I
A
E
S
I
n
t
e
r
n
at
io
n
al
Jou
r
n
al
of
A
r
t
if
ic
ia
l
I
n
t
e
ll
ig
e
n
c
e
(
I
J
-
AI
)
V
ol
.
14
, N
o.
4
,
A
ugus
t
2025
, pp.
3366
~
3374
I
S
S
N
:
2252
-
8938
,
D
O
I
:
10.11591/
ij
a
i.
v
14
.i
4
.pp
3366
-
3374
3366
Jou
r
n
al
h
om
e
page
:
ht
tp
:
//
ij
ai
.
ia
e
s
c
or
e
.c
om
E
val
u
at
i
n
g t
h
e
i
n
f
l
u
e
n
c
e
of
f
e
at
u
r
e
se
l
e
c
t
i
on
-
b
ase
d
d
i
m
e
n
s
i
on
al
i
t
y r
e
d
u
c
t
i
on
on
se
n
t
i
m
e
n
t
an
al
ysi
s
G
ow
r
av
R
am
e
s
h
B
ab
u
K
is
h
or
e
,
B
u
k
ah
al
ly
S
om
as
h
e
k
ar
H
a
r
is
h
,
C
h
al
u
ve
gow
d
a K
an
ak
al
ak
s
h
m
i
R
oop
a
D
e
pa
r
t
m
e
nt
of
I
nf
or
m
a
t
i
on
S
c
i
e
nc
e
a
nd
E
ngi
ne
e
r
i
ng, J
S
S
S
c
i
e
nc
e
a
nd T
e
c
hnol
ogy U
ni
ve
r
s
i
t
y, M
ys
ur
u, I
ndi
a
A
r
t
ic
le
I
n
f
o
A
B
S
T
R
A
C
T
A
r
ti
c
le
h
is
to
r
y
:
R
e
c
e
iv
e
d
A
pr
2, 2024
R
e
vi
s
e
d
M
a
r
21, 2025
A
c
c
e
pt
e
d
J
un 8, 2025
As
social
media
has
become
an
integral
part
of
digital
medium,
the
us
age
of
the same h
as incre
ased multi
-
fold in recent
years. With
increase in us
a
ge, the
sentiment
analysis
of
such
data
has
emerged
as
one
of
the
most
sought
research
domains
.
At
the
same
time,
social
media
texts
are
known
t
o
pose
variety
of
challenges
during
the
analysis,
thus
making
pre
-
processing
one
of
the importan
t steps. The
aim of
this work is
to perfor
m sentiment a
nal
ysis on
social media text, w
hile
handling the
noise effec
tively
in
the
data. Thi
s study
is
performed
on
a
multi
-
class
twitter
sentiment
dataset.
Firstly,
we
apply
several
text
cleaning
techniques
in
order
to
eliminate
noise
and
redu
ndancy
in
the
data.
In
addition,
we
examine
the
influence
of
regularized
l
ocality
preserving
indexing
(RLPI)
technique
combined
with
the
well
-
known
word
weighting
methods.
The
findings
obtained
from
experiment
indicat
e
that,
RLPI
outperforms
other
algorithms
in
feature
selection
and
when
paired
with
long
short
-
term
memory
(LSTM),
the
combination
outperform
s
other
classifi
cation m
odels t
hat are di
scussed.
K
e
y
w
o
r
d
s
:
S
e
nt
im
e
nt
a
na
ly
s
is
P
r
e
-
pr
oc
e
s
s
in
g
D
im
e
ns
io
na
li
ty
r
e
duc
ti
on
F
e
a
tu
r
e
s
e
le
c
ti
on
C
la
s
s
if
ic
a
ti
on
This is an
open
acce
ss artic
le unde
r the
CC BY
-
SA
license.
C
or
r
e
s
pon
di
n
g A
u
th
or
:
C
ha
lu
ve
gow
da
K
a
n
a
ka
la
ks
hm
i
R
oopa
D
e
pa
r
tm
e
nt
of
I
nf
or
m
a
ti
on S
c
ie
nc
e
a
nd E
ngi
ne
e
r
in
g, J
S
S
S
c
ie
n
c
e
a
nd
T
e
c
hnol
ogy Unive
r
s
it
y
M
ys
ur
u, 570006, Ka
r
na
ta
ka
, I
ndi
a
E
m
a
il
:
c
kr
@
js
s
s
tu
ni
v.i
n
1.
I
N
T
R
O
D
U
C
T
I
O
N
N
ow
-
a
-
da
ys
s
oc
ia
l
m
e
di
a
ha
s
ga
r
ne
r
e
d
a
lo
t
of
a
tt
e
nt
io
n.
I
t
is
a
m
ul
ti
m
e
di
a
pl
a
tf
or
m
w
he
r
e
pe
opl
e
c
a
n
s
ha
r
e
or
c
ons
um
e
in
f
or
m
a
ti
on
in
a
ny
f
or
m
a
t
th
a
t
th
e
y
w
a
n
t,
be
it
im
a
ge
,
vi
de
o,
a
udi
o
,
or
te
xt
.
T
ha
nks
to
it
s
in
s
ta
nt
a
ne
ous
gl
ob
a
l
a
c
c
e
s
s
ib
il
it
y,
it
ha
s
b
e
c
om
e
a
vi
ta
l
p
a
r
t
of
di
gi
ta
l
m
e
di
a
.
A
s
pe
opl
e
s
ta
r
te
d
us
in
g
s
oc
ia
l
m
e
di
a
in
la
r
ge
num
be
r
s
,
th
e
ne
e
d
to
a
na
ly
z
e
th
e
s
a
m
e
be
c
a
m
e
ne
c
e
s
s
a
r
y.
T
he
a
n
a
ly
s
is
s
ta
r
te
d
ta
ki
ng
pl
a
c
e
on
a
ll
pos
s
ib
le
a
s
pe
c
t
s
.
I
f
one
s
e
c
ti
on
of
r
e
s
e
a
r
c
h
c
om
m
u
ni
ty
f
oc
us
e
d
on
th
e
opt
im
a
l
u
s
e
of
c
om
put
in
g
r
e
s
our
c
e
s
, t
he
ot
he
r
s
e
c
ti
on f
oc
us
e
d on the
e
f
f
e
c
ti
ve
i
nf
or
m
a
ti
on r
e
tr
ie
va
l
te
c
hni
que
s
f
or
t
he
s
a
m
e
.
O
ne
of
th
e
t
r
e
ndi
ng
a
r
e
a
s
in
in
f
o
r
m
a
ti
on
r
e
t
r
ie
va
l
is
s
e
nt
im
e
nt
a
na
ly
s
is
,
w
he
r
e
th
e
gi
ve
n
da
ta
is
a
na
ly
z
e
d
in
or
de
r
to
obt
a
in
th
e
in
te
nde
d
opi
ni
on
or
e
m
ot
io
n.
T
he
r
e
a
r
e
m
a
ny
w
a
y
s
to
e
xpr
e
s
s
s
e
nt
im
e
nt
s
.
T
he
m
os
t
popula
r
m
e
th
ods
t
o c
a
te
gor
iz
e
t
he
m
i
s
e
it
he
r
ba
s
e
d on pola
r
it
y or
ba
s
e
d on e
m
ot
io
n. W
he
n i
t
c
om
e
s
to
pol
a
r
it
y,
th
e
s
e
nt
im
e
nt
s
m
ig
ht
be
one
a
m
ong
pos
it
iv
e
,
ne
g
a
ti
ve
,
or
ne
ut
r
a
l.
S
uc
h
la
be
ll
in
g
is
be
s
t
s
ui
te
d
w
he
n
th
e
a
im
of
th
e
a
na
ly
s
i
s
is
to
g
e
t
th
e
in
f
e
r
e
nc
e
onl
y a
t
hi
gh
e
r
le
ve
l.
O
n
th
e
ot
he
r
ha
nd,
f
or
e
m
ot
io
n,
th
e
r
e
is
w
id
e
r
a
nge
of
te
r
m
s
to
e
xpr
e
s
s
,
s
u
c
h
a
s
ha
ppy, s
a
d,
s
a
r
c
a
s
ti
c
,
ir
oni
c
,
a
nd
m
e
ta
phor
ic
a
l
;
a
nd
s
uc
h
s
e
nt
im
e
nt
la
be
ll
in
g w
or
ks
be
s
t
w
he
n t
he
a
na
ly
s
is
c
a
ll
s
f
or
t
he
i
nf
e
r
e
nc
e
of
pa
r
ti
c
ul
a
r
opi
ni
on.
I
n
r
e
c
e
nt
ye
a
r
s
,
s
e
nt
im
e
nt
a
na
ly
s
is
on
th
e
s
oc
ia
l
m
e
di
a
te
xt
ha
s
ga
in
e
d
a
lo
t
of
m
om
e
nt
um
.
W
he
th
e
r
it
is
a
na
ly
z
in
g
a
m
a
z
on
r
e
vi
e
w
s
f
or
m
a
r
ke
t
r
e
s
e
a
r
c
h,
or
a
na
ly
z
in
g
twe
e
ts
to
ga
uge
a
udi
e
nc
e
s
e
nt
im
e
nt
,
th
e
r
e
s
e
a
r
c
h
is
be
in
g
c
ondu
c
te
d
on
a
ll
c
onc
e
iv
a
bl
e
f
r
ont
s
.
A
lt
ho
ugh
s
oc
ia
l
m
e
di
a
is
w
id
e
ly
r
e
c
ogni
z
e
d
a
s
a
va
lu
a
bl
e
da
ta
s
our
c
e
,
th
e
te
xt
da
ta
c
ol
le
c
te
d
f
r
om
th
e
s
e
pl
a
tf
or
m
s
c
a
n
ha
ve
a
num
be
r
of
is
s
ue
s
.
I
s
s
ue
s
li
ke
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
E
v
al
uat
in
g t
he
i
nf
lu
e
nc
e
of
f
e
at
ur
e
s
e
le
c
ti
on bas
e
d di
m
e
n
s
io
nal
it
y
…
(
G
ow
r
av
R
am
e
s
h B
abu K
is
hor
e
)
3367
e
m
oj
is
,
ha
s
ht
a
g
s
,
e
m
oj
i
s
,
m
im
ic
ki
ng
s
pok
e
n
w
or
d
pr
ol
ong
a
ti
ons
,
m
is
s
p
e
ll
in
gs
,
a
nd
s
pe
c
ia
l
c
ha
r
a
c
te
r
s
oc
c
a
s
io
na
ll
y
c
a
u
s
e
noi
s
e
in
th
e
da
ta
,
th
us
m
a
ki
ng
it
di
f
f
ic
ul
t
to
pr
oc
e
s
s
it
di
r
e
c
tl
y.
P
r
oc
e
s
s
in
g
a
nd
a
na
ly
z
in
g
s
oc
ia
l
m
e
di
a
te
xt
s
c
a
n
a
ls
o
be
di
f
f
ic
ul
t
be
c
a
us
e
of
th
e
ir
non
-
u
ni
f
or
m
na
tu
r
e
,
a
s
th
e
y
don'
t
a
lw
a
ys
a
dhe
r
e
to
li
ngui
s
ti
c
nor
m
s
.
U
s
ua
ll
y,
s
uc
h
i
s
s
ue
s
a
r
e
not
e
nc
ount
e
r
e
d
in
ot
he
r
s
ta
nda
r
d
s
our
c
e
s
,
s
uc
h
a
s
ne
w
s
pa
pe
r
s
or
e
-
books
,
a
s
th
e
y
a
dhe
r
e
to
la
ngua
ge
s
ta
nda
r
ds
.
T
he
r
e
f
or
e
,
pr
e
-
pr
oc
e
s
s
in
g
s
te
ps
s
uc
h
a
s
te
xt
c
le
a
ni
ng
or
di
m
e
ns
io
na
li
ty
r
e
duc
ti
on
be
c
om
e
s
ne
c
e
s
s
a
r
y,
in
or
de
r
to
ha
ndl
e
s
upe
r
f
lu
ous
or
hi
gh
-
di
m
e
ns
io
na
l
s
oc
ia
l
m
e
di
a
te
xt
da
ta
[
1]
.
A
ddi
ti
ona
ll
y,
it
is
c
r
it
ic
a
l
th
a
t
th
e
m
ode
l
b
e
a
bl
e
to
unde
r
s
ta
nd
th
e
c
ont
e
xt
a
nd
s
e
nt
im
e
nt
f
r
om
s
hor
t
-
te
xt
s
, a
s
s
oc
ia
l
m
e
di
a
pl
a
tf
or
m
s
a
ls
o i
m
pos
e
l
im
it
s
o
n t
he
numbe
r
of
w
or
ds
.
P
r
e
-
pr
oc
e
s
s
in
g
of
a
te
xt
in
vol
ve
s
s
e
ve
r
a
l
im
por
ta
nt
s
te
ps
,
w
he
r
e
w
it
h
e
a
c
h
s
te
p
th
e
le
a
s
t
im
por
ta
nt
pa
r
t
of
th
e
da
ta
is
dr
oppe
d.
S
om
e
ti
m
e
s
,
m
or
e
th
a
n
th
e
a
na
ly
s
is
,
pr
e
-
pr
oc
e
s
s
in
g
it
s
e
lf
ta
ke
s
m
or
e
ti
m
e
[
2]
.
T
he
r
e
m
ova
l
of
s
pe
c
ia
l
s
ym
bol
s
a
nd
s
to
p
-
w
or
ds
r
e
duc
e
s
th
e
di
m
e
ns
io
na
li
ty
in
th
e
te
r
m
s
pa
c
e
[
3]
.
H
ow
e
ve
r
,
c
e
r
ta
in
c
le
a
ni
ng
pr
oc
e
dur
e
s
do
not
r
e
qui
r
e
th
e
c
om
pl
e
te
r
e
m
ova
l
of
th
e
te
r
m
f
r
o
m
th
e
da
ta
,
a
s
f
or
e
xa
m
pl
e
,
le
m
m
a
ti
z
a
ti
on
a
nd
s
te
m
m
in
g
m
e
r
e
ly
r
e
qui
r
e
th
e
te
r
m
to
be
r
e
duc
e
d
to
it
s
ba
s
ic
f
or
m
s
.
I
t
a
ls
o
gr
e
a
tl
y
a
id
s
i
n
r
e
m
ovi
ng
r
e
dunda
nc
y
a
nd
noi
s
e
,
s
o
th
a
t
onl
y
th
e
m
os
t
im
por
ta
nt
c
om
pone
nt
s
a
r
e
le
f
t
f
or
f
ur
th
e
r
a
na
ly
s
is
.
O
f
te
n,
e
ve
n
a
f
te
r
c
le
a
ni
ng
th
e
in
put
te
xt
,
th
e
f
in
a
l
c
or
pus
s
iz
e
w
il
l
s
ur
pa
s
s
th
e
pr
oc
e
s
s
in
g
c
a
pa
bi
li
ty
of
th
e
s
ys
te
m
.
S
o,
in
or
de
r
to
r
e
duc
e
th
e
di
m
e
ns
io
na
li
ty
of
in
put
f
u
r
th
e
r
m
or
e
,
f
e
a
tu
r
e
e
ngi
ne
e
r
in
g
is
pe
r
f
or
m
e
d.
F
e
a
tu
r
e
e
ngi
ne
e
r
in
g
te
c
hni
que
s
a
r
e
us
e
d
m
a
in
ly
to
e
xt
r
a
c
t
or
s
e
le
c
t
m
os
t
r
e
le
va
nt
s
e
t
of
f
e
a
tu
r
e
s
.
I
n
c
a
s
e
of
te
xt
, t
he
f
ir
s
t
a
nd f
or
e
m
os
t
ta
s
k i
s
f
e
a
tu
r
e
s
e
xt
r
a
c
ti
on, whe
r
e
t
he
t
e
xt
i
s
r
e
pr
e
s
e
nt
e
d i
n m
a
c
hi
ne
unde
r
s
ta
nda
bl
e
num
e
r
ic
a
l
f
or
m
.
S
ubs
e
que
nt
ly
,
f
e
a
tu
r
e
s
e
le
c
ti
on
is
e
m
pl
oy
e
d
t
o
is
ol
a
te
th
e
m
os
t
s
ig
ni
f
ic
a
nt
f
e
a
tu
r
e
s
,
w
hos
e
c
ont
r
ib
ut
io
n i
s
m
or
e
dur
in
g t
he
c
la
s
s
if
ic
a
ti
on.
F
ig
ur
e
1
s
how
s
th
e
c
a
te
gor
ie
s
of
di
m
e
ns
io
na
li
ty
r
e
duc
ti
o
n
te
c
hni
que
s
.
G
e
ne
r
a
ll
y,
in
f
e
a
tu
r
e
e
xt
r
a
c
ti
on,
th
e
or
ig
in
a
l
s
e
t
of
f
e
a
tu
r
e
s
is
tr
a
ns
f
or
m
e
d
to
ge
t
a
le
s
s
e
r
num
be
r
of
m
e
a
ni
ngf
ul
a
nd
r
e
le
va
nt
f
e
a
tu
r
e
s
e
t.
S
om
e
of
th
e
w
e
ll
-
known
a
lg
or
it
hm
s
a
r
e
pr
in
c
ip
a
l
c
om
pone
nt
a
na
ly
s
is
(
P
C
A
)
a
nd
t
-
di
s
tr
ib
ut
e
d
s
to
c
ha
s
ti
c
ne
ig
hbor
e
m
be
ddi
ng
(t
-
S
N
E
)
. I
n
f
e
a
tu
r
e
s
e
le
c
ti
on, s
ubs
e
ts
of
f
e
a
tu
r
e
s
a
r
e
s
e
le
c
te
d f
r
om
t
he
or
ig
in
a
l
s
e
t
of
f
e
a
tu
r
e
s
by
e
li
m
in
a
ti
ng
th
e
r
e
dunda
nt
or
ir
r
e
le
va
nt
on
e
s
.
S
om
e
w
e
ll
-
known
m
e
th
ods
a
r
e
r
e
c
ur
s
iv
e
f
e
a
tu
r
e
e
li
m
in
a
ti
on
(
R
F
E
)
, c
or
r
e
la
ti
on a
nd mut
ua
l
in
f
or
m
a
ti
on
-
ba
s
e
d a
lg
or
it
hm
s
.
F
ig
ur
e
1. C
a
te
gor
ie
s
of
di
m
e
ns
io
na
li
ty
r
e
duc
ti
on t
e
c
hni
que
s
I
n
th
is
pa
pe
r
,
w
e
e
va
lu
a
te
th
e
pe
r
f
o
r
m
a
nc
e
of
f
e
a
tu
r
e
s
e
le
c
ti
on
te
c
hni
que
s
w
he
n
pa
ir
e
d
w
it
h
r
e
gul
a
r
iz
e
d
lo
c
a
li
ty
pr
e
s
e
r
vi
ng
in
de
xi
n
g
(
R
L
P
I
)
a
lg
or
it
hm
.
A
ls
o,
e
xa
m
in
e
th
e
be
ha
vi
or
of
th
e
s
e
le
c
te
d
s
e
t
of
f
e
a
tu
r
e
s
w
it
h
va
r
io
us
ne
ur
a
l
ne
twor
k
-
ba
s
e
d
c
la
s
s
if
ic
a
ti
on
m
ode
ls
.
T
he
pur
pos
e
of
th
is
s
tu
dy
is
to
ga
in
a
de
e
pe
r
unde
r
s
ta
ndi
ng
on
us
in
g
va
r
io
us
pr
e
-
pr
oc
e
s
s
in
g
te
c
hni
que
s
in
c
om
bi
na
ti
on
w
it
h
R
L
P
I
di
m
e
ns
io
na
li
ty
r
e
duc
ti
on t
e
c
hni
que
t
ha
t
a
f
f
e
c
t
th
e
pe
r
f
or
m
a
nc
e
s
e
nt
im
e
nt
c
la
s
s
if
ic
a
ti
on. T
he
pr
im
a
r
y
f
oc
us
of
t
hi
s
r
e
s
e
a
r
c
h i
s
on
f
e
a
tu
r
e
s
e
le
c
ti
on
a
ppr
oa
c
he
s
a
nd
th
e
ir
e
f
f
e
c
t
on
s
e
nt
im
e
nt
t
e
xt
c
la
s
s
if
ic
a
ti
on
pe
r
f
or
m
a
nc
e
.
T
he
f
ol
lo
w
in
g
di
s
c
us
s
io
n
pr
ovi
de
s
s
om
e
in
it
ia
l
in
s
ig
ht
s
on
th
e
pr
om
in
e
nt
f
e
a
tu
r
e
s
e
le
c
ti
on
te
c
hni
qu
e
s
a
nd
th
e
ir
im
pa
c
t
on
s
e
nt
im
e
nt
c
la
s
s
if
ic
a
ti
on.
T
e
r
m
f
r
e
que
nc
y
-
in
ve
r
s
e
doc
um
e
nt
f
r
e
que
n
c
y
(
TF
-
I
D
F
)
is
one
of
th
e
w
e
ll
-
known
f
e
a
tu
r
e
e
xt
r
a
c
ti
on
m
e
th
ods
,
w
he
r
e
it
i
s
ge
ne
r
a
ll
y
u
s
e
d
f
or
e
xt
r
a
c
ti
ng
num
e
r
i
c
a
l
f
e
a
tu
r
e
s
out
of
t
e
xt
da
ta
[
4]
.
H
ow
e
v
e
r
,
P
a
ti
l
a
nd
A
ti
que
[
5]
s
how
s
how
f
e
a
tu
r
e
s
e
le
c
ti
on
c
a
n
be
im
pl
e
m
e
nt
e
d
w
it
h
T
F
-
I
D
F
,
by
a
ddi
ng
th
r
e
s
hol
d
pa
r
a
m
e
te
r
s
to
th
e
te
r
m
s
in
or
de
r
to
s
e
le
c
t
th
e
ke
y
te
r
m
s
.
W
hi
l
e
Q
u
e
t
al
.
[
6]
pr
opos
e
d
a
n
im
pr
ove
d
T
F
-
I
D
F
a
ppr
oa
c
h
by
in
c
lu
di
ng
doc
um
e
nt
’
s
r
e
la
ti
on
w
it
h
m
ul
ti
-
c
la
s
s
in
f
or
m
a
ti
on,
a
nd
ba
s
e
d
on
th
e
w
e
ig
ht
s
obt
a
in
e
d,
th
e
to
p
K
voc
a
bul
a
r
y
te
r
m
s
f
or
e
a
c
h
doc
um
e
nt
a
r
e
id
e
nt
if
ie
d.
L
i
e
t
al
.
[
7]
a
ppl
ie
d
r
e
gul
a
r
iz
e
d
le
a
s
t
s
qu
a
r
e
s
-
m
ul
ti
a
ngl
e
r
e
gr
e
s
s
io
n
a
nd
s
hr
in
ka
ge
(
R
L
S
-
M
A
R
S
)
m
ode
l
to
de
te
r
m
in
e
th
e
le
a
s
t
s
ig
ni
f
ic
a
nt
f
e
a
tu
r
e
s
.
T
he
pr
opos
e
d
m
e
th
od
a
s
s
ig
ns
le
s
s
w
e
ig
ht
to
th
e
le
a
s
t
s
ig
ni
f
ic
a
n
t
f
e
a
tu
r
e
s
.
A
c
c
or
di
ng
to
W
a
ng
a
nd
Z
ha
ng
[
8]
,
a
f
e
a
tu
r
e
s
e
le
c
ti
on
m
e
th
od
i
s
pr
e
s
e
nt
e
d
ba
s
e
d
on
T
F
-
I
D
F
by
c
om
bi
ni
ng
it
w
it
h
K
ul
lb
a
c
k
–
L
e
ib
le
r
(
KL
)
di
ve
r
ge
nc
e
,
w
he
r
e
by
c
on
s
id
e
r
in
g
th
e
m
ut
ua
l
in
f
or
m
a
ti
on
a
s
th
e
c
r
it
e
r
io
n,
th
e
a
ut
hor
s
pr
opos
e
d
a
n
im
pr
ove
d
c
la
s
s
if
ic
a
ti
on
a
ppr
oa
c
h.
S
ong
e
t
al
.
[
9]
in
t
r
oduc
e
d
a
n
e
nt
r
opy
in
de
x
a
lo
ng
w
it
h
T
F
-
I
D
F
in
or
de
r
to
ge
t
th
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
, V
ol
.
14
, N
o.
4
,
A
ugus
t
2025
:
3366
-
3374
3368
e
nt
r
opy
in
f
or
m
a
ti
on
of
a
te
r
m
w
it
h
-
in
a
nd
a
m
ong
th
e
c
la
s
s
e
s
,
w
hi
c
h
w
il
l
th
e
n
be
us
e
d
f
or
te
xt
f
e
a
tu
r
e
s
e
le
c
ti
on.
N
a
f
is
a
nd
A
w
a
ng
[
10]
pr
opos
e
s
a
two
s
ta
ge
f
e
a
tu
r
e
s
e
le
c
ti
on
a
ppr
oa
c
h.
W
he
r
e
in
f
ir
s
t
s
ta
ge
,
th
e
va
r
ia
nc
e
obt
a
in
e
d
f
or
e
nt
ir
e
T
F
-
I
D
F
m
a
tr
ix
is
us
e
d
a
s
th
r
e
s
hol
d
to
s
e
le
c
t
th
e
f
e
a
tu
r
e
s
.
T
he
n
in
s
e
c
ond
s
ta
ge
,
th
e
s
uppor
t
ve
c
to
r
m
a
c
hi
ne
(
S
V
M
)
-
R
F
E
i
s
a
ppl
ie
d on the
ne
w
f
e
a
tu
r
e
s
e
t
to
re
-
e
va
lu
a
te
t
he
f
e
a
tu
r
e
s
.
T
he
a
ut
hor
s
in
[
11]
pr
opos
e
d
a
f
e
a
tu
r
e
s
e
l
e
c
ti
on
m
e
th
od
ba
s
e
d
on
th
e
c
om
bi
na
ti
on
of
in
f
or
m
a
ti
on
ga
in
a
nd
di
ve
r
ge
nc
e
f
or
te
xt
c
a
te
gor
iz
a
ti
on
m
ode
ls
b
a
s
e
d
on
s
t
a
ti
s
ti
c
s
,
w
h
e
r
e
it
c
hoo
s
e
s
e
ve
r
y
f
e
a
tu
r
e
b
a
s
e
d
on
a
c
om
bi
na
ti
on
of
in
f
or
m
a
ti
on
ga
in
a
nd
nove
lt
y
c
r
it
e
r
ia
r
e
s
ul
ti
ng
in
r
e
duc
e
d
r
e
dunda
n
c
y
a
m
ong
th
e
s
e
le
c
te
d
f
e
a
tu
r
e
s
.
T
he
b
e
ha
vi
or
of
th
e
in
f
or
m
a
ti
on
ga
in
-
ba
s
e
d
f
e
a
tu
r
e
s
e
le
c
ti
on
m
e
th
od
c
om
bi
ne
d
w
it
h
th
e
ge
ne
ti
c
a
lg
or
it
hm
is
de
m
on
s
tr
a
te
d
in
[
12]
,
de
m
ons
tr
a
ti
ng
th
e
m
e
th
od
th
a
t
lo
w
e
r
s
th
e
te
xt
ve
c
to
r
'
s
di
m
e
ns
io
n.
S
ha
ng
e
t
al
.
[
13]
pr
opos
e
d
a
m
a
xi
m
iz
in
g
gl
oba
l
in
f
or
m
a
ti
on
g
a
in
a
ppr
oa
c
h,
w
hi
c
h
is
a
n
e
nha
n
c
e
d
ve
r
s
io
n
of
in
f
or
m
a
ti
on
ga
in
a
lg
or
it
hm
.
A
lo
ng
w
it
h
a
voi
di
ng
th
e
r
e
dunda
nc
y
in
th
e
f
e
a
tu
r
e
s
,
gl
oba
l
in
f
or
m
a
ti
on
ga
i
n
m
e
tr
ic
is
s
a
id
to
be
m
or
e
in
f
or
m
a
ti
ve
,
di
s
ti
nc
ti
ve
a
nd
a
l
s
o
pe
r
f
or
m
f
a
s
te
r
w
he
n
c
om
pa
r
e
d
to
th
e
tr
a
di
ti
ona
l
in
f
or
m
a
ti
on
ga
in
.
P
e
r
e
ir
a
e
t
al
.
[
14]
di
s
c
us
s
e
s
th
e
p
e
r
f
or
m
a
nc
e
of
in
f
or
m
a
ti
on
ga
in
ba
s
e
d
f
e
a
tu
r
e
s
e
le
c
ti
on,
a
nd
c
om
pa
r
e
s
th
e
s
a
m
e
a
ga
in
s
t
ot
he
r
m
ul
ti
-
la
be
l
f
e
a
tu
r
e
s
e
le
c
ti
on
m
e
th
ods
.
O
m
uya
e
t
al
.
[
15]
p
r
opos
e
s
a
hybr
id
di
m
e
ns
io
na
li
ty
r
e
duc
ti
on
te
c
hni
que
th
a
t
us
e
s
in
f
or
m
a
ti
o
n
ga
in
a
nd
P
C
A
to
e
xt
r
a
c
t
a
nd c
hoos
e
r
e
le
va
nt
f
e
a
tu
r
e
s
.
T
h
e
a
ppr
oa
c
h'
s
e
f
f
e
c
ti
ve
ne
s
s
w
a
s
a
s
s
e
s
s
e
d a
ga
in
s
t
th
e
n
a
iv
e
B
a
y
e
s
m
ode
l,
w
he
r
e
th
e
tr
a
in
in
g
ti
m
e
is
s
hor
te
ne
d w
hi
le
e
nha
nc
in
g pe
r
f
or
m
a
nc
e
.
T
he
c
hi
-
s
qu
a
r
e
te
s
t
is
on
e
of
th
e
w
id
e
ly
us
e
d
s
ta
ti
s
ti
c
a
l
f
unc
ti
ons
a
nd
th
e
w
or
k
in
[
4]
de
m
ons
tr
a
te
s
th
e
us
e
of
c
hi
-
s
qua
r
e
te
s
t
f
or
f
e
a
tu
r
e
s
e
le
c
ti
on,
a
lo
ng
w
it
h
K
-
ne
a
r
e
s
t
ne
ig
hbor
(
K
N
N
)
a
s
th
e
c
la
s
s
if
ic
a
ti
on
a
lg
or
it
hm
.
O
n
th
e
ot
he
r
ha
nd,
Z
ha
i
e
t
al
.
[
16
]
s
how
s
it
s
a
bi
li
ty
to
e
f
f
e
c
ti
ve
ly
s
e
le
c
t
th
e
be
tt
e
r
pe
r
f
or
m
in
g
s
e
t
of
f
e
a
tu
r
e
s
th
a
n
th
e
in
f
or
m
a
ti
on
ga
in
a
lg
or
it
hm
.
J
in
e
t
al
.
[
17
]
pr
opos
e
s
a
n
e
nha
nc
e
d
ve
r
s
io
n
of
c
hi
-
s
qua
r
e
s
ta
ti
s
ti
c
s
a
ppr
oa
c
h
c
a
ll
e
d
a
s
te
r
m
f
r
e
que
nc
y
a
nd
di
s
tr
ib
ut
io
n
ba
s
e
d
C
H
I
f
or
f
e
a
tu
r
e
s
e
le
c
ti
on
in
or
de
r
to
a
ddr
e
s
s
th
e
in
a
bi
li
ty
of
th
e
or
ig
in
a
l
a
ppr
oa
c
h
to
c
on
s
id
e
r
a
nd
id
e
nt
if
y
th
e
te
r
m
di
s
tr
ib
ut
io
n
in
e
a
c
h
c
la
s
s
.
Li
[
18
]
pr
opos
e
d
a
n
e
nha
nc
e
d
ve
r
s
io
n
of
c
hi
-
s
qua
r
e
a
ppr
oa
c
h
ba
s
e
d
on
C
hi
-
s
qua
r
e
r
a
nk
c
or
r
e
la
ti
on
f
a
c
to
r
iz
a
ti
on
w
he
r
e
it
is
c
la
im
e
d
th
a
t
th
e
a
lg
or
it
hm
doe
s
n
ot
ne
e
d
a
ny
pr
io
r
knowle
dge
a
nd
c
a
n
of
f
e
r
ge
ne
r
a
li
z
e
d
te
xt
c
a
te
gor
iz
a
ti
on.
H
a
r
ya
nt
o
e
t
al
.
[
19]
s
how
th
e
be
ha
vi
or
of
S
V
M
c
la
s
s
if
ie
r
upon
f
e
e
di
ng
th
e
in
put
s
w
hi
c
h a
r
e
nor
m
a
li
z
e
d
a
nd f
e
a
tu
r
e
s
a
r
e
s
e
le
c
te
d us
in
g t
he
c
hi
-
s
qua
r
e
a
ppr
oa
c
h.
S
e
l
e
t
al
.
[
20]
pr
e
s
e
nt
s
th
e
f
e
a
tu
r
e
s
e
le
c
ti
on
m
e
th
od,
w
hi
c
h
is
pe
r
f
or
m
e
d
ba
s
e
d
on
th
e
m
ut
ua
l
in
f
or
m
a
ti
on,
th
us
s
how
in
g
th
e
e
f
f
e
c
ti
ve
ne
s
s
of
th
e
a
ppr
oa
c
h
in
im
pr
ovi
ng
th
e
c
l
a
s
s
if
ic
a
ti
on
pe
r
f
or
m
a
nc
e
de
s
pi
te
of
dr
a
s
ti
c
r
e
duc
ti
on
in
th
e
num
be
r
o
f
f
e
a
tu
r
e
s
.
L
iu
e
t
al
.
[
21]
pr
opos
e
s
a
dyna
m
ic
m
ut
ua
l
in
f
o
r
m
a
ti
on
a
lg
or
it
hm
by
in
tr
oduc
in
g
a
ge
ne
r
a
l
c
r
it
e
r
io
n
f
unc
ti
on
f
or
f
e
a
tu
r
e
s
e
le
c
ti
on,
w
hi
c
h
is
e
xpe
c
te
d
to
g
e
t
m
os
t
in
f
or
m
a
ti
on
m
e
a
s
ur
e
m
e
nt
s
in
pr
e
vi
ou
s
a
lg
or
it
hm
s
to
ge
th
e
r
a
nd
w
a
s
e
va
lu
a
t
e
d
a
ga
in
s
t
va
r
io
us
e
xi
s
ti
ng
m
e
th
ods
.
A
gni
hot
r
i
e
t
al
.
[
22]
de
m
ons
tr
a
te
us
e
of
th
e
m
ut
ua
l
in
f
or
m
a
ti
on
to
obt
a
in
th
e
s
a
m
pl
e
va
r
ia
nc
e
in
or
de
r
t
o m
e
a
s
ur
e
t
he
va
r
ia
ti
ons
i
n t
e
r
m
di
s
tr
ib
ut
io
n a
nd t
o s
e
le
c
t
th
e
f
e
a
tu
r
e
s
. M
e
a
nw
hi
le
,
D
in
g a
nd
T
a
ng
[
23]
pr
e
s
e
nt
s
a
n
e
nha
nc
e
d
m
ut
ua
l
in
f
or
m
a
ti
on
m
e
th
od
by
in
tr
od
uc
in
g
th
e
f
e
a
tu
r
e
f
r
e
que
nc
y
in
c
la
s
s
a
nd
th
e
di
s
pe
r
s
io
n
of
f
e
a
tu
r
e
in
c
la
s
s
,
le
a
di
ng
to
a
n
e
f
f
ic
ie
nt
a
nd
im
pr
ove
d
te
xt
c
a
te
gor
iz
a
ti
on.
W
hi
le
D
a
r
s
ha
n
e
t
al
.
[
24]
s
how
s
th
e
a
bi
li
ty
of
R
L
P
I
to
e
f
f
e
c
ti
ve
ly
e
xt
r
a
c
t
th
e
di
s
c
r
im
in
a
ti
ve
f
e
a
tu
r
e
s
,
w
hi
c
h
in
tu
r
n
r
e
duc
e
s
th
e
c
om
pl
e
xi
ty
dur
in
g
th
e
r
e
pr
e
s
e
nt
a
ti
on
th
us
by
r
e
duc
in
g
th
e
to
ta
l
num
be
r
of
f
in
a
l
f
e
a
tu
r
e
s
e
t.
R
e
va
na
s
id
d
a
ppa
e
t
al
.
[
25]
pr
opos
e
d a
f
r
a
m
e
w
or
k ba
s
e
d on
m
e
ta
-
c
ogni
ti
ve
ne
ur
a
l
ne
twor
k
c
ons
ti
tu
ti
ng R
L
P
I
,
w
he
r
e
R
L
P
I
i
s
us
e
d a
lo
ng w
it
h
te
r
m
doc
um
e
nt
m
a
t
r
ix
(
T
D
M
)
a
s
f
e
a
tu
r
e
s
e
le
c
ti
on a
ppr
oa
c
h i
n or
de
r
t
o
r
e
duc
e
th
e
di
m
e
ns
io
na
li
ty
.
T
he
r
e
s
t
of
th
e
pa
pe
r
is
or
ga
ni
z
e
d
a
s
f
ol
lo
w
s
:
i
n
s
e
c
ti
on
3,
de
t
a
il
s
r
e
ga
r
di
ng
th
e
da
ta
s
e
t
c
on
s
id
e
r
e
d
f
or
th
e
e
xpe
r
im
e
nt
,
te
xt
c
le
a
ni
ng
a
nd
f
e
a
tu
r
e
s
e
le
c
ti
on
te
c
hni
que
s
th
a
t
a
r
e
e
m
pl
oye
d
dur
in
g
th
e
pr
e
-
pr
oc
e
s
s
in
g
s
ta
ge
,
d
e
ta
il
s
on
th
e
c
la
s
s
if
ic
a
ti
on
m
ode
ls
u
s
e
d,
f
ol
lo
w
e
d
by
th
e
w
or
ki
ng
pr
in
c
ip
le
of
th
e
e
xpe
r
im
e
nt
.
S
e
c
ti
on
4
pr
e
s
e
nt
s
th
e
e
xpe
r
im
e
nt
r
e
s
ul
t
s
a
lo
ng
w
i
th
di
s
c
us
s
io
n.
F
in
a
ll
y,
s
e
c
ti
on
5
c
onc
lu
d
e
s
th
e
w
or
k a
lo
ng w
it
h f
ut
ur
e
s
c
ope
.
2.
M
E
T
H
O
D
S
in
c
e
th
e
s
tu
dy
f
oc
us
e
s
on
t
e
xt
-
ba
s
e
d
s
e
nt
im
e
nt
a
n
a
ly
s
is
,
th
e
r
e
a
r
e
s
te
p
s
in
th
e
pr
oc
e
s
s
th
a
t
m
us
t
b
e
c
om
pl
e
te
d
in
or
de
r
to
c
le
a
n
th
e
da
ta
,
r
e
duc
e
it
s
di
m
e
ns
io
na
li
ty
,
a
nd
ge
t
it
r
e
a
dy
f
o
r
tr
a
in
in
g.
T
hi
s
s
e
c
ti
on
c
ove
r
s
t
he
s
pe
c
if
ic
s
of
t
he
da
ta
s
e
t
th
a
t
w
a
s
u
s
e
d, a
s
w
e
ll
a
s
t
he
a
ppr
oa
c
he
s
e
m
pl
oye
d f
or
e
a
c
h
s
ta
ge
.
2.1.
D
at
as
e
t
F
or
th
is
s
tu
dy
w
e
us
e
a
twi
tt
e
r
da
ta
s
e
t,
w
hi
c
h
is
c
r
e
a
te
d
by
c
o
m
bi
ni
ng
2
da
ta
s
e
t
s
w
hi
c
h
w
e
r
e
e
a
r
li
e
r
s
e
pa
r
a
te
.
O
r
ig
in
a
ll
y,
th
e
di
f
f
e
r
e
nt
ia
ti
ng
f
a
c
to
r
be
twe
e
n
th
e
two
da
ta
s
e
ts
w
a
s
th
e
ir
la
be
ll
in
g.
O
ne
da
t
a
s
e
t
w
it
h
1.6
m
il
li
on
s
a
m
pl
e
s
w
e
r
e
la
be
ll
e
d
ba
s
e
d
on
pol
a
r
it
y,
w
hi
le
th
e
ot
he
r
da
ta
s
e
t
w
it
h
a
bout
98
,
000
twe
e
t
s
a
m
pl
e
s
w
e
r
e
la
be
ll
e
d
b
a
s
e
d
on
f
e
e
li
ngs
s
uc
h
a
s
s
a
r
c
a
s
m
,
f
ig
ur
a
ti
ve
,
ir
ony
,
a
nd
r
e
gul
a
r
.
T
he
f
in
a
l
da
t
a
s
e
t
c
ons
is
t
s
of
97
,
000
s
a
m
pl
e
s
,
w
he
r
e
th
e
y
a
r
e
c
a
te
gor
iz
e
d
a
m
ong
5
s
e
nt
im
e
nt
c
la
s
s
e
s
na
m
e
ly
pos
it
iv
e
,
ne
g
a
ti
ve
,
ne
ut
r
a
l,
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
E
v
al
uat
in
g t
he
i
nf
lu
e
nc
e
of
f
e
at
ur
e
s
e
le
c
ti
on bas
e
d di
m
e
n
s
io
nal
it
y
…
(
G
ow
r
av
R
am
e
s
h B
abu K
is
hor
e
)
3369
s
a
r
c
a
s
m
,
a
nd
f
ig
ur
a
ti
ve
.
W
hi
le
c
r
e
a
ti
ng
th
e
f
in
a
l
da
ta
s
e
t,
s
a
m
pl
e
s
w
e
r
e
r
a
ndoml
y
s
e
le
c
te
d
s
uc
h
th
a
t
e
a
c
h
c
a
te
gor
y c
ont
a
in
s
s
a
m
pl
e
s
r
a
ngi
ng f
r
om
15
,
000 to 20
,
000 twe
e
ts
.
2.2.
T
e
xt
c
le
an
in
g m
e
t
h
od
s
T
hi
s
i
s
a
c
r
uc
ia
l
a
nd
w
id
e
ly
e
m
pl
oye
d
s
ta
ge
in
te
xt
-
ba
s
e
d
r
e
s
e
a
r
c
h,
s
in
c
e
it
f
a
c
il
it
a
te
s
th
e
e
xt
r
a
c
ti
on
of
us
e
f
ul
in
f
or
m
a
ti
on
f
r
om
te
xt
ua
l
da
ta
.
I
n
th
is
s
tu
dy
w
e
e
m
pl
oye
d
m
a
ny
te
xt
c
le
a
ni
ng
pr
oc
e
dur
e
s
,
a
nd
th
e
y
a
r
e
a
s
f
ol
lo
w
s
:
‒
T
e
xt
c
a
s
in
g:
t
he
s
a
m
e
w
or
d
c
a
n
b
e
pe
r
c
e
iv
e
d
a
s
a
s
in
gl
e
to
ke
n
by
c
ha
ngi
ng
it
s
c
a
s
e
w
hi
c
h
w
il
l
ot
h
e
r
w
is
e
be
c
ons
id
e
r
e
d
a
s
a
di
f
f
e
r
e
nt
to
ke
n,
s
uc
h
s
ta
nda
r
di
z
a
ti
on
of
c
a
s
e
he
lp
s
pr
e
ve
nt
r
e
dunda
nc
y
in
th
e
o
r
ig
in
a
l
c
or
pus
.
T
he
m
a
jo
r
it
y
of
th
e
ti
m
e
,
th
e
te
xt
is
c
ha
nge
d
to
lo
w
e
r
c
a
s
e
a
nd
th
e
s
a
m
e
is
f
ol
lo
w
e
d
dur
in
g
th
is
s
tu
dy a
s
w
e
ll
.
‒
R
e
m
ovi
ng
punc
tu
a
ti
on:
d
e
pe
ndi
ng
on
th
e
de
s
ig
n
a
nd
f
in
a
l
goa
l
of
th
e
m
ode
l,
punc
tu
a
ti
ons
th
a
t
a
r
e
of
te
n
us
e
d
to
in
di
c
a
te
s
e
pa
r
a
te
s
e
nt
e
nc
e
s
or
th
e
e
nd
of
s
e
nt
e
nc
e
s
s
uc
h
a
s
c
om
m
a
s
,
pe
r
io
ds
,
a
nd
s
e
m
ic
ol
ons
a
r
e
pr
e
s
e
r
ve
d or
dr
oppe
d. S
in
c
e
w
e
a
r
e
c
onc
e
nt
r
a
ti
ng mor
e
on t
he
t
oke
ns
i
n t
hi
s
i
ns
ta
nc
e
, t
h
e
punc
tu
a
ti
ons
a
r
e
dr
oppe
d.
‒
R
e
m
ovi
ng
s
pe
c
ia
l
s
ym
bol
s
:
s
in
c
e
th
e
s
tu
dy
is
pr
im
a
r
il
y
f
oc
us
e
d
on
pr
e
s
e
r
vi
ng
onl
y
th
e
im
por
ta
nt
to
ke
n
s
,
a
s
pr
e
vi
ous
ly
not
e
d,
a
ny
c
ha
r
a
c
te
r
s
ot
he
r
th
a
n
a
lp
ha
num
e
r
i
c
s
uc
h
a
s
a
m
pe
r
s
a
nd,
dol
la
r
,
pi
pe
,
a
nd
pe
r
c
e
nt
a
ge
. t
ha
t
a
r
e
known to be
of
te
n us
e
d i
n T
w
it
te
r
pos
t
s
, a
r
e
e
xc
lu
de
d.
‒
R
e
m
ovi
ng
s
to
p
w
or
ds
:
f
r
om
a
non
-
li
ngui
s
ti
c
poi
nt
of
vi
e
w
,
s
to
p
-
w
or
ds
don’
t
c
a
r
r
y
m
uc
h
in
f
o
r
m
a
ti
on
[
5
]
he
nc
e
r
e
m
ovi
ng
th
e
m
w
il
l
not
onl
y
h
e
lp
in
r
e
duc
in
g
th
e
noi
s
e
,
but
it
a
ls
o
h
e
lp
s
in
s
a
vi
ng
s
pa
c
e
.
S
to
p
-
w
or
ds
c
a
n be
i
de
nt
if
ie
d a
nd dr
oppe
d us
in
g both m
a
nua
l
a
n
d a
ut
om
a
ti
c
a
ppr
oa
c
h.
‒
S
te
m
m
in
g
or
L
e
m
m
a
ti
z
a
ti
on:
t
hi
s
is
th
e
pr
oc
e
s
s
e
s
of
r
e
duc
in
g
th
e
w
or
ds
to
th
e
ir
r
oot
f
or
m
.
I
t
w
a
s
not
ic
e
d
th
a
t
le
m
m
a
ti
z
a
ti
on
he
lp
s
be
tt
e
r
w
he
n
c
om
pa
r
e
d
to
s
te
m
m
i
ng
in
gi
vi
ng
th
e
m
e
a
ni
ngf
ul
r
oot
f
o
r
m
.
E
xa
m
pl
e
;
w
hi
le
s
te
m
m
in
g
r
e
du
c
e
s
‘
s
tu
di
e
s
’
is
r
e
du
c
e
d
‘
s
tu
d
i
’
,
l
e
m
m
a
ti
z
a
ti
on
r
e
duc
e
s
th
e
s
a
m
e
to
‘
s
tu
dy’
,
a
nd he
nc
e
i
n t
he
w
or
k l
e
m
m
a
ti
z
a
ti
on i
s
a
ppl
ie
d on the
t
e
xt
s
a
m
pl
e
s
.
‒
H
a
ndl
in
g
e
m
oj
is
:
e
m
oj
is
c
a
n
b
e
ha
ndl
e
d
in
a
num
be
r
of
w
a
y
s
,
e
it
he
r
by
r
e
m
ovi
ng
th
e
m
c
om
pl
e
te
ly
or
s
ubs
ti
tu
ti
ng t
he
m
w
it
h t
he
ir
t
e
xt
e
qui
va
le
nt
. I
n t
hi
s
s
tu
dy, e
m
ot
ic
ons
a
r
e
om
it
te
d.
‒
H
a
ndl
in
g
w
or
d
c
ont
r
a
c
ti
ons
:
i
n
th
is
a
c
ti
on,
w
e
c
onve
r
t
th
e
c
o
m
bi
ne
d
s
hor
t
f
or
m
s
of
w
or
ds
ba
c
k
to
th
e
ir
or
ig
in
a
l
f
or
m
s
.
E
xa
m
pl
e
:
‘
don’
t’
is
c
onve
r
te
d
to
‘
do
not
’
.
T
hi
s
c
a
n
a
l
s
o
be
a
c
hi
e
ve
d
in
bot
h
m
a
nua
l
a
nd
a
ut
om
a
te
d w
a
ys
.
‒
S
pe
ll
c
he
c
ki
ng:
c
he
c
ki
ng
th
e
s
pe
ll
in
g
of
th
e
to
ke
n
is
e
qua
ll
y
im
por
ta
nt
a
s
le
m
m
a
ti
z
a
ti
on,
it
he
lp
s
i
n
a
voi
di
ng unne
c
e
s
s
a
r
y a
ddi
ti
ona
l
to
ke
ns
t
ha
t
m
a
y b
e
pr
e
s
e
nt
due
t
o s
om
e
w
r
ong s
pe
ll
in
gs
.
2.3.
F
e
at
u
r
e
s
e
le
c
t
io
n
m
e
t
h
od
s
A
s
c
onve
ye
d
in
th
e
be
gi
nni
ng,
s
in
c
e
th
is
w
or
k
i
s
m
a
in
ly
f
oc
us
e
d
on
th
e
f
e
a
tu
r
e
s
e
le
c
ti
on
a
ppr
oa
c
h
f
or
di
m
e
ns
io
na
li
ty
r
e
duc
ti
on, i
t
is
ve
r
y i
m
por
ta
nt
t
o know mo
r
e
a
bout
t
he
a
ppr
oa
c
he
s
t
ha
t
a
r
e
t
he
r
e
f
or
f
e
a
tu
r
e
s
e
le
c
ti
on.
I
t
is
m
a
in
ly
c
la
s
s
if
ie
d
in
to
3
ty
pe
s
na
m
e
ly
,
f
i
lt
e
r
m
e
t
hod,
w
r
a
ppe
r
m
e
th
od
a
nd
e
m
be
dde
d
m
e
th
od.
I
n t
hi
s
s
tu
dy, we
r
e
s
tr
ic
t
th
e
e
xpe
r
im
e
nt
t
o f
il
te
r
a
nd w
r
a
ppe
r
m
e
th
ods
.
I
n
f
il
te
r
m
e
th
od,
th
e
f
e
a
tu
r
e
s
a
r
e
s
e
le
c
te
d
us
in
g
s
ta
ti
s
ti
c
a
l
te
s
ts
i
n
or
de
r
to
ge
t
th
e
c
or
r
e
la
ti
on
s
c
or
e
s
.
T
he
y a
r
e
known to be
i
ne
xp
e
ns
iv
e
a
nd f
a
s
t
a
nd s
om
e
of
t
he
t
e
c
hni
que
s
us
e
d unde
r
t
hi
s
m
e
th
od a
r
e
:
‒
TF
-
I
D
F
:
a
w
a
y
of
c
a
lc
ul
a
ti
ng
a
w
or
d'
s
w
e
ig
ht
w
it
hi
n
a
c
ol
le
c
t
io
n
of
doc
um
e
nt
s
,
ta
ki
ng
in
to
a
c
c
ount
th
e
f
a
c
t
th
a
t
s
om
e
t
e
r
m
s
a
r
e
m
or
e
c
om
m
on t
ha
n ot
he
r
s
.
T
he
w
e
ig
ht
is
c
a
lc
ul
a
te
d u
s
in
g (
1)
:
,
=
,
×
(
)
(
1)
W
he
r
e
,
is
f
r
e
que
nc
y
of
x
in
y,
is
N
um
be
r
of
doc
um
e
nt
s
c
ont
a
i
ni
ng
x
a
nd N
is
th
e
to
ta
l
num
be
r
of
doc
um
e
nt
s
.
‒
C
hi
-
s
qua
r
e
te
s
t:
t
hi
s
m
e
a
s
ur
e
[
21]
is
us
e
d
to
id
e
nt
if
y
th
e
d
e
gr
e
e
of
in
de
pe
nde
nc
e
be
twe
e
n
th
e
te
r
m
t
i
a
nd c
la
s
s
C
k
, a
nd i
t
is
gi
ve
n i
n (
2)
2
=
∗
(
∗
−
∗
)
(
+
)
(
+
)
(
+
)
(
+
)
(
2)
W
he
r
e
a
i
s
th
e
num
be
r
of
doc
um
e
nt
s
in
th
e
pos
it
iv
e
c
a
te
gor
y
th
a
t
c
ont
a
in
th
is
te
r
m
(
t
i
)
;
b
is
th
e
num
be
r
of
doc
um
e
nt
s
in
th
e
po
s
it
iv
e
c
a
te
gor
y
th
a
t
do
not
c
ont
a
in
th
is
te
r
m
(
t
i
)
;
c
is
th
e
num
be
r
of
doc
um
e
nt
s
in
th
e
ne
ga
ti
ve
c
a
te
gor
y
th
a
t
c
ont
a
in
th
is
te
r
m
(
t
i
)
;
a
nd
d
is
th
e
num
be
r
of
doc
um
e
nt
s
in
th
e
ne
ga
ti
ve
c
a
te
gor
y
th
a
t
do not c
ont
a
in
t
hi
s
t
e
r
m
(
t
i
)
;
a
nd N
i
s
t
he
t
ot
a
l
num
be
r
of
d
oc
um
e
nt
s
.
‒
I
nf
or
m
a
ti
on
ga
in
:
th
e
in
f
or
m
a
ti
on
ga
in
[
26]
pr
ovi
de
s
th
e
de
pe
nde
nc
y
be
twe
e
n
a
te
r
m
a
nd
a
c
la
s
s
a
nd
is
gi
ve
n a
s
(
3)
. W
he
r
e
a
, b, c
, d
,
a
nd N
m
e
a
n t
he
s
a
m
e
a
s
i
n (
2)
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
, V
ol
.
14
, N
o.
4
,
A
ugus
t
2025
:
3366
-
3374
3370
=
∗
∗
(
+
)
∗
(
+
)
+
∗
∗
(
+
)
∗
(
+
)
+
∗
∗
(
+
)
∗
(
+
)
+
∗
∗
(
+
)
∗
(
+
)
(
3)
‒
M
ut
ua
l
in
f
or
m
a
ti
on:
it
is
a
m
a
xi
m
um
c
la
s
s
-
ba
s
e
d
s
c
or
e
f
or
th
e
te
r
m
t
i
w
hi
c
h
is
hi
ghl
y
in
f
lu
e
nc
e
d
by
th
e
m
a
r
gi
na
l
pr
oba
bi
li
ti
e
s
,
th
a
t
a
s
s
ig
n
s
hi
ghe
r
w
e
ig
ht
f
or
th
e
r
a
r
e
te
r
m
s
a
s
c
om
p
a
r
e
d
to
th
e
c
om
m
onl
y
oc
c
ur
r
in
g
te
r
m
.
T
he
m
e
tr
ic
he
lp
s
in
m
e
a
s
ur
in
g
th
e
in
f
or
m
a
ti
on
c
ont
a
in
e
d
by
th
e
te
r
m
t
i
to
r
e
pr
e
s
e
nt
th
e
c
la
s
s
a
nd i
t
is
gi
ve
n a
s
[
22]
.
(
)
=
m
a
x
≤
≤
(
,
)
(
)
∗
(
)
(
4)
W
he
r
e
(
)
is
th
e
pr
oba
bi
li
ty
of
th
e
w
or
d
w
hi
c
h
is
(
+
)
/
,
(
)
is
th
e
pr
oba
bi
li
ty
of
c
la
s
s
gi
ve
n
as
(
+
)
/
a
nd
(
,
)
is
t
he
pr
oba
bi
li
ty
of
t
he
w
or
d
f
or
be
in
g i
n c
la
s
s
w
hi
c
h i
s
gi
ve
n by a
/N.
‒
R
L
P
I
:
is
a
m
ul
ti
s
te
p
a
lg
or
it
hm
a
ppl
ie
d
in
o
r
de
r
to
ge
t
th
e
m
e
a
ni
ngf
ul
s
e
t
of
f
e
a
tu
r
e
s
,
w
hi
c
h
in
vol
ve
s
a
dj
a
c
e
nc
y
gr
a
ph
c
on
s
tr
uc
ti
on,
E
ig
e
n
de
c
om
pos
it
io
n
a
nd
r
e
g
ul
a
r
iz
e
d
le
a
s
t
s
qua
r
e
.
R
L
P
I
e
m
be
ddi
ng
is
gi
ve
n a
s
[
27]
.
→
=
(
5)
W
he
r
e
z
i
s
a
d
-
di
m
e
ns
io
na
l
r
e
pr
e
s
e
nt
a
ti
on of
t
he
doc
um
e
nt
x a
n
d
A
is
t
he
t
r
a
ns
f
or
m
a
ti
on ma
tr
ix
.
‒
W
or
d
e
m
be
ddi
ngs
:
w
or
d
e
m
be
ddi
ng
is
a
r
e
pr
e
s
e
nt
a
ti
on
m
e
th
o
d,
w
he
r
e
a
pa
r
ti
c
ul
a
r
te
r
m
is
r
e
pr
e
s
e
nt
e
d
in
th
e
f
or
m
of
a
num
e
r
ic
a
l
ve
c
to
r
.
I
n
th
is
s
tu
dy
R
L
P
I
is
in
c
or
po
r
a
te
d
w
it
h
s
om
e
of
th
e
w
e
ll
-
known
w
or
d
e
m
be
ddi
ng me
th
ods
f
or
f
e
a
tu
r
e
s
e
le
c
ti
on i
n a
n a
tt
e
m
pt
t
o
r
e
duc
e
t
he
di
m
e
ns
io
na
li
ty
of
t
he
or
ig
in
a
l
f
e
a
tu
r
e
ve
c
to
r
s
.
I
n
th
e
w
r
a
ppe
r
m
e
th
od,
th
e
m
ode
l
i
s
tr
a
in
e
d
us
in
g
a
s
ubs
e
t
o
f
f
e
a
tu
r
e
s
,
a
nd
f
e
a
tu
r
e
a
ddi
ti
ons
a
nd
de
le
ti
ons
a
r
e
de
te
r
m
in
e
d
by
th
e
c
onc
lu
s
io
ns
de
r
iv
e
d
f
r
om
th
e
r
e
s
ul
ts
obt
a
in
e
d.
O
ne
s
uc
h
te
c
hni
que
c
ons
id
e
r
e
d
f
or
th
e
s
tu
dy
is
,
R
F
E
.
I
t
is
one
of
th
e
c
om
put
a
ti
o
na
ll
y
e
xpe
ns
iv
e
te
c
hni
qu
e
s
,
due
to
it
s
gr
e
e
dy
a
ppr
oa
c
h.
I
n
th
is
t
e
c
hni
que
,
th
e
m
ode
l
i
s
tr
a
in
e
d
it
e
r
a
ti
ve
ly
w
it
h
a
s
ub
s
e
t
of
f
e
a
tu
r
e
s
unt
il
a
ll
th
e
f
e
a
tu
r
e
s
a
r
e
e
xha
us
te
d, ul
ti
m
a
te
ly
i
de
nt
if
yi
ng t
he
be
s
t
pe
r
f
or
m
in
g s
e
t
of
f
e
a
tu
r
e
s
.
2.4. Clas
s
if
i
c
at
io
n
m
e
t
h
od
s
I
n
th
is
s
tu
dy,
s
e
nt
im
e
nt
c
l
a
s
s
if
ic
a
ti
on
i
s
pe
r
f
or
m
e
d
w
it
h
s
om
e
of
th
e
w
id
e
ly
known
ne
ur
a
l
ne
twor
k
-
ba
s
e
d
m
ode
l
s
.
W
e
a
s
s
e
s
s
th
e
c
l
a
s
s
if
ic
a
ti
on
p
e
r
f
or
m
a
nc
e
of
bot
h
ba
s
ic
a
nd
r
e
c
ur
r
e
nt
ne
ur
a
l
ne
twor
ks
(
R
N
N
)
ba
s
e
d
m
od
e
ls
.
F
ir
s
tl
y,
th
e
c
la
s
s
if
ic
a
ti
on
p
e
r
f
or
m
a
nc
e
of
ba
s
ic
f
e
e
d
f
or
w
a
r
d
ne
ur
a
l
ne
twor
k
(
F
N
N
)
m
ode
l
is
a
s
s
e
s
s
e
d.
B
e
c
a
u
s
e
of
th
e
ir
non
-
c
yc
li
c
in
f
or
m
a
ti
on
f
lo
w
,
F
N
N
s
a
r
e
hi
ghl
y
s
tr
a
ig
ht
f
o
r
w
a
r
d
a
nd
e
a
s
ie
r
to
ve
r
if
y
[
28]
.
T
he
n,
th
e
be
h
a
vi
or
of
r
a
di
a
l
ba
s
i
s
f
unc
ti
on
ne
twor
k
(
R
B
F
N
)
is
e
va
lu
a
t
e
d
a
ga
in
s
t
th
e
s
e
le
c
te
d
s
e
t
of
f
e
a
tu
r
e
s
.
I
t
is
w
id
e
ly
us
e
d
f
or
c
om
m
on
a
ppr
oxi
m
a
ti
on
pr
obl
e
m
s
,
w
he
r
e
hi
dde
n
la
ye
r
w
il
l
u
s
e
th
e
r
a
di
a
l
ba
s
is
f
unc
ti
on.
I
t
is
m
uc
h
f
a
s
te
r
w
he
n
c
om
pa
r
e
d
to
ba
c
k
pr
opa
ga
ti
on
ne
twor
k,
a
nd
c
a
n
e
ve
n
out
pe
r
f
or
m
th
e
c
la
s
s
if
ic
a
ti
on pe
r
f
or
m
a
nc
e
i
f
t
he
pr
ope
r
s
e
t
of
f
e
a
tu
r
e
s
a
r
e
s
e
le
c
te
d
[
29]
.
W
e
th
e
n
e
xa
m
in
e
th
e
c
la
s
s
if
ic
a
ti
on
pe
r
f
or
m
a
nc
e
of
m
ode
ls
th
a
t
a
r
e
de
s
ig
ne
d
f
or
s
e
que
nt
i
a
l
or
ti
m
e
s
e
r
ie
s
d
a
ta
.
F
ir
s
tl
y,
th
e
c
la
s
s
if
ic
a
ti
on
p
e
r
f
or
m
a
nc
e
of
R
N
N
is
e
va
lu
a
te
d.
T
hough
it
is
bi
t
s
lo
w
e
r
th
a
n
ba
s
ic
F
N
N
s
,
it
s
a
bi
li
ty
to
r
e
ta
in
in
f
or
m
a
ti
on
a
bout
a
s
e
que
nc
e
in
hi
dde
n
la
ye
r
s
m
a
ke
s
it
m
os
t
s
ui
ta
bl
e
f
or
pr
oc
e
s
s
in
g
s
e
que
nt
ia
l
da
t
a
s
uc
h
a
s
te
xt
.
H
ow
e
ve
r
,
th
e
va
ni
s
hi
ng
gr
a
di
e
nt
is
s
ue
in
th
e
ir
m
e
m
or
y
s
ta
te
li
m
it
s
th
e
ir
a
bi
li
ty
to
r
e
ta
in
onl
y
s
hor
t
w
in
dow
of
th
e
pr
io
r
in
pu
ts
.
I
n
or
de
r
to
ha
ndl
e
th
is
is
s
ue
,
lo
ng
s
hor
t
-
te
r
m
m
e
m
or
y
(
L
S
T
M
)
w
a
s
in
tr
oduc
e
d.
O
ne
bi
g
a
dva
nt
a
ge
of
L
S
T
M
is
it
s
r
e
la
ti
ve
in
s
e
ns
it
iv
it
y
to
ga
p
le
ngt
h,
s
o
th
e
c
la
s
s
if
ic
a
ti
on pe
r
f
or
m
a
nc
e
of
L
S
T
M
i
s
a
ls
o e
v
a
lu
a
te
d a
ga
in
s
t
th
e
s
e
le
c
te
d f
e
a
tu
r
e
s
e
t.
F
in
a
ll
y, w
e
e
va
lu
a
te
th
e
pe
r
f
or
m
a
nc
e
of
ga
te
d
r
e
c
ur
r
e
nt
uni
ts
(
G
R
U
)
.
I
t
is
a
ls
o
a
n
R
N
N
ba
s
e
d
ne
twor
k
a
nd
a
n
a
lt
e
r
na
ti
ve
to
L
S
T
M
.
B
ut
G
R
U
'
s
f
unda
m
e
nt
a
l
p
r
in
c
ip
le
is
to
upda
te
th
e
ne
t
w
or
k'
s
hi
dde
n
s
ta
te
onl
y
on
a
c
hos
e
n
s
ubs
e
t
of
ti
m
e
s
te
ps
, by me
a
n
s
of
ga
ti
ng me
th
ods
. I
t
is
s
im
pl
e
r
i
n s
tr
uc
tu
r
e
a
nd e
a
s
i
e
r
t
o t
r
a
in
t
ha
n L
S
T
M
.
2.5. E
xp
e
r
im
e
n
t
at
io
n
T
he
e
xpe
r
im
e
nt
s
e
t
-
up
s
ta
r
ts
w
it
h
tw
it
te
r
s
e
n
ti
m
e
n
t
da
ta
b
e
in
g
c
ons
id
e
r
e
d
a
s
a
n
in
p
ut
to
th
e
c
la
s
s
i
f
ic
a
t
io
n
s
ys
te
m
,
w
hi
c
h
w
i
ll
f
ir
s
t
und
e
r
go
th
e
p
r
e
-
p
r
oc
e
s
s
in
g
w
i
th
t
he
m
e
t
hods
t
ha
t
a
r
e
d
is
c
us
s
e
d
in
th
e
s
e
c
t
io
n
2.2
.
F
i
gu
r
e
2
p
r
e
s
e
n
ts
th
e
f
lo
w
di
a
gr
a
m
,
w
he
r
e
th
e
i
n
pu
t
da
ta
f
i
r
s
t
u
nde
r
g
oe
s
c
le
a
ni
ng,
f
o
ll
o
w
e
d
by
th
e
di
m
e
ns
io
na
l
it
y
r
e
d
uc
ti
o
n.
F
or
d
im
e
ns
i
ona
li
ty
r
e
duc
ti
on
,
f
i
r
s
t
in
o
r
de
r
to
o
bt
a
in
lo
c
a
l
it
y
i
n
f
o
r
m
a
ti
on
,
th
e
R
L
P
I
is
a
p
pl
ie
d
on
t
he
s
a
m
p
le
s
,
w
h
ic
h
is
th
e
n
c
ou
pl
e
d
w
it
h
t
he
f
e
a
tu
r
e
s
e
le
c
ti
on
te
c
hni
que
s
c
o
ve
r
e
d
in
s
e
c
ti
on
2
.3
o
f
th
is
w
o
r
k
. T
he
r
e
s
ul
ti
ng
s
e
t
o
f
r
e
le
va
n
t
f
e
a
tu
r
e
s
f
r
om
th
e
r
e
s
pe
c
ti
ve
c
o
m
b
in
a
ti
on
i
s
t
he
n
us
e
d
f
o
r
t
r
a
i
ni
ng
th
e
m
od
e
l
.
F
o
r
c
la
s
s
if
ic
a
ti
on,
m
os
t
c
o
m
m
on
ly
k
n
ow
n
ne
ur
a
l
ne
two
r
k
ba
s
e
d
m
o
de
ls
v
iz
.,
F
N
N
,
R
N
N
,
R
B
F
N
,
L
S
T
M
,
a
nd
G
R
U
a
r
e
us
e
d.
U
p
on
ob
ta
in
in
g
t
he
c
la
s
s
i
f
ic
a
ti
o
n
r
e
s
ul
ts
,
t
he
e
f
f
e
c
ti
ve
ne
s
s
o
f
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
E
v
al
uat
in
g t
he
i
nf
lu
e
nc
e
of
f
e
at
ur
e
s
e
le
c
ti
on bas
e
d di
m
e
n
s
io
nal
it
y
…
(
G
ow
r
av
R
am
e
s
h B
abu K
is
hor
e
)
3371
e
a
c
h
of
th
e
d
im
e
ns
i
ona
li
ty
r
e
duc
ti
on
te
c
hn
iq
u
e
s
a
n
d
c
la
s
s
if
ic
a
ti
o
n
p
e
r
f
o
r
m
a
nc
e
o
f
th
e
m
o
de
ls
a
r
e
e
va
l
ua
te
d a
nd
a
na
ly
z
e
d
.
F
ig
ur
e
2. W
or
kf
lo
w
of
te
xt
-
ba
s
e
d
s
e
nt
im
e
nt
a
na
ly
s
is
3.
R
E
S
U
L
T
S
A
N
D
D
I
S
C
U
S
S
I
O
N
D
ur
in
g t
he
di
m
e
ns
io
na
li
ty
r
e
duc
ti
on s
ta
ge
of
t
he
e
xpe
r
im
e
nt
, t
he
f
e
a
tu
r
e
s
e
le
c
ti
on w
a
s
pe
r
f
or
m
e
d f
or
s
e
ve
r
a
l
it
e
r
a
ti
ons
a
s
s
e
e
n
in
T
a
bl
e
1.
D
ur
in
g
th
is
s
tu
dy,
th
e
uppe
r
a
nd
lo
w
e
r
li
m
it
s
w
e
r
e
de
f
in
e
d
to
obt
a
in
th
e
m
os
t
r
e
le
va
nt
s
e
t
of
f
e
a
tu
r
e
s
.
W
it
h
m
in
im
um
of
300
a
nd
m
a
xi
m
um
of
700
be
in
g
th
e
e
m
pi
r
ic
a
ll
y
de
f
in
e
d
s
ta
nda
r
d
th
r
e
s
hol
d
s
f
or
th
e
num
be
r
of
f
e
a
tu
r
e
s
,
th
e
e
xpe
r
im
e
nt
s
w
e
r
e
c
a
r
r
ie
d
out
f
or
e
a
c
h
c
om
bi
na
ti
on
of
f
e
a
tu
r
e
s
e
le
c
ti
on
m
e
th
ods
.
T
a
bl
e
1
s
how
s
th
e
out
c
om
e
s
of
e
a
c
h
tr
ia
l.
I
t
c
a
n
be
obs
e
r
ve
d
f
r
om
th
e
t
a
bl
e
,
th
a
t
th
e
R
L
P
I
ha
s
s
e
le
c
te
d
a
n
in
te
r
e
s
ti
ngl
y
le
s
s
num
be
r
of
f
e
a
tu
r
e
s
i
n
e
a
c
h
tr
ia
l
w
he
n
c
om
pa
r
e
d
to
ot
he
r
m
e
th
ods
.
F
ig
ur
e
3
is
s
how
in
g
th
e
r
a
nge
of
f
e
a
tu
r
e
s
by
us
in
g
m
a
xi
m
um
a
nd
m
in
im
um
c
ount
a
s
th
e
e
xt
r
e
m
e
s
to
in
di
c
a
te
th
e
c
ount
of
f
e
a
tu
r
e
s
s
e
le
c
t
e
d by e
a
c
h of
t
he
a
ppr
oa
c
he
s
m
e
nt
io
ne
d i
n T
a
bl
e
1.
T
a
bl
e
1
.
N
um
be
r
of
f
e
a
tu
r
e
s
s
e
le
c
te
d by va
r
io
us
s
e
le
c
ti
on me
th
ods
F
e
a
t
ur
e
s
e
l
e
c
t
i
on m
e
t
hods
N
um
be
r
of
f
e
a
t
ur
e
s
s
e
l
e
c
t
e
d
T
r
i
a
l
1
T
r
i
a
l
2
T
r
i
a
l
3
T
r
i
a
l
4
T
r
i
a
l
5
T
r
i
a
l
6
TF
-
I
D
F
575
538
357
399
412
419
C
hi
-
s
qua
r
e
600
562
552
457
547
552
I
nf
or
m
a
t
i
on
ga
i
n
549
552
656
453
479
490
M
ut
ua
l
i
nf
or
m
a
t
i
on
427
477
479
360
380
411
W
or
d2V
e
c
340
342
341
353
361
379
G
l
ove
435
415
426
421
405
445
R
F
E
494
412
485
433
530
540
R
L
P
I
69
58
49
93
100
210
F
ig
ur
e
3. M
a
x a
nd mi
n numbe
r
of
f
e
a
tu
r
e
s
s
e
le
c
te
d by
e
a
c
h m
e
t
hod
0
100
200
300
400
500
600
700
N
u
m
b
e
r
of
f
e
at
u
r
e
s
F
e
at
u
r
e
s
e
l
e
c
t
i
on
m
e
t
h
od
hi
gh
l
ow
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
, V
ol
.
14
, N
o.
4
,
A
ugus
t
2025
:
3366
-
3374
3372
U
pon
s
e
le
c
ti
ng
th
e
m
in
im
um
s
e
t
of
f
e
a
tu
r
e
s
a
m
ong
each
tr
ia
l
of
e
a
c
h
f
e
a
tu
r
e
s
e
le
c
ti
on
m
e
th
ods
,
th
e
s
e
le
c
te
d
f
e
a
tu
r
e
s
s
e
ts
a
r
e
th
e
n
c
on
s
id
e
r
e
d
a
s
in
put
s
to
th
e
c
la
s
s
if
ic
a
ti
on
m
ode
ls
th
a
t
a
r
e
di
s
c
us
s
e
d
in
s
e
c
ti
on
2.4.
T
a
bl
e
2
pr
e
s
e
nt
s
th
e
c
la
s
s
if
ic
a
ti
on
r
e
s
ul
ts
of
v
a
r
io
us
f
e
a
tu
r
e
s
e
le
c
ti
on
m
e
th
ods
a
nd
ne
ur
a
l
ne
twor
k
-
ba
s
e
d
c
la
s
s
if
ie
r
s
.
T
he
r
e
s
ul
ts
a
r
e
ta
bul
a
te
d
f
or
th
e
da
ta
s
e
t
di
vi
de
d
w
it
h
50:
50
r
a
ti
os
f
or
t
r
a
in
in
g
a
nd
te
s
ti
ng
r
e
s
pe
c
ti
ve
ly
.
T
a
bl
e
3
pr
e
s
e
nt
s
th
e
r
e
s
ul
ts
e
xpe
r
im
e
nt
e
d
on
s
a
m
e
s
e
t
of
f
e
a
tu
r
e
s
e
le
c
ti
on
a
nd
c
la
s
s
if
ic
a
ti
on
m
ode
l
s
w
hi
le
th
e
r
e
s
ul
t
s
a
r
e
t
a
bul
a
te
d
f
or
th
e
d
a
t
a
s
e
t
di
vi
de
d
w
it
h
60:
40
r
a
ti
o
f
or
tr
a
in
in
g
a
nd
te
s
ti
ng r
e
s
pe
c
ti
ve
ly
.
T
a
bl
e
2
.
C
la
s
s
if
ic
a
ti
on pe
r
f
or
m
a
nc
e
f
or
50:
50 r
a
ti
o of
da
ta
s
e
t
pa
r
ti
ti
on
C
l
a
s
s
i
f
i
c
a
t
i
on
m
e
t
hod
F
e
a
t
ur
e
s
e
l
e
c
t
i
on m
e
t
hods
a
nd t
he
i
r
f
e
a
t
ur
e
c
ount
TF
-
I
D
F
357
f
e
a
t
ur
e
s
C
hi
-
s
qua
r
e
457 f
e
a
t
ur
e
s
IG
453 f
e
a
t
ur
e
s
MI
360 f
e
a
t
ur
e
s
W
2V
340 f
e
a
t
ur
e
s
G
l
ove
405 f
e
a
t
ur
e
s
R
F
E
412 f
e
a
t
ur
e
s
R
L
P
I
49
f
e
a
t
ur
e
s
F
N
N
83.42
84.92
84.18
84.96
86.23
85.85
85.42
86.98
R
N
N
84.62
84.45
84.06
85.28
86.66
85.31
85.25
87.58
RBF
-
NN
83.62
83.28
84.94
85.46
86.10
86.26
85.16
86.72
G
R
U
86.03
85.69
85.85
87.17
86.86
86.80
86.50
87.17
L
S
T
M
87.43
86.40
87.24
86.00
88.26
87.99
87.93
88.89
T
a
bl
e
3
.
C
la
s
s
if
ic
a
ti
on pe
r
f
or
m
a
nc
e
f
or
60:
40 r
a
ti
o of
da
ta
s
e
t
pa
r
ti
ti
on
C
l
a
s
s
i
f
i
c
a
t
i
on
m
e
t
hod
F
e
a
t
ur
e
s
e
l
e
c
t
i
on m
e
t
hods
a
nd t
he
i
r
f
e
a
t
ur
e
c
ount
TF
-
I
D
F
357 f
e
a
t
ur
e
s
C
hi
-
s
qua
r
e
457 f
e
a
t
ur
e
s
IG
453 f
e
a
t
ur
e
s
MI
360 f
e
a
t
ur
e
s
W
2V
340 f
e
a
t
ur
e
s
G
l
ove
405
f
e
a
t
ur
e
s
R
F
E
412 f
e
a
t
ur
e
s
R
L
P
I
49
f
e
a
t
ur
e
s
F
N
N
85.85
85.13
86.28
86.19
87.61
87.28
86.31
88.74
R
N
N
86.12
86.38
86.46
87.11
88.66
87.53
86.99
88.93
RBF
-
NN
85.25
86.57
84.62
84.12
87.94
87.16
86.77
88.59
G
R
U
86.48
88.30
86.81
88.65
90.31
89.36
88.06
90.91
L
S
T
M
88.22
91.06
90.78
90.17
91.97
91.92
91.81
92.43
F
ir
s
tl
y,
th
e
obs
e
r
va
ti
ons
in
T
a
bl
e
s
2
a
nd
3
s
how
th
e
be
ha
vi
or
of
e
a
c
h
c
la
s
s
if
ic
a
ti
on
m
ode
l
w
it
h
va
r
io
us
s
e
t
of
f
e
a
tu
r
e
s
f
r
om
di
f
f
e
r
e
nt
f
e
a
tu
r
e
s
e
le
c
ti
on
m
e
th
ods
.
I
t
c
a
n
be
s
e
e
n
th
a
t
th
e
pe
r
f
or
m
a
nc
e
of
th
e
c
la
s
s
if
ic
a
ti
on
m
ode
ls
is
be
tt
e
r
w
he
n
pa
ir
e
d
w
it
h
R
L
P
I
,
de
s
pi
t
e
s
e
le
c
ti
ng
le
a
s
t
num
be
r
of
f
e
a
tu
r
e
s
in
a
s
e
t.
I
t
de
m
ons
tr
a
te
s
th
a
t
th
e
R
L
P
I
c
a
n
c
hoo
s
e
th
e
m
os
t
di
s
ti
nc
ti
ve
a
nd
pe
r
ti
ne
nt
f
e
a
tu
r
e
s
,
w
hi
le
k
e
e
pi
ng
th
e
f
e
a
tu
r
e
c
ount
l
ow
.
I
t
c
a
n
a
ls
o
be
s
e
e
n
f
r
om
th
e
a
bove
obs
e
r
va
ti
ons
th
a
t
ir
r
e
s
pe
c
ti
ve
of
num
be
r
of
f
e
a
tu
r
e
s
,
L
S
T
M
is
c
ons
is
te
nt
ly
pe
r
f
or
m
in
g
be
tt
e
r
th
a
n
ot
he
r
c
la
s
s
if
ic
a
ti
on
m
ode
ls
.
F
in
a
ll
y,
f
r
o
m
th
e
obs
e
r
va
ti
on,
it
c
a
n
be
not
e
d
th
a
t
th
e
R
L
P
I
a
nd
L
S
T
M
c
om
bi
na
ti
on
is
out
pe
r
f
or
m
in
g
ot
he
r
c
om
bi
na
ti
ons
ir
r
e
s
pe
c
ti
ve
of
tr
a
in
-
te
s
t
s
pl
it
r
a
ti
os
.
T
he
r
e
s
ul
ts
a
l
s
o
c
onf
ir
m
th
e
f
a
c
t
th
a
t
in
or
de
r
to
ha
ndl
e
s
e
que
nt
ia
l
da
ta
s
uc
h
a
s
te
xt
a
s
in
th
is
c
a
s
e
,
L
S
T
M
i
s
be
s
t
s
ui
te
d opti
on.
4.
C
O
N
C
L
U
S
I
O
N
I
n
t
hi
s
w
or
k,
w
e
a
na
l
yz
e
t
he
in
f
lu
e
nc
e
o
f
p
r
e
-
p
r
oc
e
s
s
in
g
te
c
h
ni
q
ue
s
.
M
a
in
ly
,
th
e
f
e
a
t
ur
e
s
e
le
c
ti
on
s
ta
ge
w
hi
c
h
is
in
te
nde
d
f
or
r
e
duc
in
g
th
e
d
im
e
ns
i
ona
li
ty
,
on
t
he
ov
e
r
a
ll
c
la
s
s
if
ic
a
ti
on
pe
r
f
or
m
a
nc
e
.
D
ur
in
g
th
e
e
xpe
r
i
m
e
nt
,
R
L
P
I
w
a
s
in
c
o
r
po
r
a
te
d
a
l
ong
w
i
th
va
r
io
us
f
e
a
tu
r
e
s
e
le
c
ti
on
te
c
h
ni
que
s
in
or
de
r
t
o
ob
ta
in
th
e
le
a
s
t
nu
m
be
r
o
f
m
os
t
r
e
le
va
nt
a
n
d
di
s
ti
nc
t
iv
e
s
e
t
o
f
f
e
a
t
ur
e
s
.
T
he
c
la
s
s
i
f
ic
a
t
io
n
pe
r
f
or
m
a
n
c
e
s
of
ne
u
r
a
l
ne
tw
or
k
-
ba
s
e
d
m
ode
ls
a
r
e
e
va
l
ua
t
e
d
a
ga
i
ns
t
m
i
ni
m
u
m
f
e
a
tu
r
e
s
e
ts
,
w
h
ic
h
a
r
e
ob
ta
i
ne
d
by
di
f
f
e
r
e
n
t
f
e
a
tu
r
e
s
e
le
c
t
io
n
m
e
th
ods
.
R
e
s
u
lt
s
s
h
ow
th
a
t
t
he
c
o
m
b
in
a
ti
on
of
R
L
P
I
i
n
i
ts
s
i
m
p
le
s
t
f
o
r
m
a
nd
L
S
T
M
o
ut
p
e
r
f
o
r
m
a
ll
th
e
ot
he
r
c
o
m
b
in
a
ti
ons
in
bo
th
f
e
a
tu
r
e
s
e
le
c
ti
on
a
nd
s
e
nt
im
e
nt
c
la
s
s
if
ic
a
ti
o
n
r
e
s
pe
c
t
iv
e
ly
.
T
he
r
e
s
u
lt
s
onc
e
a
g
a
in
a
f
f
i
r
m
th
e
f
a
c
t
th
a
t
t
h
e
L
S
T
M
is
one
a
m
o
ng
t
he
be
s
t
s
ui
te
d
m
ode
ls
f
or
h
a
nd
li
ng
s
e
q
ue
n
ti
a
l
da
ta
.
I
t
w
a
s
o
bs
e
r
ve
d
t
ha
t
,
t
he
va
r
ia
n
c
e
b
e
twe
e
n
m
i
ni
m
u
m
a
n
d
m
a
xi
m
um
nu
m
be
r
of
f
e
a
tu
r
e
s
w
a
s
a
lm
os
t
s
a
m
e
in
e
a
c
h
f
e
a
tu
r
e
s
e
le
c
ti
on
a
pp
r
oa
c
he
s
.
S
e
n
ti
m
e
n
t
c
la
s
s
i
f
ic
a
t
io
n
w
o
ul
d
be
ne
f
i
t
m
or
e
f
r
om
a
n
e
nha
nc
e
d
m
e
t
hod
f
o
r
ob
ta
i
ni
ng
t
he
id
e
a
l
nu
m
be
r
o
f
f
e
a
tu
r
e
s
w
h
i
le
ke
e
p
in
g
th
e
m
os
t
r
e
le
va
n
t
te
r
m
s
.
A
be
tt
e
r
di
m
e
ns
i
ona
li
ty
r
e
duc
ti
on
m
e
t
ho
d
is
a
ls
o
ne
e
de
d
,
w
h
ic
h
c
a
n
lo
w
e
r
th
e
f
in
a
l
d
im
e
ns
i
ona
li
ty
of
f
e
a
t
ur
e
s
w
hi
le
m
a
i
nt
a
in
in
g c
on
te
x
t.
F
U
N
D
I
N
G
I
N
F
O
R
M
A
T
I
O
N
T
hi
s
r
e
s
e
a
r
c
h
r
e
c
e
iv
e
d
no
s
pe
c
if
ic
gr
a
nt
f
r
om
a
ny
f
undi
ng
a
ge
nc
y
in
th
e
publ
ic
,
c
om
m
e
r
c
ia
l,
or
not
-
f
or
-
p
r
of
it
s
e
c
to
r
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
E
v
al
uat
in
g t
he
i
nf
lu
e
nc
e
of
f
e
at
ur
e
s
e
le
c
ti
on bas
e
d di
m
e
n
s
io
nal
it
y
…
(
G
ow
r
av
R
am
e
s
h B
abu K
is
hor
e
)
3373
A
U
T
H
O
R
C
O
N
T
R
I
B
U
T
I
O
N
S
S
T
A
T
E
M
E
N
T
T
hi
s
jo
ur
na
l
us
e
s
th
e
C
ont
r
ib
ut
or
R
ol
e
s
T
a
xonomy
(
C
R
e
di
T
)
to
r
e
c
ogni
z
e
in
di
vi
dua
l
a
ut
hor
c
ont
r
ib
ut
io
ns
, r
e
duc
e
a
ut
hor
s
hi
p di
s
put
e
s
,
a
nd f
a
c
il
it
a
te
c
ol
la
bo
r
a
ti
on.
N
am
e
o
f
A
u
t
h
or
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
G
ow
r
a
v
R
a
m
e
s
h
B
a
bu
K
is
hor
e
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
B
uka
ha
ll
y
S
om
a
s
he
ka
r
H
a
r
is
h
✓
✓
✓
✓
✓
✓
✓
✓
C
ha
lu
ve
gow
da
K
a
na
ka
la
ks
hm
i
R
oopa
✓
✓
✓
✓
✓
✓
✓
C
:
C
onc
e
pt
ua
l
i
z
a
t
i
on
M
:
M
e
t
hodol
ogy
So
:
So
f
t
w
a
r
e
Va
:
Va
l
i
da
t
i
on
Fo
:
Fo
r
m
a
l
a
na
l
ys
i
s
I
:
I
nve
s
t
i
ga
t
i
on
R
:
R
e
s
our
c
e
s
D
:
D
a
t
a
C
ur
a
t
i
on
O
:
W
r
i
t
i
ng
-
O
r
i
gi
na
l
D
r
a
f
t
E
:
W
r
i
t
i
ng
-
R
e
vi
e
w
&
E
di
t
i
ng
Vi
:
Vi
s
ua
l
i
z
a
t
i
on
Su
:
Su
pe
r
vi
s
i
on
P
:
P
r
oj
e
c
t
a
dm
i
ni
s
t
r
a
t
i
on
Fu
:
Fu
ndi
ng a
c
qui
s
i
t
i
on
C
O
N
F
L
I
C
T
O
F
I
N
T
E
R
E
S
T
S
T
A
T
E
M
E
N
T
A
ll
a
ut
hor
s
de
c
la
r
e
t
ha
t
th
e
y ha
ve
no
c
onf
li
c
ts
of
i
nt
e
r
e
s
t.
D
A
T
A
A
V
A
I
L
A
B
I
L
I
T
Y
D
a
ta
s
ha
r
in
g i
s
not
a
ppl
ic
a
bl
e
t
o t
hi
s
a
r
ti
c
le
a
s
no n
e
w
da
ta
w
e
r
e
c
r
e
a
te
d i
n t
hi
s
s
tu
dy.
RE
F
E
R
E
N
C
E
S
[
1]
V
.
D
ogr
a
,
A
.
S
i
ngh,
S
.
V
e
r
m
a
,
K
a
vi
t
a
,
N
.
Z
.
J
ha
nj
hi
,
a
nd
M
.
N
.
T
a
l
i
b,
“
U
nde
r
s
t
a
ndi
ng
of
da
t
a
pr
e
pr
oc
e
s
s
i
ng
f
or
di
m
e
ns
i
ona
l
i
t
y
r
e
duc
t
i
on
us
i
ng
f
e
a
t
ur
e
s
e
l
e
c
t
i
on
t
e
c
hni
que
s
i
n
t
e
xt
c
l
a
s
s
i
f
i
c
a
t
i
on,”
i
n
I
nt
e
l
l
i
ge
nt
C
om
put
i
ng
and
I
nnov
at
i
on
on
D
at
a
Sc
i
e
nc
e
,
S
i
nga
por
e
:
S
pr
i
nge
r
, 2021, pp. 455
–
464
, doi
:
10.1007/
978
-
981
-
16
-
3153
-
5_48.
[
2]
M
.
A
na
nda
r
a
j
a
n,
C
.
H
i
l
l
,
a
nd
T
.
N
ol
a
n,
P
r
ac
t
i
c
al
t
e
x
t
anal
y
t
i
c
s
:
m
ax
i
m
i
z
i
ng
t
he
v
al
ue
of
t
e
x
t
dat
a
.
C
ha
m
:
S
pr
i
nge
r
,
2019
,
doi
:
10.1007/
978
-
3
-
319
-
95663
-
3.
[
3]
S
.
V
i
j
a
ya
r
a
ni
,
J
.
I
l
a
m
a
t
hi
,
a
nd
N
i
t
hya
,
“
P
r
e
pr
oc
e
s
s
i
ng
t
e
c
hni
que
s
f
or
t
e
xt
m
i
ni
ng
-
a
n
ove
r
vi
e
w
,”
I
nt
e
r
nat
i
onal
J
our
nal
of
C
om
put
e
r
Sc
i
e
n
c
e
&
C
om
m
uni
c
at
i
on N
e
t
w
or
k
s
, vol
. 5, no. 1, pp. 7
–
16, 2015.
[
4]
Y
.
D
.
K
i
r
a
na
a
nd
S
.
A
l
F
a
r
a
by,
“
S
e
nt
i
m
e
nt
a
na
l
ys
i
s
of
be
a
ut
y
pr
oduc
t
r
e
vi
e
w
s
us
i
ng
t
he
K
-
ne
a
r
e
s
t
ne
i
ghbor
(
K
N
N
)
a
nd
T
F
-
I
D
F
m
e
t
hods
w
i
t
h
c
hi
-
s
qua
r
e
f
e
a
t
ur
e
s
e
l
e
c
t
i
on,”
J
ou
r
nal
of
D
at
a
Sc
i
e
nc
e
and
I
t
s
A
ppl
i
c
at
i
ons
,
vol
.
4,
no.
1,
pp.
31
–
42,
2021,
doi
:
10.34818/
J
D
S
A
.2021.4.71.
[
5]
L
.
H
.
P
a
t
i
l
a
nd
M
.
A
t
i
que
,
“
A
nove
l
a
ppr
oa
c
h
f
or
f
e
a
t
ur
e
s
e
l
e
c
t
i
on
m
e
t
hod
T
F
-
I
D
F
i
n
doc
um
e
nt
c
l
us
t
e
r
i
ng,”
i
n
2013
3r
d
I
E
E
E
I
nt
e
r
nat
i
onal
A
dv
anc
e
C
om
put
i
ng C
onf
e
r
e
n
c
e
(
I
A
C
C
)
, 2013, pp. 858
–
862
, doi
:
10.1109/
I
A
dC
C
.2013.6514339.
[
6]
S
.
Q
u,
S
.
W
a
ng,
a
nd
Y
.
Z
ou,
“
I
m
pr
ove
m
e
nt
of
t
e
xt
f
e
a
t
ur
e
s
e
l
e
c
t
i
on
m
e
t
hod
ba
s
e
d
on
T
F
I
D
F
,”
i
n
2008
I
nt
e
r
nat
i
onal
Se
m
i
nar
on
F
ut
ur
e
I
nf
or
m
at
i
on T
e
c
hnol
ogy
and M
anage
m
e
nt
E
ngi
ne
e
r
i
ng
, 2008, pp. 79
–
81
, doi
:
10.1109/
F
I
T
M
E
.2008.25.
[
7]
X
.
L
i
,
H
.
D
a
i
,
a
nd
M
.
W
a
ng,
“
T
w
o
-
s
t
a
ge
f
e
a
t
ur
e
s
e
l
e
c
t
i
on
m
e
t
hod
f
or
t
e
xt
c
l
a
s
s
i
f
i
c
a
t
i
on,”
i
n
2009
I
nt
e
r
nat
i
onal
C
onf
e
r
e
nc
e
on
M
ul
t
i
m
e
di
a I
nf
or
m
at
i
on N
e
t
w
or
k
i
ng and Se
c
u
r
i
t
y
, 2009, pp. 234
–
238
, doi
:
10.1109/
M
I
N
E
S
.2009.127.
[
8]
B
.
W
a
ng
a
nd
S
.
Z
ha
ng,
“
A
nove
l
f
e
a
t
ur
e
s
e
l
e
c
t
i
on
a
l
gor
i
t
hm
f
o
r
t
e
xt
c
l
a
s
s
i
f
i
c
a
t
i
on
ba
s
e
d
on
T
F
I
D
F
-
w
e
i
ght
a
nd
K
L
-
di
ve
r
ge
nc
e
,”
i
n
P
r
oc
e
e
di
ngs
of
t
he
11t
h J
oi
nt
I
nt
e
r
nat
i
onal
C
om
put
e
r
C
onf
e
r
e
nc
e
, 2005, pp.
438
–
441
, doi
:
10.1142/
9789812701534_0099.
[
9]
J
.
S
ong,
M
.
X
u,
a
nd
C
.
F
a
n,
“
A
t
e
xt
f
e
a
t
ur
e
s
e
l
e
c
t
i
on
m
e
t
hod
us
i
ng
T
F
I
D
F
ba
s
e
d
on
e
nt
r
opy,”
i
n
C
om
put
at
i
onal
I
nt
e
l
l
i
ge
nc
e
,
2010, pp. 962
–
967
, doi
:
10.1142/
9789814324700_0147.
[
10]
N
. S
. M
. N
a
f
i
s
a
nd S
. A
w
a
ng,
“
A
n e
nh
a
nc
e
d hybr
i
d f
e
a
t
ur
e
s
e
l
e
c
t
i
on t
e
c
hni
que
us
i
ng t
e
r
m
f
r
e
que
nc
y
-
i
nve
r
s
e
doc
um
e
nt
f
r
e
que
nc
y
a
nd
s
uppor
t
ve
c
t
or
m
a
c
hi
ne
-
r
e
c
ur
s
i
ve
f
e
a
t
ur
e
e
l
i
m
i
na
t
i
on
f
o
r
s
e
nt
i
m
e
nt
c
l
a
s
s
i
f
i
c
a
t
i
on,”
I
E
E
E
A
c
c
e
s
s
,
vol
.
9,
pp.
52177
–
52192,
2021, doi
:
10.1109/
A
C
C
E
S
S
.2021.3069001.
[
11]
C
.
L
e
e
a
nd
G
.
G
.
L
e
e
,
“
I
nf
or
m
a
t
i
on
ga
i
n
a
nd
di
ve
r
ge
nc
e
-
ba
s
e
d
f
e
a
t
ur
e
s
e
l
e
c
t
i
on
f
or
m
a
c
hi
ne
l
e
a
r
ni
ng
-
ba
s
e
d
t
e
xt
c
a
t
e
gor
i
z
a
t
i
on,”
I
nf
or
m
at
i
on P
r
oc
e
s
s
i
ng and M
anage
m
e
nt
, vol
. 42, no. 1, pp. 155
–
165, 2006, do
i
:
10.1016/
j
.i
pm
.2004.08.006.
[
12]
S
.
L
e
i
,
“
A
f
e
a
t
ur
e
s
e
l
e
c
t
i
on
m
e
t
hod
b
a
s
e
d
on
i
nf
or
m
a
t
i
on
ga
i
n
a
nd
ge
n
e
t
i
c
a
l
gor
i
t
hm
,”
i
n
2012
I
nt
e
r
nat
i
onal
C
onf
e
r
e
nc
e
on
C
om
put
e
r
Sc
i
e
n
c
e
and E
l
e
c
t
r
oni
c
s
E
ngi
ne
e
r
i
ng
, 2012, pp. 355
–
358
, doi
:
10.1109/
I
C
C
S
E
E
.2012.97.
[
13]
C
.
S
ha
ng,
M
.
L
i
,
S
.
F
e
ng,
Q
.
J
i
a
ng,
a
nd
J
.
F
a
n,
“
F
e
a
t
ur
e
s
e
l
e
c
t
i
on
vi
a
m
a
xi
m
i
z
i
ng
gl
oba
l
i
nf
or
m
a
t
i
on
ga
i
n
f
or
t
e
xt
c
l
a
s
s
i
f
i
c
a
t
i
on,”
K
now
l
e
dge
-
B
as
e
d Sy
s
t
e
m
s
, vol
. 54, pp. 298
–
309, 2013, doi
:
10.1016/
j
.knos
y
s
.2
013.09.019.
[
14]
R
.
B
.
P
e
r
e
i
r
a
,
A
.
P
l
a
s
t
i
no,
B
.
Z
a
dr
oz
ny,
a
nd
L
.
H
.
C
.
M
e
r
s
c
hm
a
nn,
“
I
n
f
or
m
a
t
i
on
ga
i
n
f
e
a
t
ur
e
s
e
l
e
c
t
i
on
f
or
m
ul
t
i
-
l
a
be
l
c
l
a
s
s
i
f
i
c
a
t
i
on,”
J
ou
r
nal
of
I
nf
or
m
at
i
on and D
at
a M
anage
m
e
nt
, vol
. 6, no. 1, pp. 48
–
48, 2015.
[
15]
E
.
O
.
O
m
uya
,
G
.
O
.
O
ke
yo,
a
nd
M
.
W
.
K
i
m
w
e
l
e
,
“
F
e
a
t
ur
e
s
e
l
e
c
t
i
on
f
or
c
l
a
s
s
i
f
i
c
a
t
i
on
us
i
ng
pr
i
nc
i
pa
l
c
om
pone
nt
a
na
l
y
s
i
s
a
nd
i
nf
or
m
a
t
i
on ga
i
n,”
E
x
pe
r
t
Sy
s
t
e
m
s
w
i
t
h A
ppl
i
c
at
i
ons
, vol
. 174, 2021, doi
:
10.10
16/
j
.e
s
w
a
.2021.114765.
[
16]
Y
.
Z
ha
i
,
W
.
S
ong,
X
.
L
i
u,
L
.
L
i
u,
a
nd
X
.
Z
ha
o,
“
A
c
hi
-
s
qua
r
e
s
t
a
t
i
s
t
i
c
s
b
a
s
e
d
f
e
a
t
ur
e
s
e
l
e
c
t
i
on
m
e
t
hod
i
n
t
e
xt
c
l
a
s
s
i
f
i
c
a
t
i
on,”
i
n
2018
I
E
E
E
9t
h
I
nt
e
r
nat
i
onal
C
onf
e
r
e
nc
e
on
Sof
t
w
ar
e
E
ngi
ne
e
r
i
ng
and
Se
r
v
i
c
e
Sc
i
e
nc
e
(
I
C
SE
SS)
,
2018,
pp.
160
–
163
,
doi
:
10.1109/
I
C
S
E
S
S
.2018.8663882.
[
17]
C
.
J
i
n
e
t
al
.
,
“
C
hi
-
s
qua
r
e
s
t
a
t
i
s
t
i
c
s
f
e
a
t
ur
e
s
e
l
e
c
t
i
on
ba
s
e
d
on
t
e
r
m
f
r
e
que
n
c
y
a
nd
di
s
t
r
i
but
i
on
f
o
r
t
e
xt
c
a
t
e
gor
i
z
a
t
i
on,”
I
E
T
E
J
our
nal
of
R
e
s
e
ar
c
h
, vol
. 61, no. 4, pp. 351
–
362, 2015, doi
:
10.1080/
03772063.2015.1021385.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
, V
ol
.
14
, N
o.
4
,
A
ugus
t
2025
:
3366
-
3374
3374
[
18]
Y
.
H
.
L
i
,
“
T
e
xt
f
e
a
t
ur
e
s
e
l
e
c
t
i
on
a
l
gor
i
t
hm
ba
s
e
d
on
c
hi
-
s
qua
r
e
r
a
nk
c
or
r
e
l
a
t
i
on
f
a
c
t
or
i
z
a
t
i
on,”
J
our
nal
of
I
nt
e
r
di
s
c
i
pl
i
nar
y
M
at
he
m
at
i
c
s
, vol
. 20, no. 1, pp. 153
–
160, 2017, doi
:
10.1080/
09720502.2016.1259769.
[
19]
A
.
W
.
H
a
r
ya
nt
o,
E
.
K
.
M
a
w
a
r
di
,
a
nd
M
ul
j
ono,
“
I
nf
l
ue
nc
e
of
w
or
d
no
r
m
a
l
i
z
a
t
i
on
a
nd
c
hi
-
s
qua
r
e
d
f
e
a
t
ur
e
s
e
l
e
c
t
i
on
on
s
uppor
t
ve
c
t
or
m
a
c
hi
ne
(
S
V
M
)
t
e
xt
c
l
a
s
s
i
f
i
c
a
t
i
on,”
i
n
2018
I
nt
e
r
nat
i
onal
Se
m
i
nar
o
n
A
ppl
i
c
at
i
on
f
or
T
e
c
hnol
ogy
of
I
nf
or
m
at
i
on
and
C
om
m
uni
c
at
i
on
, 2018, pp. 229
–
233
, doi
:
10.1109/
I
S
E
M
A
N
T
I
C
.2018.8549748.
[
20]
İ
.
S
e
l
,
A
.
K
a
r
c
i
,
a
nd
D
.
H
a
nba
y,
“
F
e
a
t
ur
e
s
e
l
e
c
t
i
on
f
or
t
e
xt
c
l
a
s
s
i
f
i
c
a
t
i
on
us
i
ng
m
ut
ua
l
i
nf
or
m
a
t
i
on,”
i
n
2019
I
nt
e
r
nat
i
onal
A
r
t
i
f
i
c
i
al
I
nt
e
l
l
i
ge
nc
e
and D
at
a P
r
oc
e
s
s
i
ng Sy
m
po
s
i
um
(
I
D
A
P
)
, 2019, pp. 1
–
4
,
doi
:
10.1109/
I
D
A
P
.2019.8875927.
[
21]
H
.
L
i
u,
J
.
S
un,
L
.
L
i
u,
a
nd
H
.
Z
ha
ng,
“
F
e
a
t
ur
e
s
e
l
e
c
t
i
on
w
i
t
h
dyna
m
i
c
m
ut
ua
l
i
nf
or
m
a
t
i
on,”
P
at
t
e
r
n
R
e
c
ogni
t
i
on
,
vol
.
42,
no.
7,
pp. 1330
–
1339, 2009, doi
:
10.1016/
j
.pa
t
c
og.2008.10.028.
[
22]
D
.
A
gni
hot
r
i
,
K
.
V
e
r
m
a
,
a
nd P
.
T
r
i
pa
t
hi
,
“
M
ut
ua
l
i
n
f
or
m
a
t
i
on
us
i
ng
s
a
m
pl
e
va
r
i
a
nc
e
f
or
t
e
xt
f
e
a
t
ur
e
s
e
l
e
c
t
i
on,”
i
n
P
r
oc
e
e
di
ngs
of
t
he
3r
d
I
nt
e
r
nat
i
onal
C
onf
e
r
e
n
c
e
on
C
om
m
uni
c
at
i
on
and
I
nf
or
m
at
i
on
P
r
oc
e
s
s
i
ng
,
2017,
pp.
39
–
44
,
doi
:
10.1145/
3162957.3163054.
[
23]
X
.
D
i
ng
a
nd
Y
.
T
a
ng,
“
I
m
pr
ove
d
m
ut
ua
l
i
nf
or
m
a
t
i
on
m
e
t
hod
f
or
t
e
xt
f
e
a
t
ur
e
s
e
l
e
c
t
i
on,”
i
n
2013
8t
h
I
nt
e
r
nat
i
onal
C
onf
e
r
e
nc
e
o
n
C
om
put
e
r
Sc
i
e
n
c
e
&
E
duc
at
i
on
, 2013, pp. 163
–
166
, doi
:
10.1109/
I
C
C
S
E
.2013.
6553903.
[
24]
H
.
K
.
D
a
r
s
ha
n,
A
.
R
.
S
ha
nka
r
,
B
.
S
.
H
a
r
i
s
h,
a
nd
K
.
H
.
M
.
K
um
a
r
,
“
E
xpl
oi
t
i
ng
R
L
P
I
f
or
s
e
nt
i
m
e
nt
a
na
l
y
s
i
s
on
m
ovi
e
r
e
vi
e
w
s
,
”
J
our
nal
of
A
dv
anc
e
s
i
n I
nf
or
m
at
i
on T
e
c
hnol
ogy
, vol
. 10, no. 1, pp. 14
–
19, 2019
, doi
:
10.12720/
j
a
i
t
.10.1.14
-
19.
[
25]
M
.
B
.
R
e
va
na
s
i
dda
ppa
,
B
.
S
.
H
a
r
i
s
h,
a
nd
S
.
V
.
A
.
K
um
a
r
,
“
M
e
t
a
-
c
ogni
t
i
ve
ne
ur
a
l
ne
t
w
or
k
ba
s
e
d
s
e
que
nt
i
a
l
l
e
a
r
ni
ng
f
r
a
m
e
w
or
k
f
or
t
e
xt
c
a
t
e
gor
i
z
a
t
i
on,”
P
r
oc
e
di
a C
om
put
e
r
Sc
i
e
n
c
e
, vol
. 132, pp. 1503
–
1511,
2018, doi
:
10.1016/
j
.pr
oc
s
.2018.05.104.
[
26]
M
. L
a
n, C
.
L
.
T
a
n, J
. S
u, a
nd Y
.
L
u,
“
S
upe
r
vi
s
e
d
a
nd t
r
a
di
t
i
ona
l
t
e
r
m
w
e
i
ght
i
n
g m
e
t
hods
f
or
a
ut
om
a
t
i
c
t
e
xt
c
a
t
e
gor
i
z
a
t
i
on,”
I
E
E
E
T
r
ans
ac
t
i
ons
on P
at
t
e
r
n A
nal
y
s
i
s
and M
ac
hi
ne
I
nt
e
l
l
i
ge
nc
e
, vol
. 31, no. 4, pp. 721
–
735, 2009, doi
:
10.1109/
T
P
A
M
I
.2008.110.
[
27]
D
.
C
a
i
,
X
.
H
e
,
W
.
V
.
Z
ha
ng,
a
nd
J
.
H
a
n,
“
R
e
gul
a
r
i
z
e
d
l
oc
a
l
i
t
y
pr
e
s
e
r
vi
ng
i
nde
xi
ng
vi
a
s
pe
c
t
r
a
l
r
e
gr
e
s
s
i
on,”
i
n
P
r
oc
e
e
di
ngs
of
t
he
s
i
x
t
e
e
nt
h
A
C
M
c
onf
e
r
e
nc
e
on
C
onf
e
r
e
n
c
e
on
i
nf
or
m
at
i
on
and
k
now
l
e
dge
m
anage
m
e
nt
,
2007,
pp.
741
–
750
,
doi
:
10.1145/
1321440.1321544.
[
28]
I
.
M
okr
i
š
a
nd
L
.
S
kova
j
s
ová
,
“
F
e
e
d
-
f
or
w
a
r
d
a
nd
s
e
l
f
-
or
ga
ni
z
i
ng
ne
ur
a
l
ne
t
w
or
ks
f
or
t
e
xt
doc
um
e
nt
r
e
t
r
i
e
va
l
,”
A
c
t
a
E
l
e
c
t
r
ot
e
c
hni
c
a e
t
I
nf
or
m
at
i
c
a
, vol
. 8, no. 2, pp. 3
–
10, 2008.
[
29]
Z
.
W
a
ng,
Y
.
H
e
,
a
nd
M
.
J
i
a
ng,
“
A
c
om
pa
r
i
s
on
a
m
ong
t
hr
e
e
n
e
ur
a
l
ne
t
w
or
ks
f
or
t
e
xt
c
l
a
s
s
i
f
i
c
a
t
i
on,”
i
n
2006
8t
h
i
nt
e
r
nat
i
onal
C
onf
e
r
e
nc
e
on Si
gnal
P
r
oc
e
s
s
i
ng
, 2006
, doi
:
10.1109/
I
C
O
S
P
.2006.345923.
B
I
O
G
R
A
P
H
I
E
S
O
F
A
U
T
H
O
R
S
Gowrav
Ramesh
Babu
Kishore
received
his
B.E
.
degree
in
infor
mation
science
and engin
eering
from Maharaja Ins
titute o
f Technology,
Mys
uru
, India. and M.Tech
.
degree in
data
scienc
e
from
the
Department
of
Information
Science
and
Engin
eering,
JSS
Science
an
d
Technology
University,
India.
Presently
he
is
a
r
esearch
scholar
in
the
Department
of
Information
Science
and
Engineering,
JSS
Science
and
Technology
University,
India.
He
can
be contacted at email: kkishorkumar12@gmail.com or kishore_gr@
jssstuniv.in.
Bukahally
Somashekar
Harish
obtained
his
Ph.D.
in
computer
scien
ce
from
University
of
Mysore,
India.
Presently
he
is
working
as
a
Professo
r
in
the
Department
of
Information
Science
and
Engineering,
JSS
Science
and
Technology
U
niversity,
India.
He
wa
s
a
visiting
researcher
at
DIBRIS
-
Department
of
Informatics,
Bio
En
gineering,
Robotics
and
System
Engineering,
University
of
Genova,
Italy.
He
has
been
invited
as
a
resource
person
to
deliver
various
technical
talks
on
data
mining,
image
processing,
patt
ern
recognition,
and
soft
computi
ng
.
He
is
serving
as
a
reviewer
for
internatio
nal
conferenc
es
and
journals
.
He
has
published
articles
in
more
than
100+
i
nternational
reputed
peer
reviewed
journals
and
conferences
proceedings
.
He
successful
ly
executed
AICTE
-
RPS
project,
which
was
sanctioned
by
AICTE,
Government
of
India.
His
area
of
interest
inc
ludes
machine
learning,
text
mining,
and
computational
intelligence
.
He
can
be
contacted
at
email:
bsharish@
jssstuniv.in.
Chaluvegowd
a
Kanakala
kshmi
Roopa
received
her
B.E
.
degree
in
information
science
and
engineering
and
M.Tech
.
degree
in
computer
engineer
ing
from
Visvesvaraya
Technological
University,
Belagavi,
Karnataka,
India.
She
com
pleted
her
Ph.D.
from
University
of
Mysore,
India.
She
is
currently
working as
an
associate
professor
at
JSS
Science
and Technol
ogy Uni
versity.
She is
serving
as reviewer
for many
conferences an
d journal
s. She
is
a
lifetime
member
of
ISTE
and
CSI.
Her
area
of
research
includes
medical
image
analysis,
biometrics, and text mining
. She can be contacted a
t email: ckr@
jssstuniv.in.
Evaluation Warning : The document was created with Spire.PDF for Python.