I
nte
rna
t
io
na
l J
o
urna
l o
f
E
lect
rica
l a
nd
Co
m
pu
t
er
E
ng
ineering
(
I
J
E
CE
)
Vo
l.
15
,
No
.
3
,
J
u
n
e
20
25
,
p
p
.
3396
~
3
4
0
9
I
SS
N:
2088
-
8
7
0
8
,
DOI
:
1
0
.
1
1
5
9
1
/ijece.
v
15
i
3
.
pp
3
3
9
6
-
3
4
0
9
3396
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ec
e.
ia
esco
r
e.
co
m
Artif
i
cia
l int
e
llig
ence f
o
r au
tom
a
tic
mo
dera
tion o
f
te
x
tual
co
ntent
in
o
nline
cha
ts and so
cia
l n
etworks
So
lo
m
iia
L
ia
s
k
o
v
s
k
a
1
,
Rex
B
a
ca
rr
a
2
,
Yev
hen M
a
rt
y
n
3
,
V
o
lo
dy
m
y
r
B
a
idy
ch
4
,
J
a
m
il Abeda
lra
him
J
a
m
il Al
s
a
y
a
y
deh
5
1
D
e
p
a
r
t
me
n
t
o
f
A
r
t
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
,
Lv
i
v
P
o
l
y
t
e
c
h
n
i
c
N
a
t
i
o
n
a
l
U
n
i
v
e
r
si
t
y
,
L
v
i
v
,
U
k
r
a
i
n
e
2
D
e
p
a
r
t
me
n
t
o
f
G
e
n
e
r
a
l
E
d
u
c
a
t
i
o
n
a
n
d
F
o
u
n
d
a
t
i
o
n
,
R
a
b
d
a
n
A
c
a
d
e
m
y
,
A
b
u
D
h
a
b
i
,
U
n
i
t
e
d
A
r
a
b
Em
i
r
a
t
e
s
3
D
e
p
a
r
t
me
n
t
o
f
I
n
f
o
r
mat
i
o
n
Te
c
h
n
o
l
o
g
i
e
s a
n
d
E
l
e
c
t
r
o
n
i
c
C
o
mm
u
n
i
c
a
t
i
o
n
s
S
y
st
e
ms,
Lv
i
v
S
t
a
t
e
U
n
i
v
e
r
si
t
y
o
f
Li
f
e
S
a
f
e
t
y
L
v
i
v
,
U
k
r
a
i
n
e
4
D
e
p
a
r
t
me
n
t
I
n
f
o
r
ma
t
i
o
n
Te
c
h
n
o
l
o
g
i
e
s Kh
me
l
n
y
t
sk
y
i
N
a
t
i
o
n
a
l
U
n
i
v
e
r
si
t
y
,
K
h
me
l
n
y
t
s
k
,
U
k
r
a
i
n
e
5
D
e
p
a
r
t
me
n
t
o
f
E
n
g
i
n
e
e
r
i
n
g
T
e
c
h
n
o
l
o
g
y
,
F
a
k
u
l
t
i
Te
k
n
o
l
o
g
i
D
a
n
K
e
j
u
r
u
t
e
r
a
a
n
El
e
k
t
r
o
n
i
k
D
a
n
K
o
m
p
u
t
e
r
,
U
n
i
v
e
r
s
i
t
i
Te
k
n
i
k
a
l
M
a
l
a
y
si
a
M
e
l
a
k
a
,
M
e
l
a
k
a
,
M
a
l
a
y
s
i
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Ma
r
1
1
,
2
0
2
4
R
ev
is
ed
Feb
1
7
,
2
0
2
5
Acc
ep
ted
Ma
r
5
,
2
0
2
5
Th
e
a
rti
c
le
e
x
p
lo
re
s
fu
n
d
a
m
e
n
tal
tec
h
n
iq
u
e
s
fo
r
c
o
n
v
e
rti
n
g
tex
t
in
t
o
n
u
m
e
rica
l
d
a
ta
fo
r
m
a
c
h
in
e
lea
rn
in
g
a
lg
o
rit
h
m
s.
It
m
e
ti
c
u
l
o
u
sl
y
e
x
a
m
in
e
s
v
a
rio
u
s
m
e
th
o
d
s,
in
c
l
u
d
i
n
g
wo
r
d
v
e
c
to
r
re
p
re
se
n
tati
o
n
v
ia
n
e
u
ra
l
n
e
two
r
k
s
li
k
e
Wo
r
d
2
Ve
c
,
a
n
d
e
x
p
lain
s
t
h
e
p
rin
c
ip
les
b
e
h
in
d
li
n
e
a
r
m
o
d
e
l
s
su
c
h
a
s
lo
g
isti
c
re
g
re
ss
io
n
a
n
d
s
u
p
p
o
rt
v
e
c
to
r
m
a
c
h
in
e
s.
C
o
n
v
o
lu
ti
o
n
a
l
n
e
u
ra
l
n
e
two
rk
s
(CNN
)
a
n
d
l
o
n
g
sh
o
rt
-
term
m
e
m
o
ry
(LS
TM
)
m
e
th
o
d
s
a
re
a
lso
d
isc
u
ss
e
d
,
c
o
v
e
rin
g
th
e
ir
c
o
m
p
o
n
e
n
ts,
m
e
c
h
a
n
ism
s,
a
n
d
train
in
g
p
ro
c
e
ss
e
s.
Th
e
re
se
a
rc
h
e
x
ten
d
s
to
d
e
v
e
l
o
p
i
n
g
a
n
d
tes
ti
n
g
so
ftwa
re
f
o
r
s
p
a
m
d
e
tec
ti
o
n
,
h
a
te
sp
e
e
c
h
id
e
n
t
ifi
c
a
ti
o
n
,
a
n
d
r
e
c
o
g
n
izin
g
o
ffe
n
siv
e
la
n
g
u
a
g
e
.
Us
in
g
tw
o
d
a
tas
e
ts
—
o
n
e
fo
r
lab
e
led
tex
t
m
e
ss
a
g
e
s
a
n
d
a
n
o
th
e
r
fo
r
Twit
ter
p
o
sts
—
th
e
stu
d
y
a
n
a
l
y
z
e
s
d
a
ta
to
a
d
d
re
ss
c
h
a
ll
e
n
g
e
s
li
k
e
imb
a
lan
c
e
d
d
a
ta.
A
c
o
m
p
a
ra
ti
v
e
a
n
a
l
y
sis
a
m
o
n
g
li
n
e
a
r
m
o
d
e
ls,
d
e
e
p
n
e
u
ra
l
n
e
two
rk
s,
a
n
d
sin
g
le
-
lay
e
r
m
o
d
e
ls,
u
si
n
g
p
re
-
tr
a
in
e
d
b
id
irec
ti
o
n
a
l
e
n
c
o
d
e
r
re
p
re
se
n
tatio
n
s
fro
m
tran
sfo
rm
e
rs
(BERT
)
n
e
two
rk
,
re
v
e
a
ls
p
r
o
m
isin
g
re
su
lt
s.
Th
e
c
o
n
v
o
lu
ti
o
n
a
l
n
e
u
ra
l
n
e
two
rk
sta
n
d
s
o
u
t
wit
h
a
re
m
a
rk
a
b
le
a
c
c
u
ra
c
y
o
f
0
.
9
5
.
Th
e
stu
d
y
a
lso
a
d
a
p
ts
n
e
u
ra
l
n
e
two
rk
a
rc
h
i
tec
tu
re
s
fo
r
h
a
te
s
p
e
e
c
h
a
n
d
o
ffe
n
siv
e
la
n
g
u
a
g
e
c
las
sifica
ti
o
n
.
K
ey
w
o
r
d
s
:
Ar
tific
ial
in
tellig
en
ce
L
an
g
u
ag
e
class
if
icatio
n
Ma
ch
in
e
lear
n
in
g
alg
o
r
ith
m
s
Neu
r
al
n
etwo
r
k
s
Sp
am
d
etec
tio
n
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
:
J
am
il Ab
ed
alr
ah
im
J
am
il Alsay
ay
d
eh
Dep
ar
tm
en
t o
f
E
n
g
in
ee
r
in
g
T
e
ch
n
o
lo
g
y
,
Fak
u
lti T
e
k
n
o
lo
g
i
d
an
Keju
r
u
ter
aa
n
E
lek
tr
o
n
ik
d
a
n
Ko
m
p
u
te
r
(
FTKE
K)
,
Un
iv
er
s
iti T
ek
n
ik
al
Ma
lay
s
ia
Me
lak
a
7
6
1
0
0
Me
lak
a,
Ma
lay
s
ia
E
m
ail: ja
m
il@
u
tem
.
ed
u
.
m
y
1.
I
NT
RO
D
UCT
I
O
N
Du
e
to
th
e
v
ast
v
o
lu
m
e
o
f
c
o
n
ten
t
cr
ea
ted
an
d
d
is
tr
ib
u
ted
o
n
th
e
I
n
ter
n
et,
it
h
as
b
ec
o
m
e
in
cr
ea
s
in
g
ly
ch
allen
g
in
g
f
o
r
m
o
d
er
ato
r
s
to
p
r
o
m
p
tl
y
d
etec
t
an
d
r
em
o
v
e
h
ar
m
f
u
l
o
r
o
b
jectio
n
ab
l
e
co
n
ten
t
[
1
]
–
[
3
]
.
Ad
d
itio
n
ally
,
th
e
d
y
n
am
ic
n
at
u
r
e
o
f
o
n
lin
e
co
n
ten
t
an
d
u
s
er
in
ter
ac
tio
n
s
m
ak
es
it
d
if
f
icu
lt
to
estab
lis
h
s
ta
tic
r
u
les
o
r
g
u
id
elin
es
f
o
r
m
o
d
er
ato
r
s
to
ad
h
er
e
to
.
T
h
u
s
,
th
e
u
tili
za
tio
n
o
f
ar
tific
ial
in
tellig
en
ce
f
o
r
a
u
to
m
atic
m
o
d
er
atio
n
p
r
esen
ts
a
p
er
tin
en
t
an
d
cr
u
cial
s
o
lu
tio
n
to
th
is
is
s
u
e.
Ou
r
o
b
jectiv
e
is
to
in
v
esti
g
ate
th
e
ef
f
ec
tiv
en
ess
o
f
e
m
p
lo
y
in
g
n
eu
r
al
n
etwo
r
k
s
f
o
r
r
e
g
u
latin
g
tex
t
co
n
ten
t,
s
p
ec
if
ically
in
d
etec
tin
g
s
p
am
,
h
ate
s
p
ee
ch
,
an
d
ab
u
s
iv
e
lan
g
u
ag
e
u
s
in
g
m
ac
h
in
e
lear
n
in
g
alg
o
r
ith
m
s
.
T
h
is
in
v
o
lv
es
a
n
aly
zin
g
th
e
f
ea
tu
r
es
o
f
m
ac
h
in
e
lear
n
in
g
m
eth
o
d
s
f
o
r
n
atu
r
al
lan
g
u
ag
e
p
r
o
ce
s
s
in
g
task
s
,
s
u
ch
as
tex
t
class
if
icatio
n
,
d
e
v
elo
p
in
g
a
m
o
d
el
f
o
r
id
en
tif
y
in
g
s
p
am
m
ess
ag
es
b
ased
o
n
d
ee
p
n
eu
r
al
n
etwo
r
k
s
,
co
m
p
ar
in
g
th
e
r
esu
lts
o
b
tain
ed
with
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
A
r
tifi
cia
l in
tellig
en
ce
fo
r
a
u
to
ma
tic
mo
d
era
tio
n
o
f te
xtu
a
l c
o
n
ten
t …
(
S
o
l
o
miia
Lia
s
ko
vska
)
3397
class
ical
m
ac
h
in
e
lear
n
in
g
m
eth
o
d
s
,
cr
ea
tin
g
a
te
x
t
cla
s
s
if
icatio
n
m
o
d
el
f
o
r
d
etec
ti
n
g
h
ate
s
p
ee
ch
a
n
d
o
f
f
en
s
iv
e
lan
g
u
ag
e
b
ased
o
n
th
e
tr
ain
ed
b
id
ir
ec
tio
n
al
en
c
o
d
er
r
e
p
r
esen
tatio
n
s
f
r
o
m
tr
a
n
s
f
o
r
m
er
s
(
B
E
R
T
)
tr
an
s
f
o
r
m
er
m
o
d
el,
an
d
an
aly
zin
g
th
e
ac
cu
r
ac
y
an
d
e
f
f
icien
cy
o
f
th
e
d
ev
elo
p
ed
m
o
d
el.
T
o
ac
co
m
p
lis
h
th
is
,
we
p
r
o
p
o
s
e
th
e
d
ev
elo
p
m
en
t o
f
a
s
o
f
twar
e
im
p
lem
en
tatio
n
o
f
m
ac
h
in
e
lear
n
in
g
alg
o
r
ith
m
s
d
esig
n
ed
to
d
etec
t
s
p
am
an
d
h
ate
s
p
ee
ch
in
o
n
lin
e
co
m
m
u
n
icatio
n
.
Ou
r
r
esear
ch
in
teg
r
ates
m
eth
o
d
s
an
d
alg
o
r
ith
m
s
o
f
n
atu
r
a
l
lan
g
u
ag
e
p
r
o
ce
s
s
in
g
co
m
m
o
n
l
y
u
tili
ze
d
f
o
r
tex
t c
lass
if
icatio
n
.
T
h
u
s
,
i
n
o
r
d
e
r
t
o
u
t
i
l
i
z
e
t
h
e
m
o
s
t
a
p
p
r
o
p
r
i
a
t
e
a
n
d
u
p
-
to
-
d
a
t
e
i
n
f
o
r
m
a
t
i
o
n
,
a
s
w
e
l
l
a
s
e
f
f
e
c
t
i
v
e
m
e
t
h
o
d
s
,
it
i
s
n
e
c
e
s
s
a
r
y
t
o
v
e
r
i
f
y
t
h
e
r
e
l
e
v
a
n
c
e
o
f
t
h
e
d
a
t
a
i
n
v
o
l
v
e
d
i
n
o
u
r
r
e
s
e
a
r
c
h
.
A
m
o
n
g
t
h
e
s
u
b
j
e
c
t
a
r
e
a
s
t
o
w
h
i
c
h
s
c
i
e
n
t
i
f
i
c
w
o
r
k
s
c
o
r
r
e
l
a
t
i
n
g
w
i
t
h
o
u
r
c
h
o
s
e
n
s
c
i
e
n
t
i
f
i
c
d
i
r
e
c
t
i
o
n
c
a
n
b
e
l
o
n
g
a
r
e
c
o
m
p
u
t
e
r
s
c
i
e
n
c
e
,
e
n
g
i
n
e
e
r
i
n
g
,
a
n
d
m
a
t
h
e
m
a
t
i
c
s
.
I
t
i
s
e
v
i
d
e
n
t
t
h
a
t
t
h
e
c
h
o
i
c
e
i
s
q
u
i
t
e
o
b
v
i
o
u
s
a
n
d
a
l
i
g
n
s
w
i
t
h
o
u
r
s
c
i
e
n
t
i
f
i
c
o
b
j
e
c
t
i
v
e
s
t
o
i
n
v
e
s
t
i
g
a
t
e
t
h
e
e
f
f
e
c
t
i
v
e
n
e
s
s
o
f
u
s
i
n
g
n
e
u
r
a
l
n
e
t
w
o
r
k
s
t
o
r
e
g
u
l
a
t
e
t
e
x
t
u
a
l
c
o
n
t
e
n
t
u
s
i
n
g
m
a
c
h
i
n
e
l
e
a
r
n
i
n
g
m
e
t
h
o
d
s
.
W
o
r
k
s
[
4
]
–
[
8
]
d
elv
e
in
to
th
e
m
eth
o
d
s
an
d
m
ea
n
s
o
f
d
ete
ctin
g
s
o
cial
s
p
am
,
wh
ich
en
co
m
p
ass
es
illeg
al
tex
t
co
n
ten
t
[
9
]
–
[
1
3
]
,
o
f
f
en
s
iv
e
lan
g
u
ag
e,
h
ate
s
p
ee
ch
,
cy
b
e
r
b
u
lly
in
g
,
an
d
d
is
in
f
o
r
m
atio
n
.
I
n
[
1
]
,
th
e
n
ee
d
f
o
r
th
e
d
e
v
elo
p
m
e
n
t
o
f
e
f
f
ec
tiv
e
m
eth
o
d
s
f
o
r
d
etec
tin
g
s
o
cial
s
p
am
is
em
p
h
asized
,
w
h
er
ein
th
e
s
u
p
p
o
r
t
v
ec
to
r
m
eth
o
d
(
SVM)
,
r
an
d
o
m
f
o
r
est,
an
d
n
aiv
e
B
ay
esian
alg
o
r
ith
m
a
r
e
em
p
l
o
y
ed
t
o
a
d
d
r
ess
th
is
cr
u
cial
s
o
cial
p
r
o
b
lem
.
T
h
ey
p
r
o
p
o
s
e
a
co
n
ce
p
t
f
o
r
d
ev
elo
p
in
g
m
o
r
e
ac
cu
r
ate
a
n
d
co
n
tex
t
-
d
ep
en
d
en
t
h
o
s
tile
lan
g
u
ag
e
d
etec
tio
n
s
y
s
tem
s
[
1
4
]
.
Su
c
h
m
ac
h
in
e
lear
n
in
g
a
lg
o
r
ith
m
s
as
th
e
s
u
p
p
o
r
t
v
ec
t
o
r
m
eth
o
d
,
r
an
d
o
m
f
o
r
est,
B
ay
esian
class
if
ier
,
k
-
n
ea
r
est
n
eig
h
b
o
r
’
s
m
eth
o
d
,
a
n
d
m
u
ltil
ay
er
p
er
ce
p
tr
o
n
ar
e
e
x
p
lo
r
ed
in
[
3
]
.
I
n
[
4
]
,
v
ar
io
u
s
s
p
ee
ch
m
o
d
els
-
tr
an
s
f
o
r
m
er
s
lik
e
B
E
R
T
,
XL
Net,
an
d
R
o
B
E
R
T
a
ar
e
co
m
p
ar
ed
in
th
eir
ef
f
icac
y
in
d
etec
tin
g
click
b
ait
h
ea
d
lin
es.
Ad
d
itio
n
ally
,
[
6
]
u
n
d
er
s
co
r
e
s
th
e
n
ec
ess
i
ty
f
o
r
in
ter
d
is
cip
lin
ar
y
co
o
p
er
atio
n
am
o
n
g
in
f
o
r
m
atio
n
tech
n
o
lo
g
y
p
r
o
f
ess
io
n
als,
s
o
cio
lo
g
is
ts
,
an
d
leg
al
ex
p
er
ts
to
d
ev
is
e
ef
f
ec
tiv
e
an
d
eth
ical
s
o
lu
tio
n
s
to
th
e
p
r
o
b
lem
o
f
h
a
r
m
f
u
l c
o
n
ten
t.
Var
io
u
s
wo
r
k
s
o
f
f
e
r
s
o
lu
tio
n
s
to
th
e
s
p
am
p
r
o
b
lem
.
I
n
[
5
]
,
a
m
o
d
el
b
ased
o
n
th
e
c
o
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
(
C
NN)
ar
ch
itectu
r
e
is
p
r
o
p
o
s
ed
.
R
esear
ch
[
7
]
claim
s
th
at
th
e
d
ee
p
lear
n
in
g
lo
n
g
s
h
o
r
t
-
ter
m
m
em
o
r
y
(
L
STM
)
m
o
d
el
o
u
tp
er
f
o
r
m
s
o
th
e
r
m
o
d
els
in
ter
m
s
o
f
all
m
etr
ics:
ac
cu
r
ac
y
,
r
el
iab
ilit
y
,
r
ec
all,
an
d
F1
-
s
co
r
e.
I
n
[
8
]
,
th
e
em
p
h
asis
is
o
n
th
e
p
r
o
b
lem
o
f
class
if
ic
atio
n
ac
cu
r
ac
y
o
f
em
o
tio
n
al
co
lo
r
in
g
in
tex
t
d
ata,
an
d
it is
f
o
u
n
d
th
at
th
e
b
est m
o
d
el
is
a
r
an
d
o
m
f
o
r
est,
ac
h
ie
v
in
g
an
ac
cu
r
ac
y
o
f
o
v
e
r
8
0
%.
R
esear
ch
[
1
5
]
–
[
1
9
]
u
n
d
er
s
co
r
es
th
e
im
p
o
r
tan
ce
o
f
class
if
icatio
n
ac
cu
r
ac
y
an
d
s
p
am
m
ess
ag
e
d
etec
tio
n
u
s
in
g
tr
an
s
f
o
r
m
er
m
o
d
els
an
d
e
n
s
em
b
le
lear
n
i
n
g
[
2
0
]
.
P
r
o
p
o
s
ed
tr
an
s
f
o
r
m
er
m
o
d
els,
i
n
clu
d
in
g
B
E
R
T
an
d
eXtr
em
e
Gr
ad
ien
t
B
o
o
s
tin
g
(
XGBo
o
s
t)
,
ar
e
u
tili
ze
d
f
o
r
s
p
am
class
if
icatio
n
an
d
d
etec
tio
n
.
An
o
th
er
g
r
o
u
p
o
f
wo
r
k
s
p
r
o
p
o
s
es
d
ir
ec
tio
n
s
an
d
m
eth
o
d
s
f
o
r
s
o
lv
in
g
th
e
s
p
am
p
r
o
b
lem
[
2
0
]
–
[
2
4
]
.
T
h
e
au
th
o
r
s
o
f
[
1
6
]
in
tr
o
d
u
ce
a
n
ew
a
p
p
r
o
ac
h
to
tex
t
class
if
icatio
n
b
ased
o
n
C
NN
an
d
B
id
ir
ec
tio
n
al
L
STM
m
o
d
els,
wh
ich
,
in
th
eir
o
p
in
io
n
,
b
etter
ca
p
t
u
r
e
s
em
an
tic
in
f
o
r
m
atio
n
a
n
d
d
e
m
o
n
s
tr
ate
in
c
r
ea
s
ed
ac
cu
r
ac
y
f
o
r
twee
t
clas
s
if
icatio
n
.
T
h
e
wo
r
k
[
2
5
]
s
u
g
g
ests
an
ap
p
r
o
ac
h
t
h
a
t
co
m
b
in
es
a
p
r
e
-
tr
ain
e
d
tr
an
s
f
o
r
m
er
m
o
d
el
with
a
C
NN
,
wh
ile
[
2
1
]
,
[
2
2
]
,
[
2
6
]
p
r
esen
ts
a
s
p
am
d
etec
tio
n
s
y
s
tem
in
th
e
T
witter
n
etwo
r
k
in
r
ea
l
-
tim
e
alo
n
g
s
id
e
s
en
tim
en
t
an
aly
s
is
u
s
in
g
m
ac
h
in
e
lear
n
in
g
an
d
d
ee
p
lear
n
i
n
g
m
eth
o
d
s
.
T
h
e
au
t
h
o
r
s
o
f
[
1
7
]
p
r
o
p
o
s
es
a
n
ew
ap
p
r
o
ac
h
to
im
p
r
o
v
in
g
s
p
am
d
etec
tio
n
u
s
in
g
a
d
ee
p
r
ec
u
r
r
en
t
n
eu
r
al
n
etwo
r
k
,
w
h
ile
[
1
8
]
–
[
2
0
]
p
r
esen
ts
a
b
in
ar
y
class
if
ier
b
ased
o
n
m
ac
h
in
e
lear
n
in
g
.
I
n
ar
ticle
[
9
]
,
b
ased
o
n
a
c
o
m
p
r
eh
en
s
iv
e
r
ev
i
ew
o
f
m
eth
o
d
s
an
d
ev
alu
atio
n
m
etr
ics
f
o
r
d
etec
ti
n
g
s
o
cially
u
n
ac
ce
p
ta
b
le
s
tatem
en
ts
,
it
is
co
n
clu
d
ed
th
at
f
u
tu
r
e
r
esear
ch
s
h
o
u
ld
f
o
cu
s
o
n
d
e
v
elo
p
in
g
m
o
r
e
r
eliab
le
an
d
ac
cu
r
ate
m
eth
o
d
s
c
ap
ab
le
o
f
co
p
in
g
with
th
e
d
y
n
am
ics
o
f
te
x
t
d
at
a
f
lo
ws
o
n
o
n
lin
e
p
latf
o
r
m
s
[
2
7
]
,
[
2
8
]
.
On
lin
e
c
o
m
m
u
n
icatio
n
ab
u
s
e
tak
es
m
an
y
f
o
r
m
s
–
it
ca
n
b
e
cy
b
er
b
u
lly
in
g
,
m
is
in
f
o
r
m
atio
n
,
s
p
am
,
an
d
m
o
r
e.
On
lin
e
p
r
o
p
ag
an
d
a
d
eser
v
es
p
ar
ticu
lar
atten
tio
n
–
th
r
o
u
g
h
wid
esp
r
ea
d
u
s
e
o
f
f
ak
e
ac
co
u
n
ts
o
n
s
o
cial
m
ed
ia,
v
ar
io
u
s
p
o
liti
ca
l
o
r
p
u
b
lic
f
ig
u
r
es
an
d
o
r
g
an
izatio
n
s
ca
n
d
is
s
em
in
ate
d
esire
d
in
f
o
r
m
ati
o
n
to
s
h
ap
e
p
u
b
lic
o
p
in
io
n
.
T
h
is
cr
ea
tes
a
ch
allen
g
e
f
o
r
i
n
f
o
r
m
atio
n
f
ilter
in
g
an
d
co
n
tr
o
l
[
2
9
]
–
[
3
1
]
.
T
y
p
ica
lly
,
o
wn
er
s
o
f
v
ar
io
u
s
o
n
lin
e
f
o
r
u
m
s
o
r
c
h
ats
u
s
e
p
e
o
p
le
to
m
o
n
ito
r
p
u
b
lis
h
ed
co
n
ten
t,
b
u
t
th
is
m
eth
o
d
h
as
o
b
v
io
u
s
d
r
awb
ac
k
s
–
a
p
er
s
o
n
p
h
y
s
ically
ca
n
n
o
t
r
ev
iew
th
e
co
n
ten
t
o
f
h
u
n
d
r
e
d
s
o
f
m
ess
ag
es
p
o
s
ted
in
a
s
h
o
r
t
p
er
io
d
,
esp
ec
ially
d
u
r
i
n
g
m
as
s
s
p
am
attac
k
s
o
n
u
s
er
s
.
L
ar
g
e
p
latf
o
r
m
s
s
u
ch
as
T
witter
[
2
3
]
,
[
2
6
]
o
r
Face
b
o
o
k
u
s
e
s
o
f
twar
e
to
o
ls
to
d
etec
t
co
n
ten
t
th
at
v
io
lates
p
latf
o
r
m
r
u
les,
b
u
t
th
ey
also
h
av
e
lim
itatio
n
s
–
t
h
ese
to
o
ls
o
f
ten
d
o
n
o
t
co
n
s
id
er
th
e
m
ess
ag
e
co
n
tex
t
o
r
th
e
cu
ltu
r
al
b
ac
k
g
r
o
u
n
d
o
f
its
au
th
o
r
,
th
u
s
th
ey
f
r
eq
u
en
tly
b
lo
ck
co
n
te
n
t
th
at
d
o
es
n
o
t
v
io
late
co
m
m
u
n
ity
g
u
id
elin
es
[
1
4
]
,
[
1
8
]
,
[
2
0
]
,
[
2
2
]
.
T
h
is
is
s
u
e
is
h
ig
h
ly
r
elev
an
t
f
o
r
s
o
cial
n
etwo
r
k
I
n
s
tag
r
am
.
T
h
er
ef
o
r
e,
m
o
d
e
r
n
au
to
m
ated
co
n
ten
t
m
o
d
er
atio
n
to
o
ls
ar
e
n
o
t f
lawless
,
an
d
d
esp
ite
ex
is
tin
g
s
o
lu
tio
n
s
th
at
u
s
e
ar
tific
ial
in
tellig
en
ce
to
d
etec
t
illi
cit
co
n
ten
t,
th
is
ar
ea
r
eq
u
ir
es f
u
r
th
er
r
esear
ch
[
3
2
]
–
[
3
7
]
.
2.
M
E
T
H
O
D
I
n
th
is
s
cien
tific
wo
r
k
,
a
u
t
h
o
r
s
s
o
lv
e
th
e
task
o
f
tex
t
class
if
icatio
n
u
s
in
g
m
ac
h
in
e
lear
n
in
g
alg
o
r
ith
m
s
[
3
8
]
–
[
4
0
]
.
Su
ch
al
g
o
r
ith
m
s
ar
e
n
o
t
ab
le
to
wo
r
k
d
ir
ec
tly
with
tex
t
d
ata,
f
o
r
th
is
th
ey
n
ee
d
to
b
e
co
n
v
er
ted
in
to
a
n
u
m
er
ical
f
o
r
m
at
-
v
ec
to
r
s
.
T
h
er
ef
o
r
e,
we
a
n
aly
ze
d
t
h
e
m
ain
m
eth
o
d
s
[
4
1
]
–
[
4
3
]
o
f
tex
t
d
ata
r
ep
r
esen
tatio
n
[
4
4
]
:
o
n
e
-
h
o
t
v
ec
to
r
,
b
ag
o
f
wo
r
d
s
,
ter
m
f
r
eq
u
en
cy
–
in
v
er
s
e
d
o
cu
m
en
t
f
r
eq
u
en
cy
(
T
F
-
I
DF)
,
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
3
,
J
u
n
e
20
25
:
3
3
9
6
-
3
4
0
9
3398
n
-
g
r
am
s
,
as
well
as
v
ec
to
r
r
ep
r
esen
tatio
n
o
f
wo
r
d
s
an
d
i
ts
im
p
lem
en
tatio
n
th
r
o
u
g
h
t
h
e
wo
r
d
2
v
ec
m
o
d
el
[
4
5
]
–
[
4
8
]
.
I
n
p
ar
ticu
lar
,
two
m
o
d
el
ar
c
h
itectu
r
es
ar
e
in
v
o
lv
ed
in
th
e
r
esear
ch
[
4
9
]
–
[
5
2
]
:
C
o
n
tin
u
o
u
s
b
ag
o
f
wo
r
d
s
(
C
B
OW
)
an
d
Sk
ip
-
g
r
a
m
,
lin
ea
r
m
ac
h
in
e
lear
n
in
g
al
g
o
r
ith
m
s
an
d
n
eu
r
al
n
etwo
r
k
s
.
T
h
er
e
ar
e
s
ev
e
r
al
m
eth
o
d
s
,
th
e
m
o
s
t
co
m
m
o
n
o
f
wh
ich
is
“b
ag
o
f
wo
r
d
s
.
”
T
h
e
b
ag
o
f
wo
r
d
s
m
ar
k
s
th
e
p
r
esen
ce
o
f
a
wo
r
d
in
in
p
u
t
d
o
cu
m
e
n
ts
co
m
p
ar
ed
to
all
wo
r
d
s
in
t
h
e
d
ataset.
T
h
er
ef
o
r
e,
its
im
p
lem
en
tatio
n
r
eq
u
ir
es
a
d
ictio
n
a
r
y
o
f
all
u
s
ed
wo
r
d
s
an
d
a
n
i
n
d
icat
o
r
o
f
w
o
r
d
p
r
esen
ce
[
5
3
]
,
[
5
4
]
.
All
d
ata
in
p
u
tted
in
to
m
ac
h
in
e
lear
n
in
g
m
o
d
els
will th
u
s
b
e
r
ep
r
esen
ted
as n
u
m
er
ical
v
ec
to
r
s
(
1
)
:
[
1
,
2
,
3
…
]
,
(
1
)
T
h
e
p
r
e
v
io
u
s
m
eth
o
d
ca
n
b
e
i
m
p
r
o
v
e
d
b
y
r
ep
r
esen
tin
g
ea
ch
wo
r
d
in
t
h
e
v
ec
to
r
n
o
t
ju
s
t
as
0
o
r
1
,
b
u
t
r
ath
er
b
y
its
co
u
n
t
in
th
e
d
o
cu
m
en
t
o
r
its
f
r
eq
u
e
n
cy
r
elativ
e
to
t
h
e
to
tal
n
u
m
b
er
o
f
wo
r
d
s
in
th
e
tex
t.
T
h
e
m
ai
n
d
r
awb
ac
k
o
f
th
is
ap
p
r
o
ac
h
is
th
at
wo
r
d
s
ap
p
ea
r
in
g
in
ev
er
y
d
o
cu
m
e
n
t
will
h
av
e
th
e
h
ig
h
est
f
r
eq
u
en
cy
an
d
cr
ea
te
in
f
o
r
m
atio
n
al
n
o
is
e.
T
o
ad
d
r
ess
th
is
is
s
u
e,
ter
m
f
r
eq
u
en
cy
–
in
v
er
s
e
d
o
cu
m
e
n
t
f
r
eq
u
en
cy
(
T
F
-
I
DF)
ex
is
ts
–
a
m
etr
ic
t
h
at
d
eter
m
i
n
es
th
e
s
ig
n
if
ican
ce
o
f
a
wo
r
d
f
o
r
a
s
p
ec
if
ic
d
o
c
u
m
en
t
ag
a
in
s
t
its
s
ig
n
if
ican
ce
f
o
r
th
e
e
n
tire
co
r
p
u
s
.
I
t
is
lo
g
i
ca
l
to
ass
u
m
e
th
at
a
wo
r
d
a
p
p
ea
r
in
g
in
all
in
p
u
t
d
ata
will
h
a
v
e
a
lo
w
v
alu
e
f
o
r
a
s
p
ec
if
ic
d
o
cu
m
en
t,
wh
er
ea
s
a
wo
r
d
ap
p
ea
r
i
n
g
in
o
n
l
y
o
n
e
d
o
cu
m
e
n
t
will
b
etter
d
escr
ib
e
it.
T
F
-
I
DF
i
s
ca
lcu
lated
f
o
r
ea
ch
wo
r
d
,
a
n
d
th
e
h
ig
h
e
r
th
e
v
alu
e
o
f
th
e
m
etr
ic,
th
e
m
o
r
e
s
ig
n
i
f
ican
t
th
e
wo
r
d
is
f
o
r
th
e
d
o
cu
m
e
n
t.
T
h
e
f
o
r
m
u
la
f
o
r
th
e
m
etr
ic
is
as (
2
)
:
−
=
∗
l
og
(
)
(
2
)
–
ter
m
f
r
eq
u
e
n
cy
,
–
to
tal
n
u
m
b
er
o
f
d
o
cu
m
en
ts
,
–
n
u
m
b
er
o
f
d
o
cu
m
en
ts
co
n
tain
in
g
th
e
wo
r
d
.
T
h
e
ess
en
ce
o
f
o
u
r
r
esear
ch
is
th
e
u
s
e
o
f
m
ac
h
i
n
e
lear
n
in
g
alg
o
r
ith
m
s
to
d
etec
t sp
am
.
Den
s
e
v
ec
to
r
s
o
r
co
n
tex
t
v
ec
t
o
r
s
ar
e
v
ec
to
r
s
u
s
ed
to
d
escr
ib
e
a
wo
r
d
b
ased
o
n
its
r
elatio
n
s
h
ip
s
with
o
th
er
wo
r
d
s
.
Giv
en
a
s
en
ten
ce
,
we
ca
n
tak
e
a
s
p
ec
if
ic
win
d
o
w
ar
o
u
n
d
th
e
c
h
o
s
en
wo
r
d
with
a
s
ize
o
f
n
wo
r
d
s
to
r
ep
r
esen
t
its
co
n
tex
t.
W
o
r
d
s
th
at
h
a
v
e
s
im
ilar
c
o
n
tex
ts
–
m
ea
n
in
g
th
ey
s
h
ar
e
th
e
s
am
e
s
u
r
r
o
u
n
d
in
g
wo
r
d
s
as
wo
r
d
x
,
will
b
e
co
n
s
id
er
e
d
s
y
n
o
n
y
m
s
o
r
s
em
an
tically
s
im
ilar
to
wo
r
d
y
.
T
h
en
,
f
o
r
t
h
e
c
h
o
s
en
wo
r
d
,
we
ca
n
f
o
r
m
a
v
ec
t
o
r
[
1
,
2
,
3
…
]
wh
er
e
ea
c
h
v
a
r
iab
le
r
ep
r
esen
ts
th
e
f
r
eq
u
e
n
cy
o
f
ea
ch
wo
r
d
'
s
o
cc
u
r
r
en
ce
in
t
h
e
co
r
p
u
s
with
in
th
e
v
icin
ity
o
f
th
e
ch
o
s
en
wo
r
d
.
Sin
ce
wo
r
d
s
ar
e
r
ep
r
esen
ted
as
v
ec
to
r
s
,
we
ca
n
m
ea
s
u
r
e
th
e
s
im
ilar
ity
b
etwe
en
wo
r
d
s
u
s
in
g
th
e
f
o
r
m
u
la
o
f
th
e
d
o
t p
r
o
d
u
ct,
s
p
ec
if
ically
f
in
d
in
g
th
e
co
s
in
e
s
im
ilar
ity
(
3
)
:
c
os
=
∗
|
|
∗
|
|
(
3
)
wh
er
e
a
an
d
b
a
r
e
v
ec
to
r
r
ep
r
esen
tatio
n
s
o
f
wo
r
d
s
.
W
o
r
d
2
v
ec
is
a
two
-
lay
er
n
e
u
r
al
n
et
wo
r
k
th
at
p
r
o
ce
s
s
es
tex
t
b
y
“v
ec
to
r
izin
g
”
wo
r
d
s
.
I
t
tak
es
a
tex
tu
al
co
r
p
u
s
as
in
p
u
t
an
d
p
r
o
d
u
ce
s
a
s
et
o
f
d
en
s
e
v
ec
t
o
r
s
r
ep
r
esen
tin
g
wo
r
d
s
in
th
at
c
o
r
p
u
s
.
T
h
er
e
a
r
e
two
m
ain
ar
ch
it
ec
tu
r
es: C
B
O
W
an
d
s
k
ip
-
g
r
a
m
.
Au
th
o
r
s
u
s
ed
Py
th
o
n
p
r
o
g
r
a
m
m
in
g
lan
g
u
ag
e,
lib
r
a
r
ies
f
o
r
m
ac
h
in
e
lear
n
in
g
,
n
atu
r
al
lan
g
u
ag
e
p
r
o
ce
s
s
in
g
an
d
d
ata
v
is
u
aliza
tio
n
NL
T
K
,
s
k
lear
n
,
m
atp
l
o
tlib
.
p
y
p
l
o
t,
s
ea
b
o
r
n
,
n
ea
ttex
t
as
to
o
ls
f
o
r
r
esear
ch
.
As
a
d
ev
elo
p
m
en
t
e
n
v
ir
o
n
m
e
n
t,
Go
o
g
le
C
o
lab
was
u
s
ed
a
n
in
ter
ac
tiv
e
o
n
lin
e
en
v
i
r
o
n
m
en
t
f
o
r
p
er
f
o
r
m
in
g
d
ata
an
aly
s
is
an
d
v
is
u
aliza
tio
n
task
s
,
wh
ich
allo
ws
y
o
u
to
b
r
ea
k
th
e
c
o
d
e
i
n
to
s
ep
ar
ate
p
ar
ts
,
r
u
n
th
em
in
d
ep
en
d
en
tly
o
f
ea
ch
o
t
h
er
,
v
is
u
alize
s
th
e
p
r
o
ce
s
s
o
f
co
d
e
e
x
ec
u
tio
n
in
r
ea
l tim
e
an
d
g
iv
e
s
th
e
o
p
p
o
r
tu
n
ity
to
im
m
ed
iately
s
ee
th
e
r
esu
lt
ex
ec
u
tio
n
o
f
th
e
d
esire
d
p
a
r
t
o
f
th
e
p
r
o
g
r
am
,
wh
ich
g
r
ea
tly
s
im
p
lifie
s
th
eir
wr
itin
g
an
d
d
e
b
u
g
g
i
n
g
.
2
.
1
.
Cla
s
s
if
ier
o
f
lin
ea
r
m
o
d
els,
L
S
T
M
,
s
pa
m
ba
s
ed
CN
N
a
nd
B
E
RT
2
.
1
.
1
.
Da
t
a
a
na
ly
s
is
a
nd
pre
-
pro
ce
s
s
ing
T
h
e
d
ataset
u
s
ed
f
o
r
m
o
d
el
tr
ain
in
g
c
o
n
s
is
ts
o
f
5
,
5
7
4
tex
t
m
ess
ag
es,
wh
ich
ar
e
lab
eled
a
s
s
p
am
an
d
non
-
s
p
am
.
Fig
u
r
e
1
p
r
esen
ts
an
o
v
er
v
iew
o
f
th
e
d
ataset,
in
clu
d
in
g
g
en
e
r
al
s
tatis
tic
s
s
u
ch
as
wo
r
d
f
r
eq
u
en
cy
an
d
class
d
is
tr
ib
u
tio
n
.
T
h
is
v
i
s
u
aliza
tio
n
h
elp
s
to
u
n
d
er
s
tan
d
th
e
n
atu
r
e
o
f
th
e
d
ata
an
d
its
b
alan
ce
,
wh
ich
is
cr
itical
f
o
r
tr
ain
in
g
ef
f
ec
tiv
e
cl
ass
if
icatio
n
m
o
d
els.
Sto
p
-
wo
r
d
s
ar
e
wo
r
d
s
th
at
ar
e
p
r
esen
t
in
th
e
tex
t,
b
u
t
b
y
th
em
s
elv
es
d
o
n
o
t
m
ak
e
s
en
s
e,
s
u
ch
as
co
n
ju
n
ctio
n
s
,
p
r
ep
o
s
itio
n
s
,
o
t
h
er
o
f
f
icial
p
ar
ts
o
f
s
p
ee
ch
,
a
n
d
ex
clam
atio
n
s
.
Als
o
,
s
to
p
wo
r
d
s
u
s
u
ally
in
clu
d
e
wo
r
d
s
th
at
ar
e
f
o
u
n
d
in
alm
o
s
t
all
co
r
p
o
r
a
o
f
a
ce
r
tain
lan
g
u
ag
e.
B
y
th
r
o
win
g
th
em
o
u
t
,
y
o
u
ca
n
g
et
r
id
o
f
u
n
n
ec
ess
ar
y
n
o
is
e
an
d
g
i
v
e
m
o
r
e
weig
h
t
to
wo
r
d
s
t
h
at
ar
e
m
o
r
e
im
p
o
r
tan
t
an
d
h
a
v
e
a
s
ig
n
if
ican
t
im
p
ac
t
o
n
th
e
co
n
ten
t
o
f
th
e
d
o
cu
m
e
n
t.
T
h
e
NL
T
K
lib
r
ar
y
co
n
tain
s
a
b
u
ilt
-
in
lis
t o
f
s
to
p
wo
r
d
s
f
o
r
ea
ch
lan
g
u
a
g
e.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
A
r
tifi
cia
l in
tellig
en
ce
fo
r
a
u
to
ma
tic
mo
d
era
tio
n
o
f te
xtu
a
l c
o
n
ten
t …
(
S
o
l
o
miia
Lia
s
ko
vska
)
3399
Fig
u
r
e
1
.
Descr
ip
tio
n
an
d
v
is
u
aliza
tio
n
o
f
th
e
d
ata
s
et
Sin
ce
th
e
s
p
ec
if
ic
m
ea
n
in
g
o
f
wo
r
d
s
is
r
elativ
ely
u
n
im
p
o
r
ta
n
t
f
o
r
s
p
am
d
etec
tio
n
,
s
tem
m
i
n
g
ca
n
b
e
u
s
ed
.
I
t
is
m
u
ch
f
aster
an
d
ea
s
ier
to
im
p
lem
en
t.
T
h
er
e
ar
e
s
ev
er
al
s
tem
m
in
g
im
p
lem
en
ta
tio
n
s
in
th
e
NL
T
K
lib
r
ar
y
.
T
h
e
tar
g
et
v
ar
iab
le
in
th
e
d
ataset
tak
es
t
wo
s
tr
in
g
v
alu
es.
Sin
ce
th
e
m
o
d
els
ca
n
o
n
ly
wo
r
k
d
ir
ec
tly
with
n
u
m
er
ical
d
ata,
we
e
n
c
o
d
e
th
e
v
al
u
e
o
f
t
h
e
tar
g
et
v
ar
iab
le
ac
co
r
d
in
g
to
th
e
b
i
n
ar
y
class
if
icatio
n
p
r
o
b
lem
.
Usi
n
g
th
e
b
u
ilt
-
in
tr
a
in
_
test
_
s
p
lit()
f
u
n
ctio
n
o
f
t
h
e
s
k
lear
n
lib
r
a
r
y
,
we
s
p
lit
th
e
d
ata
s
et
in
to
tr
ain
in
g
an
d
test
s
am
p
les.
T
o
k
e
n
izatio
n
is
th
e
p
r
o
ce
s
s
o
f
d
iv
id
in
g
a
d
o
c
u
m
en
t
in
to
wo
r
d
co
m
p
o
n
en
ts
-
to
k
en
s
,
a
f
te
r
to
k
en
izatio
n
,
we
will
co
n
v
er
t
d
o
cu
m
en
ts
in
to
n
u
m
er
ical
v
ec
to
r
s
u
s
in
g
th
e
m
eth
o
d
s
o
f
“b
ag
o
f
wo
r
d
s
”,
n
-
g
r
am
s
an
d
TF
-
I
DF
.
T
o
d
o
th
i
s
,
we
will
u
s
e
th
e
s
k
lear
n
lib
r
ar
y
p
ac
k
ag
e
f
o
r
e
x
tr
ac
tin
g
f
ea
tu
r
es
f
r
o
m
tex
t
d
ata
an
d
class
es
f
o
r
v
ec
to
r
izatio
n
.
L
et's
in
v
o
lv
e
in
s
tan
ce
s
o
f
th
e
C
o
u
n
tVec
to
r
izer
an
d
T
f
id
f
T
r
a
n
s
f
o
r
m
er
class
es
to
cr
ea
te
n
ew
d
atasets
f
o
r
ea
c
h
f
ea
tu
r
e
ex
tr
ac
tio
n
m
et
h
o
d
.
Fo
r
ea
ch
o
f
th
e
d
ata
s
ets,
we
t
r
ain
two
lin
ea
r
m
o
d
els:
lo
g
is
tic
r
eg
r
ess
io
n
an
d
th
e
s
u
p
p
o
r
t v
ec
to
r
m
eth
o
d
.
W
e
will u
s
e
th
e
f
o
llo
win
g
m
etr
ics to
e
v
alu
ate
th
e
m
o
d
els:
−
Acc
u
r
ac
y
-
s
co
r
e
–
th
e
r
atio
o
f
th
e
n
u
m
b
er
o
f
c
o
r
r
ec
tly
p
r
e
d
icted
class
es
to
th
e
n
u
m
b
er
o
f
all
p
r
e
d
icted
d
ata,
ch
ar
ac
ter
izes th
e
ac
c
u
r
ac
y
o
f
th
e
m
o
d
el;
−
Pre
cisi
o
n
-
s
co
r
e
–
th
e
r
atio
o
f
th
e
n
u
m
b
er
o
f
c
o
r
r
ec
tly
p
r
ed
i
cted
p
o
s
itiv
e
(
y
=1
)
d
ata
to
th
e
n
u
m
b
er
o
f
all
p
r
ed
icted
p
o
s
itiv
e
d
ata,
wh
ic
h
ch
ar
ac
ter
izes
th
e
er
r
o
r
with
wh
ich
th
e
m
o
d
el
ca
n
ac
c
ep
t
d
ata
m
ar
k
ed
as
n
eg
ativ
e
as p
o
s
itiv
e;
−
R
ec
all
–
th
e
r
atio
o
f
t
h
e
n
u
m
b
er
o
f
c
o
r
r
ec
tly
p
r
ed
icted
p
o
s
itiv
e
d
ata
to
th
e
s
u
m
o
f
th
e
n
u
m
b
er
o
f
co
r
r
ec
tly
p
r
ed
icted
p
o
s
itiv
e
an
d
f
alsely
p
r
ed
icted
n
eg
ativ
e
d
ata,
c
h
ar
ac
ter
izes
th
e
m
o
d
el'
s
ab
il
i
ty
to
d
eter
m
in
e
p
o
s
itiv
e
d
ata;
−
F1
-
s
co
r
e
–
a
m
etr
ic
u
s
ed
to
ca
lcu
late
th
e
r
atio
o
f
th
e
p
r
o
p
o
r
tio
n
o
f
o
b
jects
th
at
wer
e
class
if
ied
b
y
th
e
m
o
d
el
as
p
o
s
itiv
e
an
d
r
ea
lly
wer
e
p
o
s
itiv
e
to
th
e
p
r
o
p
o
r
tio
n
o
f
f
o
u
n
d
p
o
s
itiv
e
d
ata
f
r
o
m
all
p
o
s
itiv
e
d
ata
in
th
e
s
et,
ca
lcu
lated
b
y
th
e
f
o
r
m
u
la:
=
2
+
1
∗
∗
2
∗
wh
er
e
β
–
is
th
e
weig
h
t f
o
r
m
e
tr
ics.
Af
ter
tr
ain
in
g
th
e
m
o
d
el
a
n
d
t
esti
n
g
th
e
m
o
d
el,
t
h
e
f
o
llo
win
g
m
etr
ics we
r
e
o
b
tain
e
d
:
−
Acc
u
r
ac
y
-
s
co
r
e=
0
.
9
5
2
,
p
r
ec
is
io
n
-
s
co
r
e=
0
.
9
7
,
r
ec
all=0
.
9
3
,
f
1
-
p
o
in
t=0
.
9
5
f
o
r
lo
g
is
tic
r
eg
r
ess
io
n
tr
ain
ed
o
n
“b
ag
o
f
wo
r
d
s
”;
−
Acc
u
r
ac
y
-
s
co
r
e=
0
.
9
4
,
p
r
ec
is
io
n
-
s
co
r
e=
0
.
9
6
4
,
r
ec
all=0
.
9
1
,
f
1
-
p
o
in
t=0
.
9
3
6
f
o
r
th
e
m
et
h
o
d
o
f
s
u
p
p
o
r
t
v
ec
to
r
s
tr
ain
ed
o
n
th
e
“
b
ag
o
f
wo
r
d
s
”;
−
Acc
u
r
ac
y
-
s
co
r
e=
0
.
9
4
,
p
r
ec
is
io
n
-
s
co
r
e=
0
.
9
8
8
r
ec
all=0
.
8
9
,
f
1
-
s
co
r
e=
0
.
9
3
f
o
r
lo
g
is
tic
r
eg
r
ess
io
n
tr
ain
ed
o
n
th
e
“b
ag
o
f
u
n
ig
r
am
a
n
d
b
ig
r
a
m
”;
−
Acc
u
r
ac
y
-
s
co
r
e=
0
.
9
4
7
,
p
r
ec
i
s
io
n
-
s
co
r
e=
0
.
9
8
8
,
r
ec
all=0
.
9
,
f
1
-
s
co
r
e=
0
.
9
4
f
o
r
th
e
m
eth
o
d
o
f
s
u
p
p
o
r
t
v
ec
to
r
s
tr
ain
ed
o
n
th
e
“
b
ag
o
f
u
n
ig
r
am
a
n
d
b
ig
r
am
”;
−
Acc
u
r
ac
y
-
s
co
r
e=
0
.
9
5
,
p
r
ec
is
io
n
-
s
co
r
e=
0
.
9
7
6
,
r
ec
all=0
.
9
2
2
,
f
1
-
s
co
r
e=
0
.
9
5
f
o
r
lo
g
is
tic
r
e
g
r
ess
io
n
tr
ain
ed
on
TF
-
I
DF
v
ec
to
r
s
;
−
Acc
u
r
ac
y
-
s
co
r
e=
0
.
9
6
7
9
1
4
4
3
8
5
0
2
6
7
3
8
,
p
r
ec
is
io
n
-
s
co
r
e=
0
.
9
8
,
r
ec
all=0
.
9
5
5
,
f
1
-
s
co
r
e=
0
.
9
6
6
f
o
r
th
e
s
u
p
p
o
r
t v
ec
t
o
r
m
eth
o
d
tr
ai
n
ed
o
n
TF
-
I
DF
v
ec
to
r
s
.
T
o
v
is
u
alize
th
e
q
u
ality
o
f
t
h
e
m
o
d
els,
we
will
o
u
tp
u
t
t
h
e
er
r
o
r
m
atr
ix
f
o
r
ea
ch
d
ata
s
e
t
f
o
r
ea
ch
m
o
d
el.
Fig
u
r
e
2
d
em
o
n
s
tr
ates
wh
er
e
Fig
u
r
e
2
(
a)
s
h
o
ws
er
r
o
r
m
atr
ices
f
o
r
lo
g
is
tic
r
eg
r
ess
i
o
n
an
d
Fig
u
r
e
2
(
b
)
s
h
o
ws
th
e
s
u
p
p
o
r
t
v
ec
to
r
m
et
h
o
d
th
at
we
ca
n
s
ee
th
at
b
o
th
m
o
d
els
f
o
r
b
ag
-
of
-
w
o
r
d
s
d
ata
ar
e
eq
u
ally
g
o
o
d
at
id
en
tify
in
g
n
o
n
-
s
p
am
m
ess
ag
e
s
.
B
u
t th
e
lin
ea
r
r
eg
r
ess
io
n
m
e
th
o
d
is
b
etter
at
d
ir
ec
tl
y
class
if
y
in
g
s
p
am
its
elf
.
Fig
u
r
e
3
d
em
o
n
s
tr
ates
th
at
th
e
ab
o
v
e
m
atr
ices
an
d
we
ca
n
co
n
clu
d
e
th
at
f
o
r
d
ata
s
ets
c
o
n
tain
in
g
u
n
ig
r
am
a
n
d
b
i
g
r
am
.
Fig
u
r
e
3
(
a)
s
h
o
ws
er
r
o
r
m
atr
ices
f
o
r
lo
g
is
tic
r
eg
r
ess
io
n
(
lef
t)
.
Fig
u
r
e
3
(
b
)
s
h
o
ws
th
e
s
u
p
p
o
r
t v
ec
t
o
r
m
eth
o
d
th
e
m
o
d
els g
iv
e
alm
o
s
t id
en
tical
r
esu
lts
,
b
u
t in
tu
r
n
p
r
ed
ict
less
f
alse p
o
s
itiv
e
d
ata.
W
e
ca
n
co
n
clu
d
e
th
at
th
e
d
eter
m
in
atio
n
o
f
th
e
m
ess
ag
e
lab
e
l
as
s
p
am
o
r
n
o
t
s
p
am
d
o
es
n
o
t
im
p
r
o
v
e
m
u
ch
wh
e
n
tak
i
n
g
in
t
o
ac
co
u
n
t
th
e
a
d
d
itio
n
al
c
o
n
tex
t.
B
ec
au
s
e
wh
en
u
s
in
g
th
e
n
-
g
r
am
s
d
ataset,
th
e
o
v
e
r
all
ac
cu
r
ac
y
o
f
th
e
m
o
d
el
d
id
n
o
t
im
p
r
o
v
e,
it
ev
en
d
ec
r
ea
s
ed
s
lig
h
tly
,
b
u
t
at
th
e
s
am
e
tim
e
th
e
m
o
d
el
d
o
es
less
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
3
,
J
u
n
e
20
25
:
3
3
9
6
-
3
4
0
9
3400
s
p
am
d
etec
tio
n
er
r
o
r
s
.
Fewer
f
alse
-
p
o
s
itiv
e
d
ata
a
n
d
th
e
lo
west
ac
cu
r
ac
y
a
r
e
d
em
o
n
s
tr
at
ed
b
y
m
o
d
els
tr
ain
e
d
o
n
d
ata
in
th
e
TF
-
I
DF
in
d
icato
r
f
o
r
m
at.
Fo
r
T
F
-
I
DF
f
o
r
m
at
d
ata
in
F
ig
u
r
e
4
d
em
o
n
s
tr
ates
th
at
th
e
s
u
p
p
o
r
t
v
ec
to
r
m
et
h
o
d
b
etter
class
if
ie
s
m
ess
ag
es
co
n
tain
in
g
s
p
am
,
wh
ile
allo
win
g
f
ewe
r
er
r
o
r
s
th
an
lo
g
is
tic
r
eg
r
ess
io
n
.
Fig
u
r
e
4
(
a)
s
h
o
ws
er
r
o
r
m
atr
ices f
o
r
lo
g
is
tic
r
eg
r
ess
io
n
.
Fig
u
r
e
4
(
b
)
s
h
o
ws th
e
s
u
p
p
o
r
t v
ec
to
r
m
eth
o
d
.
(
a)
(
b
)
Fig
u
r
e
2
.
E
r
r
o
r
m
atr
ices f
o
r
l
o
g
is
tic
r
eg
r
ess
io
n
(
a)
an
d
th
e
s
u
p
p
o
r
t
v
ec
to
r
m
et
h
o
d
a
n
d
(
b
)
f
o
r
th
e
“
b
ag
o
f
wo
r
d
s
”
d
at
a
s
et
(
a)
(
b
)
Fig
u
r
e
3
.
E
r
r
o
r
m
atr
ices f
o
r
l
o
g
is
tic
r
eg
r
ess
io
n
(
lef
t)
(
a)
an
d
th
e
s
u
p
p
o
r
t
v
ec
to
r
m
eth
o
d
an
d
(
b
)
(
r
ig
h
t)
f
o
r
t
h
e
“b
ag
o
f
wo
r
d
s
”
d
ata
s
et
(
a)
(
b
)
Fig
u
r
e
4
.
E
r
r
o
r
m
atr
ices f
o
r
l
o
g
is
tic
r
eg
r
ess
io
n
(
a)
an
d
th
e
s
u
p
p
o
r
t
v
ec
to
r
m
et
h
o
d
(
b
)
f
o
r
t
h
e
TF
-
I
DF
d
ataset
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
A
r
tifi
cia
l in
tellig
en
ce
fo
r
a
u
to
ma
tic
mo
d
era
tio
n
o
f te
xtu
a
l c
o
n
ten
t …
(
S
o
l
o
miia
Lia
s
ko
vska
)
3401
3.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
T
h
is
s
ec
tio
n
d
escr
ib
es
u
s
in
g
C
NN
an
d
L
STM
n
eu
r
al
n
etwo
r
k
ar
c
h
itectu
r
es
f
o
r
s
p
am
cla
s
s
if
icatio
n
.
I
n
o
r
d
er
to
tr
ain
n
e
u
r
al
n
etwo
r
k
s
o
n
o
u
r
d
ata,
it
is
n
ec
ess
ar
y
to
b
r
in
g
all
v
ec
to
r
s
,
th
at
is
,
all
d
o
cu
m
en
ts
,
to
a
f
ix
ed
len
g
th
.
As
th
e
d
im
e
n
s
io
n
v
alu
e
o
f
th
e
v
ec
to
r
s
,
a
u
th
o
r
s
tak
e
th
e
len
g
th
o
f
th
e
lar
g
est
d
o
cu
m
e
n
t.
Au
th
o
r
s
ch
an
g
e
th
e
d
im
en
s
io
n
ality
o
f
th
e
v
ec
to
r
s
u
s
in
g
th
e
p
ad
_
s
eq
u
en
ce
s
(
)
f
u
n
ctio
n
o
f
th
e
K
er
as
lib
r
ar
y
,
th
e
p
ad
d
in
g
=
ar
g
u
m
e
n
t
will
h
av
e
th
e
v
alu
e
“p
o
s
t”,
wh
ich
in
d
ica
tes
th
e
ze
r
o
v
alu
es
o
f
th
e
f
u
n
c
tio
n
,
d
u
e
to
wh
ich
we
ex
p
an
d
t
h
e
v
ec
to
r
an
d
ad
d
th
em
to
th
e
e
n
d
.
3
.
1
.
Descript
io
n o
f
t
he
net
w
o
rk
a
r
chit
ec
t
ure
E
m
b
ed
d
in
g
lay
er
:
in
p
u
t
d
ata
to
th
e
n
etwo
r
k
is
a
s
eq
u
e
n
ce
o
f
wo
r
d
s
,
wh
ich
ar
e
r
ep
r
esen
ted
as
in
teg
er
s
wo
r
d
in
d
e
x
es
in
th
e
d
ictio
n
ar
y
.
T
h
e
E
m
b
ed
d
in
g
L
a
y
er
tr
a
n
s
f
o
r
m
s
th
ese
i
n
teg
er
s
i
n
to
v
ec
t
o
r
s
o
f
g
iv
e
n
d
im
en
s
io
n
c
o
n
tain
in
g
th
e
r
e
p
r
esen
tatio
n
o
f
th
e
wo
r
d
th
r
o
u
g
h
its
co
n
tex
tu
al
r
elatio
n
s
h
ip
w
ith
o
th
er
wo
r
d
s
.
8
0
is
th
e
s
ize
o
f
th
e
len
g
th
o
f
th
e
in
p
u
t v
ec
to
r
o
f
to
k
en
s
,
a
n
d
th
e
d
im
en
s
io
n
o
f
d
e
n
s
e
v
ec
to
r
s
is
1
0
0
;
−
L
STM
:
A
lay
er
o
f
an
L
STM
n
etwo
r
k
th
at
ca
n
s
to
r
e
lo
n
g
-
ter
m
d
ep
en
d
e
n
cies
in
s
eq
u
en
tial
d
ata.
I
t
co
n
s
is
ts
o
f
n
eu
r
o
n
s
f
o
r
p
r
o
ce
s
s
in
g
s
eq
u
en
tial
in
p
u
t
d
ata
an
d
s
av
i
n
g
in
f
o
r
m
atio
n
ab
o
u
t
th
e
s
tate
o
f
th
e
n
eu
r
al
n
etwo
r
k
;
−
Glo
b
alM
ax
Po
o
lin
g
1
D:
a
lay
er
ac
tin
g
as
a
f
ilter
f
o
r
f
ea
t
u
r
es
g
en
er
ated
b
y
L
STM
,
it
s
o
u
tp
u
t
is
th
e
m
ax
im
u
m
v
alu
e
f
r
o
m
ea
ch
v
e
cto
r
o
f
f
ea
tu
r
es;
−
Dr
o
p
o
u
t
an
d
b
atch
n
o
r
m
aliza
t
io
n
:
a
d
r
o
p
o
u
t
lay
e
r
is
ap
p
lied
af
ter
th
e
L
STM
lay
er
t
o
f
ilter
th
e
n
u
m
b
er
o
f
f
ir
in
g
n
eu
r
o
n
s
to
p
r
ev
en
t
o
v
er
tr
ain
in
g
.
Af
ter
th
at,
b
atch
n
o
r
m
aliza
tio
n
is
ap
p
lied
to
s
tan
d
ar
d
ize
th
e
in
p
u
t
d
ata
to
th
e
p
r
ev
io
u
s
lay
e
r
;
−
Den
s
e:
a
f
u
lly
c
o
n
n
ec
ted
lay
e
r
ac
ce
p
ts
L
STM
o
u
t
p
u
t
d
ata
a
f
ter
p
r
o
ce
s
s
in
g
b
y
s
ev
er
al
lay
er
s
,
co
n
tain
s
8
0
n
eu
r
o
n
s
,
f
o
r
n
o
n
lin
ea
r
tr
a
n
s
f
o
r
m
atio
n
o
f
th
e
in
p
u
t
d
ata
b
y
th
e
r
ec
tifie
d
lin
ea
r
u
n
it
(
R
eL
U
)
ac
tiv
atio
n
f
u
n
ctio
n
;
−
Dr
o
p
o
u
t: r
e
p
ea
ted
r
e
m
o
v
al
o
f
a
p
ar
t o
f
n
eu
r
o
n
s
;
−
Den
s
e:
an
o
u
tp
u
t
d
e
n
s
e
lay
er
with
o
n
e
n
eu
r
o
n
an
d
a
s
ig
m
o
id
ac
tiv
atio
n
f
u
n
ctio
n
,
u
s
ed
to
ca
lcu
late
th
e
o
u
tp
u
t
p
r
o
b
a
b
ilit
y
th
at
an
o
b
je
ct
b
elo
n
g
s
to
a
class
.
Fig
u
r
e
5
p
r
esen
ts
th
e
p
er
f
o
r
m
an
ce
o
f
th
e
L
STM
-
b
ased
n
etwo
r
k
.
Fig
u
r
e
5
(
a)
s
h
o
ws
th
e
ac
cu
r
ac
y
cu
r
v
e
f
o
r
b
o
th
tr
ain
i
n
g
an
d
v
alid
atio
n
s
ets,
in
d
icatin
g
co
n
s
is
ten
t
im
p
r
o
v
em
en
t
an
d
g
o
o
d
g
en
er
aliza
tio
n
.
Fig
u
r
e
5
(
b
)
illu
s
tr
ates
th
e
lo
s
s
cu
r
v
e,
wh
ich
s
tead
ily
d
ec
r
e
ases
,
s
u
g
g
esti
n
g
th
at
th
e
m
o
d
el
is
n
o
t
o
v
er
f
itti
n
g
an
d
is
lear
n
in
g
ef
f
ec
tiv
ely
o
v
e
r
tim
e.
(
a)
(
b
)
Fig
u
r
e
5
.
Gr
a
p
h
o
f
n
etwo
r
k
lo
s
s
es a
n
d
ac
cu
r
ac
y
b
ased
o
n
L
STM
m
o
d
el
(
a)
ac
c
u
r
ac
y
p
lo
t
an
d
(
b
)
lo
s
s
p
lo
t
Fro
m
th
e
g
iv
en
g
r
ap
h
s
,
we
ca
n
s
ay
th
at
th
e
n
etwo
r
k
is
n
o
t
o
v
er
tr
ain
ed
,
b
ec
a
u
s
e
th
e
lo
s
s
es
o
n
th
e
v
alid
atio
n
d
ata
ar
e
co
n
s
tan
t
ly
d
ec
r
ea
s
in
g
.
W
e
in
itialize
th
e
co
n
v
o
lu
tio
n
al
n
e
u
r
al
n
etwo
r
k
.
Fig
u
r
e
6
d
em
o
n
s
tr
ates th
e
ar
ch
itectu
r
e
o
f
a
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etw
o
r
k
.
L
et's p
er
f
o
r
m
t
h
e
f
o
llo
win
g
d
e
s
cr
ip
tio
n
o
f
th
e
n
etwo
r
k
ar
ch
it
ec
tu
r
e:
−
E
m
b
ed
d
in
g
lay
er
:
th
e
lay
er
v
ec
to
r
izes
th
e
in
p
u
t
d
ata
in
to
d
en
s
e
co
n
tex
t
v
ec
to
r
s
.
T
h
e
s
ize
o
f
th
e
in
p
u
t
v
ec
to
r
s
is
8
0
,
a
n
d
th
e
d
im
en
s
io
n
o
f
t
h
e
d
e
n
s
e
v
ec
to
r
s
is
5
0
.
C
o
n
v
o
lu
tio
n
al
lay
er
:
A
c
o
n
v
o
lu
tio
n
al
lay
e
r
co
n
tain
in
g
6
4
f
ilter
s
is
ap
p
lied
,
with
a
o
n
e
-
d
im
e
n
s
io
n
al
k
e
r
n
el
o
f
d
im
e
n
s
io
n
3
an
d
a
R
eL
U
ac
tiv
atio
n
f
u
n
ctio
n
.
T
h
is
lay
e
r
p
e
r
f
o
r
m
s
a
co
n
v
o
lu
tio
n
o
p
e
r
atio
n
o
n
th
e
in
p
u
t
s
eq
u
en
ce
th
u
s
f
o
r
m
in
g
a
f
ea
t
u
r
e
m
ap
f
o
r
th
e
v
ec
t
o
r
.
Glo
b
alM
ax
Po
o
lin
g
1
D:
T
h
e
in
p
u
t
d
ata
ar
e
f
e
atu
r
e
m
ap
s
f
r
o
m
th
e
co
n
v
o
lu
t
io
n
al
lay
er
,
th
e
cu
r
r
en
t
lay
er
in
tu
r
n
s
elec
ts
t
h
e
m
ax
im
u
m
v
alu
e
f
r
o
m
ea
c
h
f
ea
tu
r
e
m
ap
,
th
u
s
r
ed
u
cin
g
th
e
d
ata
v
o
lu
m
e
an
d
h
ig
h
lig
h
tin
g
th
e
m
o
s
t
im
p
o
r
tan
t
in
f
o
r
m
atio
n
.
Dr
o
p
o
u
t
a
n
d
B
atch
No
r
m
aliza
tio
n
:
r
em
o
v
in
g
p
a
r
t
o
f
t
h
e
n
eu
r
o
n
s
an
d
s
tan
d
ar
d
izin
g
th
e
in
p
u
t
d
ata
to
s
p
ee
d
u
p
an
d
r
eg
u
late
th
e
n
etwo
r
k
.
Den
s
e:
th
e
o
u
tp
u
t
d
ata
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
3
,
J
u
n
e
20
25
:
3
3
9
6
-
3
4
0
9
3402
af
ter
“scr
ee
n
in
g
”
an
d
s
tan
d
ar
d
izatio
n
p
ass
es
th
r
o
u
g
h
a
f
u
ll
y
co
n
n
ec
te
d
d
en
s
e
lay
er
with
2
5
6
n
e
u
r
o
n
s
an
d
R
eL
U
ac
tiv
atio
n
f
u
n
ctio
n
.
T
h
is
lay
er
p
er
f
o
r
m
s
a
n
o
n
-
lin
ea
r
tr
an
s
f
o
r
m
atio
n
o
f
th
e
in
p
u
t
d
at
a,
en
ab
lin
g
th
e
n
etwo
r
k
t
o
lear
n
co
m
p
lex
d
at
a
r
elatio
n
s
h
ip
s
.
Dr
o
p
o
u
t
an
d
b
atch
n
o
r
m
aliza
tio
n
:
r
ep
ea
ted
l
y
r
em
o
v
in
g
p
a
r
t
o
f
th
e
n
e
u
r
o
n
s
an
d
s
tan
d
a
r
d
izin
g
th
e
in
p
u
t
d
ata
to
s
p
ee
d
u
p
a
n
d
r
e
g
u
late
th
e
n
etwo
r
k
.
Fig
u
r
e
7
(
a)
d
em
o
n
s
tr
ates
th
at
p
lo
t
th
e
a
cc
u
r
ac
y
a
n
d
Fig
u
r
e
7
(
b
)
d
e
m
o
n
s
tr
ates
th
at
lo
s
s
es
o
f
th
e
co
n
v
o
lu
tio
n
al
n
etwo
r
k
.
T
h
is
m
o
d
el
is
also
n
o
t
o
v
e
r
tr
ain
ed
;
o
n
e
ca
n
s
ay
th
at
th
e
a
cc
u
r
ac
y
v
al
u
es
f
o
r
th
e
tr
ain
in
g
d
ata
ar
e
r
elev
an
t
f
o
r
th
e
en
tire
d
ataset
s
in
ce
th
ey
c
o
in
cid
e
with
th
e
m
o
d
el'
s
ac
cu
r
ac
y
f
o
r
th
e
v
alid
atio
n
d
ata.
T
h
e
ac
cu
r
ac
y
o
f
th
e
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
ex
ce
ed
s
t
h
e
a
cc
u
r
ac
y
o
f
th
e
lo
n
g
s
h
o
r
t
-
ter
m
m
em
o
r
y
n
etwo
r
k
.
Fig
u
r
e
8
d
em
o
n
s
tr
ates a
co
m
p
r
eh
en
s
iv
e
co
m
p
ar
is
o
n
,
let'
s
o
u
tp
u
t th
e
co
n
f
u
s
io
n
m
atr
ices f
o
r
th
e
d
ee
p
n
etwo
r
k
s
an
d
ca
lcu
late
th
e
k
ey
m
etr
ics.
Fig
u
r
e
6
.
Ar
c
h
itectu
r
e
o
f
a
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
(
a)
(
b
)
Fig
u
r
e
7
.
Plo
t o
f
ac
cu
r
ac
y
an
d
lo
s
s
f
o
r
a
co
n
v
o
lu
tio
n
al
n
eu
r
a
l n
etwo
r
k
(
a
)
ac
cu
r
ac
y
an
d
(
b
)
lo
s
s
(
a)
(
b
)
Fig
u
r
e
8
.
T
h
e
co
n
f
u
s
io
n
m
atr
ices f
o
r
(
a)
L
STM
an
d
(
b
)
C
NN
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
A
r
tifi
cia
l in
tellig
en
ce
fo
r
a
u
to
ma
tic
mo
d
era
tio
n
o
f te
xtu
a
l c
o
n
ten
t …
(
S
o
l
o
miia
Lia
s
ko
vska
)
3403
Me
tr
ics f
o
r
th
e
m
o
d
els:
Acc
u
r
ac
y
Sco
r
e:
0
.
9
2
,
Pre
cisi
o
n
Sco
r
e:
0
.
8
8
,
R
ec
all: 0
.
9
8
,
F1
-
Sco
r
e:
0
.
9
2
8
f
o
r
L
STM
.
Acc
u
r
ac
y
Sco
r
e:
0
.
9
5
,
Pre
cisi
o
n
Sco
r
e:
0
.
9
6
,
R
ec
all: 0
.
9
4
,
F1
-
Sco
r
e:
0
.
9
5
f
o
r
C
NN.
3
.
2
.
P
a
ra
m
et
er
s
a
na
ly
s
is
,
a
nd
re
s
ults
B
E
R
T
i
s
a
l
a
n
g
u
a
g
e
r
e
p
r
e
s
e
n
t
a
t
i
o
n
m
o
d
e
l
d
e
s
i
g
n
e
d
t
o
c
r
e
a
t
e
b
i
d
i
r
e
c
t
i
o
n
al
r
e
p
r
e
s
e
n
ta
t
i
o
n
s
f
o
r
d
e
e
p
n
e
u
r
a
l
n
e
t
w
o
r
k
s
o
n
r
a
w
,
u
n
a
n
n
o
t
a
t
e
d
t
e
x
t
b
y
c
o
m
b
i
n
i
n
g
l
e
f
t
a
n
d
r
i
g
h
t
c
o
n
te
x
t
s
i
n
al
l
l
a
y
e
r
s
.
Al
l
m
o
d
el
s
i
n
t
h
e
B
E
R
T
f
a
m
i
l
y
u
s
e
a
p
a
r
t
i
a
l
i
m
p
l
e
m
e
n
t
a
ti
o
n
o
f
t
r
a
n
s
f
o
r
m
e
r
m
o
d
e
l
s
,
n
a
m
e
l
y
e
n
c
o
d
e
r
s
,
a
s
t
h
e
n
e
t
w
o
r
k
'
s
o
u
t
p
u
t
i
s
a
l
a
n
g
u
a
g
e
m
o
d
e
l
.
B
E
R
T
is
p
r
e
-
t
r
a
i
n
e
d
o
n
“
d
i
r
t
y
”
t
e
x
t
d
a
t
a
,
s
o
t
h
e
r
e
i
s
n
o
n
e
e
d
t
o
p
e
r
f
o
r
m
p
r
e
p
r
o
c
e
s
s
i
n
g
t
h
a
t
w
as
u
s
e
d
f
o
r
li
n
e
a
r
m
o
d
e
ls
a
n
d
n
e
u
r
a
l
n
e
t
w
o
r
k
s
.
De
n
s
e
l
a
y
e
r
:
A
f
u
l
l
y
c
o
n
n
e
c
t
e
d
l
a
y
e
r
c
o
n
s
is
t
i
n
g
o
f
o
n
e
n
e
u
r
o
n
a
n
d
a
l
o
g
i
s
ti
c
a
ct
i
v
a
ti
o
n
f
u
n
c
t
i
o
n
t
o
r
e
t
u
r
n
t
h
e
p
r
o
b
a
b
i
l
it
y
o
f
a
n
o
b
j
e
c
t
b
el
o
n
g
i
n
g
t
o
a
cl
a
s
s
.
N
e
x
t
s
t
e
p
i
s
t
o
t
r
ai
n
t
h
e
m
o
d
e
l
a
n
d
o
u
t
p
u
t
t
h
e
c
o
n
f
u
s
i
o
n
m
a
t
r
i
x
F
i
g
u
r
e
9
d
e
m
o
n
s
tr
a
t
e
s
l
o
s
s
a
n
d
a
c
c
u
r
ac
y
p
l
o
ts
s
i
m
il
a
r
l
y
t
o
d
e
e
p
n
e
t
w
o
r
k
s
.
Fig
u
r
e
9
.
T
h
e
lo
s
s
an
d
ac
c
u
r
ac
y
p
lo
t f
o
r
th
e
B
E
R
T
-
b
ased
m
o
d
el
B
E
R
T
i
s
tr
ain
ed
u
s
in
g
m
ask
ed
to
k
en
p
r
e
d
ictio
n
an
d
n
e
x
t
s
en
ten
ce
p
r
ed
ictio
n
.
T
h
r
o
u
g
h
t
h
is
tr
ain
in
g
p
r
o
ce
s
s
,
B
E
R
T
ac
q
u
ir
es
co
n
tex
tu
al,
laten
t
r
ep
r
esen
tatio
n
s
o
f
to
k
en
s
b
ased
o
n
th
eir
co
n
tex
t.
Fig
u
r
e
1
0
d
em
o
n
s
tr
ates th
e
lo
s
s
an
d
ac
c
u
r
ac
y
p
l
o
t f
o
r
th
e
B
E
R
T
-
b
ased
m
o
d
el.
Fig
u
r
e
1
0
.
T
h
e
lo
s
s
an
d
ac
cu
r
a
cy
p
lo
t f
o
r
th
e
B
E
R
T
-
b
ased
m
o
d
el
3
.
3
.
Co
m
pa
riso
n a
nd
dis
cus
s
io
n
Fo
r
class
if
icatio
n
,
we
will
u
s
e
th
e
ar
ch
itectu
r
es
o
f
m
o
d
els
f
o
r
s
p
am
class
if
icatio
n
,
b
u
t
s
in
ce
we
h
av
e
th
r
ee
class
es
in
th
e
d
ataset,
f
o
r
ea
ch
o
f
th
e
n
etwo
r
k
s
,
we
n
ee
d
to
ch
an
g
e
th
e
o
u
tp
u
t
lay
er
an
d
th
e
lo
s
s
f
u
n
ctio
n
.
As
th
e
o
u
t
p
u
t
lay
e
r
,
we
u
s
ed
a
d
en
s
e
co
n
n
ec
ted
lay
er
with
o
n
e
n
e
u
r
o
n
an
d
a
lo
g
is
tic
ac
tiv
atio
n
f
u
n
ctio
n
,
t
h
u
s
o
b
tain
in
g
th
e
p
r
o
b
ab
ilit
y
o
f
a
n
o
b
ject
b
el
o
n
g
i
n
g
to
th
e
tar
g
et
class
.
Af
ter
m
ak
in
g
ch
a
n
g
es to
th
e
ar
ch
itectu
r
e,
we
will
tr
ain
d
e
ep
lear
n
in
g
m
o
d
els
an
d
d
em
o
n
s
tr
ate
th
e
lo
s
s
an
d
ac
cu
r
ac
y
g
r
ap
h
.
Fig
u
r
e
1
1
d
em
o
n
s
tr
ates
ac
cu
r
ac
y
a
n
d
lo
s
s
p
lo
t
f
o
r
th
e
L
STM
-
b
ased
m
o
d
el.
Fig
u
r
e
1
2
d
em
o
n
s
tr
ate
s
ac
cu
r
ac
y
an
d
lo
s
s
p
lo
t f
o
r
th
e
C
NN
-
b
ased
m
o
d
el
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
3
,
J
u
n
e
20
25
:
3
3
9
6
-
3
4
0
9
3404
Fig
u
r
e
1
1
.
Acc
u
r
ac
y
an
d
l
o
s
s
p
lo
t f
o
r
th
e
L
STM
-
b
ased
m
o
d
el
Fig
u
r
e
1
2
.
Acc
u
r
ac
y
an
d
l
o
s
s
p
lo
t f
o
r
th
e
C
NN
-
b
ased
m
o
d
el
T
o
an
aly
ze
th
e
ac
cu
r
ac
y
o
f
th
e
m
o
d
els,
we
will
ap
p
ly
a
c
o
n
f
u
s
io
n
m
atr
ix
.
Fig
u
r
e
1
3
d
e
m
o
n
s
tr
ates
co
n
f
u
s
io
n
m
atr
ices
f
o
r
L
ST
M
Fig
u
r
e
1
3
(
a
)
an
d
C
NN
Fi
g
u
r
e
1
3
(
b
)
.
T
h
e
r
esu
lts
o
f
th
e
co
n
f
u
s
io
n
m
atr
ices
allo
w
u
s
to
u
n
d
er
s
tan
d
th
e
p
e
r
f
o
r
m
a
n
ce
o
f
th
e
m
o
d
els.
B
o
th
m
o
d
els
m
ak
e
th
e
m
o
s
t
er
r
o
r
s
wh
en
class
if
y
in
g
h
ate
s
p
ee
ch
,
o
f
ten
m
is
tak
in
g
it
f
o
r
o
f
f
en
s
iv
e
lan
g
u
ag
e.
Sp
ec
if
ically
,
th
e
ac
cu
r
ac
y
in
class
if
y
in
g
p
o
s
ts
co
n
tain
in
g
h
ate
s
p
ee
ch
is
lo
wer
th
an
th
e
ac
cu
r
ac
y
in
cla
s
s
if
y
in
g
p
o
s
ts
co
n
tain
in
g
o
f
f
en
s
iv
e
lan
g
u
a
g
e
o
r
“n
o
r
m
al”
p
o
s
ts
.
T
h
e
m
o
d
els
p
er
f
o
r
m
b
est
at
d
is
tin
g
u
is
h
in
g
o
f
f
e
n
s
iv
e
lan
g
u
ag
e
f
r
o
m
r
eg
u
lar
p
o
s
ts
.
Ov
er
all,
co
n
s
id
er
in
g
all
er
r
o
r
v
alu
es,
it
ca
n
b
e
co
n
clu
d
ed
th
at
th
is
im
p
lem
en
tatio
n
o
f
th
e
c
o
n
v
o
lu
ti
o
n
al
n
eu
r
al
n
etwo
r
k
h
an
d
les th
e
task
b
etter
th
a
n
th
e
lo
n
g
s
h
o
r
t
-
ter
m
m
e
m
o
r
y
n
et
wo
r
k
.
(
a)
(
b
)
Fig
u
r
e
1
3
.
C
o
n
f
u
s
io
n
m
atr
ices f
o
r
(
a
)
L
STM
an
d
(
b
)
C
NN
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
A
r
tifi
cia
l in
tellig
en
ce
fo
r
a
u
to
ma
tic
mo
d
era
tio
n
o
f te
xtu
a
l c
o
n
ten
t …
(
S
o
l
o
miia
Lia
s
ko
vska
)
3405
L
e
t'
s
c
o
m
p
u
t
e
t
h
e
m
et
r
i
cs
o
f
th
e
m
o
d
e
l
s
f
o
r
e
a
c
h
c
l
a
s
s
f
o
r
c
o
m
p
a
r
i
s
o
n
,
a
n
d
w
e
c
a
n
s
e
e
t
h
e
r
e
s
u
l
ts
i
n
T
a
b
l
e
s
1
a
n
d
2
.
T
h
es
e
t
a
b
l
es
p
r
e
s
e
n
t
t
h
e
p
e
r
f
o
r
m
a
n
c
e
o
f
th
e
L
S
T
M
a
n
d
C
N
N
m
o
d
e
l
s
a
c
r
o
s
s
h
a
t
e
s
p
e
e
c
h
,
o
f
f
e
n
s
i
v
e
l
a
n
g
u
a
g
e
,
a
n
d
r
e
g
u
la
r
p
o
s
t
s
.
T
a
b
l
e
3
f
u
r
t
h
e
r
d
e
m
o
n
s
t
r
a
t
es
t
h
e
cl
as
s
i
f
i
c
a
ti
o
n
m
et
r
i
c
s
f
o
r
t
h
e
B
E
R
T
-
b
a
s
e
d
m
o
d
e
l
,
a
ll
o
w
i
n
g
f
o
r
a
c
o
m
p
r
e
h
e
n
s
i
v
e
c
o
m
p
a
r
i
s
o
n
a
m
o
n
g
a
l
l
t
h
r
e
e
a
p
p
r
o
a
c
h
e
s
.
T
a
b
l
e
1
s
h
o
w
s
m
et
r
i
cs
f
o
r
e
a
c
h
c
l
as
s
o
f
p
u
b
l
i
c
at
i
o
n
f
o
r
t
h
e
L
S
T
M
m
o
d
e
l
.
T
a
b
l
e
2
s
h
o
ws
f
o
r
e
a
c
h
c
l
a
s
s
f
o
r
p
u
b
l
i
c
a
ti
o
n
f
o
r
t
h
e
C
NN
m
o
d
e
l
.
T
ab
le
1
.
Me
tr
ics f
o
r
ea
ch
class
o
f
p
u
b
licatio
n
s
f
o
r
th
e
L
STM
m
o
d
el
A
c
c
u
r
a
c
y
-
sc
o
r
e
P
r
e
c
i
s
i
o
n
-
sc
o
r
e
R
e
c
a
l
l
F1
-
sc
o
r
e
H
a
t
e
sp
e
e
c
h
0
.
7
5
0
.
7
8
0
.
6
2
0
.
6
9
O
f
f
e
n
si
v
e
l
a
n
g
u
a
g
e
0
.
7
5
0
.
8
0
.
7
1
0
.
7
5
R
e
g
u
l
a
r
p
o
st
s
0
.
7
5
0
.
8
0
.
7
4
5
0
.
7
7
T
ab
le
2
.
Me
tr
ics f
o
r
ea
ch
class
o
f
p
u
b
licatio
n
s
f
o
r
th
e
C
NN
m
o
d
el
A
c
c
u
r
a
c
y
-
sc
o
r
e
P
r
e
c
i
s
i
o
n
-
sc
o
r
e
R
e
c
a
l
l
F1
-
sc
o
r
e
H
a
t
e
sp
e
e
c
h
0
.
7
6
0
.
7
2
0
.
6
8
0
.
7
0
O
f
f
e
n
si
v
e
l
a
n
g
u
a
g
e
0
.
7
6
0
.
8
2
0
.
7
4
0
.
7
8
R
e
g
u
l
a
r
p
o
st
s
0
.
7
6
0
.
8
1
0
.
8
1
0
.
8
1
T
ab
le
3
.
Me
tr
ics f
o
r
ea
ch
class
o
f
p
u
b
licatio
n
s
f
o
r
th
e
m
o
d
el
b
ased
o
n
B
E
R
T
A
c
c
u
r
a
c
y
-
S
c
o
r
e
P
r
e
c
i
s
i
o
n
-
sc
o
r
e
R
e
c
a
l
l
F1
-
sc
o
r
e
H
a
t
e
sp
e
e
c
h
0
.
3
0
.
3
3
0
.
3
0
.
3
1
O
f
f
e
n
si
v
e
l
a
n
g
u
a
g
e
0
.
3
0
.
3
5
0
.
3
2
0
.
3
3
R
e
g
u
l
a
r
p
o
st
s
0
.
3
0
.
3
2
0
.
3
3
0
.
3
2
T
h
e
p
r
o
p
o
s
ed
ar
ch
itectu
r
e
b
a
s
ed
o
n
B
E
R
T
f
ield
s
r
at
h
er
p
o
o
r
r
esu
lts
.
W
e
ca
n
ass
u
m
e
t
h
at
th
is
is
r
elate
d
to
th
e
n
o
n
lin
ea
r
ity
o
f
d
ep
en
d
e
n
cies
in
tex
tu
al
d
ata
s
in
ce
o
u
r
n
etwo
r
k
is
ess
en
tially
eq
u
iv
alen
t
to
a
lin
ea
r
m
o
d
el
with
a
s
in
g
le
E
m
b
ed
d
in
g
lay
er
.
I
n
th
e
co
n
tex
t
o
f
s
p
am
d
etec
tio
n
,
a
c
o
m
p
ar
ativ
e
s
tu
d
y
was
co
n
d
u
cte
d
f
o
r
lin
ea
r
m
o
d
els,
d
ee
p
n
eu
r
al
n
etwo
r
k
s
,
a
n
d
s
in
g
le
-
lay
er
m
o
d
els
u
s
in
g
th
e
p
r
e
-
tr
ain
e
d
B
E
R
T
n
etwo
r
k
.
A
d
d
itio
n
al
d
atasets
wer
e
cr
ea
ted
f
o
r
d
if
f
er
en
t
tex
t
r
ep
r
esen
tatio
n
tech
n
iq
u
es,
n
a
m
ely
b
ag
-
of
-
w
o
r
d
s
,
n
-
g
r
am
s
,
a
n
d
TF
-
I
DF
.
As
a
r
e
s
u
lt,
th
r
ee
p
air
s
o
f
lo
g
is
tic
r
eg
r
ess
io
n
an
d
s
u
p
p
o
r
t
v
ec
to
r
m
a
ch
in
e
m
o
d
els
wer
e
tr
ain
ed
.
All
m
o
d
els
ac
h
iev
ed
r
ea
s
o
n
ab
ly
h
ig
h
o
v
er
all
ac
c
u
r
ac
y
,
with
lo
g
is
tic
r
eg
r
ess
io
n
p
er
f
o
r
m
in
g
b
etter
in
id
en
tify
in
g
s
p
am
f
o
r
th
e
s
tan
d
ar
d
b
ag
-
of
-
w
o
r
d
s
,
wh
ile
th
e
s
u
p
p
o
r
t
v
ec
to
r
m
ac
h
i
n
e
h
ad
h
ig
h
er
m
etr
ics
f
o
r
TF
-
I
DF
.
T
h
e
lo
west
o
v
er
all
ac
cu
r
ac
y
was
o
b
s
er
v
e
d
f
o
r
th
e
TF
-
I
DF
d
ata
f
o
r
m
at,
alth
o
u
g
h
t
h
e
g
ap
i
n
all
m
etr
ics
f
o
r
th
e
th
r
ee
d
atasets
is
n
o
t
s
ig
n
if
ican
t.
Dee
p
m
o
d
el
s
an
d
th
e
B
E
R
T
-
b
ased
m
o
d
el
wer
e
th
en
tr
ain
ed
.
T
h
e
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
m
o
d
el
d
em
o
n
s
tr
ated
th
e
h
i
g
h
est ac
cu
r
ac
y
with
a
v
alu
e
o
f
0
.
9
5
.
Fo
r
th
e
class
if
icatio
n
o
f
h
ate
s
p
ee
ch
an
d
o
f
f
en
s
iv
e
lan
g
u
ag
e,
we
u
s
ed
th
e
s
am
e
n
eu
r
al
n
etwo
r
k
ar
ch
itectu
r
es
as
f
o
r
s
p
am
,
a
d
ap
tin
g
th
eir
o
u
tp
u
t
lay
e
r
f
o
r
m
u
lti
-
class
class
if
icat
io
n
task
s
.
Ag
ain
,
th
e
co
n
v
o
l
u
tio
n
al
n
e
u
r
al
n
etwo
r
k
ac
h
iev
ed
t
h
e
h
ig
h
est
ac
cu
r
ac
y
-
0
.
7
6
,
wh
ile
th
e
B
E
R
T
-
b
a
s
ed
m
o
d
el
s
h
o
wed
v
er
y
lo
w
r
esu
lts
-
0
.
3
.
4.
CO
NCLU
SI
O
N
T
h
e
m
ain
tech
n
iq
u
es
f
o
r
r
ep
r
e
s
en
tin
g
tex
t
in
n
u
m
er
ical
f
o
r
m
at
f
o
r
m
ac
h
in
e
lear
n
in
g
alg
o
r
it
h
m
s
wer
e
in
v
esti
g
ated
,
an
aly
zin
g
th
eir
c
h
ar
ac
ter
is
tics
,
wo
r
k
in
g
p
r
in
ci
p
les,
ad
v
an
tag
es,
a
n
d
d
is
ad
v
a
n
tag
es.
T
h
e
m
eth
o
d
o
f
wo
r
d
v
ec
to
r
r
e
p
r
esen
tatio
n
u
s
in
g
n
e
u
r
al
n
etwo
r
k
s
,
ex
em
p
lifie
d
b
y
th
e
wo
r
d
2
v
ec
m
o
d
el,
was
d
etailed
.
Fo
r
th
e
ch
o
s
en
lin
ea
r
m
o
d
els
—
lo
g
is
tic
r
eg
r
ess
io
n
an
d
s
u
p
p
o
r
t
v
ec
to
r
m
ac
h
in
es
—
an
ex
p
lan
at
io
n
o
f
th
eir
wo
r
k
i
n
g
p
r
in
cip
les
an
d
m
ath
em
atica
l
f
o
u
n
d
atio
n
s
was
p
r
o
v
id
e
d
.
T
h
e
d
escr
ip
tio
n
o
f
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
s
an
d
th
e
lo
n
g
s
h
o
r
t
-
ter
m
m
e
m
o
r
y
m
eth
o
d
in
clu
d
ed
th
ei
r
b
asic
ar
ch
itectu
r
al
co
m
p
o
n
e
n
ts
,
o
p
er
atio
n
al
p
r
in
ci
p
les,
an
d
tr
ain
in
g
p
r
o
ce
s
s
es.
T
h
e
s
p
ec
if
icity
o
f
u
s
in
g
c
o
n
v
o
lu
tio
n
a
l la
y
er
s
f
o
r
tex
t
u
al
d
ata
was a
ls
o
d
is
cu
s
s
ed
.
A
d
ataset
was
s
elec
ted
f
o
r
ea
ch
class
if
icatio
n
task
.
T
h
e
r
ese
ar
ch
wo
r
k
in
clu
d
es
a
d
etailed
d
escr
ip
tio
n
o
f
th
e
d
ata
p
r
ep
r
o
ce
s
s
in
g
an
d
f
ea
tu
r
e
e
x
tr
ac
tio
n
p
r
o
ce
s
s
u
s
in
g
v
ar
io
u
s
m
eth
o
d
s
.
C
o
r
r
esp
o
n
d
in
g
im
p
lem
en
tatio
n
s
o
f
m
ac
h
in
e
lear
n
in
g
alg
o
r
ith
m
s
wer
e
tr
ain
ed
f
o
r
ea
c
h
d
ataset,
an
d
m
o
d
el
p
er
f
o
r
m
a
n
ce
r
esu
lts
wer
e
d
em
o
n
s
tr
ated
.
I
t
was
f
o
u
n
d
th
at
lo
g
is
tic
r
eg
r
ess
io
n
an
d
s
u
p
p
o
r
t
v
ec
to
r
m
a
ch
in
es
ca
n
class
if
y
s
p
am
with
h
ig
h
ac
c
u
r
ac
y
,
an
d
d
if
f
er
e
n
t
d
ata
r
ep
r
esen
tatio
n
s
m
in
im
ally
af
f
ec
t
th
e
m
o
d
el
r
esu
lts
.
Fro
m
th
e
r
esear
ch
f
in
d
in
g
s
,
it
was
co
n
clu
d
ed
th
at
d
etec
tin
g
s
p
am
in
m
ess
ag
es
is
wea
k
ly
d
ep
en
d
e
n
t
o
n
th
e
s
em
an
tic
co
n
ten
t o
f
th
e
tex
t; f
r
eq
u
e
n
tly
u
s
ed
wo
r
d
s
ca
n
b
e
cr
u
cial
i
n
d
i
ca
to
r
s
o
f
s
p
am
.
Evaluation Warning : The document was created with Spire.PDF for Python.