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
.
1
,
Feb
r
u
ar
y
20
25
,
p
p
.
980
~
994
I
SS
N:
2088
-
8
7
0
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijece.
v
15
i
1
.
pp
9
8
0
-
9
9
4
980
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ec
e.
ia
esco
r
e.
co
m
Product
reviews a
na
ly
sis
t
o
ex
t
ra
ct
sentimen
tal insig
hts wi
th
cla
ss
confidence
r
a
te
using
self
-
o
rg
a
nizing
ma
p neur
a
l net
wo
rk
Sa
ra
Ahs
a
in,
Ya
s
y
n E
ly
u
s
ufi
,
M
’
ha
m
ed
Ait
K
bir
I
n
t
e
l
l
i
g
e
n
t
A
u
t
o
m
a
t
i
o
n
a
n
d
B
i
o
M
e
d
G
e
n
o
mi
c
s
L
a
b
o
r
a
t
o
r
y
(
I
A
B
L)
,
T
e
c
h
n
i
c
a
l
S
c
i
e
n
c
e
s
a
n
d
M
e
d
i
c
a
l
S
c
i
e
n
c
e
s (S
T
S
M
)
D
o
c
t
o
r
a
l
C
e
n
t
e
r
,
A
b
d
e
l
m
a
l
e
k
Essa
a
d
i
U
n
i
v
e
r
s
i
t
y
,
Té
t
o
u
a
n
,
M
o
r
o
c
c
o
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Ap
r
2
2
,
2
0
2
4
R
ev
is
ed
Au
g
1
4
,
2
0
2
4
Acc
ep
ted
Au
g
2
0
,
2
0
2
4
Cu
sto
m
e
r
d
a
ta
a
n
a
ly
sis
h
e
lp
s
c
o
m
p
a
n
ies
to
u
n
d
e
rsta
n
d
c
u
st
o
m
e
r
in
ten
ti
o
n
s
a
n
d
b
e
h
a
v
io
rs
b
e
tt
e
r.
T
h
is
stu
d
y
i
n
tro
d
u
c
e
s
a
n
a
n
a
ly
sis
o
f
p
r
o
d
u
c
t
r
e
v
iew
s
to
h
e
lp
m
a
n
a
g
e
rs
a
d
o
p
t
a
m
o
re
e
fficie
n
t
stra
teg
y
to
e
x
trac
t
v
a
lu
a
b
le
k
n
o
wle
d
g
e
a
n
d
h
e
l
p
d
e
tec
t
se
g
m
e
n
t
o
f
c
u
s
to
m
e
rs
th
a
t
n
e
e
d
a
s
p
e
c
ial
a
tt
e
n
ti
o
n
a
n
d
p
ro
d
u
c
ts
t
h
a
t
n
e
e
d
imp
r
o
v
e
m
e
n
t
o
r
wit
h
t
h
e
m
o
st
imp
a
c
t.
T
h
e
u
se
d
d
a
tas
e
t
is
a
se
t
o
f
Am
a
z
o
n
re
v
iew
s
d
iv
id
e
d
in
to
m
u
lt
i
p
le
c
a
teg
o
ries
;
e
a
c
h
re
v
iew
h
a
s
a
targ
e
t
c
o
l
u
m
n
c
a
ll
e
d
‘
o
v
e
ra
ll
’
th
a
t
tak
e
s
a
v
a
l
u
e
b
e
twe
e
n
1
a
n
d
5
(
c
u
sto
m
e
r'
s
sa
ti
sfa
c
ti
o
n
).
Ba
se
d
o
n
th
e
‘
o
v
e
ra
ll
’
c
o
lu
m
n
,
m
u
lt
ip
le
lab
e
li
n
g
m
e
t
h
o
d
s
h
a
v
e
b
e
e
n
u
se
d
a
n
d
c
o
m
p
a
re
d
to
g
e
t
a
b
in
a
ry
tar
g
e
t
v
a
riab
le,
p
o
siti
v
e
o
r
n
e
g
a
ti
v
e
,
th
a
t
a
ffe
c
ts
a
c
las
s
to
a
re
v
iew
.
T
h
is
d
a
tas
e
t
c
o
n
tai
n
s
m
o
re
t
h
a
n
o
n
e
m
il
li
o
n
re
v
iew
s
a
n
d
c
a
n
g
i
v
e
c
o
m
p
a
n
ie
s
g
re
a
t
in
sig
h
t
i
n
to
p
ro
d
u
c
ts’
q
u
a
li
ty
a
n
d
c
u
sto
m
e
rs’
re
ten
ti
o
n
.
T
h
is
wo
rk
h
a
s
m
a
teria
li
z
e
d
b
y
u
si
n
g
c
u
sto
m
e
r
se
g
m
e
n
tatio
n
a
n
d
c
o
m
p
e
ti
ti
v
e
l
e
a
rn
in
g
wi
th
se
lf
-
o
rg
a
n
izi
n
g
m
a
p
(
S
OM
)
M
o
d
e
l
a
n
d
a
d
o
p
ti
n
g
a
n
e
w
a
p
p
ro
a
c
h
to
e
x
p
lo
re
t
h
e
g
e
n
e
ra
ted
n
e
t
wo
rk
/ma
p
,
it
is
b
a
se
d
o
n
c
lu
ste
rin
g
a
n
d
m
a
p
n
o
d
e
s
lab
e
ll
i
n
g
u
si
n
g
a
m
a
jo
ri
ty
v
o
t
in
g
p
ro
c
e
ss
.
Th
e
re
su
lt
s
s
h
o
w
th
a
t
t
h
e
p
r
o
p
o
se
d
d
u
a
l
a
p
p
r
o
a
c
h
c
o
m
b
in
i
n
g
th
e
p
rio
r
k
n
o
wle
d
g
e
,
re
late
d
t
o
su
p
e
r
v
ise
d
lea
rn
in
g
,
a
n
d
t
h
e
c
o
m
p
e
ti
t
iv
e
lea
rn
in
g
a
b
il
it
ies
e
n
h
a
n
c
e
s th
e
S
OM
m
o
d
e
l’s
c
a
p
a
b
il
it
ies
.
K
ey
w
o
r
d
s
:
C
o
m
p
etitiv
e
lear
n
in
g
C
u
s
to
m
er
b
eh
av
io
r
Per
s
o
n
alize
d
m
ar
k
etin
g
Pro
d
u
ct
r
ec
o
m
m
en
d
atio
n
s
Self
-
o
r
g
an
izin
g
m
ap
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
:
Ah
s
ain
Sar
a
I
n
t
e
l
l
i
g
e
n
t
A
u
t
o
m
a
t
i
o
n
a
n
d
B
i
o
M
e
d
G
e
n
o
m
i
cs
L
a
b
o
r
a
t
o
r
y
(
I
AB
L
)
,
T
e
c
h
n
i
c
al
S
ci
e
n
c
es
a
n
d
M
e
d
i
c
a
l
S
c
i
e
n
ce
s
(
S
T
S
M
)
D
o
c
t
o
r
a
l
C
e
n
t
e
r
,
A
b
d
e
l
m
a
l
e
k
E
s
s
aa
d
i
U
n
i
v
e
r
s
i
t
y
T
éto
u
an
,
M
o
r
o
cc
o
E
m
ail: sar
a.
ah
s
ain
@
etu
.
u
ae
.
ac
.
m
a
1.
I
NT
RO
D
UCT
I
O
N
C
u
s
t
o
m
e
r
a
n
a
l
y
s
is
is
b
e
c
o
m
i
n
g
a
n
i
n
c
r
e
as
i
n
g
l
y
c
o
m
m
o
n
s
o
u
r
c
e
o
f
i
n
f
o
r
m
a
t
i
o
n
.
Al
s
o
,
s
t
r
e
n
g
t
h
e
n
i
n
g
l
o
n
g
-
t
e
r
m
c
u
s
t
o
m
e
r
r
e
l
a
ti
o
n
s
h
i
p
s
t
h
r
o
u
g
h
i
n
d
i
v
i
d
u
a
l
i
z
e
d
e
-
c
o
m
m
e
r
c
e
o
f
f
e
r
s
w
i
l
l
d
e
p
e
n
d
o
n
t
h
e
c
o
m
p
a
n
y
’
s
a
b
i
l
i
t
y
t
o
u
s
e
c
u
s
t
o
m
e
r
d
a
t
a
t
o
d
e
v
e
l
o
p
a
n
d
c
o
n
t
r
o
l
c
u
s
t
o
m
e
r
i
n
t
e
r
a
c
t
i
o
n
s
.
I
n
t
h
e
s
a
m
e
c
o
n
t
e
x
t
,
t
h
e
r
e
v
o
l
u
t
i
o
n
o
f
p
r
e
d
i
c
t
i
v
e
a
n
a
l
y
t
ic
s
c
o
n
s
i
s
t
s
o
f
e
x
p
l
o
i
t
i
n
g
t
h
e
m
a
s
s
o
f
d
a
t
a
n
e
w
l
y
a
v
a
i
l
a
b
l
e
t
o
es
t
a
b
li
s
h
p
r
e
d
i
c
t
i
v
e
h
y
p
o
t
h
e
s
es
f
r
o
m
p
r
e
v
i
o
u
s
c
u
s
t
o
m
e
r
b
e
h
av
i
o
r
a
n
d
i
n
t
e
r
ac
t
i
o
n
s
.
D
e
v
el
o
p
i
n
g
r
o
b
u
s
t
s
e
n
t
i
m
e
n
t
a
n
al
y
s
is
m
o
d
e
ls
t
a
i
l
o
r
e
d
t
o
p
r
o
d
u
c
t
r
e
v
i
e
w
s
i
s
c
r
u
c
i
al
t
o
e
n
h
a
n
c
i
n
g
c
u
s
t
o
m
e
r
l
o
y
a
l
t
y
a
n
d
p
r
o
d
u
c
t
q
u
a
l
i
t
y
.
T
h
e
s
e
m
o
d
e
l
s
a
r
e
i
n
v
a
l
u
a
b
l
e
t
o
o
l
s
f
o
r
u
n
d
e
r
s
t
a
n
d
i
n
g
c
u
s
t
o
m
e
r
s
e
n
t
i
m
e
n
ts
a
n
d
i
m
p
r
o
v
i
n
g
p
r
o
d
u
c
t
q
u
a
l
i
t
y
.
V
a
r
i
o
u
s
s
ta
t
is
ti
ca
l
a
n
d
d
a
t
a
m
i
n
i
n
g
t
e
c
h
n
i
q
u
e
s
h
a
v
e
b
e
e
n
e
x
t
e
n
s
iv
e
l
y
r
e
s
e
a
r
c
h
e
d
i
n
l
i
te
r
a
t
u
r
e
t
o
c
r
a
f
t
p
r
e
d
i
c
t
i
v
e
m
o
d
e
l
s
t
o
l
e
v
e
r
a
g
e
i
n
s
i
g
h
ts
e
x
t
r
a
c
t
e
d
f
r
o
m
c
u
s
t
o
m
e
r
f
e
e
d
b
a
c
k
a
n
d
i
d
e
n
t
i
f
y
b
o
t
h
s
t
r
e
n
g
th
s
a
n
d
a
r
e
a
s
t
h
at
r
e
q
u
i
r
e
e
n
h
a
n
c
e
m
e
n
t
t
o
e
n
s
u
r
e
d
e
e
p
e
r
c
u
s
t
o
m
e
r
e
n
g
a
g
e
m
e
n
t
a
n
d
t
o
a
u
g
m
e
n
t
o
v
e
r
a
l
l
p
r
o
d
u
c
t
s
at
i
s
f
a
c
t
i
o
n
.
T
h
e
m
a
c
h
i
n
e
l
ea
r
n
i
n
g
a
n
d
a
r
t
i
f
i
c
i
al
i
n
t
e
l
li
g
e
n
c
e
l
a
n
d
s
c
a
p
e
is
v
a
s
t
,
th
e
s
a
m
e
as
t
h
a
t
o
f
t
h
e
g
r
o
wi
n
g
f
i
e
l
d
o
f
c
u
s
t
o
m
e
r
a
n
al
y
s
is
.
As
b
u
s
in
ess
es
ex
p
lo
r
e
t
h
e
p
o
s
s
ib
ilit
ie
s
o
f
in
te
g
r
atin
g
th
e
s
e
tech
n
o
lo
g
ies,
lev
er
ag
i
n
g
m
o
d
els
to
s
tr
en
g
th
en
cu
s
to
m
er
r
elatio
n
s
h
ip
s
an
d
p
r
o
v
id
e
p
e
r
s
o
n
ali
ze
d
e
-
co
m
m
er
ce
ex
p
er
ien
ce
s
b
ec
o
m
es
cr
u
cial.
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
P
r
o
d
u
ct
r
ev
iew
s
a
n
a
lysi
s
to
ex
tr
a
ct
s
en
timen
ta
l in
s
ig
h
ts
w
ith
cla
s
s
co
n
fid
en
ce
…
(
S
a
r
a
A
h
s
a
in
)
981
E
-
co
m
m
er
ce
lan
d
s
ca
p
e
k
ee
p
s
g
r
o
win
g
b
y
u
s
in
g
a
d
v
an
ce
d
ar
tific
ial
in
tellig
en
ce
(
AI
)
tec
h
n
iq
u
es
s
u
ch
as
s
elf
-
o
r
g
an
izin
g
m
ap
(
SOM
)
m
o
d
els
f
o
r
s
en
tim
en
t
an
aly
s
is
r
ep
r
esen
ts
a
cr
itical
f
r
o
n
tier
in
u
n
d
er
s
tan
d
in
g
an
d
r
esp
o
n
d
in
g
to
cu
s
to
m
er
n
ee
d
s
.
T
h
is
u
n
s
u
p
er
v
is
ed
n
eu
r
al
n
etwo
r
k
m
o
d
el
is
a
p
r
o
m
is
in
g
s
o
lu
tio
n
f
o
r
a
d
ee
p
er
u
n
d
er
s
tan
d
i
n
g
o
f
cu
s
to
m
er
r
ev
iews
b
y
f
ac
ilit
atin
g
th
e
d
i
s
co
v
er
y
o
f
p
atter
n
s
an
d
co
r
r
elatio
n
s
th
at
m
ig
h
t
u
n
co
v
e
r
ex
citin
g
i
n
s
ig
h
ts
in
th
e
d
ata.
Su
p
er
v
is
ed
lear
n
in
g
m
o
d
els,
lik
e
class
if
icatio
n
,
r
eg
r
ess
io
n
,
o
r
en
s
em
b
le
lear
n
i
n
g
m
eth
o
d
s
,
h
av
e
b
ec
o
m
e
estab
lis
h
ed
to
o
ls
f
o
r
l
ev
er
ag
in
g
lab
eled
d
ata
t
o
m
ak
e
p
r
ed
ictio
n
s
o
r
u
n
c
o
v
er
r
elatio
n
s
h
ip
s
.
Su
p
er
v
is
ed
lear
n
in
g
o
f
f
er
s
a
r
an
g
e
o
f
b
en
ef
its
d
u
e
to
its
v
er
s
atility
,
as
i
t
in
clu
d
es
alg
o
r
ith
m
s
d
esig
n
e
d
to
tack
le
d
if
f
e
r
en
t
task
s
(
b
in
ar
y
class
if
icatio
n
/m
u
lti
-
class
cla
s
s
if
icatio
n
)
.
T
h
is
ty
p
e
o
f
m
o
d
el,
alth
o
u
g
h
p
r
ac
tica
l,
h
ea
v
ily
d
e
p
en
d
s
o
n
h
av
in
g
lab
eled
tr
ain
in
g
d
ata
av
ailab
le.
T
h
e
s
u
cc
ess
o
f
th
is
ap
p
r
o
ac
h
r
elies
o
n
th
e
q
u
ality
an
d
r
ep
r
esen
tativ
en
ess
o
f
th
e
d
ata
u
s
ed
f
o
r
tr
ai
n
in
g
.
On
th
e
o
th
e
r
h
an
d
,
u
n
s
u
p
er
v
is
ed
lear
n
in
g
tec
h
n
iq
u
es,
s
u
ch
as
clu
s
ter
in
g
a
n
d
d
im
e
n
s
io
n
ality
r
ed
u
ctio
n
,
o
f
f
er
a
r
an
g
e
o
f
ap
p
r
o
ac
h
es
th
at
h
elp
id
e
n
tify
p
a
tter
n
s
,
s
tr
u
ctu
r
es,
o
r
r
elatio
n
s
h
ip
s
f
r
o
m
d
ata.
T
h
is
q
u
ality
m
ak
es
it
ex
ce
p
tio
n
ally
s
k
illed
in
r
ev
ea
lin
g
co
n
ce
al
ed
in
s
ig
h
ts
,
r
ec
o
g
n
izin
g
clu
s
t
er
s
,
an
d
s
im
p
lify
in
g
d
ata.
T
h
is
alg
o
r
ith
m
f
in
d
s
a
p
p
licatio
n
in
d
o
m
ain
s
s
u
ch
as
an
o
m
aly
d
etec
tio
n
,
p
atter
n
r
ec
o
g
n
itio
n
,
an
d
ex
p
lo
r
ato
r
y
d
ata
an
aly
s
is
.
Alth
o
u
g
h
u
n
s
u
p
er
v
is
ed
alg
o
r
ith
m
s
h
av
e
b
e
n
ef
its
,
ev
alu
atin
g
th
e
ir
p
er
f
o
r
m
an
ce
ca
n
b
e
ch
allen
g
i
n
g
b
ec
a
u
s
e
th
ey
r
eq
u
ir
e
n
u
m
er
ical
i
n
p
u
t
a
n
d
h
av
e
n
o
clea
r
p
er
f
o
r
m
an
ce
m
etr
ics.
Fu
r
th
er
m
o
r
e
,
in
ter
p
r
etin
g
t
h
e
r
esu
lts
ca
n
b
e
co
m
p
lex
s
in
ce
th
ese
alg
o
r
ith
m
s
id
en
tify
p
atter
n
s
with
o
u
t g
u
i
d
an
ce
.
Mo
d
els
b
ased
o
n
ar
tific
ial
n
eu
r
al
n
etwo
r
k
s
(
ANN)
h
a
v
e
also
s
ee
n
g
r
ea
t
s
u
cc
ess
;
f
o
r
i
n
s
tan
ce
,
a
m
u
ltil
ay
er
p
e
r
ce
p
tr
o
n
(
ML
P)
i
s
a
s
u
p
er
v
is
ed
lear
n
in
g
alg
o
r
it
h
m
th
at
is
a
f
ee
d
f
o
r
war
d
ty
p
e
o
f
n
eu
r
al
n
etwo
r
k
,
wh
ich
m
ea
n
s
th
at
th
e
in
f
o
r
m
a
tio
n
tr
av
els
in
o
n
e
d
ir
ec
tio
n
f
r
o
m
th
e
in
p
u
t
lay
er
th
r
o
u
g
h
s
ev
er
al
h
id
d
e
n
lay
er
s
to
g
et
to
t
h
e
o
u
tp
u
t
la
y
er
wit
h
o
u
t
an
y
cy
cles
o
r
lo
o
p
s
.
ML
Ps
ar
e
m
o
s
tly
k
n
o
wn
f
o
r
th
eir
ca
p
ab
ilit
y
to
lear
n
f
r
o
m
co
m
p
lex
m
ap
p
i
n
g
s
an
d
n
o
n
-
lin
ea
r
r
elatio
n
s
h
ip
s
in
d
a
ta.
Ho
wev
er
,
th
is
m
o
d
el
h
as
ce
r
tain
d
r
awb
ac
k
s
,
s
u
ch
as
o
v
er
f
itti
n
g
o
r
u
n
d
e
r
f
i
ttin
g
,
b
ec
au
s
e
it
ca
n
ca
p
tu
r
e
n
o
is
e
in
d
ata,
wh
ich
m
ig
h
t
r
e
q
u
ir
e
r
eg
u
lar
izatio
n
an
d
ca
r
ef
u
l
co
n
s
id
er
atio
n
o
f
p
ar
am
eter
s
.
I
t
also
r
eq
u
ir
es
a
la
r
g
e
am
o
u
n
t
o
f
la
b
eled
d
ata,
m
ak
in
g
it
a
b
it&
h
a
r
d
to
u
s
e
in
all
d
o
m
ain
s
b
ec
a
u
s
e
th
e
d
ata
m
ay
n
o
t
alwa
y
s
b
e
r
e
ad
ily
av
ailab
le.
Fu
r
th
e
r
m
o
r
e
,
ML
P
m
o
d
els
ca
n
b
e
co
m
p
u
tatio
n
ally
ex
p
e
n
s
iv
e,
wh
ich
wo
u
ld
d
em
an
d
s
u
b
s
tan
tia
l r
eso
u
r
ce
s
[
1
]
,
[
2
]
.
A
s
elf
-
o
r
g
an
izin
g
m
ap
(
SOM
)
is
a
wid
ely
u
s
ed
m
o
d
el
i
n
u
n
s
u
p
er
v
is
ed
lear
n
in
g
task
s
.
T
h
e
d
ata
ca
n
b
e
ex
p
lo
r
e
d
to
ex
tr
ac
t
r
eg
u
lar
ities
o
win
g
to
th
e
co
m
p
etitiv
e
lear
n
in
g
alg
o
r
ith
m
b
eh
i
n
d
th
e
m
o
d
el.
I
n
th
is
co
n
tex
t,
T
s
ai
s
u
g
g
ests
u
s
in
g
t
wo
h
y
b
r
id
m
o
d
els
th
at
co
m
b
i
n
e
two
d
is
tin
ct
n
eu
r
al
n
etwo
r
k
m
eth
o
d
s
to
p
r
ed
ict
ch
u
r
n
[
3
]
T
h
ese
tech
n
i
q
u
es
ar
e
b
ac
k
-
p
r
o
p
ag
atio
n
n
eu
r
al
n
et
wo
r
k
s
(
ANN)
an
d
s
elf
-
o
r
g
an
i
zin
g
m
ap
s
(
SOM
)
;
th
e
f
ir
s
t
m
o
d
el
em
p
lo
y
s
a
m
e
th
o
d
to
r
ed
u
ce
d
ata
b
y
f
ilter
i
n
g
o
u
t
tr
ain
in
g
d
ata
th
at'
s
n
o
t
r
ep
r
esen
tativ
e,
an
d
th
en
th
e
r
esu
lt
is
f
ed
to
th
e
p
r
ed
ictin
g
m
o
d
el
u
s
in
g
th
e
s
ec
o
n
d
tech
n
iq
u
e
.
T
h
e
r
esu
lts
s
h
o
w
th
at
th
e
co
m
b
in
ed
ANN
h
y
b
r
id
m
o
d
el
o
u
tp
er
f
o
r
m
s
th
e
o
th
e
r
m
et
h
o
d
s
.
C
u
ad
r
o
s
et
a
l.
p
r
o
p
o
s
es
in
[
4
]
a
s
eg
m
en
tatio
n
f
r
am
ewo
r
k
,
wh
er
e
th
e
c
u
s
to
m
er
life
tim
e
v
alu
e,
cu
s
to
m
er
lo
y
alty
ca
lc
u
latio
n
,
an
d
clien
t
s
eg
m
e
n
t
b
u
ild
in
g
ar
e
d
o
n
e
u
s
in
g
a
s
elf
-
o
r
g
an
ized
m
ap
.
I
n
th
e
s
a
m
e
co
n
tex
t
Asma
r
a
et
a
l.
u
s
ed
in
[
5
]
SOM
m
o
d
el
to
an
aly
ze
in
ter
ac
tio
n
s
am
o
n
g
b
ir
d
d
iv
e
r
s
ity
,
s
p
atial
d
is
tr
ib
u
tio
n
,
an
d
lan
d
u
s
e
ty
p
es in
th
e
Ken
y
ir
lan
d
s
ca
p
e
i
n
Ma
lay
s
ia.
I
n
o
u
r
w
o
r
k
,
we
aim
to
s
tu
d
y
p
o
s
s
ib
ilit
ies
p
r
o
v
id
e
d
b
y
u
n
s
u
p
er
v
is
ed
m
o
d
els,
esp
ec
ially
th
e
SOM
m
o
d
el,
ap
p
lied
to
p
r
o
d
u
ct
r
ev
iew
d
ata
to
g
ai
n
in
s
ig
h
t
in
to
p
r
o
d
u
cts
th
at
th
e
cu
s
to
m
er
s
en
d
o
r
s
e.
T
h
is
wo
r
k
is
p
ar
t
o
f
a
s
er
ies
o
f
r
esear
ch
co
n
d
u
cte
d
b
y
o
u
r
team
with
in
th
e
c
o
n
tex
t
o
f
th
e
d
ev
elo
p
m
en
t
o
f
c
u
s
to
m
er
p
r
o
f
ilin
g
v
ia
t
h
eir
ac
co
u
n
t
ac
ti
v
ity
b
y
an
s
wer
in
g
s
p
ec
if
ic
q
u
e
s
tio
n
s
to
b
u
ild
an
e
f
f
ec
tiv
e
e
-
c
o
m
m
er
ce
p
latf
o
r
m
(
E
x
:
wh
eth
er
it
is
a
f
ak
e
ac
co
u
n
t,
d
etec
tin
g
th
eir
p
r
ef
er
e
n
ce
s
v
ia
s
en
tim
en
t
an
aly
s
is
;
r
ec
o
m
m
en
d
in
g
p
r
o
d
u
cts;
d
etec
tin
g
ch
u
r
n
)
.
I
n
th
is
v
e
r
y
co
n
te
x
t
a
n
ew
ap
p
r
o
ac
h
to
cu
s
to
m
er
p
r
o
d
u
ct
ap
p
r
ec
iati
o
n
an
d
s
en
tim
en
tal
in
s
ig
h
ts
is
p
r
o
p
o
s
ed
in
o
r
d
er
t
o
ex
tr
ac
t
s
en
tim
en
tal
in
s
ig
h
ts
u
s
in
g
SOM
n
eu
r
al
n
etwo
r
k
.
I
n
th
e
f
ir
s
t
s
ec
tio
n
o
f
th
is
wo
r
k
,
we
p
r
esen
t
r
elate
d
wo
r
k
s
f
o
u
n
d
in
th
e
liter
atu
r
e.
T
h
e
s
ec
o
n
d
s
ec
tio
n
p
r
esen
ts
o
u
r
ap
p
r
o
ac
h
to
class
if
y
in
g
cu
s
to
m
er
s
in
to
tw
o
ca
teg
o
r
ies
an
d
p
r
e
d
ictin
g
p
o
s
itiv
e
f
r
o
m
n
eg
ativ
e
r
ev
iews.
I
n
th
e
th
ir
d
s
ec
tio
n
,
we
p
r
esen
t a
n
d
d
is
cu
s
s
th
e
r
esu
lts
o
b
tain
ed
.
T
h
e
co
n
cl
u
s
io
n
co
m
es in
th
e
last
s
ec
tio
n
.
2.
SE
L
F
O
RG
ANIZ
I
NG
M
AP
USE
I
N
L
I
T
E
RA
T
UR
E
A
SOM
is
an
ar
tific
ial
n
e
u
r
al
n
etwo
r
k
alg
o
r
ith
m
f
o
r
clu
s
te
r
in
g
a
n
d
v
is
u
alizin
g
m
u
lti
-
d
i
m
en
s
io
n
al
d
ata
in
lo
wer
-
d
im
en
s
io
n
al
s
p
ac
e.
SOM
is
a
ty
p
e
o
f
u
n
s
u
p
er
v
is
ed
lear
n
in
g
al
g
o
r
ith
m
.
I
t
u
s
es
a
g
r
id
o
f
n
eu
r
o
n
s
o
r
n
o
d
es
a
r
r
an
g
ed
in
two
d
im
en
s
io
n
s
,
wh
er
e
ea
ch
n
o
d
e
r
ep
r
esen
ts
a
p
r
o
to
ty
p
e
o
r
a
clu
s
ter
co
r
r
esp
o
n
d
i
n
g
to
a
r
eg
io
n
o
f
th
e
i
n
p
u
t
s
p
ac
e
[
6
]
.
T
h
e
SOM
is
tr
ain
ed
b
y
iter
ati
v
ely
ad
ju
s
tin
g
th
e
weig
h
ts
o
f
t
h
ese
n
o
d
es
to
m
atc
h
th
e
in
p
u
t
d
ata.
T
h
e
ad
ju
s
tm
e
n
t
is
m
ad
e
u
s
in
g
a
tech
n
iq
u
e
ca
lled
co
m
p
etitiv
e
lear
n
in
g
,
with
o
u
t
b
esto
win
g
lab
els,
wh
er
e
t
h
e
n
o
d
e
with
t
h
e
clo
s
est
weig
h
t
t
o
th
e
in
p
u
t
d
ata
is
c
h
o
s
en
as
th
e
win
n
er
,
an
d
its
weig
h
ts
ar
e
u
p
d
ated
t
o
b
e
e
v
en
clo
s
er
to
th
e
in
p
u
t
[
7
]
.
SOM
s
h
av
e
b
ee
n
wid
ely
u
s
ed
in
m
an
y
f
ield
s
,
s
u
c
h
as
p
atter
n
r
ec
o
g
n
itio
n
[
8
]
,
[
9
]
,
d
ata
m
i
n
in
g
[
1
0
]
,
im
ag
e
p
r
o
ce
s
s
in
g
[
1
1
]
,
an
d
d
a
ta
v
is
u
aliza
tio
n
[
1
2
]
,
[
1
3
]
.
T
h
e
y
ar
e
b
en
ef
icial
f
o
r
d
im
en
s
io
n
ality
r
ed
u
ctio
n
an
d
ex
p
lo
r
ato
r
y
d
ata
an
aly
s
is
,
as
t
h
ey
ca
n
p
r
o
v
id
e
a
lo
w
-
d
im
en
s
io
n
al
r
ep
r
esen
tatio
n
o
f
c
o
m
p
l
ex
d
atasets
,
m
ak
in
g
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
.
1
,
Feb
r
u
ar
y
20
25
:
9
8
0
-
994
982
it
ea
s
ier
to
u
n
d
er
s
tan
d
an
d
in
ter
p
r
et
th
e
u
n
d
e
r
ly
in
g
p
atter
n
s
an
d
r
elatio
n
s
h
ip
s
in
th
e
d
ata.
Neisar
i
et
a
l.
[
1
4
]
u
s
ed
a
m
ix
o
f
u
n
s
u
p
er
v
is
ed
le
ar
n
in
g
(
SOM
)
an
d
c
o
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
s
(
C
NN)
to
class
if
y
r
ev
iews
in
o
r
d
er
t
o
d
etec
t
s
p
am
r
ev
iews,
wh
ich
r
esu
lted
in
0
.
8
7
%
ac
cu
r
ac
y
b
y
co
m
b
i
n
in
g
t
h
e
two
m
o
d
els.
SOM
h
as
also
b
ee
n
a
p
p
lied
t
o
tem
p
er
atu
r
e
a
n
d
p
r
ec
ip
itatio
n
p
atter
n
s
o
v
er
C
h
in
a
in
th
is
s
tu
d
y
[
1
5
]
to
co
m
p
ar
e
b
etwe
en
2
0
2
1
an
d
2
0
2
2
p
atter
n
s
.
Dala
l
et
a
l.
[
1
6
]
h
as
u
s
ed
SOM
m
o
r
e
s
p
ec
if
ically
,
as
well
as
an
ad
ap
tiv
e
m
o
v
in
g
s
elf
-
o
r
g
an
izin
g
m
ap
a
n
d
f
u
zz
y
k
-
m
ea
n
clu
s
ter
in
g
,
f
o
r
b
r
ain
tu
m
o
r
s
eg
m
en
tatio
n
,
f
o
c
u
s
in
g
m
ain
ly
o
n
e
x
tr
ac
tin
g
th
e
tu
m
o
r
r
eg
io
n
s
.
Z
h
en
g
tian
et
a
l.
[
1
7
]
SOM
clu
s
ter
in
g
ab
ilit
ies
wer
e
u
s
ed
to
s
elec
t
r
elev
an
t
f
ea
tu
r
es
f
r
o
m
d
ata
b
ef
o
r
e
ap
p
ly
in
g
d
i
f
f
er
en
t
m
o
d
els
s
u
ch
as
(
K
-
NN,
SVM,
an
d
d
ec
is
io
n
tr
ee
s
)
th
en
,
it
co
m
p
ar
ed
t
h
e
r
esu
lts
to
o
th
er
f
ea
tu
r
e
s
elec
tio
n
m
eth
o
d
s
an
d
p
r
o
v
e
d
a
s
ig
n
if
ican
t
in
c
r
ea
s
e
i
n
th
e
ac
cu
r
ac
y
o
f
th
e
m
o
d
els.
An
g
u
lo
-
Sau
ce
d
o
et
a
l.
[
1
8
]
h
as
also
u
s
ed
v
ar
ian
t
s
o
f
SOM
in
s
tr
u
ctu
r
al
h
ea
lth
m
o
n
ito
r
in
g
to
class
if
y
d
am
ag
es.
Ad
d
itio
n
ally
,
Yu
an
et
a
l.
[
1
9
]
h
as
u
s
ed
t
h
e
SOM
alg
o
r
ith
m
to
p
r
ed
ict
th
e
p
atien
t
o
u
tco
m
e
an
d
r
esp
o
n
s
e
to
th
e
r
ap
y
b
y
ap
p
ly
in
g
ce
ll
s
eg
m
e
n
tatio
n
,
s
y
s
tem
atic
class
if
icatio
n
,
an
d
in
s
ilico
ce
ll
lab
elin
g
o
n
a
n
im
ag
e
d
atab
ase
o
f
b
r
ea
s
t c
an
ce
r
.
Z
h
en
g
tian
et
a
l.
[
1
7
]
u
s
ed
S
OM
clu
s
ter
in
g
m
eth
o
d
s
o
n
a
b
in
ar
y
class
if
icatio
n
p
r
o
b
lem
f
o
r
f
ea
t
u
r
e
s
elec
tio
n
an
d
th
e
n
a
p
p
lied
d
if
f
er
en
t
m
o
d
els,
s
u
ch
as
d
ec
is
io
n
tr
ee
s
,
r
esu
ltin
g
in
0
.
7
5
%
a
c
cu
r
ac
y
.
Ho
wev
er
,
it
ac
h
iev
ed
a
h
ig
h
e
r
ac
cu
r
ac
y
w
h
en
ap
p
ly
in
g
SVM
0
.
8
5
%.
On
th
e
o
t
h
er
h
an
d
,
th
e
p
ap
er
[
1
8
]
u
s
es
v
ar
ian
ts
o
f
th
e
SOM
m
o
d
el
ca
lled
c
o
u
n
te
r
p
r
o
p
ag
atio
n
ar
tific
ial
n
eu
r
al
n
etwo
r
k
(
C
PANN
)
,
s
u
p
er
v
is
e
d
Ko
h
o
n
en
(
SKN
)
,
an
d
X
–
Y
f
u
s
ed
Ko
h
o
n
en
(
XYF
)
,
wh
ich
h
as
r
esu
lted
in
a
n
o
v
er
all
o
f
0
.
7
4
%
ac
cu
r
ac
y
u
s
in
g
SKN
an
d
0
.
7
3
%
u
s
in
g
SYF.
Ad
d
itio
n
ally
,
th
is
p
ap
er
[
1
9
]
h
as
u
s
ed
a
n
SOM
m
o
d
el
o
f
4
9
n
o
d
es
with
5
0
0
0
iter
atio
n
s
.
B
ased
o
n
o
n
ly
th
e
t
o
p
f
iv
e
f
ea
tu
r
es,
it a
c
h
iev
ed
0
.
7
6
% p
r
ec
is
io
n
,
0
.
7
9
% r
ec
all,
0
.
7
8
% F1
,
a
n
d
0
.
7
0
% AU
C
.
I
n
th
e
p
r
esen
t
p
ap
er
,
we
p
r
es
en
t
a
SOM
-
b
ased
m
o
d
el
to
p
r
ed
ict
cu
s
to
m
er
p
r
o
f
iles
th
at
ar
e
m
o
r
e
lik
ely
to
h
av
e
a
p
o
s
itiv
e
f
e
elin
g
to
wa
r
d
s
a
p
r
o
d
u
ct
th
an
th
o
s
e
wh
o
d
o
n
o
t.
T
h
e
d
ataset
u
s
ed
in
th
is
p
a
p
er
is
a
co
llectio
n
o
f
Am
az
o
n
r
e
v
iews
th
at
h
as
b
ee
n
co
llected
s
in
c
e
1
9
9
6
u
p
to
2
0
1
8
an
d
is
d
iv
id
ed
in
to
m
u
ltip
le
ca
teg
o
r
ies.
T
h
e
ca
teg
o
r
ies
u
s
e
d
f
o
r
th
is
ex
p
er
im
e
n
t
co
m
b
i
n
e
th
r
ee
ca
teg
o
r
ies:
‘
Ma
g
az
in
es
an
d
Su
b
s
cr
ip
tio
n
s
’
,
‘
So
f
twar
e’
,
a
n
d
‘
B
ea
u
ty
’
with
1
2
f
ea
tu
r
es.
Sin
ce
t
h
e
f
o
cu
s
i
s
o
n
t
h
e
n
atu
r
al
lan
g
u
a
g
e
p
r
o
c
ess
in
g
asp
ec
t
o
f
th
e
d
ataset,
we
o
n
ly
k
ep
t
th
e
tex
t
d
ata
th
at
was
n
ee
d
ed
.
B
ef
o
r
e
ev
alu
atin
g
m
o
d
els,
th
o
r
o
u
g
h
s
tep
s
wer
e
u
s
ed
,
s
u
ch
as d
ata
p
r
e
-
p
r
o
ce
s
s
in
g
,
r
o
b
u
s
t scalin
g
,
an
d
f
ea
tu
r
e
s
elec
tio
n
.
L
ik
e
m
an
y
m
ac
h
in
e
lear
n
in
g
a
lg
o
r
ith
m
s
,
SOM
-
b
ased
m
o
d
els
also
r
e
q
u
ir
e
p
ar
am
eter
tu
n
in
g
(
lear
n
in
g
r
ates,
n
eig
h
b
o
r
h
o
o
d
f
u
n
ctio
n
s
)
,
wh
ich
ca
n
b
e
ess
en
tial
to
th
e
f
in
al
r
esu
lts
.
T
h
u
s
,
s
y
s
tem
atic
ex
p
er
im
en
tatio
n
an
d
cr
o
s
s
-
v
alid
atio
n
ar
e
cr
u
c
ial
f
o
r
d
ete
r
m
in
in
g
o
p
tim
al
p
ar
am
eter
s
.
An
o
t
h
er
ch
alle
n
g
e
is
th
e
s
u
b
jectiv
e
in
ter
p
r
etatio
n
o
f
SOM
m
ap
s
,
m
ea
n
in
g
th
at
r
ele
v
an
t
clu
s
ter
s
o
r
p
atter
n
s
ca
n
r
eq
u
ir
e
an
e
x
p
er
t
o
r
r
is
k
b
ein
g
wr
o
n
g
ly
id
e
n
tifie
d
o
r
ig
n
o
r
e
d
.
Qu
an
titativ
e
m
ea
s
u
r
es a
n
d
v
a
lid
atio
n
tech
n
iq
u
es sh
o
u
l
d
b
e
u
s
ed
to
en
h
an
ce
th
e
o
b
jectiv
ity
o
f
m
ap
in
ter
p
r
etati
o
n
.
3.
T
H
E
P
RO
P
O
SE
D
AP
P
RO
A
CH
F
O
R
CUST
O
M
E
R
R
E
T
E
NT
I
O
N
AND
P
RO
DU
CT
AP
P
RE
CI
AT
I
O
N
T
h
is
wo
r
k
is
a
co
n
tin
u
atio
n
o
f
th
e
r
esear
ch
wo
r
k
ca
r
r
ied
o
u
t
in
[
2
0
]
t
h
is
s
tag
e
we
ar
e
i
n
ter
ested
to
r
ev
iews
b
in
ar
y
class
if
icatio
n
.
T
h
e
im
p
ac
t
o
f
u
s
in
g
th
e
SO
M
m
o
d
el
will
b
e
ass
ess
ed
to
h
elp
th
e
f
in
an
cial
o
r
g
an
izatio
n
m
ak
e
d
ec
is
io
n
s
a
b
o
u
t
th
e
s
u
itab
le
s
tr
ateg
y
to
d
i
s
ce
r
n
cu
s
to
m
er
s
en
tim
en
ts
an
d
r
ef
in
e
th
e
o
v
e
r
all
p
r
o
d
u
cts
q
u
ality
b
ased
o
n
s
en
tim
en
tal
in
s
ig
h
ts
ex
tr
ac
ted
v
ia
r
ev
iew
an
aly
s
is
.
T
h
e
f
o
llo
wi
n
g
s
ec
tio
n
s
will
b
e
d
ev
o
ted
to
th
e
p
r
e
p
r
o
ce
s
s
in
g
,
m
o
d
elin
g
an
d
ev
alu
atio
n
p
h
ases
.
So
m
e
o
f
th
e
k
ey
co
n
tr
ib
u
ti
o
n
s
o
f
th
is
r
esear
ch
ar
e:
a.
E
x
ten
s
iv
e
d
ataset
u
s
e:
T
h
e
d
ataset
in
clu
d
es
Am
az
o
n
r
ev
iews
d
iv
id
ed
in
to
ca
teg
o
r
ies,
f
o
cu
s
in
g
o
n
s
p
ec
if
ic
o
n
es
lik
e
'
Ma
g
az
in
es
an
d
Su
b
s
cr
ip
tio
n
s
,
'
'
So
f
twar
e,
'
an
d
'B
e
au
ty
.
'
T
h
is
co
m
p
r
eh
en
s
iv
e
d
at
aset
p
r
o
v
id
es
a
d
etailed
an
aly
s
is
o
f
cu
s
to
m
er
s
atis
f
ac
tio
n
an
d
s
en
tim
en
ts
;
b.
C
o
m
b
in
in
g
th
e
p
r
io
r
k
n
o
wled
g
e
an
d
th
e
cl
u
s
ter
in
g
ca
p
ab
ili
ties
:
T
h
e
r
esear
ch
em
p
lo
y
s
a
SOM
m
o
d
el
to
s
eg
m
en
t
an
d
p
r
ed
ict
cu
s
to
m
e
r
p
r
o
f
iles
b
ased
o
n
th
eir
s
en
tim
en
ts
to
war
d
p
r
o
d
u
cts,
aid
in
g
in
id
en
tify
in
g
cu
s
to
m
er
clu
s
ter
s
a
n
d
th
eir
ass
o
ciate
d
s
en
tim
en
ts
.
T
h
is
ca
n
b
e
r
ea
ch
e
d
b
y
co
m
b
i
n
in
g
th
e
p
r
io
r
k
n
o
wled
g
e
,
r
ev
iews
ar
e
lab
elled
,
an
d
th
e
co
m
p
etitiv
e
lear
n
in
g
p
o
wer
p
r
o
v
id
ed
b
y
th
e
m
o
d
el.
T
h
e
SOM
m
o
d
el
is
f
ir
s
t
tr
ain
ed
in
an
u
n
s
u
p
er
v
is
ed
m
a
n
n
er
f
o
r
in
itial c
lu
s
ter
in
g
,
f
o
ll
o
wed
b
y
s
u
p
e
r
v
is
ed
lab
elin
g
a
n
d
class
if
icatio
n
u
s
in
g
a
m
ajo
r
ity
v
o
tin
g
p
r
o
ce
s
s
.
T
h
is
d
u
al
ap
p
r
o
ac
h
e
n
h
an
ce
s
th
e
m
o
d
el'
s
p
r
ed
ictiv
e
ca
p
ab
ilit
ies.
c.
I
m
p
r
o
v
ed
cu
s
to
m
er
an
d
p
r
o
d
u
ct
in
s
ig
h
t: th
e
m
ap
g
en
er
ated
b
y
th
e
SOM
m
o
d
el
is
ex
p
lo
r
ed
to
d
etec
t n
o
d
es
with
s
o
m
e
p
ar
tic
u
lar
p
r
o
f
ile
s
.
T
h
is
co
n
tr
ib
u
tes
o
n
a
b
et
ter
u
n
d
er
s
tan
d
in
g
cu
s
to
m
er
s
'
in
ten
tio
n
s
a
n
d
b
eh
av
io
r
s
,
all
o
ws
p
er
s
o
n
alize
d
p
r
o
d
u
ct
r
ec
o
m
m
e
n
d
atio
n
s
a
n
d
tar
g
ete
d
m
ar
k
etin
g
s
tr
ateg
ies
an
d
r
ed
u
ce
s
th
e
n
ee
d
f
o
r
g
en
er
ic
m
ar
k
etin
g
ca
m
p
aig
n
s
.
T
h
is
o
f
f
er
s
al
s
o
v
alu
ab
le
in
s
ig
h
ts
f
o
r
m
an
ag
er
s
to
i
d
en
tify
wh
ich
p
r
o
d
u
cts
n
ee
d
i
m
p
r
o
v
e
m
en
t
an
d
wh
ich
h
a
v
e
th
e
m
o
s
t
s
ig
n
if
ican
t
im
p
ac
t,
th
er
e
b
y
a
id
in
g
in
p
r
o
d
u
ct
q
u
ality
en
h
a
n
ce
m
en
t.
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
P
r
o
d
u
ct
r
ev
iew
s
a
n
a
lysi
s
to
ex
tr
a
ct
s
en
timen
ta
l in
s
ig
h
ts
w
ith
cla
s
s
co
n
fid
en
ce
…
(
S
a
r
a
A
h
s
a
in
)
983
d.
Ob
jectiv
e
in
ter
p
r
etatio
n
o
f
S
OM
m
ap
s
:
T
o
ad
d
r
ess
th
e
s
u
b
jectiv
ity
in
in
ter
p
r
etin
g
SO
M
m
ap
s
,
th
e
s
tu
d
y
s
u
g
g
ests
u
s
in
g
q
u
an
titativ
e
m
ea
s
u
r
es
an
d
v
alid
atio
n
tec
h
n
iq
u
es,
en
s
u
r
in
g
ac
c
u
r
ate
i
d
en
tific
atio
n
o
f
r
elev
an
t c
lu
s
ter
s
.
e.
T
h
o
r
o
u
g
h
m
o
d
el
e
v
alu
atio
n
a
n
d
f
in
e
-
tu
n
in
g
:
T
h
e
r
esear
c
h
i
n
v
o
lv
es
ex
ten
s
iv
e
p
r
ep
r
o
ce
s
s
in
g
,
d
ata
s
ca
lin
g
,
f
ea
tu
r
e
s
elec
tio
n
,
an
d
p
ar
am
e
ter
in
itializatio
n
to
en
s
u
r
e
a
g
o
o
d
a
n
d
a
p
p
r
o
p
r
iate
im
p
lem
en
tatio
n
o
f
t
h
e
SOM
m
o
d
el.
I
n
f
ac
t,
d
etailed
f
in
e
-
tu
n
in
g
is
n
ee
d
ed
,
wh
ich
is
ac
co
m
p
lis
h
ed
b
y
s
y
s
tem
atic
ex
p
er
im
en
tatio
n
an
d
cr
o
s
s
-
v
alid
atio
n
to
o
p
tim
ize
p
ar
am
eter
s
s
u
ch
as lea
r
n
i
n
g
r
ates a
n
d
n
eig
h
b
o
r
h
o
o
d
f
u
n
cti
o
n
s
.
T
h
e
f
in
d
in
g
s
o
f
th
is
s
tu
d
y
ca
n
h
elp
f
in
an
cial
o
r
g
an
izatio
n
s
m
ak
e
in
f
o
r
m
e
d
d
ec
is
io
n
s
ab
o
u
t
cu
s
to
m
er
s
en
tim
en
t
an
aly
s
is
an
d
p
r
o
d
u
c
t
q
u
ality
im
p
r
o
v
em
en
t,
le
v
er
a
g
in
g
s
en
tim
en
tal
in
s
ig
h
ts
d
er
i
v
ed
f
r
o
m
cu
s
to
m
er
r
ev
iews.
T
o
p
r
o
v
i
d
e
a
clea
r
u
n
d
er
s
tan
d
in
g
o
f
th
e
r
esear
ch
p
r
o
ce
s
s
an
d
its
k
ey
co
m
p
o
n
e
n
ts
,
th
e
d
iag
r
am
in
Fig
u
r
e
1
illu
s
tr
ates
th
e
m
eth
o
d
o
lo
g
y
e
m
p
lo
y
e
d
in
th
is
s
tu
d
y
.
T
h
e
m
o
d
el
f
o
cu
s
es
o
n
p
r
o
c
ess
in
g
th
e
d
ata
an
d
m
ak
e
it r
ea
d
y
to
b
e
u
s
ed
b
y
th
e
d
if
f
er
en
t a
l
g
o
r
ith
m
s
wh
ile
al
s
o
en
s
u
r
in
g
a
h
ig
h
e
f
f
ec
tiv
en
e
s
s
.
Fig
u
r
e
1
o
u
tlin
es
th
e
p
r
o
ce
s
s
u
s
ed
to
class
if
y
s
en
tim
en
t
s
in
Am
az
o
n
p
r
o
d
u
ct
r
e
v
ie
ws
th
r
o
u
g
h
m
ac
h
in
e
lear
n
in
g
.
I
t
s
tar
ts
with
d
ata
co
llectio
n
,
ex
p
licitly
g
a
th
er
in
g
a
n
Am
az
o
n
r
ev
iew
d
ataset.
T
h
is
d
ata
th
e
n
g
o
es
th
r
o
u
g
h
a
p
r
e
p
r
o
ce
s
s
in
g
s
tag
e,
wh
ich
in
clu
d
es
ex
p
a
n
d
in
g
c
o
n
tr
ac
tio
n
s
,
co
n
v
er
tin
g
tex
t
to
lo
wer
ca
s
e,
r
em
o
v
in
g
d
ig
its
an
d
p
u
n
ct
u
atio
n
,
elim
in
atin
g
s
to
p
wo
r
d
s
,
a
n
d
lem
m
atizin
g
t
h
e
tex
t
to
its
b
ase
f
o
r
m
s
.
Af
te
r
p
r
ep
r
o
ce
s
s
in
g
,
th
e
d
ata
is
lab
eled
with
VADE
R
,
a
p
r
e
-
tr
ai
n
ed
m
ac
h
in
e
-
lear
n
in
g
to
o
l
k
n
o
wn
f
o
r
s
en
tim
en
t
an
aly
s
is
.
Nex
t,
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
D
F)
was
u
s
ed
to
tr
an
s
f
o
r
m
tex
t
d
ata
in
to
n
u
m
er
ical
f
ea
t
u
r
es
f
o
r
f
ea
tu
r
e
ex
tr
ac
tio
n
.
T
h
ese
f
ea
t
u
r
es
ar
e
s
elec
ted
u
s
in
g
th
e
lig
h
t
g
r
ad
ien
t
b
o
o
s
tin
g
m
ac
h
in
e
(
L
GB
M)
to
p
in
p
o
in
t
th
e
m
o
s
t
r
elev
an
t
o
n
es
f
o
r
cl
ass
if
icatio
n
.
Fin
ally
,
th
e
d
ata
is
clas
s
if
ied
u
s
in
g
m
ac
h
in
e
lear
n
in
g
m
o
d
els,
s
u
ch
as
a
s
elf
-
o
r
g
a
n
izin
g
m
ap
,
a
n
d
c
o
m
p
ar
e
d
to
th
e
class
if
icatio
n
an
d
r
e
g
r
ess
io
n
tr
ee
s
(
C
AR
T
)
d
ec
is
io
n
tr
ee
,
wh
ich
ca
teg
o
r
izes
th
e
r
ev
ie
ws
as
p
o
s
itiv
e
o
r
n
eg
ativ
e
,
th
er
eb
y
i
d
en
tify
in
g
cu
s
to
m
er
s
en
tim
en
ts
to
war
d
s
th
e
p
r
o
d
u
cts.
T
h
is
r
esear
ch
p
r
o
v
id
es
a
s
tr
u
ctu
r
ed
m
et
h
o
d
f
o
r
a
n
aly
zin
g
cu
s
to
m
er
r
ev
iews
u
s
in
g
t
h
e
SOM
m
o
d
el,
en
ab
lin
g
b
etter
cu
s
to
m
er
s
eg
m
en
t
atio
n
,
p
er
s
o
n
alize
d
r
ec
o
m
m
en
d
atio
n
s
,
an
d
im
p
r
o
v
em
en
ts
in
p
r
o
d
u
ct
q
u
ality
.
Fig
u
r
e
1
.
T
h
e
m
eth
o
d
o
lo
g
y
d
iag
r
am
o
f
th
is
r
esear
ch
3
.
1
.
Da
t
a
s
et
L
ar
g
e
d
atasets
ca
n
h
elp
b
u
ild
m
o
d
els
th
at
ac
h
iev
e
b
etter
ac
cu
r
ac
y
an
d
o
t
h
er
p
er
f
o
r
m
a
n
ce
s
d
u
e
to
th
eir
ab
ilit
y
to
p
r
o
v
i
d
e
v
ar
io
u
s
r
ep
r
esen
tativ
e
s
am
p
les.
T
h
is
allo
ws
th
e
m
o
d
el
to
co
llect
a
b
r
o
ad
er
r
an
g
e
o
f
p
atter
n
s
.
Als
o
lear
n
in
g
f
r
o
m
ex
am
p
les
h
elp
s
r
ed
u
ce
o
v
er
f
itti
n
g
,
as
it
m
in
im
izes
m
em
o
r
izin
g
s
p
ec
if
i
c
in
s
tan
ce
s
o
r
n
o
is
e,
an
d
th
en
im
p
r
o
v
e
m
o
d
el
r
o
b
u
s
tn
ess
ag
ain
s
t
o
u
tlier
s
an
d
v
ar
iatio
n
s
,
an
d
f
ac
ilit
ate
r
ar
e
ev
en
t
d
etec
tio
n
.
T
h
e
d
ataset
u
s
ed
to
tr
ain
th
e
m
o
d
el
c
o
n
ta
in
s
p
r
o
d
u
ct
r
e
v
iews
an
d
m
etad
ata
f
r
o
m
Am
az
o
n
,
in
clu
d
in
g
4
5
9
4
3
6
So
f
twar
e
r
ev
iews,
8
9
6
8
9
Ma
g
az
in
e
s
u
b
s
cr
ip
tio
n
s
,
an
d
3
7
1
3
4
5
B
ea
u
t
y
r
ev
iews
s
p
an
n
in
g
f
r
o
m
1
9
9
6
to
2
0
1
8
[
2
1
]
.
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
.
1
,
Feb
r
u
ar
y
20
25
:
9
8
0
-
994
984
T
ab
le
1
s
h
o
ws
th
e
d
if
f
er
en
t
d
is
tr
ib
u
tio
n
s
o
f
th
e
Am
az
o
n
d
ataset,
in
clu
d
in
g
th
e
n
u
m
b
er
o
f
r
ev
iews
it
with
h
o
ld
s
.
T
o
f
in
d
th
e
b
est
-
r
ev
iewe
d
p
r
o
d
u
cts
an
d
th
e
m
o
s
t
will
in
g
cu
s
to
m
er
s
to
r
e
-
p
u
r
ch
ase
an
d
h
elp
th
e
d
ec
is
io
n
-
m
ak
in
g
p
r
o
ce
s
s
b
e
a
war
e
o
f
th
e
p
r
ec
au
tio
n
s
th
e
f
in
an
cial
o
r
g
an
izatio
n
s
h
o
u
ld
u
n
d
er
tak
e,
we
d
ec
id
e
d
to
ap
p
ly
t
h
e
SOM
m
o
d
el
an
d
e
v
alu
ate
th
e
p
er
f
o
r
m
a
n
ce
.
T
a
b
le
1
s
h
o
ws th
e
f
ea
tu
r
es
u
s
ed
.
T
h
is
d
ataset
h
as
1
2
f
ea
tu
r
es,
with
d
if
f
er
e
n
t
d
ata
t
y
p
es
d
esc
r
ib
ed
in
d
etail
in
T
ab
le
2
.
T
h
is
r
esear
ch
p
ap
er
wo
r
k
s
o
n
a
n
at
u
r
al
lan
g
u
ag
e
p
r
o
ce
s
s
in
g
task
,
wh
ich
lead
s
to
a
f
o
cu
s
o
n
ly
o
n
tex
t
d
at
a.
T
h
e
f
in
al
d
ataset
m
ix
es
th
r
ee
s
u
b
-
d
atasets
b
elo
n
g
in
g
to
th
r
ee
ca
teg
o
r
ies
in
o
r
d
er
to
wo
r
k
o
n
th
e
class
if
icatio
n
o
f
r
ev
iews
b
y
r
atin
g
.
T
ab
le
1
.
Am
az
o
n
r
e
v
iews d
is
tr
ib
u
tio
n
p
er
ca
teg
o
r
y
D
a
t
a
s
e
t
N
u
mb
e
r
o
f
r
e
v
i
e
w
s
A
l
l
r
e
v
i
e
w
s
2
3
3
.
1
m
i
l
l
i
o
n
M
o
v
i
e
s
a
n
d
TV
8
7
6
5
5
6
8
S
o
f
t
w
a
r
e
4
5
9
4
3
6
B
o
o
k
s
5
1
3
1
1
6
2
1
C
e
l
l
p
h
o
n
e
s
a
n
d
a
c
c
e
ss
o
r
i
e
s
10
0
6
3
2
5
5
D
i
g
i
t
a
l
M
u
si
c
1
5
8
4
0
8
2
M
a
g
a
z
i
n
e
su
b
scr
i
p
t
i
o
n
s
89
689
B
e
a
u
t
y
3
7
1
3
4
5
G
r
o
c
e
r
y
a
n
d
g
o
u
r
me
t
f
o
o
d
5
0
7
4
1
6
0
H
o
me
a
n
d
K
i
t
c
h
e
n
21
9
2
8
5
6
8
T
ab
le
2
.
Data
s
et
f
ea
tu
r
es
F
e
a
t
u
r
e
n
a
me
S
i
g
n
i
f
i
c
a
t
i
o
n
Ty
p
e
O
v
e
r
a
l
l
R
a
t
i
n
g
o
f
t
h
e
p
r
o
d
u
c
t
F
l
o
a
t
v
e
r
i
f
i
e
d
Tr
u
e
i
f
t
h
e
p
u
r
c
h
a
se
w
a
s
v
e
r
i
f
i
e
d
B
o
o
l
e
a
n
R
e
v
i
e
w
t
i
me
R
a
w
d
a
t
e
t
i
me
o
f
t
h
e
r
e
v
i
e
w
D
a
t
e
R
e
v
i
e
w
e
r
I
D
Th
e
I
D
o
f
t
h
e
r
e
v
i
e
w
e
r
S
t
r
i
n
g
A
si
n
I
D
o
f
t
h
e
p
r
o
d
u
c
t
S
t
r
i
n
g
S
t
y
l
e
D
i
c
t
i
o
n
a
r
y
o
f
t
h
e
p
r
o
d
u
c
t
me
t
a
d
a
t
a
,
e
.
g
.
,
“
F
o
r
mat
”
i
s
“
H
a
r
d
c
o
v
e
r
”
A
r
r
a
y
R
e
v
i
e
w
e
r
n
a
me
N
a
me
o
f
t
h
e
r
e
v
i
e
w
e
r
S
t
r
i
n
g
R
e
v
i
e
w
t
e
x
t
Te
x
t
o
f
t
h
e
r
e
v
i
e
w
S
t
r
i
n
g
S
u
mm
a
r
y
S
u
mm
a
r
y
o
f
t
h
e
r
e
v
i
e
w
S
t
r
i
n
g
U
n
i
x
r
e
v
i
e
w
t
i
me
U
n
i
x
t
i
me
o
f
t
h
e
r
e
v
i
e
w
U
n
i
x
t
i
me
V
o
t
e
H
e
l
p
f
u
l
v
o
t
e
s
o
f
r
e
v
i
e
w
s
b
y
o
t
h
e
r
r
e
v
i
e
w
e
r
s
N
u
mb
e
r
I
mag
e
A
t
t
a
c
h
e
d
i
m
a
g
e
t
o
t
h
e
r
e
v
i
e
w
A
r
r
a
y
Data
s
cien
tis
t
s
n
ee
d
an
a
p
p
r
o
p
r
iate
d
ataset
to
p
e
r
f
o
r
m
well
to
g
ain
i
n
s
ig
h
t
in
to
c
u
s
to
m
er
s
'
b
eh
av
i
o
r
.
T
h
is
d
ataset
s
h
o
u
ld
co
n
tain
s
u
f
f
icien
t
s
am
p
les
f
r
o
m
d
if
f
er
en
t
d
o
cu
m
e
n
t
ca
teg
o
r
ies.
I
t
e
n
s
u
r
es
th
e
u
s
ag
e
o
f
r
ea
l
-
wo
r
ld
r
e
v
iews c
ateg
o
r
ize
d
in
to
d
if
f
er
en
t
p
r
o
d
u
ct
ca
teg
o
r
ies af
ter
p
u
r
c
h
ase.
3
.
2
.
Da
t
a
pre
-
pro
ce
s
s
ing
Data
p
r
ep
r
o
ce
s
s
in
g
an
d
clea
n
in
g
ar
e
ess
en
tial
s
tep
s
f
o
r
a
d
ataset
b
ased
o
n
tex
t
an
aly
s
is
.
I
r
r
elev
an
t
in
f
o
r
m
atio
n
lik
e
HT
ML
tag
s
,
p
u
n
ctu
atio
n
,
o
r
s
p
ec
ial
ch
ar
ac
ter
s
s
h
o
u
ld
b
e
r
em
o
v
ed
.
Ad
d
r
ess
in
g
s
tan
d
ar
d
tex
t
p
r
ep
r
o
ce
s
s
in
g
task
s
s
u
ch
as
r
em
o
v
in
g
s
to
p
wo
r
d
s
,
n
o
r
m
alizi
n
g
tex
t
(
lo
wer
ca
s
in
g
,
s
tem
m
in
g
,
lem
m
atiza
tio
n
)
,
an
d
h
a
n
d
lin
g
s
p
ellin
g
m
is
tak
es o
r
ab
b
r
e
v
iatio
n
s
ca
n
im
p
r
o
v
e
th
e
d
ataset'
s
q
u
ality
.
T
o
tr
ain
m
ac
h
in
e
lear
n
in
g
m
o
d
els
ac
cu
r
ately
,
th
e
d
ata
m
u
s
t
b
e
clea
n
ed
an
d
p
r
e
p
r
o
ce
s
s
ed
.
I
r
r
elev
a
n
t
d
ata,
s
u
ch
as
n
u
ll
a
n
d
p
o
o
r
ly
f
o
r
m
atted
d
ata,
s
p
ec
ial
ch
ar
ac
ter
s
,
p
u
n
ctu
atio
n
,
s
h
o
u
ld
b
e
d
is
ca
r
d
ed
.
Ad
d
itio
n
ally
,
o
th
er
s
tep
s
wer
e
u
s
ed
,
s
u
ch
as
lo
wer
ca
s
in
g
,
lem
m
atiza
tio
n
,
an
d
s
tem
m
in
g
.
C
h
an
g
in
g
th
e
ef
f
ec
tiv
en
ess
o
f
o
n
e
o
r
m
o
r
e
o
f
th
ese
s
tep
s
ca
n
s
ig
n
if
ican
tly
in
cr
ea
s
e
th
e
m
o
d
el'
s
ac
cu
r
ac
y
.
First,
th
e
f
lo
w
was
in
itialize
d
b
y
f
o
r
m
attin
g
th
e
d
ataset'
s
attr
ib
u
tes
to
f
it
th
e
p
a
p
er
’
s
n
ee
d
s
.
I
t
co
m
b
in
es
th
e
r
ev
iew
tex
t
an
d
th
e
titl
e
in
to
o
n
e
co
lu
m
n
,
an
d
th
en
we
ad
d
e
d
th
e
p
r
o
d
u
ct
ca
teg
o
r
y
co
lu
m
n
b
ased
o
n
wh
ich
ca
te
g
o
r
y
th
e
r
ev
iew
b
el
o
n
g
s
to
.
B
ef
o
r
e
p
er
f
o
r
m
in
g
ex
p
lo
r
ato
r
y
d
ata
a
n
aly
s
is
(
E
DA)
an
d
co
n
v
er
tin
g
th
e
d
ataset
to
a
f
o
r
m
at
th
at
is
ad
eq
u
ate
f
o
r
m
o
d
els,
th
e
f
o
llo
win
g
tr
an
s
f
o
r
m
atio
n
p
ip
elin
e
was
ad
o
p
ted
:
i)
clea
n
in
g
an
d
f
ea
t
u
r
e
en
g
i
n
ee
r
in
g
,
ii)
clea
n
in
g
s
to
p
wo
r
d
s
,
iii)
r
em
o
v
in
g
n
u
lls
,
iv
)
r
em
o
v
i
n
g
p
u
n
ctu
atio
n
;
lab
el
en
co
d
in
g
,
a
n
d
v
)
lem
m
atizin
g
.
An
E
DA
h
as
b
ee
n
p
er
f
o
r
m
ed
to
co
m
p
r
eh
en
s
iv
ely
u
n
d
e
r
s
tan
d
th
e
d
ataset
an
d
ass
ess
d
ata
q
u
ality
.
I
t
aid
s
in
f
ea
tu
r
e
s
elec
tio
n
,
v
alid
ates a
s
s
u
m
p
tio
n
s
,
in
f
o
r
m
s
m
o
d
el
s
elec
tio
n
an
d
d
esig
n
,
an
d
h
elp
s
d
etec
t o
u
tlier
s
.
I
t
allo
ws
m
o
r
e
p
r
o
f
o
u
n
d
in
s
ig
h
ts
in
to
th
e
d
ataset’
s
ch
a
r
ac
te
r
is
tics
an
d
r
elatio
n
s
h
ip
s
.
Fig
u
r
e
2
s
h
o
ws
th
at
th
e
o
r
ig
in
al
d
ataset
was
d
iv
id
e
d
i
n
to
5
r
atin
g
v
alu
es;
ea
ch
tar
g
et
v
alu
e
ca
n
b
e
co
n
s
id
er
ed
as
a
class
;
it
is
also
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
P
r
o
d
u
ct
r
ev
iew
s
a
n
a
lysi
s
to
ex
tr
a
ct
s
en
timen
ta
l in
s
ig
h
ts
w
ith
cla
s
s
co
n
fid
en
ce
…
(
S
a
r
a
A
h
s
a
in
)
985
u
n
ev
en
l
y
d
is
tr
ib
u
te
d
b
etwe
en
th
e
class
es
an
d
r
e
v
iews
with
a
p
o
s
itiv
e
m
ea
n
in
g
th
at
is
ca
te
g
o
r
ized
as
th
e
class
‘
p
o
s
itiv
e’
r
ep
r
esen
ts
th
e
m
ajo
r
ity
class
with
o
v
er
th
an
7
5
% o
f
th
e
d
ataset.
Fig
u
r
e
2.
Dis
tr
ib
u
tio
n
o
f
class
‘
o
v
er
all’
in
th
e
d
ataset
T
h
is
p
ap
er
co
v
e
r
s
d
iv
er
s
e
n
at
u
r
al
lan
g
u
ag
e
p
r
o
ce
s
s
in
g
tech
n
iq
u
es
,
f
o
cu
s
in
g
o
n
class
if
y
in
g
r
ev
iews
b
y
r
atin
g
.
T
o
d
o
th
at,
two
cla
s
s
es
(
p
o
s
itiv
e/
n
eg
ativ
e)
wer
e
co
m
b
in
e
d
(
9
,
5
0
0
r
an
d
o
m
r
e
v
iews
wer
e
ch
o
s
en
f
r
o
m
ea
c
h
class
)
.
Her
e
is
an
e
x
am
p
le
o
f
a
r
ev
iew
b
ef
o
r
e
an
d
af
ter
ap
p
l
y
in
g
th
e
clea
n
in
g
p
i
p
elin
e:
−
My
h
u
s
b
an
d
wan
ted
to
r
ea
d
i
n
g
a
b
o
u
t
th
e
Neg
r
o
B
aseb
all
an
d
th
is
a
g
r
ea
t
a
d
d
itio
n
to
h
is
lib
r
ar
y
\
n
Ou
r
lib
r
ar
y
d
o
es
not
h
a
v
e
in
f
o
r
m
at
io
n
s
o
th
is
b
o
o
k
is
h
is
s
tar
t.
T
h
an
k
y
o
u
−
H
u
s
b
an
d
wan
ts
to
r
ea
d
ab
o
u
t
n
eg
r
o
b
aseb
all
th
is
g
o
o
d
ad
d
itio
n
to
lib
r
ar
y
d
o
es n
o
t h
a
v
e
in
f
o
r
m
atio
n
b
o
o
k
s
s
tar
t th
an
k
y
o
u
So
m
e
ad
d
itio
n
al
e
x
am
p
les we
r
e
ad
d
ed
in
T
ab
le
3
b
y
ca
teg
o
r
y
af
ter
th
e
p
r
e
-
p
r
o
ce
s
s
in
g
s
tep
.
T
ab
le
3
.
E
x
am
p
le
o
f
r
ev
iews a
f
ter
d
ata
p
r
e
-
p
r
o
ce
s
s
in
g
Te
x
t
r
e
v
i
e
w
C
a
t
e
g
o
r
y
C
a
n
s
o
f
t
w
a
r
e
c
o
m
p
a
n
y
p
l
e
a
se
c
o
me
o
u
t
a
l
t
e
r
n
a
t
i
v
e
q
u
i
c
k
b
o
o
k
s
s
o
s
ma
l
l
b
u
s
i
n
e
ss
o
w
n
e
r
b
e
r
e
l
e
a
se
f
r
o
m
c
o
n
st
a
n
t
f
o
r
c
e
u
p
g
r
a
d
e
a
n
d
r
e
mo
v
a
l
o
f
b
a
s
i
c
f
u
n
c
t
i
o
n
a
l
i
t
y
Th
e
p
r
o
d
u
c
t
b
e
m
e
d
i
o
c
r
e
a
t
g
o
o
d
a
n
d
q
u
i
c
k
b
o
o
k
s
r
e
mo
v
e
b
a
si
c
f
u
n
c
t
i
o
n
a
l
i
t
y
f
r
o
m
t
h
e
b
a
se
p
r
o
d
u
c
t
o
n
e
a
c
h
u
p
g
r
a
d
e
a
n
d
f
o
r
c
e
i
t
s
c
u
st
o
mer
t
o
p
a
y
e
x
t
r
a
f
o
r
i
t
s
h
o
r
r
e
n
d
o
u
s
b
u
si
n
e
s
s p
r
a
c
t
i
c
e
,
m
e
d
i
o
c
r
e
p
r
o
d
u
c
t
,
i
n
b
u
s
i
n
e
ss
b
e
c
a
u
se
o
f
l
a
c
k
o
f
c
o
mp
e
t
i
t
i
o
n
S
o
f
t
w
a
r
e
I
t
i
s a
g
o
o
d
p
r
o
d
u
c
t
b
u
t
t
h
e
o
n
l
y
t
h
i
n
g
b
e
i
t
j
u
s
t
sm
e
l
l
r
e
a
l
l
y
b
a
d
i
f
y
o
u
r
l
o
o
k
f
o
r
a
sm
o
o
t
h
r
e
m
o
v
a
l
t
h
e
n
t
h
i
s
b
e
i
t
g
o
o
d
p
r
o
d
u
c
t
B
e
a
u
t
y
Of
c
o
u
r
se,
I
w
i
l
l
h
a
v
e
t
o
w
a
i
t
a
y
e
a
r
t
o
se
e
w
h
a
t
h
a
p
p
e
n
b
u
t
w
h
e
n
I
t
r
y
t
o
c
a
n
c
e
l
a
u
t
o
r
e
n
e
w
a
l
o
n
a
n
o
t
h
e
r
mag
a
z
i
n
e
w
h
a
t
a
h
a
ssl
e
w
i
l
l
w
a
i
t
a
n
d
see
h
o
w
t
h
i
s
o
n
e
g
o
a
u
t
o
r
e
n
e
w
a
l
M
a
g
a
z
i
n
e
su
b
s
c
r
i
p
t
i
o
n
Fig
u
r
e
3
s
h
o
ws
th
e
r
esu
lts
o
f
m
ix
in
g
d
if
f
er
en
t
ca
teg
o
r
ies
o
f
r
ev
iews
to
g
et
m
o
r
e
ac
cu
r
ate
o
u
tco
m
es.
T
h
r
ee
d
ataset
ca
teg
o
r
ies
wer
e
co
m
b
i
n
ed
to
d
iv
er
s
if
y
th
e
t
y
p
e
o
f
clien
ts
b
y
c
h
o
o
s
in
g
r
e
v
i
ews
th
at
b
elo
n
g
t
o
d
iv
er
s
e
ce
n
ter
s
o
f
i
n
ter
est an
d
th
u
s
ca
p
tu
r
e
th
e
n
u
a
n
ce
s
o
f
la
n
g
u
ag
e
u
s
e
th
r
o
u
g
h
d
if
f
er
e
n
t t
y
p
es o
f
clien
ts
.
W
o
r
d
d
is
tr
ib
u
tio
n
f
r
eq
u
en
cy
p
er
ca
teg
o
r
y
h
as
also
b
ee
n
v
i
s
u
alize
d
to
ex
p
lo
r
e
th
is
d
atas
et
f
u
r
th
er
.
T
h
e
f
iv
e
m
o
s
t
f
r
e
q
u
en
t
w
o
r
d
s
b
y
ea
c
h
ca
teg
o
r
y
a
r
e
d
is
p
lay
ed
in
Fig
u
r
es
4
,
5
,
a
n
d
6
.
T
h
e
r
esu
lts
s
h
o
w
th
at
m
o
s
t
to
p
wo
r
d
s
ar
e
r
elate
d
to
th
e
ca
teg
o
r
y
s
u
b
ject.
E
x
a
m
p
les
o
f
So
f
twar
e
in
clu
d
e
s
o
f
twar
e,
h
ar
d
war
e,
an
d
co
m
p
u
ter
.
T
h
e
class
‘
o
v
er
all’
la
b
eled
th
e
d
ata
a
n
d
was
s
et
as
th
e
tar
g
et
co
lu
m
n
.
Data
lab
elin
g
en
ab
les
th
e
m
o
d
el
to
d
if
f
e
r
en
tiate
p
atter
n
s
an
d
c
o
n
n
ec
ti
o
n
s
with
in
t
h
e
d
ata
d
u
r
i
n
g
th
e
last
co
m
p
ar
is
o
n
p
h
ase.
T
h
e
Am
az
o
n
r
ev
iew
d
ataset
u
s
ed
in
th
is
s
tu
d
y
lack
s
a
d
ef
au
lt
l
ab
el
co
lu
m
n
.
Ho
wev
er
,
a
h
u
m
an
an
n
o
tato
r
ca
n
in
f
er
th
e
g
e
n
er
al
s
en
tim
en
t
th
r
o
u
g
h
th
e'
o
v
er
all'
co
lu
m
n
.
Nev
er
th
eless
,
th
is
r
esear
ch
d
o
e
s
n
o
t
s
o
lely
r
ely
o
n
m
an
u
al
lab
elin
g
an
d
e
v
alu
ate
s
well
-
estab
li
s
h
ed
m
ac
h
in
e
le
ar
n
in
g
a
n
d
tex
t
m
i
n
in
g
tech
n
i
q
u
es
p
r
e
-
tr
ain
ed
o
n
E
n
g
lis
h
v
o
ca
b
u
lar
y
.
A
h
s
ain
et
a
l.
[
2
2
]
,
a
co
m
p
ar
is
o
n
b
etwe
e
n
m
an
u
al
lab
elin
g
,
Vad
er
to
o
l
[
2
3
]
a
n
d
T
ex
tB
lo
b
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
.
1
,
Feb
r
u
ar
y
20
25
:
9
8
0
-
994
986
to
o
l
[
2
4
]
u
s
ed
o
n
th
e
s
am
e
d
ataset,
co
n
clu
d
e
d
th
at
th
e
r
esu
lts
o
f
t
h
ese
th
r
ee
m
eth
o
d
s
wer
e
clo
s
e.
Fo
r
th
is
p
ap
er
'
s
co
n
tin
u
atio
n
,
th
e
d
at
aset
was
lab
eled
u
s
in
g
th
e
Vad
er
to
o
l
an
d
was
class
if
ied
in
to
two
class
es
(
n
eg
ativ
e
an
d
p
o
s
itiv
e
r
ev
iews).
Fig
u
r
e
3.
Dis
tr
ib
u
tio
n
o
f
class
‘
ca
teg
o
r
y
’
af
ter
m
i
x
in
g
d
atasets
Fig
u
r
e
4
.
T
h
e
f
iv
e
m
o
s
t f
r
eq
u
e
n
t w
o
r
d
s
in
th
e
‘
s
o
f
twar
e’
ca
teg
o
r
y
Fig
u
r
e
5
.
T
h
e
f
iv
e
f
r
eq
u
en
t w
o
r
d
s
o
f
th
e
ca
te
g
o
r
y
‘
m
ag
az
in
e
s
u
b
s
cr
ip
tio
n
’
Fig
u
r
e
6
.
T
h
e
f
iv
e
t
o
p
f
r
eq
u
e
n
t w
o
r
d
s
o
f
th
e
ca
teg
o
r
y
‘
b
ea
u
t
y
’
3
.
2
.
1
.
F
ea
t
ure
e
x
t
ra
ct
i
o
n
W
ith
in
n
atu
r
al
lan
g
u
a
g
e
p
r
o
ce
s
s
in
g
(
NL
P),
th
e
f
ea
tu
r
e
e
x
tr
a
ctio
n
s
tep
is
a
f
u
n
d
am
e
n
tal
p
r
o
ce
s
s
th
at
aid
s
m
ac
h
in
e
lear
n
in
g
s
y
s
tem
s
in
d
ec
ip
h
e
r
in
g
tex
tu
al
d
ata
b
y
co
n
v
er
tin
g
it
in
t
o
n
u
m
er
ic
al
r
ep
r
esen
tatio
n
s
.
T
h
is
n
o
t
o
n
ly
f
ac
ilit
ates
m
ac
h
in
e
co
m
p
r
e
h
en
s
io
n
o
f
tex
tu
al
in
f
o
r
m
atio
n
b
u
t
also
s
er
v
es
to
h
av
e
a
m
in
im
alis
t
d
ata
r
ep
r
esen
tatio
n
.
T
h
is
s
tep
is
im
p
o
r
tan
t
b
ec
au
s
e
o
f
th
e
r
o
le
it
p
lay
s
in
m
itig
atin
g
th
e
ch
allen
g
es
o
f
h
ig
h
d
im
en
s
io
n
ality
,
wh
ich
r
esu
lts
in
th
e
en
h
an
ce
m
en
t
o
f
co
m
p
u
tatio
n
al
ef
f
icien
cy
.
I
t
also
co
n
tr
ib
u
tes
to
th
e
g
en
er
aliza
tio
n
ca
p
ab
ilit
ies o
f
th
e
f
r
am
ewo
r
k
.
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
P
r
o
d
u
ct
r
ev
iew
s
a
n
a
lysi
s
to
ex
tr
a
ct
s
en
timen
ta
l in
s
ig
h
ts
w
ith
cla
s
s
co
n
fid
en
ce
…
(
S
a
r
a
A
h
s
a
in
)
987
T
h
is
s
tu
d
y
u
s
ed
th
e
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
(TF
-
I
DF)
alg
o
r
ith
m
.
T
F
-
I
DF
is
a
wid
ely
ad
o
p
ted
tech
n
i
q
u
e
f
o
r
tr
an
s
f
o
r
m
in
g
tex
tu
al
co
n
ten
t
i
n
to
a
s
tr
u
ctu
r
ed
r
ep
r
esen
tatio
n
wh
ile
en
s
u
r
in
g
th
e
s
ig
n
if
ican
ce
o
f
w
o
r
d
s
ac
r
o
s
s
th
e
en
tire
d
o
cu
m
en
t.
T
h
e
p
ap
er
aim
s
to
en
ca
p
s
u
late
th
e
im
p
o
r
tan
ce
o
f
wo
r
d
s
with
in
th
e
tex
t a
n
d
co
n
tr
ib
u
te
to
th
e
o
v
e
r
all
ef
f
icac
y
o
f
th
e
f
ea
tu
r
e
ex
tr
ac
tio
n
p
r
o
ce
s
s
.
T
h
e
T
F
-
I
DF
alg
o
r
ith
m
,
as
in
d
icate
d
b
y
th
e
p
a
p
er
s
[
2
5
]
–
[
2
7
]
ca
p
tu
r
es
a
wo
r
d
'
s
u
n
iq
u
en
ess
b
y
co
m
p
ar
in
g
its
o
cc
u
r
r
e
n
ce
in
a
d
o
cu
m
e
n
t
an
d
its
p
r
e
v
alen
ce
ac
r
o
s
s
v
ar
io
u
s
d
o
c
u
m
en
ts
.
T
h
e
ter
m
“
TF
”
s
tan
d
s
f
o
r
T
e
r
m
Fre
q
u
en
c
y
,
th
e
n
u
m
b
er
o
f
tim
es
a
wo
r
d
ap
p
ea
r
s
in
a
d
o
cu
m
en
t;
“
I
DF
”
s
tan
d
s
f
o
r
I
n
v
e
r
s
e
Do
cu
m
e
n
t
Fre
q
u
en
cy
,
th
e
n
u
m
b
er
o
f
d
o
c
u
m
en
ts
in
wh
ich
a
s
p
ec
if
ic
w
o
r
d
h
as a
p
p
e
ar
ed
[
2
8
]
.
−
(
)
=
(
,
)
∗
(
)
(
1
)
TF
-
I
DF
s
elec
tiv
ely
ca
p
tu
r
es
t
er
m
s
th
at
ar
e
f
r
eq
u
e
n
t
with
in
a
s
in
g
u
lar
r
e
v
iew
b
u
t
n
o
t
ac
r
o
s
s
th
e
en
tire
t
y
o
f
r
ev
iews
[
2
2
]
.
3
.
2
.
2
.
F
ea
t
ure
s
elec
t
io
n
T
h
e
f
ea
tu
r
e
s
elec
tio
n
s
tep
is
cr
u
cial
to
g
ettin
g
in
ter
esti
n
g
r
es
u
lts
.
As
d
etailed
in
th
e
p
r
ev
io
u
s
s
ec
tio
n
,
th
e
n
u
m
b
er
o
f
f
ea
tu
r
es
o
u
t
g
r
o
ws
th
e
n
u
m
b
er
o
f
r
e
v
iews.
T
h
is
im
p
lies
th
at
th
e
m
o
d
el
tr
a
in
in
g
will
r
e
q
u
ir
e
a
lar
g
e
co
m
p
u
tatio
n
al
ca
p
ac
ity
an
d
will
p
r
o
b
a
b
ly
h
a
v
e
f
ewe
r
r
esu
lts
b
ec
au
s
e
a
lar
g
e
p
ar
t
o
f
th
e
f
ea
tu
r
es
is
g
en
er
ic
an
d
d
o
es
n
o
t in
f
lu
e
n
ce
th
e
p
r
ed
ictio
n
o
f
th
e
cu
s
to
m
er
'
s
o
v
er
all
ap
p
r
ec
iatio
n
.
B
ased
o
n
th
e
r
esu
lts
o
f
t
h
e
f
e
atu
r
e
s
elec
tio
n
s
tep
in
th
e
p
ap
er
[
2
0
]
,
lig
h
t
g
r
ad
ie
n
t
b
o
o
s
ted
m
ac
h
in
e
L
ig
h
tGB
M
r
ep
r
esen
ts
o
n
e
o
f
th
e
lig
h
test
a
n
d
m
o
s
t
ef
f
ec
tiv
e
m
eth
o
d
s
f
o
r
f
ea
tu
r
e
s
elec
tio
n
an
d
class
if
icatio
n
.
T
h
e
r
esu
lts
in
th
e
p
a
p
er
[
2
2
]
a
d
em
o
n
s
tr
atio
n
o
f
L
ig
h
tGB
M
was
ap
p
lied
to
th
is
d
ata.
T
h
e
in
itial
co
r
p
u
s
co
n
s
is
ted
o
f
2
5
,
3
7
4
f
ea
tu
r
es,
with
a
tar
g
et
class
co
n
s
is
tin
g
o
f
p
o
s
itiv
e
o
r
n
eg
ativ
e
r
ev
ie
ws.
L
ig
h
tGB
M
was
u
s
ed
to
s
elec
t
th
e
b
est
n
u
m
b
e
r
o
f
f
ea
tu
r
es.
I
t
s
elec
ts
an
in
c
r
ea
s
in
g
n
u
m
b
er
o
f
th
e
m
o
s
t
r
elev
an
t
f
ea
tu
r
es
to
f
ee
d
th
e
r
ev
iews'
co
r
r
esp
o
n
d
i
n
g
v
ec
to
r
s
to
t
h
e
m
o
d
els:
C
AR
T
,
SVM,
an
d
ML
P.
Fig
u
r
e
7
s
h
o
ws
th
at
th
e
r
esu
lts
s
tag
n
ate
at
a
co
r
p
u
s
o
f
1
0
0
f
ea
t
u
r
es.
Fig
u
r
e
7
.
B
est f
ea
tu
r
es u
s
in
g
L
ig
h
tGB
M
3
.
2
.
3
.
H
o
w
s
elf
-
o
r
g
a
nizin
g
ma
p m
o
del w
o
r
k
s
SOM
m
o
d
el
u
s
es
an
u
n
s
u
p
e
r
v
is
ed
lear
n
i
n
g
t
y
p
e
ca
lled
co
m
p
etitiv
e
lear
n
in
g
,
a
s
elf
-
o
r
g
an
izin
g
p
r
o
ce
d
u
r
e
u
s
ed
to
m
atch
ea
ch
in
p
u
t
v
ec
to
r
with
a
n
eu
r
o
n
in
a
2
D
g
r
id
/m
ap
o
f
n
e
u
r
o
n
s
[
2
9
]
.
T
h
e
k
e
y
id
ea
o
f
th
e
SOM
m
o
d
el
is
th
at
n
o
d
es
lo
ca
ted
clo
s
e
to
ea
ch
o
th
er
i
n
th
e
m
ap
h
av
e
weig
h
t
v
ec
to
r
s
co
r
r
esp
o
n
d
in
g
t
o
d
ata
s
am
p
les
s
i
tu
ated
clo
s
e
to
ea
ch
o
th
er
in
th
e
d
ata
s
p
ac
e.
T
r
ain
in
g
s
am
p
les
ar
e
in
tr
o
d
u
c
ed
o
n
e
at
a
tim
e
to
th
e
iter
ativ
e
tr
ain
in
g
alg
o
r
ith
m
to
u
p
d
ate
weig
h
ts
,
in
itialized
r
an
d
o
m
ly
,
an
d
m
o
v
e
t
h
e
c
o
r
r
esp
o
n
d
in
g
v
ec
to
r
s
to
war
d
d
en
s
e
r
e
g
io
n
s
o
f
th
e
d
a
ta
s
p
ac
e.
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
.
1
,
Feb
r
u
ar
y
20
25
:
9
8
0
-
994
988
Du
r
in
g
th
e
tr
ain
in
g
p
r
o
ce
s
s
,
th
e
SOM
p
r
o
g
r
ess
iv
ely
m
ap
s
th
e
h
ig
h
er
-
d
im
en
s
io
n
al
in
p
u
t
s
p
ac
e
to
a
2
D
m
ap
,
p
r
eser
v
in
g
th
e
to
p
o
l
o
g
ical
p
r
o
p
er
ties
o
f
th
e
in
p
u
t d
a
ta.
T
h
is
m
ea
n
s
th
at
s
im
ilar
o
r
n
eig
h
b
o
r
in
g
in
p
u
ts
will
b
e
r
e
p
r
esen
ted
b
y
n
ea
r
b
y
n
o
d
es
o
n
th
e
m
a
p
.
As
a
r
esu
lt,
SOM
ca
n
b
e
u
s
ed
t
o
clu
s
te
r
an
d
v
is
u
alize
th
e
r
elatio
n
s
h
ip
s
am
o
n
g
th
e
in
p
u
t
d
ata.
T
ak
in
g
an
in
p
u
t
d
ata
o
f
s
ize
(
m
,
n
)
wh
er
e
m
is
th
e
n
u
m
b
er
o
f
tr
ain
in
g
s
am
p
les
an
d
n
is
th
e
n
u
m
b
e
r
o
f
f
ea
tu
r
es.
W
e
s
tar
t
with
b
y
th
e
in
itializatio
n
o
f
th
e
weig
h
t’
s
m
atr
ix
,
o
f
s
ize
(
n
,
L
,
C
)
,
wh
er
e
L
×
C
is
th
e
n
u
m
b
er
o
f
n
o
d
es/n
eu
r
o
n
s
o
f
th
e
m
ap
,
L
an
d
C
ar
e
th
e
n
u
m
b
er
o
f
lin
es
an
d
co
l
u
m
n
s
o
f
th
e
m
a
p
.
T
h
en
,
iter
atin
g
o
v
er
t
h
e
in
p
u
t
d
ata,
ea
ch
tr
ain
in
g
s
am
p
le
u
p
d
ates
th
e
win
n
in
g
n
o
d
e/n
e
u
r
o
n
,
wei
g
h
t
v
ec
to
r
with
t
h
e
s
h
o
r
test
d
is
tan
ce
f
r
o
m
th
e
tr
ain
in
g
s
am
p
le,
a
n
d
its
n
eig
h
b
o
r
s
[
3
0
]
.
As
th
e
tr
ain
in
g
p
r
o
g
r
ess
es,
th
e
SOM
o
r
g
an
izes
its
elf
s
o
th
at
n
ea
r
b
y
n
eu
r
o
n
s
o
n
th
e
g
r
i
d
r
esp
o
n
d
to
s
im
ilar
in
p
u
t
v
ec
to
r
s
.
T
h
is
m
ea
n
s
th
at
n
eig
h
b
o
r
i
n
g
n
eu
r
o
n
s
r
ef
lect
r
elatio
n
s
h
ip
s
b
etwe
e
n
in
p
u
t
v
ec
to
r
s
.
T
h
e
n
eu
r
o
n
th
at
m
o
s
t
clo
s
ely
m
at
ch
es
th
e
p
r
esen
te
d
in
p
u
t
p
att
er
n
is
r
ef
e
r
r
ed
t
o
as
th
e
win
n
er
n
e
u
r
o
n
o
r
b
est
m
atch
in
g
u
n
it.
T
h
is
n
eu
r
o
n
,
alo
n
g
with
its
n
ei
g
h
b
o
r
s
as
d
ef
in
ed
b
y
th
e
alg
o
r
ith
m
[
5
]
,
u
p
d
ates
its
weig
h
t
v
ec
to
r
s
b
ased
o
n
th
e
SOM
lear
n
in
g
r
u
les as (
2
)
:
(
+
1
)
=
(
)
+
(
)
ℎ
(
)
[
(
)
−
(
)
]
(
2
)
Her
e,
(
t)
r
ep
r
esen
ts
th
e
weig
h
t
b
etwe
en
n
o
d
e
i,
in
th
e
in
p
u
t
lay
er
,
a
n
d
n
o
d
e
j,
i
n
th
e
o
u
t
p
u
t
lay
er
,
at
th
e
iter
atio
n
tim
e
t.
α
(
t)
is
th
e
le
ar
n
in
g
r
ate,
wh
ich
d
ec
r
ea
s
es
o
v
er
tim
e.
ℎ
(
t)
is
th
e
n
eig
h
b
o
r
h
o
o
d
f
u
n
ctio
n
,
wh
ich
d
ef
i
n
es
th
e
s
ize
o
f
t
h
e
n
eig
h
b
o
r
h
o
o
d
ar
o
u
n
d
th
e
win
n
in
g
n
o
d
e
to
b
e
u
p
d
ate
d
d
u
r
i
n
g
t
h
e
lear
n
in
g
p
r
o
ce
s
s
.
I
n
th
e
f
in
al
s
tag
e,
th
e
weig
h
t v
ec
to
r
s
o
f
all
ac
tiv
ated
n
eu
r
o
n
s
ar
e
u
p
d
ated
ac
c
o
r
d
in
g
ly
[
5
]
.
T
h
e
m
ap
p
r
o
v
id
es
a
v
is
u
al
r
ep
r
esen
tatio
n
o
f
t
h
e
r
elat
io
n
s
h
ip
s
b
etwe
en
in
p
u
t
v
ec
to
r
s
.
T
h
e
n
eig
h
b
o
r
in
g
n
e
u
r
o
n
s
o
n
th
e
m
ap
co
r
r
esp
o
n
d
to
s
im
ilar
in
p
u
ts
,
wh
ich
allo
ws f
o
r
v
is
u
aliza
ti
o
n
o
f
clu
s
ter
s
in
th
e
d
ata.
E
ac
h
class
was
g
iv
en
a
s
y
m
b
o
l
‘
x
’
,
‘
o
’
,
o
r
‘
r
ec
ta
n
g
le’
with
d
if
f
er
e
n
t
co
l
o
r
i
n
ten
s
ities
,
g
r
e
y
s
ca
le
lev
els.
T
h
e
s
y
m
b
o
ls
‘
o
’
an
d
th
e
‘
r
ec
t
an
g
le’
wer
e
attr
ib
u
ted
to
th
e
c
lass
0
o
r
1
,
r
esp
ec
tiv
ely
;
th
e
s
y
m
b
o
l
‘
x
’
s
h
o
wed
eq
u
al
v
o
tin
g
f
o
r
b
o
th
class
es (
ca
n
b
e
ca
lled
n
eu
tr
al)
.
4.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
4
.
1
.
Usi
ng
s
elf
-
o
rg
a
nizin
g
m
a
p mo
del
T
h
e
SOM
m
ap
s
ize
is
f
i
x
ed
,
s
p
ec
if
y
in
g
th
e
n
u
m
b
er
o
f
r
o
ws
an
d
co
lu
m
n
s
eq
u
al
to
1
5
in
t
h
e
p
r
esen
t
ex
p
er
im
en
t.
Fig
u
r
e
8
s
h
o
ws
t
h
e
n
eu
r
o
n
lo
ca
tio
n
s
in
th
e
m
ap
an
d
th
e
cl
u
s
ter
'
s
to
p
o
lo
g
y
,
r
esu
ltin
g
f
r
o
m
t
h
e
v
o
tin
g
p
r
o
ce
s
s
,
wh
en
all
th
e
tr
ain
in
g
s
am
p
les
a
r
e
v
en
tilated
o
v
er
t
h
e
m
a
p
ac
c
o
r
d
in
g
to
th
e
win
n
er
n
e
u
r
o
n
co
m
p
u
tin
g
r
u
le.
SOM
is
f
ir
s
t
tr
ain
ed
in
an
u
n
s
u
p
er
v
is
ed
m
an
n
er
to
m
ap
in
p
u
t
d
ata
in
t
o
clu
s
ter
s
,
g
r
o
u
p
in
g
th
em
b
y
f
ea
tu
r
e
v
ec
to
r
s
im
ilar
ity
.
T
h
en
,
in
a
s
u
p
er
v
is
ed
m
an
n
er
,
lab
els
ar
e
allo
ca
ted
to
SOM
n
o
d
es,
ea
ch
clu
s
ter
with
an
in
d
ep
en
d
en
t
la
b
el.
T
h
e
tr
ain
in
g
s
am
p
les
ar
e
v
en
tilated
o
v
er
t
h
e
m
ap
ac
c
o
r
d
in
g
to
th
e
win
n
er
n
eu
r
o
n
co
m
p
u
tin
g
r
u
le
u
s
ed
i
n
th
e
tr
ain
in
g
p
h
ase.
Af
ter
t
h
is
,
th
e
m
ajo
r
ity
v
o
tin
g
is
ap
p
li
ed
to
d
eter
m
i
n
e
th
e
m
o
s
t
co
m
m
o
n
lab
el
f
o
r
d
ata
s
am
p
les
as
s
o
ciate
d
with
ea
ch
n
o
d
e.
New
d
ata
is
cla
s
s
if
ied
b
y
ass
ig
n
in
g
th
e
n
ea
r
est n
o
d
e
la
b
el
to
it.
T
h
e
SOM
n
eu
r
al
n
etwo
r
k
m
o
d
el
is
im
p
lem
en
ted
u
s
in
g
t
h
e
Min
i
SOM
lib
r
ar
y
o
f
Scik
it
-
lear
n
.
T
h
e
lib
r
ar
y
allo
ws
u
s
er
s
to
ex
p
lo
r
e
h
o
w
o
f
ten
n
eu
r
o
n
s
h
av
e
wo
n
th
e
co
m
p
etitio
n
.
W
h
en
p
ass
in
g
th
r
o
u
g
h
th
e
d
ataset
s
am
p
les,
f
in
e
-
tu
n
in
g
is
m
ad
e
to
g
iv
e
all
n
eu
r
o
n
s
th
e
c
h
an
ce
to
p
ar
ticip
ate
in
co
n
test
s
.
T
h
e
in
itializatio
n
m
eth
o
d
s
ets
em
p
ir
ical
p
ar
am
eter
s
s
u
ch
as
n
etwo
r
k
d
im
e
n
s
io
n
s
:
(
5
×
5
)
,
(
1
0
×
1
0
)
,
(
1
5
×
1
5
)
,
an
d
(
2
0
×
2
0
)
,
as
well
as
th
e
n
u
m
b
er
o
f
tr
ain
in
g
c
y
cles
(
1
0
0
)
.
T
h
e
clu
s
ter
s
will
b
e
m
ap
p
ed
to
th
e
p
r
o
b
lem
class
u
s
in
g
th
e
tr
ain
i
n
g
s
am
p
les
lab
els.
T
h
e
r
esu
lt
is
a
m
esh
g
r
id
th
at
illu
s
tr
ate
s
th
e
n
eig
h
b
o
r
h
o
o
d
r
elatio
n
s
h
ip
s
b
etwe
en
n
eu
r
o
n
s
.
T
h
e
alg
o
r
ith
m
em
p
lo
y
s
a
win
n
er
-
tak
es
-
all
ap
p
r
o
ac
h
,
wh
er
e
th
e
n
eu
r
o
n
-
r
elate
d
weig
h
t
v
ec
to
r
,
th
e
clo
s
est
to
t
h
e
in
p
u
t
p
atter
n
(
th
e
f
ittes
t
n
e
u
r
o
n
)
,
is
u
p
d
ated
alo
n
g
with
i
ts
n
eig
h
b
o
r
s
.
Af
ter
tr
ain
in
g
,
th
e
n
etwo
r
k
ca
n
b
e
u
s
ed
f
o
r
lab
elin
g
a
n
d
class
if
icatio
n
task
s
.
T
h
e
lear
n
i
n
g
r
ate
in
itial
v
alu
e
is
α
(
0
)
=0
.
9
,
a
d
ec
ay
f
u
n
ctio
n
is
u
s
ed
to
r
ed
u
ce
th
e
lear
n
in
g
r
ate
o
v
er
iter
atio
n
s
[
3
1
]
,
as illu
s
tr
ated
in
Fig
u
r
e
8
.
(
)
=
(
0
)
(
1
−
_
+
1
)
(3
)
T
h
e
s
am
e
p
r
o
ce
s
s
es
h
av
e
b
ee
n
r
u
n
o
n
d
if
f
er
en
t
d
im
en
s
io
n
s
b
ef
o
r
e
ch
o
o
s
in
g
t
h
e
o
p
tim
al
g
en
er
ated
m
ap
to
en
s
u
r
e
th
at
th
e
m
ap
’
s
co
n
f
ig
u
r
atio
n
ef
f
ec
tiv
ely
m
e
ets
th
e
an
aly
tical
g
o
als
an
d
p
r
o
v
id
es
r
o
b
u
s
t
an
d
in
ter
p
r
etab
le
r
esu
lts
.
I
n
Fig
u
r
e
9
.
d
if
f
er
e
n
t
d
im
e
n
s
io
n
s
wer
e
p
r
o
ce
s
s
ed
in
Fig
u
r
e
9
(
a)
a
n
ex
am
p
le
o
f
5
×
5
d
im
en
s
io
n
s
,
in
Fig
u
r
e
9
(
b
)
th
e
d
im
en
s
io
n
s
wer
e
s
lig
h
tly
in
cr
ea
s
ed
to
1
0
×
1
0
,
th
en
1
5
×
1
5
in
Fig
u
r
e
9
(
c)
an
d
f
in
ally
,
th
e
r
esu
lts
wer
e
also
p
r
o
ce
s
s
ed
o
n
a
2
0
×2
0
m
ap
.
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
P
r
o
d
u
ct
r
ev
iew
s
a
n
a
lysi
s
to
ex
tr
a
ct
s
en
timen
ta
l in
s
ig
h
ts
w
ith
cla
s
s
co
n
fid
en
ce
…
(
S
a
r
a
A
h
s
a
in
)
989
Fig
u
r
e
8
.
Dec
ay
f
u
n
cti
o
n
o
f
th
e
lear
n
in
g
r
ate
(
a)
(
b
)
(
c)
(
d
)
Fig
u
r
e
9
.
SOM
m
ap
o
u
tp
u
t w
ith
m
u
ltip
le
s
izes: (
a)
5
×
5
,
(
b
)
1
0
×
1
0
,
(
c)
1
5
×
15
,
an
d
(
d
)
2
0
×
20
C
h
o
o
s
in
g
th
e
p
r
o
p
er
d
im
en
s
i
o
n
ality
f
o
r
a
SOM
h
o
ld
s
g
r
ea
t
im
p
o
r
tan
ce
as
it
d
ir
ec
tly
in
f
l
u
en
ce
s
th
e
m
ap
’
s
ca
p
ac
ity
to
ac
cu
r
ately
r
ep
r
esen
t
an
d
d
is
p
lay
co
m
p
lex
d
ata
r
elatio
n
s
h
ip
s
.
T
h
e
d
ec
is
io
n
to
g
o
with
15
-
d
im
e
n
s
io
n
SOM
in
s
tead
o
f
o
th
er
alter
n
ativ
es,
s
u
ch
as
5
,
1
0
,
o
r
2
0
,
ca
n
b
e
in
f
lu
e
n
ce
d
b
y
v
ar
i
o
u
s
f
ac
to
r
s
.
T
h
e
d
ataset
s
ize
an
d
lev
el
o
f
co
m
p
lex
ity
ar
e
n
o
t
ad
e
q
u
ate
ly
ca
p
tu
r
ed
b
y
s
m
aller
m
ap
s
,
s
u
ch
as
5
o
r
10
d
im
en
s
io
n
s
,
y
et
it
d
o
es
n
ot
r
eq
u
ir
e
t
h
e
ad
d
itio
n
al
g
r
a
n
u
lar
ity
an
d
in
ter
p
r
etatio
n
co
m
p
lex
ity
t
h
at
a
2
0
-
d
im
e
n
s
io
n
al
m
ap
wo
u
ld
p
r
o
v
id
e.
T
h
e
1
5
-
d
im
e
n
s
io
n
m
ap
is
al
s
o
co
n
s
id
er
ed
to
b
e
a
m
o
r
e
m
an
ag
ea
b
le
an
d
v
is
u
ally
ac
ce
s
s
ib
le
m
ap
co
m
p
ar
ed
to
lar
g
er
m
ap
s
.
T
h
e
g
o
al
is
to
s
p
o
t
p
atter
n
s
th
at
co
u
ld
b
e
ch
allen
g
in
g
t
o
Evaluation Warning : The document was created with Spire.PDF for Python.