
6
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
N
e
u
ra
l
I
n
f
o
rm
a
t
i
o
n
Pro
c
e
ss
i
n
g
S
y
s
t
e
m
,
2
0
2
2
,
p
p
.
2
5
0
7
–
2
5
2
1
.
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
E
n
h
a
n
ce
d
fr
a
mewo
r
k
fo
r
d
etec
tin
g
V
ietn
a
mese
h
a
te
a
n
d
o
ffen
s
ive
s
p
a
n
s
(
Din
h
-
Ho
n
g
V
u
)
971
[
1
6
]
K
.
V
T
r
a
n
,
H
.
P
.
P
h
a
n
,
K
.
V
N
g
u
y
e
n
,
a
n
d
N
.
L
.
T
.
N
g
u
y
e
n
,
“
V
i
C
L
E
V
R
:
a
v
i
s
u
a
l
r
e
a
s
o
n
i
n
g
d
a
t
a
s
e
t
a
n
d
h
y
b
r
i
d
m
u
l
t
i
m
o
d
a
l
f
u
s
i
o
n
m
o
d
e
l
f
o
r
v
i
s
u
a
l
q
u
e
s
t
i
o
n
a
n
s
w
e
r
i
n
g
i
n
V
i
e
t
n
a
m
e
s
e
,
”
M
u
l
t
i
m
e
d
i
a
S
y
s
t
e
m
s
,
v
o
l
.
3
0
,
n
o
.
4
,
2
0
2
4
,
d
o
i
:
1
0
.
1
0
0
7
/
s
0
0
5
3
0
-
0
2
4
-
0
1
3
9
4
-
w.
[
1
7
]
S
.
C
h
o
w
d
h
u
r
y
a
n
d
B
.
S
o
n
i
,
“
R
-
V
Q
A
:
a
r
o
b
u
s
t
v
i
s
u
a
l
q
u
e
st
i
o
n
a
n
sw
e
r
i
n
g
m
o
d
e
l
,
”
K
n
o
w
l
e
d
g
e
-
B
a
s
e
d
S
y
st
e
m
s
,
v
o
l
.
3
0
9
,
2
0
2
5
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
k
n
o
s
y
s.
2
0
2
4
.
1
1
2
8
2
7
.
[
1
8
]
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
Am
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
,
2
0
1
9
,
v
o
l
.
1
,
p
p
.
4
1
7
1
–
4
1
8
6
.
[
1
9
]
A
.
R
a
d
f
o
r
d
,
K
.
N
a
r
a
s
i
mh
a
n
,
T
.
S
a
l
i
man
s
,
a
n
d
I
.
S
u
t
s
k
e
v
e
r
,
"
I
mp
r
o
v
i
n
g
l
a
n
g
u
a
g
e
u
n
d
e
r
st
a
n
d
i
n
g
b
y
g
e
n
e
r
a
t
i
v
e
p
r
e
-
t
r
a
i
n
i
n
g
,"
O
p
e
n
A
I
,
2
0
1
8
.
[
O
n
l
i
n
e
]
.
A
v
a
i
l
a
b
l
e
:
h
t
t
p
s
:
/
/
o
p
e
n
a
i
.
c
o
m
/
r
e
s
e
a
r
c
h
/
l
a
n
g
u
a
g
e
-
u
n
su
p
e
r
v
i
s
e
d
[
2
0
]
D
.
Q
.
N
g
u
y
e
n
a
n
d
A
.
T.
N
g
u
y
e
n
,
“
P
h
o
B
E
R
T
:
p
r
e
-
t
r
a
i
n
e
d
l
a
n
g
u
a
g
e
mo
d
e
l
s fo
r
V
i
e
t
n
a
mes
e
,
”
i
n
Pr
o
c
e
e
d
i
n
g
s o
f
E
MN
L
P
Fi
n
d
i
n
g
s
,
2
0
2
0
,
p
p
.
1
0
3
7
–
1
0
4
2
,
d
o
i
:
1
0
.
1
8
6
5
3
/
v
1
/
2
0
2
0
.
f
i
n
d
i
n
g
s
-
e
m
n
l
p
.
9
2
.
[
2
1
]
S
.
Y
u
a
n
,
A
.
M
a
r
o
n
i
k
o
l
a
k
i
s
,
a
n
d
H
.
S
c
h
ü
t
z
e
,
“
S
e
p
a
r
a
t
i
n
g
h
a
t
e
s
p
e
e
c
h
a
n
d
o
f
f
e
n
si
v
e
l
a
n
g
u
a
g
e
c
l
a
ss
e
s
v
i
a
a
d
v
e
r
s
a
r
i
a
l
d
e
b
i
a
si
n
g
,
”
i
n
Pro
c
e
e
d
i
n
g
s
o
f
S
i
x
t
h
W
o
rks
h
o
p
o
n
O
n
l
i
n
e
Ab
u
se
a
n
d
H
a
rm
s (WO
AH
)
,
2
0
2
2
,
p
p
.
1
–
10
,
d
o
i
:
1
0
.
1
8
6
5
3
/
v
1
/
2
0
2
2
.
w
o
a
h
-
1
.
1
.
[
2
2
]
M
.
Z
a
m
p
i
e
r
i
,
S
.
R
o
se
n
t
h
a
l
,
P
.
N
a
k
o
v
,
A
.
D
mo
n
t
e
,
a
n
d
T.
R
a
n
a
s
i
n
g
h
e
,
“
O
f
f
e
n
sE
v
a
l
2
0
2
3
:
o
f
f
e
n
si
v
e
l
a
n
g
u
a
g
e
i
d
e
n
t
i
f
i
c
a
t
i
o
n
i
n
t
h
e
a
g
e
o
f
l
a
r
g
e
l
a
n
g
u
a
g
e
m
o
d
e
l
s
,
”
N
a
t
u
r
a
l
L
a
n
g
u
a
g
e
E
n
g
i
n
e
e
r
i
n
g
,
v
o
l
.
2
9
,
n
o
.
6
,
p
p
.
1
4
1
6
–
1
4
3
5
,
2
0
2
3
,
d
o
i
:
1
0
.
1
0
1
7
/
S
1
3
5
1
3
2
4
9
2
3
0
0
0
5
1
7
.
[
2
3
]
R
.
R
a
j
a
l
a
k
s
h
m
i
,
S
.
S
e
l
v
a
r
a
j
,
a
n
d
P
.
V
a
su
d
e
v
a
n
,
“
H
o
t
t
e
st
:
h
a
t
e
a
n
d
o
f
f
e
n
s
i
v
e
c
o
n
t
e
n
t
i
d
e
n
t
i
f
i
c
a
t
i
o
n
i
n
Ta
mi
l
u
s
i
n
g
t
r
a
n
sf
o
r
mers
a
n
d
e
n
h
a
n
c
e
d
st
e
mm
i
n
g
,
”
C
o
m
p
u
t
e
r
S
p
e
e
c
h
& La
n
g
u
a
g
e
,
v
o
l
.
7
8
,
2
0
2
3
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
c
sl
.
2
0
2
2
.
1
0
1
4
6
4
.
[
2
4
]
A
.
G
a
n
d
h
i
e
t
a
l
.
,
“
H
a
t
e
s
p
e
e
c
h
d
e
t
e
c
t
i
o
n
:
a
c
o
mp
r
e
h
e
n
s
i
v
e
r
e
v
i
e
w
o
f
r
e
c
e
n
t
w
o
r
k
s,
”
Ex
p
e
r
t
S
y
st
e
m
s
,
v
o
l
.
4
1
,
n
o
.
8
,
2
0
2
4
,
d
o
i
:
1
0
.
1
1
1
1
/
e
x
sy
.
1
3
5
6
2
.
[
2
5
]
P
.
G
.
H
o
a
n
g
,
C
.
L
u
u
,
K
.
Q
.
Tr
a
n
,
K
.
V
N
g
u
y
e
n
,
a
n
d
N
.
L.
-
T
.
N
g
u
y
e
n
,
“
V
i
H
O
S
:
h
a
t
e
s
p
e
e
c
h
s
p
a
n
s
d
e
t
e
c
t
i
o
n
f
o
r
V
i
e
t
n
a
mes
e
,
”
i
n
Pro
c
e
e
d
i
n
g
s
o
f
E
AC
L
,
2
0
2
3
,
p
p
.
6
5
2
–
6
6
9
,
d
o
i
:
1
0
.
1
8
6
5
3
/
v
1
/
2
0
2
3
.
e
a
c
l
-
mai
n
.
4
7
.
B
I
O
G
RAP
H
I
E
S O
F
AUTH
O
RS
Din
h
-
H
o
n
g
Vu
re
c
e
iv
e
d
th
e
B.
S
.
d
e
g
re
e
in
I
n
fo
rm
a
ti
o
n
Tec
h
n
o
lo
g
y
fr
o
m
th
e
VN
U
Ho
Ch
i
M
i
n
h
Cit
y
Un
i
v
e
rsity
o
f
S
c
ien
c
e
,
Ho
C
h
i
M
i
n
h
Cit
y
,
Vie
tn
a
m
,
in
2
0
0
5
,
a
n
d
th
e
M
.
S
c
.
d
e
g
re
e
in
C
o
m
p
u
ter S
c
ien
c
e
fro
m
th
e
VN
U
Ho
Ch
i
M
i
n
h
Ci
ty
Un
i
v
e
rsity
o
f
S
c
ien
c
e
,
i
n
2
0
1
1
.
He
is
c
u
rre
n
tl
y
a
re
se
a
rc
h
e
r
with
NLP
-
KD
Re
se
a
rc
h
G
ro
u
p
,
To
n
Du
c
Th
a
n
g
Un
i
v
e
rsity
,
Vie
tn
a
m
.
His
re
se
a
rc
h
in
tere
sts
in
c
lu
d
e
n
a
t
u
ra
l
lan
g
u
a
g
e
p
r
o
c
e
ss
in
g
,
m
a
c
h
in
e
tran
sla
ti
o
n
,
a
n
d
tex
t
c
lu
ste
ri
n
g
.
He
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
v
u
d
i
n
h
h
o
n
g
@t
d
tu
.
e
d
u
.
v
n
.
Tu
o
n
g
Le
re
c
e
iv
e
d
h
is
P
h
.
D.
d
e
g
re
e
in
C
o
m
p
u
ter
S
c
ie
n
c
e
fro
m
S
e
jo
n
g
Un
iv
e
rsity
,
Ko
re
a
,
i
n
2
0
2
0
.
He
is
c
u
rre
n
tl
y
a
re
se
a
rc
h
e
r
a
n
d
fa
c
u
l
t
y
m
e
m
b
e
r
a
t
th
e
F
a
c
u
lt
y
o
f
In
fo
rm
a
ti
o
n
Tec
h
n
o
l
o
g
y
,
HU
T
ECH
Un
iv
e
rsity
,
Vie
tn
a
m
.
H
e
h
a
s
a
u
th
o
re
d
o
v
e
r
4
0
p
u
b
li
c
a
ti
o
n
s
in
h
i
g
h
-
imp
a
c
t
j
o
u
r
n
a
ls,
in
c
l
u
d
i
n
g
I
n
fo
rm
a
ti
o
n
S
c
ien
c
e
s,
Ex
p
e
rt
S
y
ste
m
s
wit
h
Ap
p
li
c
a
ti
o
n
s,
IEE
E
Ac
c
e
ss
,
a
n
d
E
n
g
i
n
e
e
rin
g
A
p
p
li
c
a
ti
o
n
s
o
f
Artifi
c
ial
In
tell
ig
e
n
c
e
.
His
re
se
a
rc
h
in
tere
sts
sp
a
n
m
a
c
h
in
e
lea
rn
in
g
,
im
b
a
lan
c
e
d
lea
rn
in
g
,
d
e
e
p
lea
rn
i
n
g
,
b
u
sin
e
ss
in
telli
g
e
n
c
e
,
d
a
ta
a
n
a
ly
sis,
d
a
ta
m
in
in
g
,
a
n
d
p
a
tt
e
r
n
m
in
i
n
g
.
He
c
a
n
b
e
c
o
n
ta
c
ted
a
t
e
m
a
il
:
lc.t
u
o
n
g
@
h
u
tec
h
.
e
d
u
.
v
n
.
Evaluation Warning : The document was created with Spire.PDF for Python.