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.
3
,
J
une
2025
, pp.
1839
~
1852
I
S
S
N
:
2252
-
8938
,
D
O
I
:
10.11591/
ij
a
i.
v
14
.i
3
.pp
1839
-
1852
1839
Jou
r
n
al
h
om
e
page
:
ht
tp
:
//
ij
ai
.
ia
e
s
c
or
e
.c
om
H
e
t
e
r
og
e
n
e
ou
s
se
m
an
t
i
c
gr
ap
h
e
m
b
e
d
d
i
n
g
a
ss
i
st
e
d
e
d
ge
-
se
n
si
t
i
ve
l
e
ar
n
i
n
g
f
or
c
r
oss
-
d
om
ai
n
r
e
c
om
m
e
n
d
at
i
on
P
r
avi
n
R
am
c
h
an
d
r
a
P
at
il
,
P
r
am
od
H
al
e
b
id
u
B
as
avar
aj
u
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,
A
di
c
hunc
ha
na
gi
r
i
I
ns
t
i
t
ut
e
of
T
e
c
hnol
ogy
a
f
f
i
l
i
a
t
e
d
w
i
t
h
V
i
s
ve
s
va
r
a
ya
T
e
c
hnol
ogi
c
a
l
U
ni
ve
r
s
i
t
y
,
K
a
r
na
t
a
ka
,
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
O
ct
7, 2024
R
e
vi
s
e
d
F
e
b 15, 2025
A
c
c
e
pt
e
d
M
a
r
15, 2025
In
the
digital
age,
recommendatio
n
systems
navigate
vast
altern
atives.
Content
-
based,
collaborat
ive
filtering,
deep
-
driven,
and
cross
-
d
omain
recommendatio
n
(CDR)
have
been
studied
significantly
but
face
col
d
-
start
and
data
sparsity.
Though
CDR
methods
outperform
others,
they
stru
ggle
to
optimize
user
-
item
matrices.
Recent
graph
-
based
CDR
methods
improve
efficiency
by
leverag
ing
additio
nal
user
-
item
interactio
ns;
ho
wever,
optimizing
graph
features
remains
an
open
research
area.
Moreover,
current
techniques
do
not
consider
the
impact
of
noise
items
(unrelate
d)
on
recommendatio
n
accuracy.
To
address
this
g
ap,
this
paper
deve
lops
a
heterogeneous
semantic
graph
-
embeddin
g
(HSGE)
edge
-
pruning
mod
el
that
leverag
es
user
ratings
and
item
metadata
in
the
source
and
target
dom
ains
to
recommend
items
to
target
domain
users.
To
achieve
it,
at
first
Wor
d2Vec
method
is
applied
to
explicit
and
implicit
details,
followed
by
Node2Vec
-
driven
graph
embeddin
g
matrix
generation.
Our
HSGE
method
obtain
s
user
-
user,
user
-
item,
and
item
-
item
connectio
ns
to
achieve
more
se
mantic
features.
To
improve
accuracy,
our
model
prunes
edges
that
drop
source
domain
items
and
allied
edges
unrelated
to
the
target
domain
users.
Subsequ
ently,
the
retained
HSGE
matrices
from
both
domains
are
processed
for
element
-
wise
attenti
on.
A
multi
-
layer
perceptron
with
cosine
si
milarity
processed
combined
features
matrices
to
generate
top
-
N
recommen
dations
with
superior
hit
-
rate
(HR)
and
normalized
discounted
cumulati
v
e
gain
(NDCG).
K
e
y
w
o
r
d
s
:
A
tt
e
nt
io
n
m
e
th
od
C
r
os
s
-
dom
a
in
r
e
c
om
m
e
nda
ti
on
E
dge
pr
uni
ng
H
e
te
r
oge
ne
ous
s
e
m
a
nt
ic
gr
a
ph
e
m
be
ddi
ng
Node2Vec
Word2Vec
This
is
an
open
access
article
under
the
CC
BY
-
SA
license.
C
or
r
e
s
pon
di
n
g
A
u
th
or
:
P
r
a
vi
n
R
a
m
c
ha
ndr
a
P
a
ti
l
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
A
di
c
hunc
ha
na
gi
r
i
I
ns
ti
tu
te
of
T
e
c
hnol
ogy a
f
f
il
ia
te
d w
it
h V
is
ve
s
va
r
a
ya
T
e
c
hnol
ogi
c
a
l
U
ni
ve
r
s
it
y
B
e
la
ga
vi
590018,
K
a
r
na
ta
ka
,
I
ndi
a
E
m
a
il
:
m
r
.
pa
ti
l1
234
@
gm
a
il
.c
om
1.
I
N
T
R
O
D
U
C
T
I
O
N
O
ve
r
th
e
pa
s
t
f
e
w
ye
a
r
s
,
th
e
r
e
ha
s
be
e
n
hi
gh
gr
ow
th
in
di
gi
ta
l
da
ta
a
nd
a
ll
ie
d
a
ppl
ic
a
ti
ons
s
e
r
vi
ng
va
r
io
us
pur
pos
e
s
,
in
c
lu
di
ng
bu
s
in
e
s
s
c
om
m
uni
c
a
ti
on,
a
nd
r
e
c
om
m
e
nda
ti
on
s
ys
te
m
s
.
As
r
is
in
g
di
gi
ta
l
d
a
ta
vol
um
e
r
e
qui
r
e
s
r
obus
t
a
na
ly
ti
c
s
to
he
lp
us
e
r
s
na
vi
ga
te
num
e
r
ous
opt
io
ns
,
as
th
e
c
ha
ll
e
nge
of
c
hoos
in
g
th
e
r
ig
ht
one
can
of
te
n
le
a
d
to
une
xpe
c
te
d
de
c
is
io
ns
[
1]
,
[
2
]
.
O
ve
r
th
e
r
is
in
g
popula
ti
on
a
nd
a
ll
ie
d
di
gi
ta
l
da
ta
,
r
e
c
om
m
e
nde
r
s
ys
t
e
m
s
h
a
ve
ga
in
e
d
w
id
e
s
pr
e
a
d
a
tt
e
nt
io
n
to
a
l
le
vi
a
te
c
ha
ll
e
ng
e
s
as
m
e
nt
io
ne
d
e
a
r
li
e
r
[
3]
in
onl
in
e
di
gi
ta
l
e
c
os
ys
te
m
s
[
4]
.
R
e
c
om
m
e
nde
r
s
ys
te
m
s
of
te
n
e
xpl
oi
t
us
e
r
s
’
pr
e
vi
ous
di
gi
ta
l
be
ha
vi
or
,
pur
c
ha
s
e
s
,
a
nd
pr
e
f
e
r
e
nc
e
s
to
r
e
c
om
m
e
nd
c
e
r
ta
in
s
ui
t
a
bl
e
it
e
m
s
li
ke
pr
oduc
ts
or
s
e
r
vi
c
e
s
.
R
e
c
om
m
e
nde
r
s
ys
te
m
s
a
r
e
s
e
gm
e
nt
e
d
in
to
th
r
e
e
pr
im
a
r
y
c
a
te
gor
ie
s
[
4]
.
T
he
s
e
e
xi
s
t:
i
)
c
ol
la
bor
a
ti
ve
f
il
te
r
in
g,
ii
)
c
ont
e
nt
-
ba
s
e
d
r
e
c
om
m
e
nd
a
ti
on,
a
nd
ii
i)
hybr
id
r
e
c
om
m
e
nda
ti
on
.
T
h
e
c
ol
la
bor
a
ti
ve
f
il
te
r
m
e
th
ods
[
5]
,
[
6]
de
r
iv
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, No. 3, J
une
2025
:
1839
-
1852
1840
us
e
r
’
s
pr
e
f
e
r
e
nc
e
s
by
e
xpl
oi
ti
ng
th
e
ir
in
te
r
a
c
ti
on
w
it
h
us
e
r
s
a
n
d
it
e
m
s
w
it
hout
e
xpl
oi
ti
ng
th
e
us
e
r
’
s
or
it
e
m
’
s
c
ha
r
a
c
te
r
is
ti
c
s
.
On
th
e
c
ont
r
a
r
y,
c
ont
e
nt
-
ba
s
e
d
m
e
th
od
s
id
e
nt
if
y
us
e
r
pr
e
f
e
r
e
nc
e
s
by
a
s
s
e
s
s
in
g
th
e
s
im
il
a
r
it
y
or
s
ha
r
e
d
f
e
a
tu
r
e
s
of
th
e
it
e
m
s
th
a
t
th
e
us
e
r
ha
s
br
ought
or
a
v
a
il
e
d
of
[
7]
.
H
ybr
id
r
e
c
om
m
e
nda
ti
on
m
e
th
ods
c
om
bi
ne
c
ol
la
bor
a
ti
ve
f
il
te
r
in
g
a
nd
c
ont
e
nt
-
ba
s
e
d
r
e
c
om
m
e
nde
r
s
,
but
a
c
hi
e
vi
ng
opt
im
a
l
s
ol
ut
io
n
s
is
c
ha
ll
e
ngi
ng
due
to
da
ta
s
p
a
r
s
it
y
is
s
ue
s
[
8]
,
[
9]
.
P
r
e
s
e
nt
ly
,
m
os
t
at
-
ha
nd
r
e
c
om
m
e
nd
e
r
s
ol
ut
io
ns
r
e
ly
on
c
ol
la
bor
a
ti
ve
f
il
te
r
in
g
m
e
th
ods
[
10]
th
a
t
hypo
th
e
s
iz
e
th
a
t
us
e
r
s
w
it
h
s
im
il
a
r
in
te
r
e
s
ts
or
be
ha
vi
or
s
w
oul
d
ha
ve
s
im
il
a
r
it
e
m
c
hoi
c
e
s
.
T
hough
a
hi
s
to
r
y
-
ba
s
e
d
a
ppr
oa
c
h
is
e
a
s
ie
r
to
r
e
a
li
z
e
;
ye
t,
th
e
a
ll
ie
d
da
t
a
s
pa
r
s
it
y
pr
obl
e
m
r
e
m
a
in
s
a
bot
tl
e
ne
c
k.
A
m
ong
th
e
m
os
t
e
m
pl
oye
d
it
e
m
-
ba
s
e
d
c
ol
la
bor
a
ti
ve
f
il
te
r
in
g
m
e
th
ods
[
5]
m
e
a
s
ur
e
it
e
m
s
im
il
a
r
it
y
th
r
ough
it
e
m
-
us
e
r
in
te
r
a
c
ti
ons
.
T
he
s
e
m
e
th
ods
r
e
ly
on
in
ne
r
p
r
oduc
t
(
i.
e
.,
li
ne
a
r
c
om
bi
na
ti
on)
be
twe
e
n
th
e
us
e
r
a
nd
it
e
m
la
te
nt
ve
c
to
r
in
a
one
-
hot
c
odi
ng
a
ppr
oa
c
h,
w
hi
c
h
can
limit
pe
r
f
or
m
a
nc
e
a
nd
f
a
il
to
c
a
pt
ur
e
c
om
pl
e
x
in
te
r
a
c
ti
ons
.
A
c
c
ur
a
c
y
f
or
ne
w
u
s
e
r
s
is
li
m
it
e
d
du
e
to
in
s
uf
f
ic
ie
nt
in
f
or
m
a
ti
on
a
bout
ne
w
e
nt
r
a
nt
s
[
7]
,
a
pr
obl
e
m
known
as
da
ta
s
pa
r
s
it
y
.
T
hi
s
is
s
ue
can
be
a
ddr
e
s
s
e
d
by
a
m
a
lg
a
m
a
ti
ng
th
e
s
e
m
e
th
ods
,
w
hi
c
h
e
m
pl
oy
bot
h
us
e
r
r
e
vi
e
w
s
a
nd
it
e
m
m
e
ta
da
ta
f
or
r
e
c
om
m
e
nda
ti
on
de
c
is
io
ns
.
R
e
c
e
nt
ly
,
di
f
f
e
r
e
nt
m
e
th
ods
li
ke
to
pi
c
m
ode
li
ng
[
11
]
a
nd
de
e
p
ne
twor
ks
[
12]
ha
ve
be
e
n
de
s
ig
ne
d
to
e
xpl
oi
t
c
ont
e
nt
in
f
or
m
a
ti
on
to
m
a
ke
r
e
c
om
m
e
nda
ti
on
de
c
is
io
ns
.
U
nl
ik
e
th
e
s
e
m
e
th
ods
,
c
r
o
s
s
-
dom
a
in
r
e
c
om
m
e
nda
ti
on
(
C
D
R
)
m
e
th
ods
[
13]
e
xpl
oi
t
us
e
r
pr
e
f
e
r
e
nc
e
s
a
nd
it
e
m
c
ha
r
a
c
te
r
is
ti
c
s
f
r
om
ot
he
r
dom
a
in
s
to
im
pr
ove
da
ta
in
th
e
r
e
c
e
iv
in
g
dom
a
in
f
or
r
e
c
om
m
e
nda
ti
on.
F
or
in
s
ta
nc
e
,
C
D
R
can
r
e
c
om
m
e
nd
m
ovi
e
s
ba
s
e
d
on
a
us
e
r
'
s
book
pur
c
ha
s
e
s
or
s
ugge
s
t
hot
e
ls
ba
s
e
d
on
tr
a
ve
l
hi
s
to
r
y.
T
he
c
r
os
s
-
dom
a
in
r
e
c
om
m
e
nde
r
can
e
xpl
oi
t
s
uppl
e
m
e
nt
a
r
y
de
ta
il
s
f
r
om
th
e
va
r
ie
d
dom
a
in
s
to
a
ll
e
vi
a
te
th
e
pr
obl
e
m
of
c
ol
d
s
ta
r
t
a
nd
da
ta
s
p
a
r
s
it
y
[
14]
.
A
f
e
w
a
ppr
oa
c
he
s
a
ppl
ie
d
tr
a
n
s
f
e
r
le
a
r
ni
ng
m
e
th
ods
[
1
5]
,
w
hi
c
h
w
e
r
e
la
te
r
a
ppl
ie
d
to
de
r
iv
e
c
r
o
s
s
-
dom
a
in
c
ol
la
bor
a
ti
ve
f
il
te
r
in
g
(
C
D
C
F
)
[
16
]
,
[
17]
by
tr
a
ns
f
e
r
r
in
g
in
f
or
m
a
ti
on
f
r
om
th
e
or
ig
in
a
ti
ng
dom
a
in
to
th
e
r
e
c
e
iv
in
g
dom
a
in
f
o
r
an
a
c
c
ur
a
te
r
e
c
om
m
e
nda
ti
on
ba
s
e
d
on
m
a
tr
ix
f
a
c
to
r
iz
a
ti
on
(
M
F
)
[
18]
,
[
19]
.
H
ow
e
ve
r
,
th
e
s
e
m
e
th
ods
pr
im
a
r
il
y
le
a
r
n
ove
r
th
e
s
ha
ll
ow
a
n
d
li
ne
a
r
c
ha
r
a
c
te
r
is
ti
c
s
of
th
e
us
e
r
a
nd
a
ll
ie
d
it
e
m
s
.
A
f
e
w
a
ppr
oa
c
he
s
a
ppl
y
de
e
p
le
a
r
ni
ng
m
e
th
ods
f
or
th
e
la
te
nt
f
e
a
tu
r
e
e
xt
r
a
c
ti
on
f
r
om
u
s
e
r
s
a
nd
it
e
m
s
[
1]
,
[
20]
;
ye
t,
s
uc
h
m
e
th
od
s
e
m
pl
oy
one
-
hot
ve
c
to
r
as
in
put
a
n
d
he
nc
e
of
te
n
f
a
il
in
e
m
pl
oyi
ng
c
ol
la
bor
a
ti
ve
f
e
a
tu
r
e
s
be
twe
e
n
us
e
r
s
a
nd
it
e
m
s
.
S
uc
h
a
ppr
oa
c
he
s
a
r
e
f
ound
s
ui
ta
bl
e
onl
y
f
o
r
s
in
gl
e
-
dom
a
in
r
e
c
om
m
e
nda
ti
on
ta
s
ks
[
21]
.
U
nl
ik
e
s
in
gl
e
-
dom
a
in
r
e
c
om
m
e
nda
ti
on
m
ode
ls
,
C
D
R
s
a
r
e
f
ound
m
or
e
e
f
f
ic
ie
nt
in
a
ddr
e
s
s
in
g
da
ta
s
pa
r
s
it
y
pr
obl
e
m
s
,
w
h
e
r
e
it
can
us
e
va
r
io
us
tr
a
ns
f
e
r
le
a
r
ni
ng
m
e
th
ods
[
22]
,
but
th
e
y
f
a
c
e
c
ha
ll
e
nge
s
li
ke
c
os
tl
y,
c
om
pl
e
x
c
om
put
a
ti
on
,
a
nd
unde
r
u
ti
li
z
a
ti
on
of
in
-
dom
a
in
da
ta
,
a
f
f
e
c
ti
ng
th
e
r
e
c
om
m
e
nde
r
'
s
a
c
c
ur
a
c
y
.
N
e
w
de
ve
lo
pm
e
nt
s
li
ke
C
D
R
ha
v
e
r
e
c
e
nt
ly
a
ddr
e
s
s
e
d
th
e
s
e
pr
obl
e
m
s
by
us
in
g
in
s
ig
ht
s
f
r
om
r
e
la
te
d
a
r
e
a
s
,
li
ke
r
e
c
om
m
e
ndi
ng
hot
e
ls
b
a
s
e
d
on
tr
a
ve
l
hi
s
to
r
y
or
m
ov
ie
r
e
c
om
m
e
nda
ti
ons
ba
s
e
d
on
book
pur
c
ha
s
e
s
.
D
e
e
p
le
a
r
ni
ng
a
ppr
oa
c
he
s
pr
ovi
de
m
or
e
c
om
pl
e
x
f
e
a
tu
r
e
e
xt
r
a
c
ti
on,
w
he
r
e
a
s
tr
a
ns
f
e
r
le
a
r
ni
ng
a
nd
MF
a
id
in
in
f
or
m
a
ti
on
tr
a
ns
f
e
r
a
c
r
os
s
dom
a
in
s
.
N
e
ve
r
th
e
le
s
s
,
th
e
e
f
f
ic
a
c
y
a
nd
pr
e
c
is
io
n
of
th
e
s
e
s
ugge
s
ti
ons
a
r
e
s
ti
ll
a
f
f
e
c
te
d
by
obs
ta
c
le
s
s
uc
h
as
c
om
put
a
ti
ona
l
c
om
pl
e
xi
ty
a
nd
th
e
unde
r
ut
il
iz
a
ti
on
of
dom
a
in
da
ta
.
A
li
te
r
a
tu
r
e
r
e
vi
e
w
w
il
l
e
xa
m
in
e
pr
e
vi
ous
w
or
k
a
nd
ke
y
f
ig
ur
e
s
to
hi
ghl
ig
ht
de
ve
lo
pm
e
nt
s
a
nd
is
s
ue
s
in
th
e
f
ie
ld
.
T
he
m
e
th
odol
ogy
s
e
c
ti
on
w
il
l
de
s
c
r
ib
e
th
e
s
tr
a
te
gy
a
nd
m
e
th
ods
u
s
e
d,
w
hi
le
th
e
r
e
s
ul
ts
s
e
c
ti
on
e
v
a
lu
a
te
s
f
in
di
ngs
a
nd
pot
e
nt
ia
l
im
pr
ove
m
e
nt
s
.
T
oge
th
e
r
,
th
e
s
e
s
e
c
ti
ons
c
la
r
if
y
th
e
pr
e
s
e
nt
s
ta
te
a
nd
f
ut
ur
e
of
a
dva
nc
e
d
r
e
c
om
m
e
nda
ti
on
s
ys
te
m
s
.
2.
L
I
T
E
R
A
T
U
R
E
S
U
R
V
E
Y
T
he
C
D
R
m
e
th
ods
a
r
e
c
la
s
s
if
ie
d
as
s
e
m
a
nt
ic
,
f
a
c
to
r
ia
l,
ta
g,
a
nd
gr
a
ph
-
ba
s
e
d
[
23]
–
[
25
]
.
H
ow
e
ve
r
,
a
m
ong
th
e
s
e
a
ppr
oa
c
he
s
,
th
e
gr
a
ph
-
ba
s
e
d
m
e
th
ods
h
a
ve
s
ho
w
n
be
tt
e
r
e
f
f
ic
a
c
y.
In
th
is
m
e
th
od,
us
e
r
s
a
nd
it
e
m
s
r
e
pr
e
s
e
nt
th
e
node
s
,
a
nd
th
e
ir
in
te
r
-
e
le
m
e
nt
(
s
a
y,
us
e
r
-
it
e
m
)
r
e
la
ti
ons
hi
p
is
r
e
f
e
r
r
e
d
to
as
e
dge
s
,
w
he
r
e
th
e
in
te
r
-
e
le
m
e
nt
r
e
la
ti
on
a
nd
a
ll
ie
d
in
te
r
a
c
ti
ons
can
ha
ve
a
de
c
is
iv
e
im
pa
c
t
on
th
e
r
e
c
om
m
e
nda
ti
on
a
c
c
ur
a
c
y.
S
om
e
r
e
c
e
nt
li
te
r
a
tu
r
e
di
s
c
us
s
e
s
th
e
C
D
R
s
y
s
te
m
s
e
xpl
oi
ti
n
g
r
e
vi
e
w
s
(
te
xt
-
da
ta
)
,
tr
a
ns
f
e
r
le
a
r
ni
ng,
de
e
p
le
a
r
ni
ng,
a
nd
gr
a
ph
-
e
m
be
ddi
ng
a
ppr
oa
c
he
s
.
2.1.
T
e
xt
-
d
at
a
(
r
e
vi
e
w
)
b
as
e
d
r
e
c
om
m
e
n
d
at
io
n
m
od
e
l
To
a
ddr
e
s
s
th
e
da
ta
-
s
pa
r
s
it
y
pr
obl
e
m
s
,
ut
il
iz
in
g
te
xt
da
ta
is
c
r
uc
ia
l
in
r
e
c
om
m
e
nda
ti
on
s
ys
te
m
s
.
S
r
if
i
e
t
al
.
[
26
]
ha
ve
us
e
d
to
pi
c
m
ode
li
ng
to
unc
ove
r
hi
dde
n
t
opi
c
s
in
us
e
r
r
e
vi
e
w
te
xt
.
A
MF
m
ode
l
c
a
ll
e
d
T
opi
c
M
F
w
a
s
pr
opos
e
d,
w
hi
c
h
c
om
bi
ne
d
us
e
r
r
a
ti
ngs
a
nd
r
e
vi
e
w
s
to
obt
a
in
la
te
nt
to
pi
c
s
.
S
oc
ia
l
MF
[
27]
a
nd
to
pi
c
MF
w
e
r
e
e
m
pl
oye
d
to
a
s
s
e
s
s
da
ta
f
us
io
n
e
f
f
e
c
ti
ve
ne
s
s
f
or
be
tt
e
r
r
e
c
om
m
e
nda
ti
ons
.
A
pr
oba
bi
li
s
ti
c
m
e
th
od
[
28]
c
om
bi
ne
d
an
in
te
gr
a
te
d
la
te
nt
to
pi
c
m
ode
l
w
it
h
a
r
a
ndom
w
a
lk
a
nd
r
e
s
ta
r
t
te
c
hni
que
f
or
r
e
c
om
m
e
nda
ti
on.
T
he
s
e
m
e
th
ods
of
te
n
ig
nor
e
te
xt
or
de
r
a
nd
us
e
a
ba
g
-
of
-
w
or
ds
(
B
O
W
)
m
e
th
od
to
de
r
iv
e
la
te
nt
to
pi
c
va
r
ia
bl
e
s
.
I
ns
te
a
d,
e
m
be
ddi
ng
m
e
th
ods
c
r
e
a
te
d
w
or
d
ve
c
to
r
s
,
le
a
r
ne
d
vi
a
c
onvolut
io
na
l
ne
ur
a
l
ne
twor
ks
(
C
N
N
)
f
or
pr
e
di
c
ti
on.
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
H
e
te
r
oge
ne
ou
s
s
e
m
ant
ic
gr
aph e
m
b
e
ddi
ng as
s
i
s
te
d e
dge
-
s
e
ns
it
i
v
e
l
e
ar
ni
ng
…
(
P
r
av
in
R
am
c
hand
r
a P
at
il
)
1841
2.2.
D
e
e
p
-
le
ar
n
in
g
-
b
as
e
d
r
e
c
om
m
e
n
d
at
io
n
m
od
e
l
In
th
e
p
a
s
t
,
nu
m
e
r
ou
s
e
f
f
or
t
s
h
a
ve
b
e
e
n
m
a
de
by
a
p
pl
yi
ng
de
e
p
-
le
a
r
ni
ng
m
e
t
hod
s
f
or
r
e
c
om
m
e
nda
ti
on.
Z
h
a
o
[
2
9]
a
pp
li
e
d
de
n
oi
s
i
ng
a
ut
o
-
e
n
c
od
e
r
s
to
le
a
r
n
ov
e
r
th
e
us
e
r
-
it
e
m
m
a
tr
ix
f
or
u
s
e
r
a
nd
it
e
m
pr
of
il
in
g.
X
ue
e
t
al
.
[
30]
de
ve
lo
pe
d
an
im
pr
ove
d
n
e
ur
a
l
n
e
twor
k
-
dr
iv
e
n
MF
m
ode
l
,
w
he
r
e
at
f
ir
s
t
a
u
s
e
r
-
it
e
m
m
a
tr
i
x
w
a
s
ge
n
e
r
a
t
e
d
ov
e
r
th
e
e
x
pl
ic
it
r
a
ti
ng
s
a
nd
n
on
-
pr
e
f
e
r
e
nc
e
im
pl
ic
it
r
e
vi
e
w
te
xt
.
O
nc
e
obt
a
in
in
g
th
e
us
e
r
-
it
e
m
m
a
tr
ix
,
a
d
e
e
p
ne
ur
a
l
ne
t
w
or
k
(
D
N
N
)
w
a
s
a
ppl
ie
d
to
le
a
r
n
th
e
lo
w
-
di
m
e
n
s
io
na
l
f
e
a
tu
r
e
s
pa
c
e
(
us
e
r
-
it
e
m
m
a
tr
ix
)
.
A
p
pl
yi
ng
a
b
in
a
r
y
lo
s
s
f
un
c
ti
on
,
th
e
ir
pr
opo
s
e
d
de
e
p
ne
t
w
or
k
p
e
r
f
or
m
e
d
t
he
r
e
c
om
m
e
nd
a
ti
on
.
R
e
s
e
a
r
c
h by Z
h
a
ng
e
t
al
.
[
31]
,
a
lo
ng
s
hor
t
-
t
e
r
m
m
e
m
or
y
(
L
S
T
M
)
-
ba
s
e
d
a
ppr
o
a
c
h
w
a
s
a
pp
li
e
d
to
ge
n
e
r
a
t
e
te
xt
f
e
a
tu
r
e
s
f
or
f
ur
th
e
r
r
e
c
om
m
e
nd
a
ti
on
d
e
c
i
s
io
n
s
.
R
e
s
e
a
r
c
h
by
C
he
n
e
t
al
.
[
32]
,
a
n
e
ur
a
l
a
tt
e
nt
i
ona
l
r
e
gr
e
s
s
i
on
w
a
s
a
ppl
ie
d
in
c
o
nj
un
c
ti
on
w
it
h
r
e
vi
e
w
-
l
e
ve
l
e
x
pl
a
n
a
ti
on
s
f
or
r
e
c
om
m
e
nd
a
ti
on
d
e
c
i
s
io
ns
,
f
oc
u
s
in
g
on
r
e
vi
e
w
te
xt
bu
t
ig
nor
i
ng
it
e
m
m
e
t
a
d
a
ta
,
a
nd
pr
im
a
r
il
y
a
d
dr
e
s
s
e
d
s
in
gl
e
-
dom
a
in
r
e
c
om
m
e
nd
a
ti
on
s
.
2.3.
C
r
os
s
-
d
om
ai
n
r
e
c
om
m
e
n
d
at
io
n
C
D
R
s
e
xpl
oi
t
th
e
in
f
or
m
a
ti
on
f
r
om
one
dom
a
in
to
m
a
ke
r
e
c
om
m
e
nda
ti
ons
in
ot
he
r
dom
a
in
s
.
C
he
e
m
a
e
t
al
.
[
33]
de
s
ig
ne
d
a
la
te
nt
us
e
r
pr
of
il
in
g
a
ppr
oa
c
h
th
a
t
ge
ne
r
a
te
s
a
us
e
r
pr
of
il
e
by
ga
th
e
r
in
g
d
iv
e
r
s
e
us
e
r
be
ha
vi
or
s
a
c
r
os
s
m
ul
ti
pl
e
da
t
a
dom
a
in
s
.
T
hi
s
dom
a
in
-
in
de
pe
nde
nt
pr
of
il
e
a
nd
c
ont
e
xt
ua
l
da
t
a
e
na
bl
e
r
e
le
va
nt
C
D
R
s.
T
he
tr
a
n
s
f
e
r
le
a
r
ni
ng
m
e
th
od
w
a
s
a
ppl
ie
d
to
m
a
ke
s
e
r
e
ndi
pi
to
us
it
e
m
r
e
c
om
m
e
nd
a
ti
ons
[
34]
.
A
ge
ne
r
a
li
z
e
d
c
r
os
s
dom
a
in
-
m
ul
ti
-
di
m
e
ns
io
n
te
ns
or
f
a
c
to
r
iz
a
ti
on
m
ode
l
to
ba
la
nc
e
im
pa
c
t
a
m
ong
s
t
dom
a
in
s
e
f
f
e
c
ti
ve
ly
[
35]
.
H
u
e
t
al
.
[
36
]
a
ppl
ie
d
a
tr
a
ns
f
e
r
m
e
e
ti
ng
hyb
r
id
m
ode
l
f
or
C
D
R
by
a
ppl
yi
ng
it
e
m
r
e
vi
e
w
s
a
nd
a
r
ti
c
le
ti
tl
e
s
.
T
he
y
us
e
d
a
tt
e
nt
iv
e
ly
m
in
e
d
s
e
m
a
nt
ic
f
e
a
tu
r
e
s
f
r
om
r
e
vi
e
w
te
xt
to
tr
a
ns
f
e
r
vi
ta
l
in
f
or
m
a
ti
on
f
r
om
an
or
ig
in
a
ti
ng
dom
a
in
to
pe
r
f
or
m
le
a
r
ni
ng
a
nd
r
e
c
om
m
e
nda
ti
on.
Y
e
t,
th
e
s
e
m
e
th
ods
c
oul
d
not
e
xpl
oi
t
us
e
r
r
e
vi
e
w
s
a
nd
it
e
m
m
e
ta
da
ta
to
ge
th
e
r
.
To
a
ll
e
vi
a
te
th
e
a
f
or
e
s
a
id
li
m
it
a
ti
on,
th
e
a
ut
hor
s
s
ugge
s
te
d
tr
a
ns
f
e
r
r
in
g
knowle
dge
a
m
ongs
t
th
e
v
a
r
ie
d
dom
a
in
s
to
m
a
ke
C
D
R
s
[
3]
,
[
19]
.
A
n
a
da
pt
iv
e
c
ode
book
tr
a
ns
f
e
r
le
a
r
ni
ng
(
A
C
T
L
)
w
a
s
pr
opos
e
d
th
a
t
id
e
nt
if
ie
s
th
e
s
ui
ta
bl
e
c
od
e
book
s
c
a
le
to
r
e
duc
e
c
om
put
a
ti
ona
l
c
os
t
a
nd
im
pr
ove
C
D
R
[
19]
.
Y
u
e
t
al
.
[
37]
de
ve
lo
pe
d
C
D
C
F
,
f
or
a
ddr
e
s
s
in
g
da
t
a
s
p
a
r
s
it
y
by
ut
il
iz
in
g
r
a
ti
ng
in
f
or
m
a
ti
on
f
r
om
two
a
uxi
li
a
r
y
dom
a
in
s
:
us
e
r
-
s
id
e
a
nd
it
e
m
-
s
id
e
,
bot
h
c
ha
r
a
c
te
r
iz
e
d
by
de
ns
e
r
a
ti
ng
da
ta
.
T
he
r
e
c
om
m
e
nda
ti
on
ta
s
k
is
th
e
n
r
e
f
r
a
m
e
d
as
a
c
la
s
s
if
ic
a
ti
on
pr
obl
e
m
ba
s
e
d
on
de
r
iv
e
d
in
tr
in
s
ic
f
e
a
tu
r
e
s
,
w
hi
c
h
is
s
ol
ve
d
us
in
g
an
S
V
M
.
T
he
a
lg
or
it
hm
[
38]
a
ddr
e
s
s
e
s
da
ta
s
pa
r
s
it
y
in
th
e
r
e
c
e
iv
in
g
dom
a
in
by
a
ugm
e
nt
in
g
us
e
r
a
nd
it
e
m
f
e
a
tu
r
e
s
w
it
h
la
te
nt
f
a
c
to
r
s
f
r
o
m
bot
h
a
uxi
li
a
r
y
dom
a
in
s
,
us
in
g
F
unk
-
s
in
gul
a
r
va
lu
e
de
c
om
pos
it
io
n (
S
V
D
)
to
e
xt
r
a
c
t
a
ddi
ti
ona
l
f
e
a
tu
r
e
s
.
T
w
o
-
s
id
e
d
C
D
C
F
a
lg
or
it
hm
s
[
39]
,
li
ke
two
-
s
ta
ge
s
ta
c
ke
d
e
ns
e
m
bl
e
a
ut
oe
nc
od
e
r
(
T
S
S
E
A
E
)
,
e
nha
nc
e
pe
r
f
or
m
a
nc
e
by
c
om
bi
ni
ng
us
e
r
a
nd
it
e
m
a
uxi
li
a
r
y
dom
a
in
s
,
f
r
a
m
in
g
th
e
C
D
C
F
pr
obl
e
m
as
an
e
n
s
e
m
bl
e
of
c
la
s
s
if
ie
r
s
.
T
h
e
a
ut
hor
s
de
s
ig
ne
d
th
e
f
r
a
m
e
w
or
k
in
w
hi
c
h
one
ne
ur
a
l
ne
twor
k
w
a
s
u
s
e
d
to
le
a
r
n
th
e
r
e
duc
e
d
di
m
e
ns
io
na
l
il
lu
s
tr
a
ti
on
f
r
om
th
e
one
-
hot
c
odi
ng
of
us
e
r
s
a
nd
it
e
m
s
,
w
hi
le
a
not
he
r
ne
twor
k
le
ve
r
a
ge
d
a
uxi
li
a
r
y
in
f
or
m
a
ti
on
in
a
di
f
f
e
r
e
nt
la
te
nt
s
pa
c
e
[
40]
.
R
e
c
e
nt
ly
,
a
gr
a
ph
-
ba
s
e
d
C
D
R
m
e
th
od
us
e
s
u
s
e
r
-
it
e
m
in
te
r
a
c
t
io
ns
f
r
om
th
e
or
ig
in
a
ti
ng
dom
a
in
to
r
e
c
om
m
e
nd
it
e
m
s
in
r
e
c
e
iv
in
g.
H
ow
e
ve
r
,
s
ta
te
-
of
-
th
e
-
a
r
t
gr
a
ph
-
ba
s
e
d
r
e
c
om
m
e
nde
r
s
m
a
y
in
c
lu
de
ir
r
e
le
va
nt
it
e
m
s
f
r
om
th
e
or
ig
in
a
ti
ng
dom
a
in
,
a
ddi
ng
noi
s
e
a
nd
r
e
duc
in
g
e
f
f
e
c
ti
ve
ne
s
s
.
N
one
of
th
e
e
xi
s
ti
ng
m
e
th
ods
c
oul
d
a
ddr
e
s
s
th
e
pr
obl
e
m
of
noi
s
e
it
e
m
s
.
To
a
ll
e
vi
a
te
th
is
,
we
pr
opos
e
a
he
te
r
oge
ne
ou
s
s
e
m
a
nt
ic
gr
a
ph
-
e
m
be
ddi
ng
(
H
S
G
E
)
-
a
s
s
is
te
d
m
ode
l
th
a
t
pr
une
s
ir
r
e
le
va
nt
e
dge
s
in
th
e
or
ig
in
a
ti
ng
dom
a
in
.
O
ur
pr
opos
e
d
m
e
th
od
us
e
s
th
e
W
or
d2V
e
c
s
e
m
a
nt
ic
e
m
be
ddi
ng
m
e
th
od
f
or
th
e
us
e
r
’
s
r
e
vi
e
w
,
r
a
ti
ngs
,
a
nd
it
e
m
de
ta
il
s
.
T
he
n,
th
e
N
ode
2V
e
c
c
a
pt
ur
e
s
u
s
e
r
-
us
e
r
,
u
s
e
r
-
it
e
m
,
a
nd
it
e
m
-
it
e
m
in
te
r
a
c
ti
ons
f
r
om
bot
h
dom
a
in
s
.
H
e
r
e
,
th
e
a
m
a
lg
a
m
a
ti
on
of
th
e
di
f
f
e
r
e
nt
in
te
r
a
c
ti
ons
c
ons
ti
tu
te
s
th
e
H
S
G
E
m
e
th
od.
T
he
m
ode
l
im
pr
ove
s
s
e
m
a
nt
ic
f
e
a
tu
r
e
e
m
be
ddi
ng
ve
c
to
r
s
by
r
e
m
ovi
ng
unr
e
la
te
d
it
e
m
s
f
r
om
t
he
or
ig
in
a
ti
ng
dom
a
in
f
or
be
tt
e
r
le
a
r
ni
ng
a
nd
pr
e
di
c
ti
on.
U
s
in
g
a
m
ul
ti
la
ye
r
f
e
e
df
or
w
a
r
d
ne
ur
a
l
ne
twor
k
-
m
u
lt
il
a
ye
r
pe
r
c
e
pt
r
on
(
M
F
N
N
-
M
L
P
)
w
it
h
c
os
in
e
s
im
il
a
r
it
y,
it
pr
e
di
c
ts
to
p
-
N
r
e
c
om
m
e
nda
ti
ons
in
th
e
r
e
c
e
iv
in
g
dom
a
in
,
a
c
hi
e
vi
ng
be
tt
e
r
hi
t
-
r
a
te
(
HR
)
a
nd
nor
m
a
li
z
e
d
di
s
c
ount
e
d
c
um
ul
a
ti
ve
ga
in
(
N
D
C
G
)
s
c
or
e
s
th
a
n
ot
he
r
m
e
th
ods
.
3.
M
E
T
H
O
D
T
he
pr
opos
e
d
m
e
th
od
e
nc
om
pa
s
s
e
s
a
he
t
e
r
oge
nous
s
e
m
a
nt
ic
gr
a
ph
e
m
be
ddi
ng
m
a
tr
ix
,
in
te
r
-
do
m
a
in
r
e
la
ti
ve
ne
s
s
-
dr
iv
e
n
e
dge
pr
uni
ng,
a
tt
e
nt
io
n
m
e
th
od,
a
nd
M
F
N
N
-
M
L
P
f
or
C
D
R
.
It
pr
oc
e
s
s
e
s
u
s
e
r
r
e
vi
e
w
s
a
nd
it
e
m
de
ta
il
s
in
to
in
put
la
ye
r
s
f
o
r
each
dom
a
in
,
c
r
e
a
ti
ng
s
e
m
a
nt
ic
gr
a
ph
e
m
be
ddi
ngs
of
us
e
r
-
us
e
r
,
us
e
r
-
it
e
m
,
a
nd
it
e
m
-
it
e
m
in
te
r
a
c
ti
ons
u
s
in
g
N
ode
2
V
e
c
.
T
he
s
e
e
m
be
ddi
ng
s
a
r
e
r
e
f
in
e
d
th
r
ough
e
dge
pr
uni
ng
to
r
e
m
ove
unr
e
la
te
d
it
e
m
s
,
f
ol
lo
w
e
d
by
e
le
m
e
nt
-
w
is
e
a
tt
e
nt
io
n
to
f
or
m
a
c
om
bi
ne
d
f
e
a
tu
r
e
m
ode
l.
T
hi
s
m
ode
l
is
th
e
n
us
e
d
f
or
le
a
r
ni
ng
a
nd
pr
e
di
c
ti
on
vi
a
an
M
F
N
N
-
M
L
P
w
it
h
S
of
tM
a
x
f
or
to
p
-
N
r
e
c
om
m
e
nda
ti
on
s
at
th
e
out
put
la
ye
r
.
T
he
pr
opos
e
d
m
ode
l
is
de
pi
c
te
d
in
F
ig
ur
e
1.
T
he
c
r
o
s
s
-
dom
a
in
pr
e
di
c
ti
on
m
ode
l
e
nc
om
pa
s
s
e
s
two
dom
a
in
s
,
(
or
ig
in
)
a
nd
(
r
e
c
e
iv
in
g)
,
each
w
it
h
us
e
r
s
a
nd
it
e
m
s
c
ont
a
in
in
g
e
xpl
ic
it
(
us
e
r
’
s
r
a
ti
ng
a
nd
r
e
vi
e
w
)
a
nd
im
pl
ic
it
(
it
e
m
(
s
)
de
ta
il
s
)
de
ta
il
s
.
B
y,
th
e
a
m
a
lg
a
m
a
ti
on
of
th
e
r
e
vi
e
w
a
nd
th
e
it
e
m
’
s
m
e
ta
da
ta
c
r
e
a
te
s
a
pr
of
il
e
f
or
a
us
e
r
a
im
e
d
at
im
pr
ovi
n
g
r
e
c
om
m
e
nda
ti
on
a
c
c
ur
a
c
y
in
th
e
r
e
c
e
iv
in
g
dom
a
in
.
In
r
e
a
l
-
time
da
ta
m
ode
ls
,
th
e
r
e
can
be
c
e
r
ta
in
us
e
r
s
th
a
t
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, No. 3, J
une
2025
:
1839
-
1852
1842
ove
r
la
p,
w
hi
c
h
m
e
a
ns
c
e
r
ta
in
us
e
r
s
can
e
xi
s
t
in
bot
h
th
e
or
ig
i
n
(
s
a
y,
p)
a
nd
th
e
r
e
c
e
iv
in
g
dom
a
in
(
s
a
y,
q)
,
s
e
r
vi
ng
as
a
br
id
ge
c
onne
c
ti
ng
bot
h
dom
a
in
s
.
T
hi
s
f
or
m
s
th
e
ba
s
is
f
or
th
e
c
r
os
s
-
dom
a
in
r
e
c
om
m
e
nde
r
de
s
ig
n.
T
he
im
pl
e
m
e
nt
a
ti
on
s
c
he
m
a
ti
c
of
th
e
pl
a
nn
e
d
C
D
R
m
ode
l
w
it
h
f
iv
e
m
a
in
c
om
pone
nt
s
is
s
how
n
in
F
ig
ur
e
1.
T
he
s
e
a
r
e
:
i)
in
put
la
ye
r
,
ii
)
H
S
G
E
la
ye
r
,
ii
i
)
f
e
a
tu
r
e
a
m
a
lg
a
m
a
ti
on
la
ye
r
,
iv
)
M
F
N
N
-
M
L
P
la
ye
r
,
a
nd
v)
pr
e
di
c
ti
on l
a
ye
r
or
t
he
out
put
l
a
ye
r
.
U
s
e
r
(
R
e
v
i
e
w
)
I
t
e
m
(
C
o
n
t
e
n
t
)
W
o
r
d
2
V
e
c
W
o
r
d
2
V
e
c
N
o
d
e
2
V
e
c
N
o
d
e
2
V
e
c
U
s
e
r
-
U
s
e
r
U
s
e
r
-
I
t
e
m
I
t
e
m
-
I
t
e
m
E
d
g
e
P
r
u
n
i
n
g
H
e
t
e
r
o
g
e
n
e
o
u
s
S
e
m
a
n
t
i
c
G
r
a
p
h
E
m
b
e
d
d
i
n
g
M
e
t
r
i
c
s
E
l
e
m
e
n
t
-
W
i
s
e
A
t
t
e
n
t
i
o
n
U
s
e
r
(
R
e
v
i
e
w
)
I
t
e
m
(
C
o
n
t
e
n
t
)
W
o
r
d
2
V
e
c
W
o
r
d
2
V
e
c
N
o
d
e
2
V
e
c
N
o
d
e
2
V
e
c
U
s
e
r
-
U
s
e
r
U
s
e
r
-
I
t
e
m
I
t
e
m
-
I
t
e
m
E
d
g
e
P
r
u
n
i
n
g
H
e
t
e
r
o
g
e
n
e
o
u
s
S
e
m
a
n
t
i
c
G
r
a
p
h
E
m
b
e
d
d
i
n
g
M
e
t
r
i
c
s
M
F
N
N
M
L
P
L
a
y
e
r
O
u
t
p
u
t
L
a
y
e
r
(
P
r
e
d
i
c
t
e
d
t
o
p
-
N
R
e
c
o
m
m
e
n
d
e
d
I
t
e
m
s
i
n
t
h
e
T
a
r
g
e
t
D
o
m
a
i
n
)
S
o
u
r
c
e
D
o
m
a
i
n
T
a
r
g
e
t
D
o
m
a
i
n
1
2
3
4
5
F
ig
ur
e
1.
P
r
opos
e
d
C
D
R
m
ode
l
3.1.
I
n
p
u
t
la
ye
r
T
hi
s
w
or
k
u
s
e
s
bot
h
e
xpl
ic
it
u
s
e
r
de
ta
il
s
,
li
ke
r
e
vi
e
w
s
a
nd
r
a
ti
ngs
,
a
nd
im
pl
ic
it
it
e
m
in
f
or
m
a
ti
on,
s
uc
h
as
m
e
ta
da
ta
a
nd
f
e
e
dba
c
k,
as
in
put
.
T
h
e
e
xpl
ic
it
in
f
or
m
a
ti
on
c
ons
tr
uc
te
d
a
pr
of
il
e
f
or
each
u
s
e
r
in
bot
h
s
our
c
e
as
w
e
ll
as
r
e
c
e
iv
in
g
dom
a
in
s
.
T
he
us
e
r
-
it
e
m
de
ta
il
s
w
e
r
e
th
e
n
f
e
d
as
in
put
to
th
e
s
e
m
a
nt
ic
gr
a
ph
e
m
be
ddi
ng
la
ye
r
.
3.2
.
H
e
t
e
r
oge
n
ou
s
s
e
m
an
t
ic
gr
ap
h
e
m
b
e
d
d
in
g
l
aye
r
T
he
us
e
r
’
s
r
e
vi
e
w
a
nd
it
e
m
de
ta
il
s
f
r
om
th
e
or
ig
in
a
ti
ng
(
)
a
nd
t
he
r
e
c
e
iv
in
g
dom
a
in
(
)
a
r
e
f
e
d
as
in
put
to
th
e
s
e
m
a
nt
ic
gr
a
ph
e
m
be
ddi
ng
la
ye
r
.
U
nl
ik
e
tr
a
di
t
io
na
l
one
-
hot
e
nc
odi
ng,
w
hi
c
h
s
uf
f
e
r
s
f
r
om
s
pa
r
s
it
y
a
nd
noi
s
e
is
s
u
e
s
,
th
is
nove
l
gr
a
ph
-
e
m
be
ddi
ng
m
e
th
od
le
ve
r
a
ge
s
u
s
e
r
-
us
e
r
,
it
e
m
-
us
e
r
,
a
nd
it
e
m
-
it
e
m
in
te
r
a
c
ti
ons
f
or
im
p
r
ove
d
pr
e
di
c
ti
ons
.
It
can
be
vi
ta
l
f
or
C
D
R
in
bot
h
dom
a
in
s
(
i.
e
.,
a
nd
).
O
ur
p
r
opos
e
d
m
e
th
od
us
e
s
he
te
r
oge
n
e
ous
gr
a
ph
e
m
be
ddi
ng
m
a
tr
ic
e
s
f
or
us
e
r
-
us
e
r
,
us
e
r
-
it
e
m
,
a
nd
it
e
m
-
it
e
m
in
te
r
a
c
ti
ons
in
bot
h
s
our
c
e
a
nd
r
e
c
e
iv
in
g
dom
a
in
s
.
H
e
r
e
,
th
e
us
e
r
’
s
r
a
ti
ng
a
nd
th
e
it
e
m
’
s
c
ont
e
nt
in
f
or
m
a
ti
on
r
e
la
te
d
to
each
dom
a
in
(
i.
e
.,
a
nd
)
a
r
e
le
ve
r
a
ge
d
to
c
ons
ti
tu
te
he
te
r
oge
nous
gr
a
ph
e
m
be
ddi
ng
m
e
tr
ic
s
.
U
nl
ik
e
one
-
hot
c
odi
ng,
W
or
d2V
e
c
of
f
e
r
s
r
ic
he
r
s
e
m
a
nt
ic
,
a
nd
lo
w
-
di
m
e
ns
i
ona
l
e
m
be
ddi
ng
m
e
tr
ic
s
.
We
u
s
e
d
G
e
ns
im
'
s
W
or
d2V
e
c
w
it
h
s
ki
p
-
gr
a
m
a
nd
ne
ga
ti
ve
s
a
m
pl
in
g
(
S
G
N
S
)
to
tr
a
ns
f
or
m
c
ont
e
nt
-
r
e
vi
e
w
in
f
or
m
a
ti
on
of
bot
h
a
nd
dom
a
in
s
in
to
c
or
r
e
s
ponding
s
e
m
a
nt
ic
va
lu
e
s
,
di
s
ti
nc
tl
y
a
nd
a
ls
o
to
tr
a
ns
f
or
m
each
r
e
vi
e
w
or
it
e
m
m
e
ta
da
ta
li
ke
ti
tl
e
s
,
s
ub
-
ti
tl
e
s
,
a
nd
d
e
s
c
r
ip
ti
ons
in
to
a
s
e
m
a
nt
ic
l
a
te
nt
ve
c
to
r
.
T
he
N
ode
2V
e
c
m
e
th
od
e
xt
r
a
c
te
d
H
S
G
E
m
e
tr
ic
s
f
r
om
la
te
nt
e
m
be
ddi
ng
ve
c
to
r
s
by
le
ve
r
a
gi
ng
he
te
r
oge
ne
ous
in
te
r
a
c
ti
ons
,
im
pr
ovi
ng
C
D
R
s
w
hi
l
e
a
ddr
e
s
s
i
ng
da
ta
s
pa
r
s
it
y
a
nd
c
ol
d
-
s
ta
r
t
is
s
ue
s
.
T
h
e
N
ode
2V
e
c
m
e
th
od
e
xt
r
a
c
te
d
H
S
G
E
m
e
tr
ic
s
f
r
om
la
te
nt
e
m
be
ddi
ng
ve
c
to
r
s
by
le
ve
r
a
gi
ng
he
te
r
oge
ne
ou
s
in
te
r
a
c
ti
ons
,
im
pr
ovi
ng
C
D
R
s
w
hi
le
a
ddr
e
s
s
in
g
da
ta
s
pa
r
s
it
y
a
nd
c
ol
d
-
s
ta
r
t
is
s
ue
s
.
H
S
G
E
w
a
s
obt
a
in
e
d
f
or
bot
h
s
our
c
e
a
nd
r
e
c
e
iv
in
g
dom
a
in
s
,
r
e
s
ul
ti
ng
in
two
he
te
r
oge
ne
ous
gr
a
ph
e
m
be
ddi
ng
m
a
tr
ic
e
s
.
N
ot
ic
e
a
bl
y,
th
e
H
S
G
E
gr
a
phs
obt
a
in
e
d
c
ont
a
in
e
d
nod
e
s
s
ig
ni
f
yi
ng
us
e
r
s
a
nd
it
e
m
s
in
th
e
or
ig
in
a
ti
ng
dom
a
in
,
w
hi
le
w
e
ig
ht
e
d
e
dge
s
r
e
pr
e
s
e
nt
e
d
th
e
us
e
r
’
s
r
e
vi
e
w
a
nd
m
e
ta
da
ta
c
ont
e
nt
s
im
il
a
r
it
ie
s
obt
a
in
e
d
f
or
th
e
r
e
c
e
iv
in
g
dom
a
in
,
r
e
s
pe
c
ti
ve
ly
.
It
im
pr
ove
d
th
e
qua
li
ty
of
us
e
r
a
nd
it
e
m
e
m
be
ddi
ng
m
a
tr
ic
e
s
f
or
le
a
r
ni
ng
a
nd
pr
e
di
c
ti
on
.
T
he
h
o
m
e
la
nd
s
e
c
ur
it
y
a
nd
e
m
e
r
ge
nc
y
m
a
na
ge
m
e
nt
(
H
S
E
M
)
m
ode
l
c
ons
is
t
of
th
r
e
e
f
unc
ti
ona
l
e
le
m
e
nt
s
:
i)
in
te
r
-
dom
a
in
r
e
la
te
dne
s
s
-
ba
s
e
d
e
dge
pr
uni
ng
,
ii
)
D
oc
2V
e
c
(
doc
um
e
nt
)
e
m
be
ddi
ng
,
ii
i)
s
e
m
a
nt
ic
gr
a
ph e
m
be
ddi
ng
a
nd
H
S
G
E
ge
ne
r
a
ti
on
.
T
he
de
ta
il
of
th
e
s
e
s
ub
-
c
om
pone
nt
s
is
gi
ve
n
as
f
ol
lo
w
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
H
e
te
r
oge
ne
ou
s
s
e
m
ant
ic
gr
aph e
m
b
e
ddi
ng as
s
i
s
te
d e
dge
-
s
e
ns
it
i
v
e
l
e
ar
ni
ng
…
(
P
r
av
in
R
am
c
hand
r
a P
at
il
)
1843
3.2.1.
I
n
t
e
r
-
d
om
ai
n
r
e
la
t
e
d
n
e
s
s
-
b
as
e
d
e
d
ge
p
r
u
n
in
g
T
hi
s
w
or
k
in
tr
oduc
e
s
an
in
te
r
-
dom
a
in
c
or
r
e
la
ti
on
-
dr
iv
e
n
e
dge
pr
uni
ng
m
ode
l
to
im
pr
ove
r
e
c
om
m
e
nda
ti
on
a
c
c
ur
a
c
y
by
r
e
m
ovi
ng
unr
e
la
te
d
us
e
r
-
it
e
m
e
d
ge
s
in
th
e
r
e
c
e
iv
in
g
dom
a
in
,
pr
e
ve
nt
in
g
f
a
ls
e
pos
it
iv
e
r
e
s
ul
ts
.
T
o
pr
oj
e
c
t
a
f
or
e
s
a
id
f
e
a
tu
r
e
s
to
th
e
la
te
nt
e
m
be
ddi
ng
m
a
tr
ix
or
ve
c
to
r
,
th
e
p
r
opos
e
d
m
ode
l
e
s
ti
m
a
te
s
pr
oj
e
c
ti
on
ve
c
to
r
s
in
bot
h
th
e
or
ig
in
a
ti
ng
dom
a
in
as
w
e
ll
as
th
e
r
e
c
e
iv
in
g
dom
a
in
(
th
e
ir
c
or
r
e
s
ponding
doc
um
e
nt
s
,
i.
e
.,
,
).
L
e
t,
th
e
pr
oj
e
c
ti
on
m
a
tr
ic
e
s
f
or
or
ig
in
a
ti
ng
a
nd
th
e
r
e
c
e
iv
in
g
dom
a
in
be
∈
ℝ
×
a
nd
∈
ℝ
×
.
T
he
s
e
pr
oj
e
c
ti
on
v
e
c
to
r
s
a
r
e
obt
a
in
e
d
by
p
e
r
f
or
m
in
g
c
a
noni
c
a
l
c
or
r
e
la
ti
on
a
na
ly
s
is
(
C
C
A
)
ove
r
us
e
r
’
s
s
e
m
a
nt
ic
e
m
be
ddi
ng
ve
c
to
r
out
put
(
i.
e
.,
W
or
d2V
e
c
e
m
be
ddi
ng
ve
c
to
r
f
or
us
e
r
)
in
bot
h
dom
a
in
s
.
In
ot
he
r
w
or
ds
,
le
t
th
e
us
e
r
’
s
e
m
be
ddi
ng
ve
c
to
r
in
or
ig
in
a
ti
ng
a
nd
r
e
c
e
iv
in
g
dom
a
in
s
be
(
1)
a
nd
(
2)
,
r
e
s
pe
c
ti
ve
ly
.
=
[
1
,
2
,
…
,
]
∈
ℝ
×
(
1)
=
[
1
,
2
,
…
,
]
∈
ℝ
×
(
2)
To
a
c
hi
e
ve
it,
th
e
pr
opos
e
d
m
e
th
od
pe
r
f
or
m
s
m
a
xi
m
iz
a
ti
on
of
th
e
c
os
t
f
unc
ti
on
de
f
in
e
d
in
(
3)
.
(
̂
,
̂
)
=
arg
m
a
x
,
,
√
,
√
,
(
3)
In
(
3)
,
th
e
pa
r
a
m
e
te
r
s
a
nd
r
e
pr
e
s
e
nt
s
th
e
pr
oj
e
c
ti
on
e
m
be
ddi
ng
ve
c
to
r
s
,
w
hi
c
h
c
a
n
m
a
th
e
m
a
ti
c
a
ll
y
be
de
r
iv
e
d
as
pe
r
(
4)
a
nd
(
5)
,
c
or
r
e
s
pondingl
y.
=
ℝ
×
(
4)
=
ℝ
×
(
5
)
T
he
pr
oj
e
c
ti
on
e
m
be
ddi
ng
m
e
tr
ic
s
f
or
s
our
c
e
a
nd
r
e
c
e
iv
in
g
d
om
a
in
doc
um
e
nt
s
w
e
r
e
obt
a
in
e
d
by
m
a
xi
m
iz
in
g
th
e
obj
e
c
ti
ve
f
unc
ti
on
de
f
in
e
d
in
(
3
)
.
T
he
obj
e
c
ti
ve
f
unc
ti
on
(
3)
s
ol
ve
d
e
ig
e
nva
lu
e
pr
obl
e
m
th
a
t
e
ve
nt
ua
ll
y
yi
e
ld
e
d
(
4)
a
nd
(
5)
f
or
th
e
or
ig
in
a
l
a
nd
th
e
r
e
c
e
iv
in
g
dom
a
in
s
,
r
e
s
pe
c
ti
ve
ly
.
T
h
e
e
m
be
ddi
ng
m
a
tr
ix
pe
r
dom
a
in
w
a
s
obt
a
in
e
d
as
pe
r
th
e
(
6)
a
nd
(
7)
,
c
ondi
ti
one
d
w
it
h
th
e
(
8)
a
nd
(
9)
,
c
or
r
e
s
pondingl
y.
̂
∈
ℝ
(
6)
̂
∈
ℝ
(
7)
̂
=
(
8)
̂
=
(
9)
T
he
pr
opos
e
d
m
ode
l
c
om
pa
r
e
d
th
e
it
e
m
s
in
or
ig
in
a
ti
ng
a
nd
r
e
c
e
iv
in
g
dom
a
in
in
r
e
s
p
e
c
ti
ve
e
m
be
ddi
ng
la
te
nt
s
pa
c
e
.
N
ot
ic
e
a
bl
y,
onl
y
it
e
m
s
w
it
h
hi
gh
c
or
r
e
la
ti
on
a
nd
s
im
il
a
r
it
y
w
e
r
e
r
e
ta
in
e
d.
T
he
noi
s
e
pr
one
ne
s
s
of
th
e
it
e
m
s
in
th
e
or
ig
in
a
ti
ng
dom
a
in
a
ppl
ie
s
th
e
lo
c
a
l
de
ns
it
y
f
unc
ti
on
ove
r
th
e
s
e
m
a
nt
ic
e
m
be
ddi
ng
m
e
tr
ic
s
in
s
our
c
e
doc
um
e
nt
.
In
th
is
m
a
nne
r
,
th
e
di
f
f
e
r
e
nc
e
be
twe
e
n
th
e
lo
c
a
l
de
n
s
it
y
of
it
e
m
a
nd
th
e
lo
c
a
l
de
ns
it
y
of
th
e
ne
ig
hbor
in
g
it
e
m
.
T
he
noi
s
e
de
ns
i
ty
f
or
o
r
ig
in
a
ti
ng
dom
a
in
it
e
m
s
is
e
s
ti
m
a
te
d
as
pe
r
(
10)
.
(
)
=
1
(
∑
ℎ
−
(
,
)
∈
(
)
|
(
)
|
)
(
10)
In
(
10)
,
(
)
r
e
pr
e
s
e
nt
s
a
s
e
t
of
a
dj
oi
ni
ng
it
e
m
s
(
in
th
e
vi
c
in
it
y
of
th
e
s
our
c
e
it
e
m
),
w
hi
le
(
)
be
th
e
in
ve
r
s
e
of
th
e
m
e
a
n
of
th
e
r
e
a
c
ha
bi
li
ty
di
s
ta
nc
e
vi
a
ne
a
r
e
s
t
ne
ig
hbor
s
.
A
ddi
ti
ona
ll
y,
ℎ
−
(
,
)
s
ig
ni
f
yi
ng
th
e
r
e
a
c
ha
bi
li
ty
di
s
ta
nc
e
be
tw
e
e
n
a
nd
is
m
e
a
s
ur
e
d
as
pe
r
(
11)
.
ℎ
−
(
,
)
=
(
−
(
)
,
(
,
)
)
(
11)
In
(
11)
,
−
(
)
r
e
pr
e
s
e
nt
s
th
e
di
s
ta
nc
e
in
be
twe
e
n
it
e
m
a
nd
th
e
-
th
ne
i
ghbor
in
g
it
e
m
s
.
S
im
il
a
r
ly
,
th
e
di
s
ta
nc
e
be
twe
e
n
a
nd
it
e
m
s
is
gi
ve
n
by
(
,
)
.
A
s
s
um
in
g
a
f
or
e
s
a
id
out
li
e
r
r
a
nk
(
10
)
as
th
e
(
in
te
r
-
it
e
m
)
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, No. 3, J
une
2025
:
1839
-
1852
1844
di
s
ta
nc
e
in
di
c
a
to
r
(
i.
e
.,
th
e
di
s
ta
nc
e
be
twe
e
n
a
nd
th
e
it
e
m
s
e
t,
t
he
n
th
e
lo
c
a
l
r
e
a
c
ha
bi
li
ty
de
ns
it
y
of
th
e
it
e
m
is
m
e
a
s
ur
e
d
as
pe
r
(
12)
.
(
)
=
∑
(
)
(
)
∈
(
)
|
(
)
|
(
12)
In
(
12)
,
th
e
pa
r
a
m
e
te
r
(
)
r
e
tr
ie
ve
s
th
e
e
xt
e
nt
to
w
hi
c
h
th
e
it
e
m
can
be
c
a
ll
e
d
as
th
e
no
is
e
or
th
e
out
li
e
r
it
e
m
ha
vi
ng
no
s
ig
ni
f
ic
a
nt
r
e
la
ti
ons
hi
p
w
it
h
th
e
r
e
c
e
iv
in
g
dom
a
in
us
e
r
.
E
ve
nt
ua
ll
y,
th
e
s
e
m
a
nt
ic
e
m
be
ddi
ng
f
e
a
tu
r
e
f
or
or
ig
in
a
ti
ng
dom
a
in
̂
∈
ℝ
is
obt
a
in
e
d
a
f
te
r
pr
uni
ng
th
e
noi
s
y
or
ir
r
e
le
va
nt
e
dge
s
in
or
ig
in
a
ti
ng
dom
a
in
.
N
ow
,
onc
e
obt
a
in
in
g
th
e
pr
une
d
it
e
m
s
e
m
a
nt
ic
ve
c
to
r
in
th
e
or
ig
in
a
ti
ng
do
m
a
in
,
th
e
gr
a
ph
e
m
be
ddi
ng
ve
c
to
r
s
or
m
a
tr
ic
e
s
w
e
r
e
obt
a
in
e
d
f
or
th
e
us
e
r
-
us
e
r
,
us
e
r
-
it
e
m
,
a
nd
it
e
m
-
it
e
m
in
te
r
a
c
ti
ons
.
3.2.2
.
D
oc
2V
e
c
e
m
b
e
d
d
in
g
T
he
pr
opos
e
d
m
ode
l
c
om
bi
ne
s
us
e
r
-
us
e
r
,
us
e
r
-
it
e
m
,
a
nd
it
e
m
-
it
e
m
in
te
r
a
c
ti
on
r
e
la
ti
ons
hi
ps
in
he
te
r
oge
ne
ous
s
e
m
a
nt
ic
e
m
b
e
ddi
ng
m
a
tr
ic
e
s
a
nd
is
pr
e
s
e
r
ve
d
as
doc
um
e
nt
s
,
c
r
e
a
ti
ng
uni
t
gr
a
phs
f
or
bot
h
s
our
c
e
a
nd
r
e
c
e
iv
in
g
dom
a
in
s
.
C
ont
e
nt
s
im
il
a
r
it
y
be
twe
e
n
us
e
r
s
a
nd
it
e
m
s
is
e
s
ti
m
a
te
d
us
in
g
m
ul
ti
-
s
our
c
e
c
ont
e
nt
in
f
or
m
a
ti
on.
T
he
r
e
ta
in
e
d
e
m
be
ddi
ng
m
e
tr
ic
s
a
r
e
c
on
ve
r
te
d
in
to
uni
t
doc
um
e
nt
s
f
or
each
dom
a
in
,
c
ol
le
c
ti
ng
a
s
e
m
a
nt
ic
e
m
be
ddi
ng
r
e
vi
e
w
f
or
e
a
c
h
u
s
e
r
.
B
e
in
g
a
m
ul
ti
-
s
our
c
e
doc
um
e
nt
e
m
be
ddi
ng
m
e
th
od,
th
e
s
e
m
a
nt
ic
e
m
b
e
ddi
ng
r
e
vi
e
w
∗
f
or
a
us
e
r
is
c
ol
le
c
t
e
d.
F
or
an
−
th
us
e
r
(
s
a
y,
),
th
e
a
ll
ie
d
us
e
r
pr
of
il
e
is
obt
a
in
e
d
a
nd
s
a
ve
d
in
a
s
in
gl
e
doc
um
e
nt
.
F
or
an
i
te
m
,
th
e
r
e
vi
e
w
s
∗
on
th
e
it
e
m
a
nd
(
it
e
m
)
de
ta
il
a
r
e
c
ol
le
c
te
d
in
th
e
s
a
m
e
doc
um
e
nt
+
.
S
ubs
e
que
nt
ly
,
th
e
e
m
be
ddi
ng
ve
c
to
r
s
in
a
f
or
e
s
a
id
doc
um
e
nt
s
=
{
1
,
2
,
…
,
+
}
a
r
e
s
e
gm
e
nt
e
d
by
us
in
g
to
ol
[
41]
,
w
hi
c
h
e
xt
r
a
c
ts
us
e
r
a
nd
it
e
m
ve
c
to
r
s
f
r
om
th
e
in
put
doc
um
e
nt
s
in
bot
h
s
our
c
e
a
nd
r
e
c
e
iv
in
g
dom
a
in
s
.
T
he
D
oc
2V
e
c
m
e
th
od
w
a
s
a
ppl
ie
d
to
m
a
p
in
to
th
e
te
xt
ve
c
to
r
s
a
nd
f
or
us
e
r
s
a
nd
it
e
m
s
s
e
m
a
nt
ic
e
m
b
e
dde
d
ve
c
to
r
s
.
3
.3
.
S
e
m
an
t
ic
gr
ap
h
c
on
s
t
r
u
c
t
io
n
In
th
is
w
or
k,
th
e
us
e
r
s
a
nd
it
e
m
s
w
e
r
e
m
a
ppe
d
by
e
xpl
oi
ti
n
g
th
e
ir
in
te
r
a
c
ti
on
r
e
la
ti
ons
hi
ps
.
T
he
r
e
la
ti
ve
w
e
ig
ht
of
each
in
te
r
a
c
ti
on
w
a
s
m
e
a
s
ur
e
d
in
s
ync
w
it
h
th
e
lo
c
a
l
r
e
a
c
ha
bi
li
ty
de
ns
it
y
(
12)
,
w
hi
c
h
is
de
r
iv
e
d
as
th
e
nor
m
a
li
z
e
d
r
a
ti
ngs
(
13)
.
=
(
)
(
13)
In
r
e
f
e
r
e
nc
e
to
th
e
ta
r
ge
te
d
he
te
r
oge
nous
(
s
e
m
a
nt
ic
)
gr
a
ph,
th
e
s
ynt
he
ti
c
e
dge
s
w
e
r
e
obt
a
in
e
d
in
be
twe
e
n
th
e
two
us
e
r
s
or
th
e
two
it
e
m
s
ba
s
e
d
or
c
or
r
e
s
ponding
nor
m
a
li
z
e
d
e
dge
w
e
ig
ht
s
(
13)
.
In
th
is
m
a
nne
r
,
th
e
li
ke
li
hood
(
,
)
of
th
e
e
dge
in
be
twe
e
n
th
e
two
us
e
r
s
a
nd
is
e
s
ti
m
a
te
d
by
us
in
g
(
14)
.
(
,
)
=
.
(
,
)
(
14)
In
(
14)
,
r
e
f
e
r
s
to
a
hype
r
-
pa
r
a
m
e
te
r
c
ont
r
o
ll
in
g
th
e
s
a
m
pl
in
g
li
ke
li
hood,
w
hi
le
(
,
)
r
e
pr
e
s
e
nt
s
th
e
nor
m
a
li
z
e
d
c
os
in
e
s
im
il
a
r
it
y
(
N
C
S
)
in
be
twe
e
n
a
nd
.
Be
not
e
d,
he
r
e
a
nd
r
e
pr
e
s
e
nt
th
e
us
e
r
pr
of
il
e
doc
um
e
nt
s
f
or
th
e
−
ℎ
a
nd
−
ℎ
us
e
r
.
S
im
il
a
r
ly
,
th
e
e
dge
s
(
pr
oba
bi
li
ty
)
f
o
r
it
e
m
-
it
e
m
a
r
e
a
ls
o
m
e
a
s
ur
e
d.
T
hus
,
ba
s
e
d
on
th
e
us
e
r
-
it
e
m
,
it
e
m
-
it
e
m
a
nd
us
e
r
-
us
e
r
in
te
r
a
c
ti
on
r
e
la
ti
ons
hi
p,
s
e
m
a
nt
ic
he
te
r
oge
ne
ous
gr
a
phs
a
r
e
obt
a
in
e
d
f
or
bot
h
or
ig
in
a
ti
ng
dom
a
in
a
n
d
th
e
r
e
c
e
iv
in
g
dom
a
in
.
3.
4
.
F
e
at
u
r
e
am
al
gam
at
io
n
la
ye
r
T
he
s
e
m
a
nt
ic
e
m
be
ddi
ng
m
a
tr
ix
is
e
s
ti
m
a
te
d
a
nd
c
om
bi
n
e
d
us
in
g
an
e
le
m
e
nt
-
w
is
e
a
tt
e
nt
io
n
m
e
c
ha
ni
s
m
.
T
hi
s
c
r
e
a
te
s
an
opt
im
a
l
w
e
ig
ht
s
e
t
f
or
c
om
m
on
us
e
r
s
le
a
r
ne
d
a
c
r
os
s
dom
a
in
s
,
w
it
h
c
om
bi
ne
d
e
m
be
ddi
ngs
f
or
th
e
c
om
m
on
u
s
e
r
s
U
̃
f
r
om
each
dom
a
in
r
e
pr
e
s
e
nt
s
th
e
f
e
a
tu
r
e
s
le
a
r
ne
d
ov
e
r
th
e
di
f
f
e
r
e
nt
f
r
a
c
ti
ons
.
In
a
tt
e
nt
io
n
m
e
th
ods
,
a
s
pe
c
if
ic
f
r
a
c
ti
on
of
th
e
f
e
a
tu
r
e
r
e
pr
e
s
e
nt
a
ti
on
is
s
e
l
e
c
te
d,
w
hi
c
h
a
r
e
s
ubs
e
que
nt
ly
gi
ve
n
hi
ghe
r
w
e
ig
ht
s
to
ge
ne
r
a
te
th
e
c
om
bi
ne
d
f
e
a
tu
r
e
[
42]
.
F
or
a
us
e
r
u
i
,
it
gi
ve
s
m
or
e
a
tt
e
nt
io
n
to
th
e
e
le
m
e
nt
s
pos
s
e
s
s
in
g
hi
ghe
r
in
f
or
m
a
ti
on
f
r
om
each
pa
ir
of
it
e
m
s
in
U
i
p
a
nd
U
i
q
.
T
he
pr
opo
s
e
d
e
le
m
e
nt
-
w
is
e
a
tt
e
nt
io
n
m
e
th
od
ge
ne
r
a
te
s
two
e
m
be
ddi
ng
m
a
tr
ic
e
s
U
̃
i
p
a
nd
U
̃
i
q
f
or
th
e
c
om
m
on
us
e
r
u
i
f
or
th
e
or
ig
in
a
l
a
nd
th
e
r
e
c
e
iv
in
g
dom
a
in
,
c
or
r
e
s
pondingl
y.
T
he
c
om
bi
ne
d
H
S
E
M
e
m
be
ddi
ng
f
e
a
tu
r
e
U
̃
i
a
a
f
te
r
a
tt
e
nt
io
n
m
e
c
ha
ni
s
m
f
or
a
us
e
r
u
i
in
or
ig
in
a
ti
ng
dom
a
in
is
de
r
iv
e
d
us
in
g
(
15)
.
̃
=
⊙
̃
+
(
1
−
)
⊙
(
15)
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
H
e
te
r
oge
ne
ou
s
s
e
m
ant
ic
gr
aph e
m
b
e
ddi
ng as
s
i
s
te
d e
dge
-
s
e
ns
it
i
v
e
l
e
ar
ni
ng
…
(
P
r
av
in
R
am
c
hand
r
a P
at
il
)
1845
In
(
15)
,
⊙
r
e
pr
e
s
e
nt
s
th
e
e
le
m
e
nt
-
w
is
e
a
tt
e
nt
io
n
f
unc
ti
on,
w
hi
le
∈
ℝ
×
r
e
f
e
r
s
to
th
e
w
e
ig
ht
m
a
tr
ix
r
e
tr
ie
ve
d
at
th
e
a
tt
e
nt
io
n
la
ye
r
f
or
dom
a
in
.
T
hus
,
we
obt
a
in
e
d
th
e
c
um
ul
a
ti
ve
e
m
be
ddi
ng
m
a
tr
ix
f
or
a
us
e
r
as
̃
f
or
th
e
r
e
c
e
iv
in
g
dom
a
in
.
T
he
c
om
bi
ne
d
e
m
be
ddi
ng
m
a
tr
ic
e
s
f
r
om
bot
h
dom
a
in
s
a
r
e
f
e
d
as
in
pu
t
to
th
e
M
L
P
la
ye
r
s
(
F
ig
ur
e
1)
f
or
le
a
r
ni
ng
a
nd
pr
e
di
c
ti
on.
An
M
L
P
la
ye
r
s
ni
ppe
t
is
s
how
n
a
s
f
ol
lo
w
.
3.
5
.
M
u
lt
i
-
la
ye
r
p
e
r
c
e
p
t
r
on
la
ye
r
T
he
s
tu
dy
us
e
s
a
f
ul
ly
-
c
onne
c
te
d
M
L
P
to
c
a
pt
ur
e
non
-
li
ne
a
r
u
s
e
r
-
it
e
m
in
te
r
a
c
ti
ons
in
each
dom
a
in
.
T
hi
s
m
ul
ti
-
la
ye
r
f
e
e
d
-
f
or
w
a
r
d
ne
u
r
a
l
ne
twor
k
le
a
r
ns
c
om
pl
e
x
in
te
r
a
c
ti
ons
,
unl
ik
e
tr
a
di
ti
ona
l
m
e
th
ods
th
a
t
r
e
ly
on
li
ne
a
r
e
le
m
e
nt
-
w
is
e
a
tt
e
nt
io
n.
To
le
a
r
n
ove
r
th
e
la
te
nt
f
e
a
tu
r
e
s
or
H
S
E
M
e
m
b
e
ddi
ng
m
e
tr
ic
s
of
th
e
us
e
r
s
a
nd
it
e
m
s
,
th
e
he
te
r
oge
ne
ou
s
gr
a
ph
e
m
be
ddi
ng
m
a
tr
ix
(
f
e
a
tu
r
e
s
)
f
r
om
each
dom
a
in
is
f
e
d
to
th
e
M
L
P
th
a
t
obt
a
in
e
d
th
e
la
te
nt
e
m
be
ddi
ng
m
a
tr
ix
of
th
e
−
ℎ
dom
a
in
us
in
g
(
16)
.
0
=
1
=
(
1
.
0
+
1
)
=
(
.
−
1
+
)
=
(
.
−
1
+
)
(
16)
In
(
16)
,
1
,
…
,
a
nd
1
,
…
,
r
e
pr
e
s
e
nt
th
e
w
e
ig
ht
a
nd
th
e
bi
a
s
m
a
tr
ic
e
s
f
or
th
e
M
L
P
.
H
e
r
e
,
s
ta
te
s
th
e
to
ta
l
num
be
r
of
M
L
P
la
ye
r
s
,
w
hi
le
(
∗
)
be
th
e
r
e
c
ti
f
ie
d
li
ne
a
r
uni
t
(
R
e
L
U
)
a
c
ti
va
ti
on
f
unc
ti
on.
In
(
16
)
,
∈
ℝ
×
be
th
e
us
e
r
’
s
la
te
nt
f
e
a
tu
r
e
m
a
tr
ix
f
o
r
th
e
p
-
th
dom
a
in
le
a
r
nt
by
th
e
M
L
P
m
e
th
od.
T
hus
,
th
e
la
te
nt
(
s
e
m
a
nt
ic
)
e
m
be
ddi
ng
m
a
tr
ix
f
or
it
e
m
s
pe
r
ta
in
in
g
to
th
e
−
ℎ
dom
a
in
is
obt
a
in
e
d
as
p
e
r
(
17)
.
0
=
1
=
(
1
.
0
+
1
)
=
(
.
−
1
+
)
=
(
.
−
1
+
)
(
17)
In
(
17)
,
1
,
…
,
a
nd
1
,
…
,
r
e
pr
e
s
e
nt
th
e
w
e
ig
ht
s
a
nd
th
e
bi
a
s
m
a
tr
ic
e
s
f
or
th
e
M
L
P
ne
twor
k
.
∈
ℝ
×
be
th
e
it
e
m
la
te
nt
e
m
be
ddi
ng
m
e
tr
ic
s
f
or
th
e
−
ℎ
dom
a
in
.
3.
6
.
O
u
t
p
u
t
la
ye
r
T
hi
s
la
ye
r
ha
ndl
e
s
th
e
u
s
e
r
-
it
e
m
pr
e
di
c
ti
on
f
or
th
e
r
e
c
e
iv
in
g
dom
a
in
s
.
T
he
pr
opo
s
e
d
m
ode
l
e
xe
c
ut
e
s
M
L
P
tr
a
in
in
g
ove
r
th
e
in
pu
t
c
om
bi
ne
d
us
e
r
-
it
e
m
la
te
nt
(
s
a
y,
H
S
G
E
)
m
e
tr
ic
s
,
w
he
r
e
th
e
tr
a
in
in
g
m
ode
l
is
de
s
ig
ne
d
ba
s
e
d
on
th
e
lo
s
s
in
be
twe
e
n
th
e
pr
e
di
c
te
d
a
nd
th
e
m
e
a
s
ur
e
d
us
e
r
-
it
e
m
in
te
r
a
c
ti
on
r
e
la
ti
ons
hi
p.
T
hus
,
th
e
pr
opos
e
d
m
ode
l
is
tr
a
in
e
d
by
us
in
g
th
e
f
ol
lo
w
in
g
ob
je
c
ti
ve
f
unc
ti
on
to
m
a
k
e
pr
e
di
c
ti
ons
f
or
th
e
r
e
c
e
iv
in
g
dom
a
in
.
In
(
18
)
,
(
,
̂
)
r
e
f
e
r
s
a
lo
s
s
f
unc
ti
on
be
twe
e
n
th
e
obs
e
r
ve
d
us
e
r
-
it
e
m
in
te
r
a
c
ti
ons
(
s
a
y,
)
a
nd
th
e
pr
e
di
c
te
d
us
e
r
-
it
e
m
in
te
r
a
c
ti
on
̂
.
In
(
18)
,
+
a
nd
−
r
e
pr
e
s
e
nt
s
th
e
obs
e
r
ve
d
a
nd
th
e
pr
e
di
c
te
d
us
e
r
-
it
e
m
in
te
r
a
c
ti
ons
ove
r
th
e
or
ig
in
a
ti
ng
dom
a
in
,
c
or
r
e
s
pondingl
y.
‖
‖
2
+
‖
‖
2
r
e
f
e
r
s
th
e
r
e
gul
a
r
iz
e
,
w
hi
le
r
e
pr
e
s
e
nt
s
th
e
hype
r
-
pa
r
a
m
e
te
r
c
ont
r
ol
li
ng
th
e
le
ve
l
of
s
ig
ni
f
ic
a
nc
e
of
th
e
r
e
gul
a
r
iz
e
.
m
in
,
,
⊝
∑
(
,
̂
)
∈
+
⋃
−
+
(
‖
‖
2
+
‖
‖
2
)
(
18)
To
a
ll
e
vi
a
te
ove
r
-
f
it
ti
ng
to
w
a
r
ds
+
,
a
de
f
in
it
e
a
m
ount
of
th
e
unobs
e
r
ve
d
us
e
r
-
it
e
m
in
te
r
a
c
ti
on
r
e
la
ti
ons
hi
ps
is
s
e
le
c
te
d
a
r
bi
tr
a
r
il
y
as
th
e
ne
ga
ti
ve
in
s
ta
nc
e
s
,
gi
ve
n
by
−
.
−
w
a
s
us
e
d
to
s
ubs
ti
tu
te
−
,
w
hi
c
h
he
lp
s
a
c
hi
e
vi
ng
s
w
if
t
a
nd
a
c
c
ur
a
te
le
a
r
ni
ng
ove
r
non
-
li
ne
a
r
in
put
f
e
a
tu
r
e
s
.
T
hu
s
,
us
in
g
us
e
r
’
s
r
e
vi
e
w
,
f
or
a
us
e
r
a
nd
an
it
e
m
,
th
e
c
or
r
e
s
ponding
us
e
r
-
it
e
m
in
te
r
a
c
ti
ons
w
a
s
obt
a
in
e
d
(
19)
.
=
{
∈
+
0
∈
−
ℎ
(
19)
In
t
hi
s
w
or
k,
a
nor
m
a
li
z
e
d
c
r
o
s
s
-
e
nt
r
op
y
lo
s
s
f
un
c
ti
on
w
a
s
a
ppl
i
e
d
to
pe
r
f
or
m
l
e
a
r
ni
ng
,
w
hi
c
h
is
de
r
i
ve
d
as
(
20)
.
(
,
̂
)
=
(
)
l
o
g
̂
+
(
1
−
(
)
)
l
o
g
(
1
−
̂
)
(
20)
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, No. 3, J
une
2025
:
1839
-
1852
1846
In
(
20)
,
(
)
r
e
pr
e
s
e
nt
s
th
e
m
a
xi
m
um
r
a
ti
ng
on
th
e
or
ig
in
a
ti
ng
do
m
a
in
.
In
th
is
w
or
k,
M
L
P
is
us
e
d
to
r
e
pr
e
s
e
nt
a
non
-
li
ne
a
r
a
s
s
oc
ia
ti
on
a
m
ongs
t
th
e
u
s
e
r
s
a
nd
i
te
m
s
.
L
e
t,
f
or
th
e
M
L
P
ne
twor
k
th
e
in
put
e
m
be
ddi
ng
m
a
tr
ic
e
s
f
or
th
e
us
e
r
s
a
nd
it
e
m
s
ove
r
th
e
or
ig
in
a
ti
n
g
dom
a
in
be
=
[
̃
;
]
a
nd
=
,
c
or
r
e
s
pondingl
y.
H
e
r
e
,
̃
s
ta
te
s
th
e
c
om
bi
ne
d
H
S
G
E
e
m
be
ddi
ng
m
a
tr
ix
pe
r
ta
in
in
g
to
th
e
c
om
m
on
us
e
r
s
f
o
r
th
e
dom
a
in
,
w
hi
le
th
e
e
m
be
ddi
ng
m
a
tr
ix
of
th
e
di
s
ti
nc
t
us
e
r
s
in
dom
a
in
be
.
T
hus
,
f
or
us
e
r
a
nd
it
e
m
,
th
e
c
or
r
e
s
ponding
e
m
be
ddi
ng
m
a
tr
ic
e
s
in
th
e
out
put
la
ye
r
of
th
e
M
L
P
a
r
e
obt
a
in
e
d
as
(
21
)
a
nd
(
22)
.
=
=
(
…
(
(
.
1
)
.
2
)
)
(
21)
=
=
(
…
(
(
.
1
)
.
2
)
)
(
22)
In
(
21)
a
nd
(
22)
,
R
e
L
U
a
c
ti
va
ti
on
f
unc
ti
on
(
∗
)
w
a
s
a
ppl
ie
d,
w
hi
le
1
a
nd
2
…,
a
nd
1
,
2
…
be
th
e
w
e
ig
ht
s
of
th
e
m
ul
ti
-
la
ye
r
ne
twor
ks
in
va
r
ie
d
la
ye
r
s
in
dom
a
in
f
or
a
nd
,
c
or
r
e
s
pondingl
y.
F
in
a
ll
y,
in
out
put
la
ye
r
,
th
e
pr
e
di
c
te
d
us
e
r
-
i
te
m
in
te
r
a
c
ti
on
̂
,
be
twe
e
n
t
he
us
e
r
a
nd
th
e
it
e
m
on
dom
a
in
is
gi
ve
n
by
th
e
(
23)
.
̂
,
=
(
,
)
=
.
‖
‖
‖
‖
(
23)
S
im
il
a
r
ly
,
th
e
pr
e
di
c
te
d
in
te
r
a
c
ti
on
̂
,
on
th
e
r
e
c
e
iv
in
g
do
m
a
in
w
e
r
e
obt
a
in
e
d
to
pe
r
f
or
m
to
p
-
N
r
e
c
om
m
e
nda
ti
on.
T
hus
,
th
e
c
r
os
s
-
dom
a
in
r
e
c
om
m
e
nde
r
m
ode
l
w
a
s
de
ve
lo
pe
d
u
s
in
g
th
e
a
f
or
e
m
e
nt
io
ne
d
m
e
th
ods
,
as
s
how
n
in
F
ig
ur
e
2,
w
it
h
s
im
ul
a
ti
on
r
e
s
ul
ts
a
nd
in
s
i
ght
s
di
s
c
us
s
e
d
in
th
e
s
ub
s
e
que
nt
s
e
c
ti
on.
E
x
p
l
i
c
i
t
I
m
p
l
i
c
i
t
E
x
p
l
i
c
i
t
I
m
p
l
i
c
i
t
L
a
y
e
r
1
L
a
y
e
r
1
L
a
y
e
r
1
L
a
y
e
r
1
L
a
y
e
r
2
L
a
y
e
r
2
L
a
y
e
r
2
L
a
y
e
r
2
L
a
y
e
r
3
L
a
y
e
r
3
L
a
y
e
r
3
L
a
y
e
r
3
L
a
y
e
r
X
L
a
y
e
r
X
L
a
y
e
r
X
L
a
y
e
r
X
R
e
L
U
R
e
L
U
R
e
L
U
R
e
L
U
R
e
L
U
R
e
L
U
R
e
L
U
R
e
L
U
R
e
L
U
R
e
L
U
R
e
L
U
R
e
L
U
S
o
u
r
c
e
D
o
m
a
i
n
p
T
a
r
g
e
t
D
o
m
a
i
n
q
T
r
a
i
n
i
n
g
T
r
a
i
n
i
n
g
T
o
p
-
N
P
r
e
d
i
c
t
i
o
n
T
o
p
-
N
P
r
e
d
i
c
t
i
o
n
E
l
e
m
e
n
t
-
w
i
s
e
A
t
t
e
n
t
i
o
n
1
2
3
4
5
1
I
n
p
u
t
L
a
y
e
r
2
H
S
G
M
L
a
y
e
r
3
A
t
t
e
n
t
i
o
n
L
a
y
e
r
4
M
L
P
L
a
y
e
r
5
O
u
t
p
u
t
L
a
y
e
r
F
ig
ur
e
2. P
r
opos
e
d C
D
R
m
ode
l
r
e
a
li
z
a
ti
on
4.
R
E
S
U
L
T
S
AND
D
I
S
C
U
S
S
I
O
N
4.1.
E
xp
e
r
im
e
n
t
al
d
at
as
e
t
s
In
a
li
gnm
e
nt
w
it
h
th
e
C
D
R
ta
s
k,
th
is
r
e
s
e
a
r
c
h
c
on
s
id
e
r
e
d
two
di
s
ti
nc
t
b
e
nc
hm
a
r
ks
d
a
ta
s
e
t
s
.
f
e
a
tu
r
in
g
r
e
a
l
-
time
t
r
a
ns
a
c
ti
on
de
ta
il
s
th
a
t
va
r
ie
d
s
ig
ni
f
ic
a
nt
ly
in
da
ta
na
tu
r
e
a
nd
it
e
m
ty
pe
s
.
S
pe
c
if
ic
a
ll
y,
th
e
s
tu
dy
f
oc
us
e
d
on
D
ouba
n
da
ta
s
e
ts
,
w
hi
c
h
in
c
lu
de
book
a
nd
m
ovi
e
s
ubs
e
ts
.
We
obt
a
in
e
d
th
e
us
e
r
a
nd
it
e
m
s
w
it
h
at
le
a
s
t
f
our
in
te
r
a
c
ti
ons
to
e
xpl
oi
t
m
or
e
be
ha
vi
or
a
l
a
s
pe
c
ts
,
r
e
duc
e
noi
s
e
pr
oba
bi
li
ty
,
a
nd
e
nha
nc
in
g
pr
e
di
c
ti
on
a
c
c
ur
a
c
y
in
th
e
or
ig
in
a
ti
ng
dom
a
in
by
us
in
g
th
e
M
F
N
N
-
M
L
P
la
ye
r
.
A
s
ni
ppe
t
of
th
e
da
ta
s
e
t
us
e
d
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
H
e
te
r
oge
ne
ou
s
s
e
m
ant
ic
gr
aph e
m
b
e
ddi
ng as
s
i
s
te
d e
dge
-
s
e
ns
it
i
v
e
l
e
ar
ni
ng
…
(
P
r
av
in
R
am
c
hand
r
a P
at
il
)
1847
in
th
is
w
or
k
is
gi
ve
n
in
T
a
bl
e
1.
As
de
pi
c
te
d
in
T
a
bl
e
1
,
th
e
m
ovi
e
da
ta
s
e
t
e
nc
om
p
a
s
s
e
d
a
s
e
t
of
in
te
r
a
c
ti
ons
w
it
h
2
,
718
us
e
r
s
a
nd
9
,
565
it
e
m
s
th
a
t
c
um
ul
a
ti
ve
ly
ga
ve
r
is
e
to
th
e
da
ta
s
iz
e
of
1
,
133
,
420.
S
im
il
a
r
ly
,
th
e
book
da
ta
s
e
t
po
s
s
e
s
s
e
d
2
,
718
us
e
r
s
w
it
h
6
,
777
it
e
m
s
w
it
h
a
to
t
a
l
da
ta
s
iz
e
of
96
,
041.
O
ur
c
ons
id
e
r
e
d
da
ta
s
e
ts
e
nc
om
pa
s
s
e
d
bot
h
e
xpl
ic
it
da
ta
a
nd
im
pl
ic
it
de
ta
il
s
,
w
he
r
e
th
e
f
ir
s
t
e
m
bodi
e
s
th
e
r
e
vi
e
w
s
a
nd
r
a
ti
ngs
,
w
hi
le
th
e
la
tt
e
r
c
ont
a
in
s
it
e
m
m
e
ta
da
ta
.
T
he
pr
opos
e
d
a
lg
or
it
hm
w
a
s
de
v
e
lo
pe
d
in
P
yt
hon
la
ngua
ge
,
w
he
r
e
th
e
s
im
ul
a
ti
on
e
nvi
r
onm
e
nt
e
nc
om
pa
s
s
e
d
P
yt
hon
3.4,
w
it
h
th
e
C
P
U
a
r
m
or
e
d
w
it
h
8
GB
m
e
m
or
y
r
unni
ng
ove
r
I
nt
e
l
i5
pr
oc
e
s
s
or
.
It
a
ls
o
a
ppl
ie
d
di
f
f
e
r
e
nt
P
yt
hon
li
br
a
r
ie
s
in
c
lu
di
ng
N
um
P
y
,
m
a
th
s
,
S
c
iP
y
,
G
e
ns
im
W
or
d2V
e
c
,
a
nd
T
e
ns
or
F
lo
w
.
T
he
s
im
ul
a
ti
ons
w
e
r
e
m
a
de
on
t
he
G
oogl
e
C
ol
a
b
pl
a
tf
o
r
m
w
it
h
th
e
pr
oc
e
s
s
or
c
onf
ig
ur
e
d
w
it
h
th
e
na
ti
ve
gr
a
phi
c
a
l
pr
oc
e
s
s
in
g unit
.
T
a
bl
e
1.
D
a
ta
s
pe
c
if
ic
a
ti
on
s
D
a
t
a
U
s
e
r
I
t
e
m
s
D
a
t
a
s
i
z
e
M
ovi
e
2
,
718
9
,
565
1
,
133
,
420
B
ooks
2
,
718
6
,
777
96
,
041
4.2.
S
im
u
la
t
io
n
p
ar
am
e
t
e
r
s
To
e
ns
ur
e
opt
im
a
l
pe
r
f
or
m
a
nc
e
,
th
e
pr
opos
e
d
m
ode
l
e
m
pl
oye
d
num
e
r
ous
va
lu
e
a
ddi
ti
ons
in
te
r
m
s
of
th
e
de
s
ig
n
pa
r
a
m
e
t
e
r
s
or
th
e
hype
r
pa
r
a
m
e
te
r
tu
ni
ng.
In
th
e
de
pl
oye
d
r
e
c
om
m
e
nde
r
m
ode
l,
e
s
pe
c
ia
ll
y
f
or
H
S
E
M
,
th
e
s
a
m
pl
in
g
pr
oba
bi
li
ty
w
a
s
a
s
s
ig
ne
d
as
0.05.
In
G
e
n
s
im
W
or
d2V
e
c
e
m
be
ddi
ng
of
th
e
r
a
ti
ngs
a
nd
it
e
m
s
,
we
a
s
s
ig
n
e
d
w
in
dow
s
iz
e
as
3.
T
he
N
ode
2V
e
c
a
nd
D
oc
2V
e
c
m
ode
ls
'
hype
r
-
pa
r
a
m
e
te
r
s
w
e
r
e
a
dj
us
te
d,
a
nd
an
M
F
N
N
-
dr
iv
e
n
M
L
P
w
a
s
de
s
ig
ne
d
a
nd
tu
ne
d
f
or
di
f
f
e
r
e
nt
la
ye
r
a
r
c
hi
te
c
tu
r
e
s
.
I
ni
ti
a
li
z
e
d
pa
r
a
m
e
te
r
s
w
e
r
e
G
a
us
s
ia
n
di
s
tr
ib
ut
io
n,
a
nd
tr
a
in
in
g
w
a
s
done
w
it
h
8
ne
ga
ti
ve
in
s
ta
nc
e
s
pe
r
pos
it
iv
e
in
s
ta
nc
e
,
100
e
poc
hs
,
f
ix
e
d
le
a
r
ni
ng
r
a
te
,
r
e
gul
a
r
iz
a
ti
on
pa
r
a
m
e
te
r
,
a
nd
64
ba
tc
h
s
iz
e
s
.
H
e
r
e
,
th
e
ke
y
m
ot
iv
e
be
hi
nd
lo
w
e
r
ba
tc
h
s
iz
e
w
a
s
to
im
pr
ove
(
in
tr
in
s
ic
)
f
e
a
tu
r
e
le
a
r
ni
ng
ov
e
r
th
e
he
te
r
oge
nous
e
m
be
ddi
ng
f
e
a
tu
r
e
s
w
hi
c
h
can
ha
ve
hi
gh
non
-
li
ne
a
r
it
y
ove
r
m
i
ll
io
ns
of
th
e
us
e
r
’
s
p
r
of
il
e
s
.
F
ur
th
e
r
m
o
r
e
,
we
pe
r
f
or
m
e
d
f
e
a
tu
r
e
e
m
be
ddi
ng
c
onc
e
r
ni
ng
th
e
di
f
f
e
r
e
nt
di
m
e
ns
io
ns
in
c
lu
di
ng
4,
8,
16,
32,
a
nd
64.
4.3.
E
val
u
at
io
n
p
ar
am
e
t
e
r
s
T
he
s
tu
dy
f
oc
us
e
s
on
a
to
p
-
N
r
e
c
om
m
e
nda
ti
on
ta
s
k
f
or
us
e
r
s
in
a
r
e
c
e
iv
in
g
dom
a
in
,
r
a
nki
ng
ta
r
ge
t
it
e
m
s
ba
s
e
d
on
in
te
r
a
c
ti
ons
w
it
h
uni
nt
e
r
a
c
te
d
it
e
m
s
.
A
le
a
ve
-
one
-
out
a
ppr
oa
c
h
is
a
dopt
e
d,
s
e
le
c
ti
ng
r
e
c
e
nt
in
te
r
a
c
ti
ons
w
it
h
a
te
s
t
it
e
m
f
or
each
u
s
e
r
.
T
he
pe
r
f
or
m
a
nc
e
p
a
r
a
m
e
te
r
s
a
r
e
HR
a
nd
N
D
C
G
,
w
hi
c
h
m
e
a
s
ur
e
th
e
r
e
c
a
ll
r
a
te
a
nd
th
e
qua
li
ty
of
th
e
r
a
nki
ng.
T
he
a
im
is
to
e
va
lu
a
te
th
e
e
f
f
e
c
ti
ve
ne
s
s
of
th
is
r
a
nki
ng
-
ba
s
e
d
m
e
th
od.
T
he
pr
opos
e
d
c
r
os
s
-
dom
a
in
r
e
c
om
m
e
nde
r
m
ode
l
ge
ne
r
a
te
s
a
(
to
p
-
N)
r
a
nke
d
li
s
t
of
it
e
m
s
(
in
th
e
r
e
c
e
iv
in
g
dom
a
in
)
f
or
th
e
ta
r
ge
t
u
s
e
r
.
If
a
ta
r
ge
t
it
e
m
is
in
th
is
li
s
t,
it
'
s
a
hi
t.
We
de
f
in
e
hi
t
r
a
te
(
H
R
)
as
in
(
24)
.
In
(
24)
,
th
e
de
nom
in
a
to
r
|
|
r
e
pr
e
s
e
nt
s
th
e
to
ta
l
num
be
r
of
in
te
r
a
c
ti
ons
in
th
e
te
s
t
s
e
t.
In
th
is
w
or
k,
N
D
C
G
w
a
s
a
ppl
ie
d
to
pr
ovi
de
(
or
a
s
s
e
s
s
)
th
e
hi
t
pos
it
io
n
by
a
s
s
ig
ni
ng
hi
ghe
r
s
c
or
e
s
to
th
e
hi
ts
(
e
s
pe
c
i
a
ll
y
at
th
e
to
p
K
r
a
nks
)
.
@
=
.
@
|
|
(
24)
M
a
th
e
m
a
ti
c
a
ll
y,
we
de
r
iv
e
d
N
D
C
G
@
K
as
pe
r
th
e
(
25)
.
In
(
25)
,
th
e
pa
r
a
m
e
te
r
s
ta
te
s
th
e
r
a
nke
d
r
e
le
va
nc
e
of
th
e
ta
r
ge
t
it
e
m
at
th
e
-
th
pos
it
io
n
.
T
he
r
e
f
or
e
,
in
t
hi
s
c
ont
e
xt
,
=
1
.
@
=
∑
2
−
1
l
o
g
2
(
+
1
)
=
1
(
25)
In
ot
he
r
w
is
e
c
a
s
e
,
we
la
be
l
=
0
.
A
ddi
ti
ona
ll
y,
we
a
ls
o
e
s
ti
m
a
te
d
th
e
r
oot
m
e
a
n
s
qua
r
e
e
r
r
or
(
R
M
S
E
)
pe
r
f
or
m
a
nc
e
,
w
hi
c
h
w
a
s
de
r
iv
e
d
as
pe
r
th
e
(
26)
.
In
(
26)
,
r
e
pr
e
s
e
nt
s
th
e
to
ta
l
num
be
r
of
te
s
t
r
a
ti
ngs
,
w
hi
le
(
,
)
s
ta
te
s
th
e
r
e
a
l
r
a
ti
ng
a
nd
th
e
m
e
a
s
ur
e
d
or
pr
e
di
c
t
e
d
r
a
ti
ng
be
̂
(
,
)
.
=
√
1
∑
(
(
,
)
−
̂
(
,
)
)
2
,
,
(
26)
T
hi
s
w
or
k
c
ha
r
a
c
te
r
iz
e
s
ove
r
a
ll
pe
r
f
or
m
a
nc
e
a
nd
in
vol
ve
s
bot
h
in
tr
a
-
a
nd
in
te
r
-
m
ode
l
a
s
s
e
s
s
m
e
nt
s
.
H
e
r
e
,
th
e
in
tr
a
-
m
ode
l
a
s
s
e
s
s
m
e
nt
s
ta
t
e
s
th
e
pe
r
f
or
m
a
nc
e
c
h
a
r
a
c
te
r
iz
a
ti
on
w
it
h
th
e
di
f
f
e
r
e
nt
e
m
be
ddi
ng
di
m
e
ns
io
na
l
(
s
a
y,
la
te
nt
di
m
e
n
s
io
ns
)
.
T
he
s
e
a
s
s
e
s
s
m
e
nt
s
a
r
e
c
onduc
te
d
a
c
r
os
s
di
f
f
e
r
e
nt
in
put
da
ta
s
e
ts
to
a
na
ly
z
e
pe
r
f
or
m
a
nc
e
t
r
e
nds
.
T
he
s
e
a
s
s
e
s
s
m
e
nt
s
a
r
e
c
onduc
te
d a
c
r
os
s
di
f
f
e
r
e
nt
i
nput
da
ta
s
e
ts
.
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, No. 3, J
une
2025
:
1839
-
1852
1848
4.3.1.
I
n
t
r
a
-
m
od
e
l
as
s
e
s
s
m
e
n
t
To
e
v
a
l
ua
te
th
e
e
f
f
e
c
t
iv
e
n
e
s
s
of
t
he
in
t
e
r
-
dom
a
i
n
r
e
la
te
dn
e
s
s
-
dr
iv
e
n
e
dg
e
pr
un
in
g
m
e
th
od,
we
s
im
ul
a
t
e
d
o
ur
m
o
de
l
w
it
h
a
n
d
w
i
th
o
ut
e
dg
e
pr
uni
ng
(
b
y
f
e
e
di
ng
di
r
e
c
t
H
S
G
E
m
e
tr
i
c
s
to
t
he
a
tt
e
n
ti
o
n
la
ye
r
)
.
In
a
dd
it
i
on
,
we
e
va
lu
a
t
e
d
v
a
r
i
ou
s
l
a
t
e
n
t
di
m
e
ns
io
n
s
(
4,
8,
1
6,
3
2,
a
nd
6
4)
,
w
it
h
s
im
u
la
ti
o
n
r
e
s
ul
t
s
f
or
th
e
D
oub
a
n
boo
k
a
nd
m
ovi
e
da
ta
s
e
t
s
pr
e
s
e
n
te
d
in
T
a
b
le
2.
I
t
i
s
of
t
e
n
hy
pot
h
e
s
iz
e
d
t
ha
t
t
he
hi
gh
di
m
e
ns
io
na
l
f
e
a
tu
r
e
s
of
te
n
yi
e
ld
s
u
pe
r
io
r
r
e
s
ul
ts
bu
t
in
c
r
e
a
s
e
d
c
om
put
a
t
io
n.
w
e
s
im
ul
a
te
d
our
C
D
R
m
od
e
l
u
s
in
g
f
iv
e
e
m
b
e
d
di
n
g
di
m
e
n
s
i
on
s
(
k
=
4
,
8,
1
6,
3
2,
64
)
,
w
he
r
e
is
t
h
e
e
m
be
ddi
ng
di
m
e
n
s
i
on
s
.
T
h
e
s
im
u
la
ti
o
n
r
e
s
u
lt
s
in
T
a
bl
e
2
s
ho
w
th
a
t
a
s
in
c
r
e
a
s
e
s
,
t
h
e
H
R
@
K
a
ls
o
r
i
s
e
s
.
T
r
a
in
i
ng
on
v
a
r
i
ou
s
i
n
put
d
a
t
a
s
u
bs
e
t
s
,
in
c
lu
di
ng
B
oo
k
a
nd
M
o
vi
e
,
s
h
ow
s
t
ha
t
in
c
r
e
a
s
in
g
la
te
nt
a
nd
in
t
r
in
s
i
c
f
e
a
tu
r
e
l
e
a
r
ni
ng
i
m
pr
o
ve
s
b
ot
h
H
R
a
nd
N
D
C
G
.
A
s
de
pi
c
t
e
d
i
n
T
a
bl
e
2, t
h
e
a
v
e
r
a
g
e
H
R
@
4 w
a
s
0.
56
75,
w
h
il
e
N
D
C
G
@
4 w
a
s
ob
s
e
r
v
e
d a
s
0
.34
70
. S
im
il
a
r
ly
, i
n
a
n
ot
h
e
r
t
e
s
t
c
a
s
e
w
it
h
H
S
G
E
e
m
be
ddi
ng
di
m
e
n
s
i
on
of
8
,
w
e
m
e
a
s
ur
e
d
a
v
e
r
a
g
e
H
R
@
8
a
s
0.
57
91,
w
h
il
e
N
D
C
G
@
8
w
a
s
f
ou
nd
a
s
0.3
805
.
F
o
r
k
=
1
6,
w
e
f
ou
nd
H
R
@
1
6
a
s
0.
579
9
a
nd
N
D
C
G
@
1
6
a
s
0
.3
990
.
O
n
t
he
c
o
nt
r
a
r
y
,
f
or
=
32
,
t
he
m
od
e
l
y
ie
ld
e
d H
R
@
3
2
a
s
0.5
99
8
a
nd N
D
C
G
@
3
2
a
s
0
.4
171
.
S
i
m
il
a
r
ly
,
ov
e
r
s
i
m
ul
a
t
io
n
,
it
de
pi
c
t
e
d H
R
@
6
4
of
0.6
070
a
n
d
N
D
C
G
@
64
a
s
0
.37
96
.
T
a
bl
e
3
s
ho
w
s
th
e
a
v
e
r
a
g
e
H
R
a
nd
N
D
C
G
a
c
r
o
s
s
v
a
r
i
ou
s
e
m
b
e
d
di
n
g
di
m
e
n
s
i
on
s
, w
i
th
hi
g
he
r
s
iz
e
s
l
e
a
di
ng t
o de
c
r
e
a
s
e
d R
M
S
E
,
d
e
m
o
ns
tr
a
ti
n
g
t
he
m
o
de
l
'
s
r
o
bu
s
t
ne
s
s
a
n
d
e
f
f
ic
ie
nc
y
in
pr
o
vi
d
in
g
a
c
c
ur
a
t
e
to
p
-
N
r
e
c
om
m
e
n
da
ti
o
ns
.
T
he
r
e
s
ul
ts
i
ndi
c
a
te
t
ha
t
h
ig
he
r
e
m
b
e
d
di
n
g
di
m
e
n
s
i
on
s
l
e
a
d
to
be
t
te
r
H
R
a
n
d
N
D
C
G
f
or
r
e
li
a
bl
e
C
D
R
s
.
A
c
on
s
i
s
te
nt
e
f
f
i
c
i
e
n
c
y
pa
tt
e
r
n
a
c
r
o
s
s
da
ta
s
e
t
s
(
bo
ok
s
a
n
d
m
ovi
e
s
)
c
on
f
ir
m
s
th
e
m
od
e
l'
s
s
ui
t
a
b
il
i
ty
f
or
r
e
a
l
-
t
im
e
r
e
c
om
m
e
n
da
ti
o
n
ta
s
ks
.
T
o
e
va
lu
a
te
th
e
im
pa
c
t
of
in
te
r
-
dom
a
in
e
dge
pr
uni
ng
on
to
p
-
N
r
e
c
om
m
e
nda
ti
ons
,
w
e
s
im
ul
a
te
d
our
m
ode
l
w
it
h
a
nd
w
it
hout
pr
uni
ng.
R
e
s
ul
ts
a
r
e
s
how
n
in
T
a
bl
e
4.
T
he
r
e
s
ul
t
in
T
a
bl
e
4
c
a
n
e
a
s
il
y
be
in
f
e
r
r
e
d
th
a
t
th
e
us
e
of
th
e
pr
opo
s
e
d
e
dg
e
pr
uni
ng
m
e
th
od
w
it
h
H
S
G
E
m
e
c
ha
ni
s
m
e
na
bl
e
s
s
upe
r
io
r
H
R
a
nd
c
or
r
e
s
ponding
N
D
C
G
pe
r
f
or
m
a
nc
e
.
T
he
a
ve
r
a
ge
H
R
p
e
r
f
or
m
a
nc
e
in
c
r
e
a
s
e
d
f
r
om
0.5410
w
it
hout
e
dge
-
pr
uni
ng
to
0.5866
w
it
h
e
dge
-
pr
uni
ng.
I
t
r
e
ve
a
ls
th
a
t
th
e
us
e
of
our
pr
opos
e
d
e
dge
pr
uni
ng
m
ode
l
w
it
h
H
S
G
E
e
na
bl
e
s
a
lm
os
t
9.2%
be
tt
e
r
pe
r
f
or
m
a
nc
e
(
i.
e
., H
R
)
. T
he
r
e
s
ul
t
s
c
onf
ir
m
t
ha
t
th
e
e
dge
pr
uni
ng me
th
od w
it
h t
he
H
S
G
E
m
ode
l
out
pe
r
f
or
m
s
m
e
th
ods
r
e
ly
in
g
s
ol
e
ly
on
us
e
r
-
it
e
m
in
te
r
a
c
ti
ons
,
a
s
it
r
e
duc
e
s
noi
s
e
,
s
ig
ni
f
ic
a
nt
ly
im
pr
ovi
ng
bot
h
H
R
a
nd
R
M
S
E
s
how
n
in
T
a
bl
e
s
3
a
nd
4.
T
he
H
S
G
E
-
C
D
R
m
ode
l'
s
pe
r
f
or
m
a
nc
e
w
a
s
e
va
lu
a
te
d a
ga
in
s
t
ot
he
r
l
e
a
di
ng me
th
ods
.
T
he
r
e
s
ul
ts
a
r
e
de
ta
il
e
d i
n t
he
s
ubs
e
que
nt
s
e
c
ti
on.
T
a
bl
e
2.
P
e
r
f
or
m
a
nc
e
ove
r
th
e
di
f
f
e
r
e
nt
la
te
nt
e
m
be
ddi
ng
di
m
e
ns
io
ns
L
a
t
e
n
t
di
m
e
ns
i
ons
or
e
m
be
ddi
ng
s
i
z
e
(K
)
D
a
t
a
H
R@
K
N
D
CG
@
K
RM
S
E
4
Book
0.
4531
0.
2831
1.
409
M
ovi
e
0.
6820
0.
4109
1.
389
8
Book
0.
4861
0.
2994
1.
091
M
ovi
e
0.
6721
0.
4617
1.
077
16
Book
0.
4824
0.
2993
0.
987
M
ovi
e
0.
6775
0.
4987
0.
988
32
Book
0.
5015
0.
3740
0.
938
M
ovi
e
0.
6982
0.
4603
0.
874
64
Book
0.
5126
0.
3779
0.
880
M
ovi
e
0.
7014
0.
4813
0.
871
T
a
bl
e
3.
A
ve
r
a
ge
HR
a
nd
N
D
C
G
pe
r
f
or
m
a
nc
e
ove
r
th
e
di
f
f
e
r
e
n
t
la
te
nt
di
m
e
ns
io
ns
L
a
t
e
n
t
di
m
e
ns
i
ons
or
e
m
be
ddi
ng
s
i
z
e
(K
)
A
ve
ra
ge
HR
N
D
CG
RM
S
E
4
0.
5675
0.
3470
1.
399
8
0.
5791
0.
3805
1.
084
16
0.
5799
0.
3990
0.
987
32
0.
5998
0.
4171
0.
906
64
0.
6070
0.
4296
0.
875
T
a
bl
e
4.
E
f
f
ic
a
c
y
of
e
dge
pr
uni
ng
on
H
S
G
E
d
r
iv
e
n
C
D
R
L
a
t
e
n
t
di
m
e
ns
i
ons
or
e
m
be
ddi
ng
s
i
z
e
(K
)
D
a
t
a
W
i
t
hout
e
dge
pruni
ng
With
e
dge
-
pruni
ng
H
R@
K
N
D
CG
@
K
RM
S
E
H
R@
K
N
D
CG
@
K
RM
S
E
4
Book
0.
3831
0.
2189
1.
998
0.
4531
0.
2831
1.
409
M
ovi
e
0.
5993
0.
3776
1.
849
0.
6820
0.
4109
1.
389
8
Book
0.
4473
0.
2851
1.
248
0.
4861
0.
2994
1.
091
M
ovi
e
0.
6110
0.
4294
1.
216
0.
6721
0.
4617
1.
077
16
Book
0.
4361
0.
2840
1.
198
0.
4824
0.
2993
0.
987
M
ovi
e
0.
6485
0.
4401
1.
233
0.
6775
0.
4987
0.
988
32
Book
0.
4791
0.
3392
1.
194
0.
5015
0.
3740
0.
938
M
ovi
e
0.
6620
0.
4261
1.
130
0.
6982
0.
4603
0.
874
64
Book
0.
4899
0.
3482
1.
078
0.
5126
0.
3779
0.
880
M
ovi
e
0.
6538
0.
3399
1.
004
0.
7014
0.
4813
0.
871
Evaluation Warning : The document was created with Spire.PDF for Python.