I
AE
S In
t
er
na
t
io
na
l J
o
urna
l o
f
Art
if
icia
l In
t
ellig
ence
(
I
J
-
AI
)
Vo
l.
1
5
,
No
.
4
,
A
u
g
u
s
t 2
0
2
6
,
p
p
.
3
2
4
0
~
3
2
5
1
I
SS
N:
2
2
5
2
-
8
9
3
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijai.v
15
.i
4
.
p
p
3
2
4
0
-
3
2
5
1
3240
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
i
.
ia
esco
r
e.
co
m
Reso
urce optimi
z
a
tion v
ia
ident
ifi
c
a
tion o
f
a
rticles
w
ith
a
spects
o
f
prio
rity a
nd
hi
g
h ut
ility
G
a
rim
a
Sriv
a
s
t
a
v
a
1
,
2
,
Va
is
ha
li Sin
g
h
1
,
Sa
chin K
um
a
r
3
1
D
e
p
a
r
t
me
n
t
o
f
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
En
g
i
n
e
e
r
i
n
g
,
M
a
h
a
r
i
sh
i
S
c
h
o
o
l
o
f
E
n
g
i
n
e
e
r
i
n
g
a
n
d
T
e
c
h
n
o
l
o
g
y
,
M
a
h
a
r
i
sh
i
U
n
i
v
e
r
si
t
y
o
f
I
n
f
o
r
mat
i
o
n
Te
c
h
n
o
l
o
g
y
,
L
u
c
k
n
o
w
,
I
n
d
i
a
2
D
e
p
a
r
t
me
n
t
o
f
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
En
g
i
n
e
e
r
i
n
g
,
A
mi
t
y
S
c
h
o
o
l
o
f
E
n
g
i
n
e
e
r
i
n
g
a
n
d
Te
c
h
n
o
l
o
g
y
,
A
mi
t
y
U
n
i
v
e
r
si
t
y
U
t
t
a
r
P
r
a
d
e
s
h
,
Lu
c
k
n
o
w
,
I
n
d
i
a
3
D
e
p
a
r
t
me
n
t
o
f
El
e
c
t
r
o
n
i
c
s a
n
d
C
o
m
mu
n
i
c
a
t
i
o
n
En
g
i
n
e
e
r
i
n
g
,
A
m
i
t
y
S
c
h
o
o
l
o
f
En
g
i
n
e
e
r
i
n
g
a
n
d
T
e
c
h
n
o
l
o
g
y
,
A
mi
t
y
U
n
i
v
e
r
si
t
y
U
t
t
a
r
P
r
a
d
e
s
h
,
L
u
c
k
n
o
w
,
I
n
d
i
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Ap
r
5
,
2
0
2
5
R
ev
is
ed
Ma
y
2
0
,
2
0
2
6
Acc
ep
ted
J
u
n
1
9
,
2
0
2
6
Ow
in
g
t
o
th
e
li
n
e
a
r
e
c
o
n
o
m
ic
g
ro
wt
h
,
t
h
e
m
a
n
u
fa
c
tu
ri
n
g
o
f
c
o
n
su
m
e
r
p
ro
d
u
c
ts
h
a
s
in
c
re
a
se
d
m
a
n
if
o
l
d
,
p
ro
d
u
c
ts
in
d
i
v
e
rsified
p
a
c
k
a
g
in
g
a
n
d
fo
rm
a
ts
a
re
n
o
w
a
v
a
il
a
b
le
in
a
b
u
n
d
a
n
c
e
,
so
m
e
ti
m
e
s
e
v
e
n
with
les
s
o
r
n
o
re
q
u
irem
e
n
t
,
re
su
lt
i
n
g
in
l
o
ss
e
s.
Ad
d
re
ss
in
g
t
h
e
g
a
p
t
h
a
t
e
x
ists
b
e
twe
e
n
p
ro
d
u
c
ti
v
it
y
a
n
d
re
q
u
irem
e
n
ts
c
a
n
m
in
imiz
e
lo
ss
e
s
a
n
d
u
n
n
e
c
e
ss
a
ry
b
u
r
d
e
n
o
n
m
a
n
u
fa
c
tu
r
in
g
u
n
it
s.
To
m
i
n
i
m
ize
lo
ss
e
s,
a
h
y
b
rid
a
lg
o
rit
h
m
is
p
ro
p
o
se
d
,
u
sin
g
t
h
e
a
d
v
a
n
tag
e
s
o
ffe
re
d
b
y
c
las
sifica
ti
o
n
a
n
d
h
ig
h
u
ti
l
it
y
p
a
tt
e
rn
s
to
id
e
n
ti
f
y
th
e
d
o
m
in
a
n
t
a
sp
e
c
ts
o
f
a
rti
c
les
with
a
h
ig
h
p
ro
b
a
b
il
it
y
o
f
sa
le.
Article
s
a
re
id
e
n
ti
fied
in
a
tw
o
-
s
tep
p
ro
c
e
ss
,
id
e
n
ti
fica
ti
o
n
o
f
sa
m
p
les
with
c
o
n
tex
t
a
s
p
ro
m
in
e
n
t
p
a
ra
m
e
ter
b
y
v
a
n
il
la
fe
e
d
fo
rwa
rd
n
e
u
ra
l
n
e
t
wo
r
k
(
V
NN
)
a
s
a
first
ste
p
.
T
h
e
se
c
o
n
d
ste
p
c
o
m
p
rise
s
t
h
e
d
e
term
in
a
ti
o
n
o
f
u
ti
li
t
y
p
a
tt
e
rn
m
in
in
g
in
a
rti
c
les
u
sin
g
fa
ste
r
h
ig
h
-
u
ti
li
ty
it
e
m
se
t
m
in
e
r
(F
HN
),
se
n
ti
m
e
n
t
sc
o
re
o
b
tai
n
e
d
p
r
o
v
i
d
e
s th
e
u
ti
li
t
y
p
a
tt
e
r
n
o
f
t
h
e
p
r
o
d
u
c
t.
P
ro
p
o
se
d
h
y
b
ri
d
VN
N
+
F
HN
,
a
lo
n
g
wit
h
c
o
n
v
o
lu
ti
o
n
a
l
n
e
u
ra
l
n
e
tw
o
rk
(C
NN
),
lo
n
g
sh
o
rt
-
term
m
e
m
o
ry
(LS
T
M
),
a
n
d
tran
sfo
rm
e
r
-
b
a
se
d
p
re
d
ictio
n
m
o
d
e
l
(TP
M
)
we
re
u
se
d
fo
r
a
ss
e
ss
in
g
t
h
e
u
ti
li
t
y
-
d
ri
v
e
n
p
ro
d
u
c
t
a
n
a
ly
sis.
T
h
e
h
y
b
rid
V
NN
+
F
HN
p
ro
p
o
se
d
d
isp
lay
s
t
h
e
b
e
st
a
d
a
p
tab
il
it
y
b
y
e
x
trac
ti
n
g
1
5
8
p
a
tt
e
rn
s
with
a
u
ti
li
t
y
o
f
1
6
2
.
8
8
u
n
it
s
.
T
h
e
a
lg
o
r
it
h
m
o
u
t
p
e
rfo
rm
s
CNN
in
u
ti
li
t
y
,
LS
TM
a
n
d
TP
M
i
n
sp
e
e
d
,
a
n
d
CNN
in
p
a
tt
e
rn
c
o
u
n
t
,
m
a
k
i
n
g
i
t
a
b
e
tt
e
r
c
h
o
ice
fo
r
re
so
u
rc
e
o
p
ti
m
iza
ti
o
n
.
K
ey
w
o
r
d
s
:
Fas
ter
h
ig
h
-
u
tili
ty
item
s
et
m
in
er
Fre
q
u
en
t item
s
et
m
in
in
g
Hig
h
u
tili
ty
p
atter
n
Sco
r
e
T
r
ip
les
Van
illa n
eu
r
al
n
etwo
r
k
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
:
Gar
im
a
Sriv
astav
a
Dep
ar
tm
en
t o
f
C
o
m
p
u
ter
Scie
n
ce
E
n
g
in
ee
r
in
g
,
Ma
h
ar
is
h
i Sch
o
o
l o
f
E
n
g
in
ee
r
in
g
an
d
T
ec
h
n
o
lo
g
y
Ma
h
ar
is
h
i U
n
iv
er
s
ity
o
f
I
n
f
o
r
m
atio
n
T
ec
h
n
o
lo
g
y
L
u
ck
n
o
w,
I
n
d
ia
E
m
ail:
g
s
r
iv
astav
a1
@
lk
o
.
am
ity
.
ed
u
1.
I
NT
RO
D
UCT
I
O
N
Ad
d
r
ess
in
g
tar
g
ets,
r
eq
u
ir
em
en
t
an
d
av
ailab
ilit
y
with
co
n
s
u
m
er
p
r
o
d
u
cts
is
a
tr
ick
y
ta
s
k
,
lin
ea
r
ly
p
r
o
p
o
r
tio
n
al
r
elatio
n
b
etwe
en
m
an
u
f
ac
tu
r
in
g
,
s
u
p
p
ly
c
h
ain
a
n
d
r
eq
u
ir
em
en
t
ca
n
m
in
im
ize
lo
s
s
es
an
d
in
cr
ea
s
e
f
in
an
cial
b
en
e
f
its
.
Fre
q
u
en
tly
a
wid
e
g
ap
e
x
is
ts
b
etwe
en
p
r
o
d
u
ctiv
ity
an
d
r
eq
u
ir
em
e
n
t,
g
ettin
g
ac
c
u
r
ate
f
ee
d
b
ac
k
a
n
d
ad
d
r
ess
in
g
th
e
g
ap
ef
f
ec
tiv
ely
ca
n
m
in
i
m
ize
b
u
r
d
e
n
o
n
m
a
n
u
f
ac
tu
r
in
g
u
n
its
.
R
atin
g
,
r
ec
o
m
m
en
d
atio
n
s
,
r
ev
iews
an
d
f
ee
d
b
ac
k
ca
n
p
r
o
v
id
e
q
u
ite
in
s
ig
h
tf
u
l
d
ata
ab
o
u
t
p
r
o
d
u
cts
in
d
em
an
d
a
n
d
ca
n
s
tr
ea
m
lin
e
p
r
o
p
o
r
tio
n
ality
b
et
wee
n
d
em
an
d
a
n
d
s
u
p
p
l
y
.
Pleth
o
r
a
o
f
d
iv
er
s
e
r
ec
o
m
m
en
d
at
io
n
s
y
s
tem
s
b
ased
o
n
o
p
in
io
n
m
in
in
g
tech
n
iq
u
es
ar
e
av
ailab
le
f
o
r
r
eso
u
r
ce
o
p
tim
izatio
n
,
h
o
wev
e
r
,
th
e
r
e
ex
is
ts
a
wid
e
g
ap
in
tar
g
eted
ap
p
r
o
ac
h
es
em
p
lo
y
in
g
h
y
b
r
id
alg
o
r
ith
m
s
ass
is
ted
b
y
s
tatis
tical
tech
n
iq
u
es
[
1
]
.
C
u
r
r
en
tly
,
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
R
eso
u
r
ce
o
p
timiz
a
tio
n
via
id
e
n
tifi
ca
tio
n
o
f a
r
ticles w
ith
a
s
p
ec
ts
o
f p
r
io
r
ity
…
(
Ga
r
ima
S
r
i
va
s
ta
va
)
3241
r
ec
o
m
m
en
d
er
s
y
s
tem
s
(
R
S)
ar
e
b
ein
g
u
tili
ze
d
ex
ten
s
iv
ely
in
s
ev
er
al
in
d
u
s
tr
ies,
in
clu
d
in
g
o
n
lin
e
s
h
o
p
p
in
g
,
s
o
cial
m
ed
ia,
elec
tr
o
n
ic
n
ews
ch
an
n
els,
o
n
lin
e
m
u
s
ic
p
o
r
tals
,
tr
av
el
p
lan
n
in
g
,
an
d
en
ter
tai
n
m
en
t.
T
h
is
is
d
u
e
to
th
e
g
r
o
wth
an
d
d
iv
er
s
if
icati
o
n
o
f
p
r
o
d
u
cts an
d
s
er
v
ices
[
2
]
,
[
3
]
.
T
r
ad
itio
n
al
RS
f
o
cu
s
o
n
ac
cu
r
ac
y
an
d
s
im
ilar
ity
,
b
u
t
th
e
y
f
ail
to
in
co
r
p
o
r
ate
item
p
r
o
f
itab
ilit
y
,
asp
ec
t
-
lev
el
p
r
ef
e
r
en
ce
s
,
o
r
r
ea
l
u
tili
ty
co
n
s
tr
ain
ts
.
L
i
k
ewise,
class
ical
p
atter
n
-
m
in
in
g
tech
n
iq
u
es
s
u
ch
as
ap
r
io
r
i
o
r
f
r
e
q
u
en
t
p
atter
n
(
FP
)
-
g
r
o
wth
ca
p
tu
r
e
o
n
l
y
f
r
e
q
u
en
cy
,
n
o
t
u
tili
ty
,
lead
in
g
to
th
e
o
m
is
s
io
n
o
f
h
i
g
h
-
v
a
lu
e
b
u
t
lo
w
-
f
r
e
q
u
e
n
c
y
p
atte
r
n
s
.
Hi
g
h
-
u
tili
ty
it
em
s
et
m
i
n
i
n
g
ad
d
r
ess
es
t
h
is
g
ap
b
y
m
o
d
eli
n
g
u
ti
lit
y
th
r
o
u
g
h
q
u
a
n
t
ita
ti
v
e
f
a
ct
o
r
s
s
u
c
h
as
p
r
o
f
it
,
w
ei
g
h
t,
an
d
i
m
p
o
r
t
a
n
ce
.
R
e
ce
n
t
r
ese
a
r
c
h
s
h
o
ws
t
h
at
c
o
m
b
in
in
g
h
i
g
h
-
u
ti
lit
y
p
a
tte
r
n
m
i
n
i
n
g
(
H
UPM)
wit
h
asp
ec
t
-
b
as
ed
o
p
i
n
i
o
n
m
in
in
g
en
a
b
les
m
o
r
e
m
e
an
i
n
g
f
u
l f
e
at
u
r
e
-
u
tili
ty
ass
o
ciatio
n
s
[
4
]
,
[
5
]
,
h
y
b
r
i
d
R
S
–
HUPM
m
o
d
els
[
6
]
–
[
8
]
th
er
ef
o
r
e
o
u
tp
e
r
f
o
r
m
ea
r
lier
s
in
g
le
-
o
b
jectiv
e
ap
p
r
o
ac
h
es
b
y
s
im
u
ltan
eo
u
s
ly
ca
p
tu
r
in
g
wh
at
u
s
er
s
p
r
ef
er
,
wh
at
y
ield
s
v
alu
e
,
an
d
wh
ic
h
item
asp
ec
ts
d
r
iv
e
p
r
io
r
ity
,
m
ak
in
g
th
em
i
n
cr
ea
s
in
g
ly
s
u
itab
le
f
o
r
p
r
o
d
u
ct
s
elec
tio
n
,
ca
talo
g
o
p
tim
izatio
n
,
an
d
p
er
s
o
n
alize
d
d
ec
is
io
n
s
u
p
p
o
r
t.
Ma
n
u
f
ac
tu
r
in
g
o
f
co
n
s
u
m
er
p
r
o
d
u
cts
h
as
in
cr
ea
s
ed
o
v
er
th
e
last
d
ec
ad
e,
p
r
o
d
u
cts
in
d
iv
er
s
if
ied
p
ac
k
ag
in
g
an
d
f
o
r
m
ats
ar
e
n
o
w
av
ailab
le
in
ab
u
n
d
an
ce
,
s
o
m
etim
es
ev
en
with
less
o
r
n
o
r
eq
u
ir
em
en
t
r
esu
ltin
g
in
lo
s
s
es.
A
wid
e
g
ap
ex
is
ts
b
etwe
en
p
r
o
d
u
ctiv
ity
an
d
r
eq
u
ir
em
en
t,
ad
d
r
ess
in
g
th
e
g
ap
ef
f
ec
tiv
ely
ca
n
m
in
im
ize
lo
s
s
es
an
d
u
n
n
ec
e
s
s
ar
y
b
u
r
d
en
o
n
m
an
u
f
ac
tu
r
i
n
g
u
n
its
.
Pleth
o
r
a
o
f
d
iv
e
r
s
e
r
ec
o
m
m
en
d
atio
n
s
y
s
tem
s
b
ased
o
n
o
p
in
io
n
m
in
in
g
tech
n
iq
u
es
ar
e
av
ailab
le
f
o
r
r
eso
u
r
ce
o
p
tim
izatio
n
,
h
o
w
ev
er
,
th
er
e
ex
is
ts
a
g
ap
in
tar
g
eted
ap
p
r
o
ac
h
es e
m
p
lo
y
in
g
h
y
b
r
id
alg
o
r
ith
m
s
.
W
o
r
k
p
r
o
p
o
s
ed
em
p
l
o
y
s
a
h
y
b
r
id
ap
p
r
o
ac
h
o
f
class
if
icatio
n
an
d
h
ig
h
u
tili
ty
p
atter
n
to
i
d
en
tify
th
e
d
o
m
in
an
t
asp
ec
ts
o
f
ar
ticles
with
h
ig
h
s
ale
p
r
o
b
ab
ilit
y
.
T
h
e
f
ir
s
t
s
tep
co
m
p
r
is
es
o
f
id
en
tific
atio
n
a
n
d
class
if
icatio
n
o
f
s
am
p
les
with
co
n
tex
t
as
p
r
o
m
in
e
n
t
p
ar
am
e
ter
,
th
e
s
ec
o
n
d
s
tep
tak
es
o
u
tp
u
t
o
f
f
ir
s
t
s
tep
as
in
p
u
t,
s
tep
co
m
p
r
is
es
o
f
o
p
in
i
o
n
ass
is
ted
m
in
in
g
f
o
r
d
eter
m
i
n
in
g
h
i
g
h
u
tili
ty
p
atter
n
in
ar
ti
cles,
s
co
r
e
r
atin
g
is
esti
m
ated
f
o
r
id
en
tify
in
g
s
am
p
les
with
h
ig
h
u
tili
ty
f
r
o
m
tr
ip
les
ex
tr
ac
ted
f
r
o
m
th
e
lab
eled
co
r
p
u
s
.
E
-
co
m
m
er
ce
item
s
h
a
v
e
b
ee
n
u
s
ed
to
v
alid
ate
th
e
p
r
o
p
o
s
ed
m
eth
o
d
[
9
]
,
[
1
0
]
.
Pap
er
is
o
r
g
an
ized
as
s
ec
tio
n
2
p
r
esen
ts
o
p
in
io
n
m
in
in
g
tech
n
iq
u
es
alo
n
g
with
id
en
tific
ati
o
n
o
f
h
ig
h
u
tili
ty
p
atter
n
s
f
o
r
d
ev
elo
p
in
g
tar
g
eted
ap
p
r
o
ac
h
f
o
r
r
eso
u
r
ce
o
p
tim
izatio
n
.
S
ec
tio
n
3
p
r
esen
ts
o
b
jectiv
es
id
en
tifie
d
f
r
o
m
r
esear
ch
g
ap
s
o
b
s
er
v
ed
in
liter
atu
r
e
s
u
r
v
ey
.
S
ec
tio
n
4
d
is
cu
s
s
es
p
r
o
p
o
s
ed
m
eth
o
d
o
l
o
g
y
em
p
lo
y
ed
to
ac
h
iev
e
th
e
o
b
j
ec
tiv
e
.
S
ec
tio
n
5
p
r
esen
ts
th
e
d
is
cu
s
s
io
n
an
d
th
e
r
esu
lts
ac
h
iev
ed
f
r
o
m
th
e
p
r
o
p
o
s
ed
alg
o
r
ith
m
an
d
f
in
all
y
s
ec
tio
n
6
p
r
esen
ts
th
e
co
n
clu
s
io
n
.
2.
L
I
T
E
R
AT
U
RE
SU
RVE
Y
A
tar
g
eted
ap
p
r
o
ac
h
f
o
r
in
v
es
tm
en
t
in
ar
ticle
m
an
u
f
ac
tu
r
in
g
ca
n
o
p
tim
ize
lo
t
o
f
r
eso
u
r
ce
allo
ca
tio
n
,
ef
f
o
r
t
an
d
tim
e.
Pr
o
p
o
s
ed
m
eth
o
d
o
lo
g
y
is
an
ef
f
o
r
t
to
id
en
tify
ar
ticles
o
f
h
ig
h
u
tili
ty
an
d
h
en
ce
f
o
cu
s
p
r
o
d
u
ctio
n
u
n
its
o
n
co
m
m
o
d
i
ties
o
f
lar
g
er
in
ter
est.
A
r
elate
d
s
tu
d
y
was
co
n
d
u
cted
to
in
v
esti
g
ate
asp
ec
ts
o
f
ar
ticles in
h
ig
h
d
em
a
n
d
a
n
d
f
u
r
th
er
to
n
ar
r
o
w
d
o
wn
th
e
r
an
g
e.
2
.
1
.
Asp
ec
t
-
ba
s
ed
s
ent
im
ent
a
na
ly
s
is
T
h
e
th
r
ee
m
ain
s
u
b
task
s
o
f
asp
ec
t
-
b
ased
s
en
tim
en
t
a
n
al
y
s
is
(
AB
SA
)
in
clu
d
e
asp
ec
t
-
s
en
tim
en
t
class
if
icatio
n
,
asp
ec
t
-
ter
m
ex
t
r
ac
tio
n
,
a
n
d
asp
ec
t
ca
teg
o
r
y
d
etec
tio
n
.
AB
SA
tar
g
ets
d
ee
p
s
en
tim
en
t
r
esear
ch
with
in
a
s
p
ec
if
ic
en
tity
[
1
1
]
.
I
m
p
licit
elem
en
ts
ca
n
b
e
d
if
f
icu
lt
to
id
e
n
tify
;
h
o
we
v
e
r
,
it
is
o
f
ex
tr
em
e
im
p
o
r
tan
ce
.
Un
f
o
r
tu
n
ately
,
lit
tle
co
n
s
id
er
atio
n
h
as
b
ee
n
g
i
v
en
in
r
ec
e
n
t
s
tu
d
ies
to
war
d
s
th
e
ex
tr
ac
tio
n
o
f
im
p
licit
elem
en
ts
.
Fro
m
v
ar
io
u
s
r
esear
ch
,
th
e
au
th
o
r
s
ex
tr
a
cted
th
e
d
ata
an
d
s
u
m
m
ar
ize
d
th
e
is
s
u
es
o
f
th
e
s
en
tim
en
t
an
aly
s
is
(
SA
)
class
if
icatio
n
tech
n
iq
u
es
[
1
2
]
.
R
esear
ch
th
at
v
iews
th
e
im
p
ac
t
o
f
d
if
f
er
e
n
t
p
r
o
d
u
ct
ca
teg
o
r
ies in
tr
o
d
u
ce
s
an
ap
p
r
o
ac
h
f
o
r
f
r
au
d
u
len
t r
e
v
iews id
en
tific
atio
n
u
s
in
g
th
e
AB
SA
m
eth
o
d
[
1
3
]
.
Ad
d
itio
n
ally
,
talk
i
n
g
-
h
ea
d
att
en
tio
n
a
n
d
a
m
o
d
el
was
p
r
o
p
o
s
ed
b
ased
o
n
a
lo
ca
l
c
o
n
t
ex
t
f
o
c
u
s
m
ec
h
an
is
m
,
wh
ic
h
im
p
r
o
v
es
SA
r
esu
lts
co
m
p
ar
e
d
to
ex
is
tin
g
b
aselin
e
m
o
d
els
[
1
4
]
.
A
h
eter
o
g
en
eo
u
s
asp
ec
t
g
r
ap
h
n
eu
r
al
n
etwo
r
k
(
HAGN
N)
,
wh
ich
ca
p
tu
r
es
r
elatio
n
s
h
ip
s
b
etwe
en
s
en
ten
ce
s
an
d
asp
ec
ts
to
im
p
r
o
v
e
m
o
d
el
p
er
f
o
r
m
a
n
ce
,
was
p
r
o
p
o
s
ed
[
1
5
]
.
T
h
e
p
er
f
o
r
m
an
ce
o
f
th
e
u
n
if
ied
f
r
am
ewo
r
k
is
s
u
b
-
o
p
tim
al.
A
m
o
d
el
ca
lled
r
esid
u
al
atten
tio
n
g
atin
g
–
th
r
ee
-
ch
an
n
el
g
r
ap
h
co
n
v
o
lu
tio
n
al
n
etwo
r
k
(
R
AG
-
T
C
GC
N
)
f
o
r
AB
SA
,
wh
ich
u
s
es
a
th
r
ee
-
ch
an
n
el
g
r
ap
h
co
n
v
o
lu
tio
n
al
n
etwo
r
k
to
lear
n
im
p
o
r
tan
t
in
f
o
r
m
atio
n
f
r
o
m
s
en
ten
ce
s
[
1
6
]
.
I
n
o
r
d
er
t
o
ass
ess
th
e
s
en
tim
en
t
p
o
lar
ity
o
f
ce
r
tain
asp
ec
ts
in
tex
t
r
ev
iews,
th
e
s
tu
d
y
s
u
g
g
ests
u
s
in
g
a
d
ep
en
d
e
n
cy
g
r
ap
h
co
n
v
o
lu
tio
n
al
n
etwo
r
k
m
o
d
el
f
o
r
AB
SA
[
1
7
]
.
2
.
2
.
H
i
g
h ut
ility
ba
s
ed
o
n c
o
nte
x
t
identif
ica
t
io
n
C
o
n
tex
t
-
b
ased
SA
h
as
b
ee
n
ap
p
r
o
ac
h
ed
in
a
v
a
r
iety
o
f
way
s
,
f
r
o
m
th
e
u
s
ag
e
o
f
co
n
tex
tu
alize
d
lex
ico
n
s
to
th
e
h
y
b
r
id
izatio
n
o
f
lex
ico
n
s
an
d
co
n
tex
t
-
awa
r
e
m
ac
h
in
e
lear
n
i
n
g
alg
o
r
ith
m
s
.
Dif
f
er
en
t
s
o
r
t
o
f
co
n
tex
t
th
at
is
lin
g
u
is
tic,
u
s
er
p
r
o
f
ile,
s
o
cial,
g
e
o
g
r
a
p
h
ical,
a
n
d
tem
p
o
r
al.
M
ay
b
e
em
p
lo
y
e
d
in
a
s
in
g
leto
n
as
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
5
,
No
.
4
,
Au
g
u
s
t 2
0
2
6
:
3
2
4
0
-
3
2
5
1
3242
well
as
in
co
m
b
in
atio
n
to
p
r
o
d
u
ce
b
etter
o
u
tco
m
es
[
1
8
]
.
I
n
o
r
d
er
to
g
et
co
n
tex
t
in
f
o
r
m
ati
o
n
f
o
r
t
h
e
s
en
tim
en
t
class
if
icatio
n
p
r
o
b
lem
,
a
u
n
iq
u
e
n
eu
r
al
n
etwo
r
k
-
b
ased
m
o
d
el
was p
u
t f
o
r
th
[
1
9
]
.
Acc
o
r
d
in
g
t
o
Har
i
et
a
l.
[
2
0
]
,
a
n
o
v
el
f
ea
tu
r
e
f
u
s
io
n
tec
h
n
iq
u
e
f
o
r
m
u
ltimo
d
al
SA
a
n
d
co
n
tex
t
m
o
d
ellin
g
im
p
r
o
v
ed
class
if
icatio
n
r
esu
lts
b
y
a
m
ar
g
in
o
f
1
-
2
p
er
ce
n
t
f
o
r
all
co
m
b
in
ati
o
n
s
o
f
m
o
d
alities
.
A
s
tu
d
y
th
at
em
p
lo
y
ed
th
e
T
wi
tter
r
an
k
alg
o
r
ith
m
to
i
d
en
tif
y
p
r
o
m
in
en
t
u
s
er
s
f
o
r
a
to
p
i
c
an
d
ex
tr
ac
te
d
th
e
em
o
tio
n
t
h
at
was
m
o
s
t
in
f
lu
en
tial
f
o
r
th
at
to
p
ic
[
2
1
]
,
[
2
2
]
f
o
c
u
s
es
o
n
r
ec
o
g
n
izin
g
f
ee
l
in
g
s
f
r
o
m
im
p
licit
p
h
r
ases
th
at
d
o
n
o
t
co
n
tain
em
o
tio
n
al
ter
m
s
.
Alg
o
r
ith
m
is
a
m
u
lti
-
f
ea
tu
r
e
n
eu
r
al
n
etwo
r
k
m
o
d
el
f
o
r
s
en
tim
en
t
r
ec
o
g
n
itio
n
in
im
p
licit
s
en
ten
ce
s
th
at
co
n
s
id
er
s
co
n
tex
t
u
al,
s
y
n
tactic,
an
d
s
em
an
tic
in
f
o
r
m
atio
n
,
am
o
n
g
o
th
er
f
ac
to
r
s
.
I
n
o
r
d
er
to
d
eter
m
i
n
e
th
e
s
en
tim
en
t
p
o
lar
ity
o
f
a
n
asp
ec
t
in
a
p
ar
ticu
lar
co
n
t
ex
t,
Z
h
u
et
a
l.
[
2
3
]
s
u
g
g
ests
u
s
in
g
a
co
n
tex
t
-
e
n
h
a
n
ce
d
atten
tio
n
n
etwo
r
k
(
C
E
-
AT
T
)
f
o
r
asp
ec
t
-
b
ased
s
en
tim
en
t
ca
teg
o
r
izatio
n
.
Hig
h
-
u
tili
ty
p
atter
n
an
aly
s
is
f
o
llo
win
g
th
e
in
itial
in
tr
o
d
u
cti
o
n
o
f
h
ig
h
u
tili
ty
item
s
et
m
in
i
n
g
(
HUI
M)
in
[
4
]
,
a
n
u
m
b
er
o
f
m
eth
o
d
s
,
in
clu
d
in
g
as
UM
in
in
g
[
5
]
,
i
n
c
r
em
en
tal
h
ig
h
u
tili
ty
p
atter
n
m
in
in
g
(
I
HUP
)
[
7
]
,
an
d
u
tili
ty
p
atter
n
-
g
r
o
wth
(
UP
-
Gr
o
wth
)
[
8
]
,
wer
e
p
u
t
f
o
r
th
.
Sin
ce
t
h
e
d
o
wn
war
d
-
clo
s
u
r
e
attr
ib
u
te
[
8
]
ca
n
n
o
t
b
e
u
s
ed
to
r
ed
u
ce
th
e
s
ea
r
ch
s
p
ac
e,
th
e
is
s
u
e
ass
o
ciate
d
with
HUI
M
is
s
ee
n
to
b
e
m
o
r
e
c
h
allen
g
in
g
t
h
an
th
e
is
s
u
e
o
f
f
r
eq
u
en
t
item
s
et
m
in
i
n
g
(
FIM
)
.
T
h
e
u
tili
ty
o
f
an
ite
m
s
et
is
n
eith
er
m
o
n
o
to
n
ic
n
o
r
an
ti
-
m
o
n
o
t
o
n
ic
in
HUI
M
,
wh
ich
is
th
e
r
ea
s
o
n
f
o
r
th
e
d
o
wn
war
d
clo
s
u
r
e
p
r
o
p
e
r
ty
'
s
n
ee
d
th
at
it
h
o
ld
s
.
T
h
e
tr
an
s
ac
tio
n
-
weig
h
ted
d
o
wn
war
d
clo
s
u
r
e
(
T
W
DC
)
f
ea
tu
r
e
was
p
r
esen
ted
in
[
2
4
]
as
a
s
o
lu
tio
n
t
o
th
is
p
r
o
b
le
m
,
an
d
m
o
s
t
o
f
th
e
s
u
b
s
eq
u
en
t
r
esear
ch
h
av
e
m
a
d
e
u
s
e
o
f
th
is
p
r
o
p
er
t
y
to
r
e
d
u
ce
th
e
s
ea
r
ch
s
p
ac
e
an
d
im
p
r
o
v
e
tim
e
ef
f
icien
c
y
.
T
ab
le
1
p
r
esen
ts
th
e
ev
o
lu
tio
n
o
f
o
p
in
io
n
m
in
in
g
tec
h
n
iq
u
es
o
v
er
th
e
y
ea
r
s
.
Ap
ar
t
f
r
o
m
d
esig
n
i
n
g
,
cr
ea
t
in
g
th
e
p
r
o
to
t
y
p
e,
g
ettin
g
r
aw
m
ater
ial,
f
ab
r
icatin
g
,
ad
v
er
tis
in
g
,
tr
an
s
p
o
r
tatio
n
an
d
s
ales,
an
d
m
ar
k
et
n
e
ed
also
p
lay
s
a
cr
u
cial
r
o
le.
W
ith
in
n
o
v
atio
n
s
in
m
ac
h
in
e
lear
n
in
g
alg
o
r
ith
m
s
,
s
ev
er
al
wo
r
k
s
wi
th
s
im
p
le
class
if
ier
s
s
u
ch
as
r
eg
r
ess
io
n
,
s
u
p
p
o
r
t
v
ec
to
r
m
ac
h
in
e
(
SVM)
,
a
n
d
n
aiv
e
B
ay
es
to
ad
v
a
n
ce
d
tech
n
iq
u
es
lik
e
r
an
d
o
m
f
o
r
est,
b
ag
g
in
g
,
b
o
o
s
tin
g
to
n
eu
r
al
n
etwo
r
k
s
h
a
v
e
b
ee
n
ap
p
lied
f
o
r
e
n
h
an
ce
d
r
esu
lts
.
Ho
wev
er
,
a
wid
e
g
ap
s
till
ex
is
ts
in
o
p
tim
izatio
n
o
f
r
eso
u
r
ce
s
ass
i
s
ted
b
y
o
p
in
io
n
m
in
in
g
an
d
u
tili
ty
-
b
ased
s
elec
tio
n
.
Fu
r
t
h
er
t
h
e
im
p
ac
t
o
f
u
tili
ty
p
atter
n
o
n
o
p
tim
izatio
n
o
f
r
eso
u
r
ce
allo
ca
tio
n
in
p
r
o
d
u
ct
life
cy
cle
is
s
till
a
p
o
ten
tial
ar
ea
wh
ich
is
u
n
ex
p
lo
r
ed
.
R
es
ea
r
ch
g
ap
s
id
en
tifie
d
ca
n
b
e
s
u
m
m
ar
ized
as f
o
llo
ws:
−
W
id
e
g
ap
ex
is
ts
d
u
e
to
lim
ited
wo
r
k
in
a
r
ea
o
f
o
p
tim
izatio
n
o
f
r
eso
u
r
ce
s
ass
is
ted
b
y
o
p
in
io
n
m
in
in
g
.
−
I
m
p
ac
t
o
f
u
tili
ty
p
atter
n
o
n
r
eso
u
r
ce
allo
ca
tio
n
in
p
r
o
d
u
ct
life
cy
cle
is
s
t
ill
a
p
o
ten
tial
ar
ea
wh
ich
is
u
n
ex
p
l
o
r
ed
.
−
Hy
b
r
id
r
ec
o
m
m
en
d
e
r
s
th
at
id
en
tify
p
r
o
d
u
cts
with
p
r
io
r
it
y
an
d
h
ig
h
u
tili
ty
p
atter
n
ca
n
h
av
e
m
ajo
r
f
in
an
cial
im
p
ac
t.
T
h
e
wo
r
k
p
r
o
p
o
s
ed
is
an
ef
f
o
r
t to
ad
d
r
ess
g
ap
s
id
en
tifie
d
,
u
s
e
o
f
u
tili
ty
p
atter
n
s
o
f
h
i
g
h
f
r
e
q
u
en
cy
p
r
o
d
u
cts in
tan
d
em
with
class
if
icatio
n
f
o
r
ef
f
ec
tiv
e
r
eso
u
r
ce
u
tili
za
tio
n
a
n
d
h
en
ce
f
o
c
u
s
p
r
o
d
u
ctio
n
u
n
i
ts
o
n
co
m
m
o
d
ities
o
f
lar
g
er
in
ter
est.
T
ab
le
1
.
Su
m
m
a
r
y
o
f
th
e
e
v
o
l
u
tio
n
o
f
o
p
in
i
o
n
m
in
i
n
g
tech
n
iq
u
es
S
.
n
o
C
a
t
e
g
o
r
y
R
e
f
e
r
e
n
c
e
s
M
e
t
h
o
d
o
l
o
g
y
1
H
y
b
r
i
d
mo
d
e
l
[
2
5
]
–
[
3
0
]
H
y
b
r
i
d
b
i
d
i
r
e
c
t
i
o
n
a
l
g
a
t
e
d
r
e
c
u
r
r
e
n
t
u
n
i
t
(
B
G
R
U
)
/
b
i
d
i
r
e
c
t
i
o
n
a
l
l
o
n
g
s
h
o
r
t
-
t
e
r
m m
e
m
o
r
y
(
B
i
LS
TM
)
-
g
a
t
e
d
r
e
c
u
r
r
e
n
t
u
n
i
t
(
G
R
U
)
mo
d
e
l
s,
B
ER
T
-
G
R
U
2
H
i
g
h
u
t
i
l
i
t
y
m
i
n
i
n
g
[
5
]
–
[
8
]
,
[
2
3
]
,
[
3
1
]
–
[
3
4
]
U
M
i
n
i
n
g
,
I
H
U
P
,
UP
-
G
r
o
w
t
h
,
TWD
C
,
m
o
d
i
f
i
e
d
e
f
f
i
c
i
e
n
t
f
r
e
q
u
e
n
t
(
M
EF
)
-
H
U
I
M
,
i
n
c
r
e
m
e
n
t
a
l
h
i
g
h
a
v
e
r
a
g
e
-
u
t
i
l
i
t
y
i
t
e
mse
t
mi
n
i
n
g
(
I
n
c
H
A
U
I
M
)
,
EH
N
L,
h
i
g
h
a
v
e
r
a
g
e
-
u
t
i
l
i
t
y
i
t
e
ms
e
t
(
H
A
U
I
)
-
M
i
n
e
r
+
+
,
e
f
f
i
c
i
e
n
t
h
i
g
h
-
u
t
i
l
i
t
y
i
t
e
mset
mi
n
i
n
g
(
EFI
M
)
FP
-
t
r
e
e
p
r
o
j
e
c
t
i
o
n
3
H
y
b
r
i
d
n
a
t
u
r
e
i
n
s
p
i
r
e
d
a
l
g
o
r
i
t
h
ms
[
3
5
]
–
[
4
0
]
H
y
b
r
i
d
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
s
(
C
N
N
s)
-
l
o
n
g
s
h
o
r
t
-
t
e
r
m
mem
o
r
y
(
LSTM
)
-
S
V
M
,
p
a
r
t
i
c
l
e
sw
a
r
m
o
p
t
i
m
i
z
a
t
i
o
n
(
PSO
)
S
V
M
,
n
a
i
v
e
B
a
y
e
s
-
P
S
O
,
H
U
I
M
+
P
S
O
+
C
S
A
4
A
sp
e
c
t
s
e
n
t
i
m
e
n
t
c
l
a
ss
i
f
i
c
a
t
i
o
n
[
1
5
]
–
[
1
7
]
,
[
4
1
]
–
[
4
4
]
HAGNN
,
R
A
G
-
TC
G
C
N
,
d
e
p
e
n
d
e
n
c
y
g
r
a
p
h
C
N
N
,
g
a
t
e
d
r
e
c
u
r
r
e
n
t
u
n
i
t
s,
k
n
o
w
l
e
d
g
e
-
a
w
a
r
e
d
e
p
e
n
d
e
n
c
y
g
r
a
p
h
n
e
t
w
o
r
k
,
p
r
o
mp
t
k
n
o
w
l
e
d
g
e
t
u
n
i
n
g
,
r
e
t
r
i
e
v
a
l
c
o
n
t
r
a
st
i
v
e
l
e
a
r
n
i
n
g
5
C
o
n
t
e
x
t
b
a
s
e
d
[
1
9
]
,
[
2
3
]
,
[
4
5
]
–
[
4
7
]
N
e
u
r
a
l
n
e
t
w
o
r
k
m
o
d
e
l
,
CE
-
A
TT
,
c
o
n
t
e
x
t
-
se
n
si
t
i
v
e
m
u
l
t
i
-
t
i
e
r
d
e
e
p
l
e
a
r
n
i
n
g
f
r
a
mew
o
r
k
,
S
V
M
w
i
t
h
g
r
i
d
s
e
a
r
c
h
,
t
r
a
n
sf
o
r
mer
m
o
d
e
l
s
R
o
B
ER
T
D
i
st
i
l
B
ER
T
6
C
o
l
l
a
b
o
r
a
t
i
v
e
f
i
l
t
e
r
i
n
g
w
i
t
h
u
t
i
l
i
t
y
mi
n
i
n
g
[
4
8
]
–
[
5
0
]
A
l
t
e
r
n
a
t
i
n
g
l
e
a
st
sq
u
a
r
e
(
A
LS)
w
i
t
h
a
t
t
e
n
t
i
o
n
m
e
c
h
a
n
i
s
ms
a
n
d
Bi
LST
M
,
G
l
o
V
e
+
B
i
LST
M
,
B
a
y
e
si
a
n
h
y
b
r
i
d
c
o
l
l
a
b
o
r
a
t
i
v
e
f
i
l
t
e
r
i
n
g
(
B
H
C
F
)
7
H
i
g
h
u
t
i
l
i
t
y
w
i
t
h
n
e
g
a
t
i
v
e
u
n
i
t
p
r
o
f
i
t
s
[
5
1
]
–
[
5
3
]
F
H
N
,
g
l
o
b
a
l
h
i
g
h
-
u
t
i
l
i
t
y
mi
n
i
n
g
(
G
H
U
M
)
,
t
o
p
-
k
h
i
g
h
u
t
i
l
i
t
y
i
t
e
ms
e
t
w
i
t
h
p
o
s
i
t
i
v
e
o
r
n
e
g
a
t
i
v
e
p
r
o
f
i
t
s
(
T
K
N
)
3.
O
B
J
E
CT
I
V
E
S
Ad
d
r
ess
in
g
th
e
id
en
tifie
d
lim
itatio
n
s
,
th
is
wo
r
k
aim
s
to
d
esig
n
a
h
y
b
r
id
al
g
o
r
ith
m
t
h
at
in
teg
r
ates
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
R
eso
u
r
ce
o
p
timiz
a
tio
n
via
id
e
n
tifi
ca
tio
n
o
f a
r
ticles w
ith
a
s
p
ec
ts
o
f p
r
io
r
ity
…
(
Ga
r
ima
S
r
i
va
s
ta
va
)
3243
H
U
P
M
wi
th
co
n
tex
t
-
awa
r
e
class
if
icatio
n
to
en
ab
le
ta
r
g
eted
in
v
estme
n
t
d
ec
is
io
n
s
in
m
an
u
f
ac
tu
r
i
n
g
.
B
ased
o
n
th
e
r
esear
ch
g
ap
s
,
th
e
o
b
jectiv
es o
f
th
e
p
r
o
p
o
s
ed
w
o
r
k
ar
e
d
ef
in
e
d
as f
o
llo
ws
:
−
T
o
d
esig
n
a
n
alg
o
r
ith
m
f
o
r
tar
g
eted
in
v
estme
n
t
b
y
co
m
b
in
in
g
u
tili
ty
m
in
in
g
an
d
o
p
in
io
n
-
b
ased
class
if
icatio
n
,
ac
h
iev
in
g
g
r
ea
t
er
th
an
1
0
–
1
5
% r
ed
u
ctio
n
in
f
alse
-
p
o
s
itiv
e
p
r
o
d
u
ct
s
elec
tio
n
s
.
−
Use
u
tili
ty
-
p
atter
n
an
aly
s
is
to
g
u
id
e
p
r
o
d
u
ct
life
-
c
y
cle
r
eso
u
r
ce
allo
ca
tio
n
b
y
en
f
o
r
c
in
g
m
in
im
u
m
u
tili
ty
th
r
esh
o
ld
s
.
−
I
m
p
lem
en
t
a
h
y
b
r
id
r
ec
o
m
m
en
d
er
th
at
in
teg
r
ates
r
atin
g
s
,
r
ev
iews,
an
d
b
eh
a
v
io
r
al
d
at
a
to
ac
h
iev
e
im
p
r
o
v
em
e
n
t in
r
ec
o
m
m
en
d
at
io
n
p
r
ec
is
io
n
a
n
d
s
tr
o
n
g
er
r
ec
all
o
f
h
ig
h
-
u
tili
ty
p
r
o
d
u
cts
.
4.
P
RO
P
O
SE
D
M
E
T
H
O
D
Pro
p
o
s
ed
wo
r
k
em
p
lo
y
s
a
h
y
b
r
id
ap
p
r
o
ac
h
o
f
class
if
icatio
n
an
d
h
ig
h
u
tili
ty
p
atter
n
to
id
en
tify
th
e
d
o
m
in
an
t
asp
ec
ts
o
f
ar
ticles
with
h
ig
h
s
ale
p
r
o
b
ab
ilit
y
.
Fi
g
u
r
e
1
d
ep
icts
th
e
f
lo
w
g
r
ap
h
d
esig
n
o
f
t
h
r
ee
s
tep
p
r
o
p
o
s
ed
m
o
d
el,
f
lo
w
g
r
ap
h
f
u
n
ctio
n
s
in
th
r
ee
lay
er
s
,
f
ir
s
t
l
ay
er
ac
q
u
i
r
es
s
am
p
les
an
d
n
o
r
m
alize
s
th
e
d
ataset
f
o
r
o
p
tim
al
ex
ec
u
tio
n
,
s
ec
o
n
d
s
tep
co
m
p
r
is
es
o
f
id
en
tific
ati
o
n
an
d
class
if
icatio
n
o
f
s
am
p
l
es
with
co
n
tex
t
as
p
r
o
m
in
e
n
t
p
ar
a
m
eter
,
v
a
n
i
l
la
f
ee
d
f
o
r
war
d
n
e
u
r
al
n
etwo
r
k
(
V
NN)
with
d
en
s
e
co
n
n
ec
tio
n
s
i
s
em
p
lo
y
ed
f
o
r
th
e
class
if
icatio
n
an
d
in
t
h
ir
d
,
ar
t
icles
id
en
tifie
d
f
r
o
m
s
ec
o
n
d
s
tep
is
s
u
b
jecte
d
to
u
tili
ty
p
a
tter
n
m
in
in
g
u
s
in
g
tr
ip
les
f
o
r
ex
tr
ac
tin
g
c
o
m
m
o
d
ities
an
d
asp
ec
t
o
f
h
ig
h
d
e
m
a
n
d
.
E
-
c
o
m
m
er
ce
item
s
h
av
e
b
ee
n
u
s
ed
to
v
alid
ate
th
e
p
r
o
p
o
s
ed
m
eth
o
d
.
T
h
e
f
o
llo
win
g
s
tep
s
ar
e
em
p
lo
y
ed
f
o
r
th
e
task
.
Fig
u
r
e
1
.
Flo
w
g
r
ap
h
d
esig
n
o
f
two
-
s
tep
p
r
o
p
o
s
ed
m
o
d
el
4
.
1
.
Da
t
a
s
et
T
h
e
p
r
o
d
u
ct.
js
o
n
f
ile,
is
u
s
ed
f
o
r
th
e
s
u
g
g
ested
wo
r
k
,
o
b
ta
in
ed
f
r
o
m
GitHu
b
.
Sam
p
le
ac
q
u
ir
ed
h
as
th
e
p
r
o
d
u
ct
n
a
m
es,
d
escr
ip
tio
n
s
,
an
d
ca
teg
o
r
ies.
T
h
e
p
r
o
d
u
ct.
js
o
n
f
ile
co
n
tain
s
a
to
tal
o
f
5
1
,
6
4
6
p
r
o
d
u
cts
lis
ted
with
d
ataset
co
n
tain
s
m
u
lticlas
s
ca
teg
o
r
ies.
T
h
e
s
u
g
g
ested
tech
n
iq
u
e
ca
n
id
e
n
tify
an
d
f
o
r
ec
ast
1
,
8
0
2
class
es
s
in
ce
th
er
e
ar
e
1
,
8
0
2
n
o
n
-
o
v
er
lap
p
i
n
g
class
es.
T
h
e
d
ataset
th
at
is
b
ein
g
s
tu
d
ied
is
s
h
o
wn
i
n
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
5
,
No
.
4
,
Au
g
u
s
t 2
0
2
6
:
3
2
4
0
-
3
2
5
1
3244
F
ig
u
r
e
2
,
with
th
e
f
r
eq
u
en
cy
o
f
o
cc
u
r
r
en
ce
o
f
item
s
,
ca
teg
o
r
y
co
u
n
t,
an
d
a
h
ig
h
lig
h
ted
r
ed
lin
e
th
at
s
ep
ar
ates
all
ca
teg
o
r
ies
with
a
co
u
n
t
o
f
m
o
r
e
th
an
5
0
0
f
r
o
m
th
o
s
e
wit
h
o
u
t
[
5
4
]
.
Fig
u
r
e
2
.
Fre
q
u
en
cy
o
f
o
cc
u
r
r
en
ce
o
f
item
s
4
.
2
.
Dee
p neura
l net
wo
rk
a
s
s
is
t
ed
prio
rit
y
co
nte
x
t
identif
ica
t
io
n
T
o
ad
ap
t
m
ac
h
in
e
alg
o
r
ith
m
s
f
o
r
p
r
ed
ictio
n
,
th
e
d
ataset
u
n
d
er
in
v
esti
g
atio
n
is
d
iv
id
ed
,
to
k
en
ized
,
p
r
ep
r
o
ce
s
s
ed
,
an
d
tr
an
s
f
o
r
m
e
d
in
to
ter
m
f
r
eq
u
en
cy
–
in
v
er
s
e
d
o
cu
m
en
t
f
r
eq
u
e
n
cy
(
TF
-
I
DF
)
.
C
NNs
wo
r
k
o
n
s
p
atial
p
atter
n
ex
tr
ac
tio
n
an
d
L
STM
s
ca
p
tu
r
es
s
eq
u
en
ti
al
d
ep
en
d
e
n
cies
wh
ile
VNNs
ac
q
u
ir
e
a
laten
t
d
is
tr
ib
u
tio
n
en
ab
lin
g
b
etter
g
en
er
aliza
tio
n
,
r
e
d
u
ce
d
o
v
er
f
itti
n
g
,
an
d
im
p
r
o
v
es
in
ter
o
p
er
ab
ilit
y
th
r
o
u
g
h
p
r
o
b
a
b
ilis
tic
o
u
tp
u
ts
,
s
u
p
p
o
r
ts
g
en
er
ativ
e
ca
p
ab
ilit
y
a
n
d
w
o
r
k
b
etter
with
s
m
aller
d
atasets
allo
win
g
th
em
t
o
p
r
ed
ict
m
is
s
in
g
d
ata,
cr
ea
te
s
y
n
th
etic
s
am
p
les,
an
d
d
is
co
v
er
d
ee
p
er
h
id
d
en
p
atter
n
s
.
T
a
s
k
s
wh
ich
in
v
o
lv
e
u
n
ce
r
tain
ty
m
o
d
elin
g
,
n
o
is
e
t
o
ler
an
ce
,
f
ea
t
u
r
e
d
is
co
v
e
r
y
,
a
n
d
p
r
o
b
ab
ilis
tic
d
ec
is
io
n
-
m
ak
in
g
,
VNNs
o
u
tclass
C
NNs
an
d
L
STM
s
.
Owin
g
to
r
ea
s
o
n
s
m
en
tio
n
e
d
a
b
o
v
e
V
NN
was
u
s
ed
in
th
e
d
ee
p
lear
n
i
n
g
m
o
d
el,
w
h
ich
is
tr
ain
ed
f
o
r
3
0
ep
o
c
h
s
an
d
p
l
o
ts
th
e
lo
s
s
es
an
d
ar
ea
u
n
d
er
th
e
cu
r
v
e
(
AUC
)
.
Fig
u
r
e
3
d
is
p
lay
s
th
e
VNN
class
if
icatio
n
p
ar
am
eter
s
f
o
r
t
h
e
p
r
o
p
o
s
ed
m
o
d
el.
Fig
u
r
e
3
.
VNN
class
if
icatio
n
p
ar
am
eter
s
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
R
eso
u
r
ce
o
p
timiz
a
tio
n
via
id
e
n
tifi
ca
tio
n
o
f a
r
ticles w
ith
a
s
p
ec
ts
o
f p
r
io
r
ity
…
(
Ga
r
ima
S
r
i
va
s
ta
va
)
3245
VNN
em
p
lo
y
ed
is
with
d
en
s
e
co
n
n
ec
tio
n
s
f
o
r
class
if
icatio
n
[
5
5
]
.
A
d
e
n
s
ely
lin
k
ed
n
etwo
r
k
em
p
lo
y
s
f
ea
tu
r
e
r
eu
s
e;
it
p
er
m
its
an
o
u
tp
u
t
at
o
n
e
la
y
er
to
b
e
r
eu
s
ed
s
ev
er
al
tim
es
at
s
u
cc
ee
d
in
g
lev
els
[
5
5
]
,
[
5
6
]
.
T
h
e
q
u
ality
o
f
em
b
ed
d
in
g
p
air
s
co
llected
ac
r
o
s
s
th
e
d
ata
s
et
is
ac
cu
m
u
lated
to
r
ef
lect
th
is
aim
in
a
lo
s
s
f
u
n
ctio
n
£
g
iv
en
b
y
(
1
)
.
Den
o
t
in
g
th
e
q
u
ality
o
f
a
p
ar
ticu
lar
p
air
o
f
e
m
b
ed
d
in
g
s
with
ij,
th
e
o
v
er
all
lo
s
s
is
d
ef
in
ed
as
(
1
)
.
£
=
∑
∑
(
)
(
1
)
W
o
r
k
p
r
o
p
o
s
ed
esti
m
ates
b
y
esti
m
atin
g
alg
eb
r
aic
s
u
m
s
f
o
r
all
d
escr
ip
to
r
s
,
th
e
alg
eb
r
aic
s
u
m
m
ay
g
et
escalate
d
f
r
o
m
er
r
o
r
s
g
en
e
r
ated
d
u
e
to
u
n
c
o
r
r
elate
d
d
escr
i
p
to
r
s
th
at
ca
n
in
f
lu
en
ce
th
e
e
s
tim
ated
s
im
ilar
ity
,
ev
en
if
er
r
o
r
g
en
er
ate
d
h
as
a
n
o
n
-
s
ig
n
if
ica
n
t
in
d
iv
id
u
al
s
im
ilar
ity
.
W
eig
h
ts
is
s
o
tak
en
th
at
it
eq
u
alize
s
th
e
ef
f
ec
t o
f
e
v
er
y
e
n
tity
to
th
e
s
i
m
ilar
ity
esti
m
ates.
T
h
e
cu
m
u
l
ativ
e
r
ep
r
esen
tatio
n
ca
n
b
e
r
e
p
r
esen
ted
in
(
2
)
.
ζ
=
∑
α
(
x
)
(
)
=
α
(
2
)
T
h
e
p
r
o
ce
d
u
r
e
to
s
o
lv
e
a
n
o
p
t
im
izatio
n
p
r
o
b
lem
to
esti
m
ate
th
e
weig
h
ts
α
is
k
n
o
wn
as
g
e
n
er
alize
d
m
ax
p
o
o
lin
g
.
W
o
r
k
p
r
o
p
o
s
ed
ev
alu
ates
alg
eb
r
aic
s
u
m
s
f
o
r
all
d
escr
ip
to
r
s
,
th
e
alg
e
b
r
aic
s
u
m
m
ay
g
et
escalate
d
f
r
o
m
er
r
o
r
s
g
en
e
r
ated
d
u
e
to
u
n
c
o
r
r
elate
d
d
escr
i
p
to
r
s
th
at
ca
n
in
f
lu
en
ce
th
e
e
s
tim
ated
s
im
ilar
ity
,
ev
en
if
e
r
r
o
r
g
en
er
ated
h
as
a
n
o
n
-
s
ig
n
if
ican
t
i
n
d
iv
id
u
al
s
im
ilar
ity
.
Fig
u
r
e
4
d
e
p
icts
th
e
lo
s
s
es
in
cu
r
r
ed
f
o
r
t
h
e
p
r
o
p
o
s
ed
m
o
d
el,
wh
ich
ar
e
ap
p
r
o
x
im
ately
0
.
0
4
f
o
r
v
alid
atio
n
an
d
less
th
an
0
.
0
1
f
o
r
tr
ain
i
n
g
.
Fig
u
r
e
5
d
ep
icts
th
e
AUC,
th
e
c
u
r
v
e
p
r
o
v
id
es
th
e
p
r
ec
is
io
n
with
wh
ich
th
e
alg
o
r
ith
m
ca
n
id
e
n
tify
d
iv
er
s
if
ied
class
es,
it
is
ap
p
r
o
x
im
ately
0
.
9
7
f
o
r
v
alid
a
tio
n
an
d
0
.
9
9
f
o
r
tr
ain
in
g
,
r
esu
lts
ac
h
iev
ed
s
h
o
w
th
at
co
r
p
u
s
h
as
b
ee
n
lab
elled
with
ac
ce
p
tab
le
er
r
o
r
r
atio
[
5
7
]
,
[
5
8
]
.
Fig
u
r
e
4
.
L
o
s
s
es in
cu
r
r
ed
f
o
r
t
h
e
p
r
o
p
o
s
ed
m
o
d
el
Fig
u
r
e
5
.
AUC r
ep
r
esen
ts
th
e
p
r
ec
is
io
n
with
wh
ich
th
e
alg
o
r
ith
m
ca
n
id
e
n
tify
d
iv
er
s
if
ied
class
es
4
.
3
.
P
a
t
t
er
n
m
ini
ng
f
o
r
hig
h ut
ility
VNN
p
r
o
v
id
es
a
p
r
o
ce
s
s
ed
d
a
taset
wh
er
e
co
m
m
o
d
ities
with
co
n
tex
t
h
av
in
g
h
ig
h
p
r
io
r
ity
h
av
e
b
ee
n
lab
elled
,
co
m
m
o
d
ities
ex
tr
ac
t
ed
f
r
o
m
s
tag
e
o
n
e
will
b
e
s
u
b
j
ec
ted
to
th
e
m
in
in
g
p
r
o
ce
s
s
to
id
en
tify
ar
ticles
o
f
h
ig
h
u
tili
ty
.
W
o
r
k
f
ir
s
t
estab
lis
h
es
th
e
p
r
im
ar
y
en
titi
es
o
f
HUPM
.
L
et
I
b
e
f
in
ite
clas
s
o
f
u
n
iq
u
e
item
s
an
d
D
g
iv
es
g
r
o
u
p
o
f
tr
a
n
s
ac
tio
n
s
in
a
d
atab
ase.
I
t
is
d
is
tin
g
u
is
h
ed
b
y
th
e
tr
a
n
s
ac
tio
n
id
f
o
r
ea
ch
tr
an
s
ac
tio
n
,
.
C
o
m
m
o
n
ly
,
th
e
f
o
r
m
er
h
as
an
in
ter
n
al
u
tili
ty
,
(
,
)
,
an
d
th
e
latter
h
as
ex
ter
io
r
u
tili
ty
,
(
)
.
(
)
r
ep
r
esen
ts
th
e
u
tili
ty
v
alu
e
o
f
th
e
ar
ticle
X
ac
r
o
s
s
d
atab
ase
D,
wh
ich
is
esti
m
ated
b
y
s
u
m
m
in
g
to
g
eth
e
r
all
th
e
co
m
p
o
n
en
ts
o
f
(
,
)
s
o
th
at
T
k
h
as X
as m
em
b
er
,
g
iv
en
in
(
3
)
.
(
,
)
=
(
)
×
(
,
)
(
3
)
Hig
h
u
tili
ty
p
atter
n
ex
tr
ac
tio
n
s
ee
k
s
to
id
en
tify
all
h
ig
h
u
til
ity
ar
ticles
in
th
e
d
atab
ase
D
th
at
h
av
e
u
tili
ty
lar
g
er
th
an
o
r
eq
u
al
t
o
m
in
im
u
m
u
tili
ty
,
g
i
v
en
th
e
d
atab
ase
D
an
d
th
e
m
in
im
al
u
tili
ty
cr
iter
io
n
.
Asp
ec
t
-
b
ased
o
p
in
io
n
m
in
in
g
co
n
s
tr
u
cts
a
s
et
h
av
in
g
tr
ip
l
es:
(
ai,
s
j,
s
c)
.
Her
e,
s
j
r
ef
er
s
to
th
e
ass
o
ciate
d
s
en
tim
en
t
f
o
r
asp
ec
t
ai
an
d
s
c
is
th
e
s
en
tim
en
t
s
co
r
e
.
T
h
e
p
r
o
ce
d
u
r
e
to
e
x
tr
ac
t
tr
ip
les
(
ai,
s
j,
s
c)
is
g
iv
e
n
in
Alg
o
r
ith
m
1
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
5
,
No
.
4
,
Au
g
u
s
t 2
0
2
6
:
3
2
4
0
-
3
2
5
1
3246
Alg
o
r
ith
m
1
.
Hig
h
p
r
io
r
ity
p
ar
am
eter
ex
tr
ac
tio
n
al
g
o
r
ith
m
I
n
p
u
t:
co
r
p
u
s
Ou
tp
u
t:
h
ig
h
p
r
io
r
ity
p
ar
am
ete
r
ex
tr
ac
tio
n
Step
1
:
asp
ec
t
ass
is
ted
o
p
in
io
n
m
in
in
g
.
Step
2
: Q
←
C
o
r
p
u
s
Step
3
: E
←
β (
Nu
ll Set)
Step
4
: f
o
r
τ
ε
Q
Step
5
: Q
; E
←
Acq
u
ir
e
T
r
i
p
les(τ)
Step
6
: E
←
E
Ụ
E
Q
#
R
ev
iew
+
asp
ec
t +
s
co
r
e
→
E
Step
7
: E
n
d
Step
8
: CT
←
C
lu
s
ter
T
r
ip
les(
E
)
#
C
lu
s
ter
o
f
tr
ip
les =
r
ev
iew
+
asp
ec
t
Step
9
: D
_
s
et
←
C
T
#
Data
b
ase
o
f
tr
ip
les
to
tr
a
n
s
ac
tio
n
s
Step
1
0
: M
UA
←I
m
p
lem
en
tF
HN_
Alg
o
r
ith
m
(
D_
s
et)
#
Ma
x
im
u
m
u
tili
ty
asp
ec
t
Sam
p
le
r
ev
iews
ar
e
in
clu
d
ed
i
n
Fig
u
r
e
6
f
r
o
m
w
h
ich
tr
ip
les
ar
e
to
b
e
ex
tr
ac
te
d
.
T
o
e
x
tr
ac
t
tr
ip
les
in
a
p
ar
ticu
lar
r
ev
iew,
th
e
wo
r
k
p
r
o
p
o
s
ed
f
o
llo
wed
th
e
p
r
o
ce
d
u
r
e
o
u
tlin
ed
in
[
1
3
]
,
[
1
4
]
,
wh
ich
is
f
u
r
th
er
ex
p
lain
ed
as
f
o
llo
ws:
i)
C
o
r
p
u
s
is
d
iv
id
ed
d
o
wn
in
to
s
en
ten
ce
s
.
ii)
NL
P to
k
en
izer
is
em
p
lo
y
e
d
f
o
r
th
e
p
r
o
ce
s
s
as f
o
llo
ws
:
−
An
in
d
iv
id
u
al
asp
ec
t
is
lab
ele
d
f
o
r
ev
e
r
y
n
o
u
n
an
d
n
o
u
n
p
h
r
ase.
I
n
a
d
d
itio
n
,
ev
er
y
wo
r
d
i
n
a
n
o
u
n
co
r
p
u
s
is
co
n
v
er
ted
to
its
lem
m
atize
d
s
h
ap
e,
an
d
all
lem
m
atize
d
wo
r
d
f
o
r
m
s
ar
e
co
n
n
ec
te
d
to
g
eth
er
s
o
th
at
in
f
lectio
n
al
v
ar
ian
ts
o
f
n
o
u
n
s
an
d
n
o
u
n
c
o
r
p
u
s
ca
n
b
e
co
m
b
in
ed
i
n
s
u
b
s
eq
u
en
t
s
tag
es.
−
Sen
tim
en
t le
x
ico
n
s
ar
e
u
s
ed
t
o
an
n
o
tate
wo
r
d
s
with
em
o
tio
n
s
.
−
Sen
tim
en
t
wo
r
d
s
,
an
n
o
tated
n
o
u
n
s
,
an
d
n
o
u
n
p
h
r
ases
ar
e
ar
r
an
g
ed
in
th
e
s
en
ten
ce
'
s
o
r
d
er
o
f
o
cc
u
r
r
e
n
ce
.
−
E
v
er
y
p
air
o
f
n
o
u
n
s
an
d
s
en
ti
m
en
t w
o
r
d
s
th
at
c
o
-
o
cc
u
r
m
ak
es a
n
ew
tr
ip
le.
iii)
T
h
e
t
r
ip
le
s
ets u
n
io
n
is
ac
q
u
ir
ed
,
wh
ich
is
f
o
r
m
ed
f
o
r
e
v
er
y
r
ev
iew
s
en
ten
ce
.
Fig
u
r
e
7
p
r
o
v
id
es
co
r
r
esp
o
n
d
i
n
g
tr
i
p
les
th
at
ar
e
r
etr
ie
v
ed
u
s
in
g
th
e
p
r
e
v
io
u
s
ly
d
escr
ib
ed
p
r
o
ce
d
u
r
e.
E
x
tr
ac
ted
tr
ip
les
ar
e
co
n
v
er
te
d
in
to
tr
a
n
s
ac
tio
n
s
with
u
tili
ty
v
alu
es
in
t
h
is
s
tep
.
I
tem
s
h
av
e
b
o
th
in
ter
n
al
a
n
d
ex
ter
io
r
u
s
ef
u
ln
ess
in
u
tili
ty
-
b
ased
p
atter
n
m
in
in
g
.
Fig
u
r
e
6
.
Sam
p
le
r
ev
iews
Fig
u
r
e
7
.
C
o
r
r
esp
o
n
d
i
n
g
tr
ip
le
s
A
r
ev
iew
a
n
d
a
tr
an
s
ac
tio
n
ar
e
r
elate
d
,
with
elem
e
n
ts
f
r
o
m
a
r
ev
iew
ta
k
en
f
r
o
m
th
e
f
ir
s
t
s
tep
,
m
atch
in
g
th
in
g
s
f
r
o
m
a
tr
an
s
ac
tio
n
in
th
e
th
ir
d
.
Fo
r
ev
e
r
y
o
b
ject,
th
e
wo
r
k
p
r
o
p
o
s
es
to
g
iv
e
it
an
ex
ter
n
al
u
tili
ty
.
T
h
is
esti
m
ate,
wh
ich
is
d
o
m
ain
-
d
e
p
en
d
e
n
t
an
d
n
o
t
i
n
clu
d
ed
in
th
e
r
ev
iew
d
ata,
m
ay
b
e
a
p
r
o
d
u
ce
r
'
s
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
R
eso
u
r
ce
o
p
timiz
a
tio
n
via
id
e
n
tifi
ca
tio
n
o
f a
r
ticles w
ith
a
s
p
ec
ts
o
f p
r
io
r
ity
…
(
Ga
r
ima
S
r
i
va
s
ta
va
)
3247
ch
o
ice
b
ec
au
s
e
o
f
ch
ea
p
m
an
u
f
ac
tu
r
in
g
co
s
ts
,
o
r
it
co
u
ld
b
e
a
cu
s
to
m
er
'
s
p
r
ef
er
en
ce
.
T
ab
le
2
d
ep
icts
th
e
tr
an
s
f
o
r
m
atio
n
m
o
d
el
u
s
ed
f
o
r
tr
an
s
f
o
r
m
atio
n
o
f
t
r
ip
les to
tr
an
s
ac
tio
n
s
.
T
ab
le
2
.
T
r
a
n
s
f
o
r
m
atio
n
m
o
d
e
l
U
t
i
l
i
t
y
p
a
t
t
e
r
n
mi
n
i
n
g
A
sp
e
c
t
a
ssi
s
t
e
d
o
p
i
n
i
o
n
mi
n
i
n
g
I
t
e
m (I
i
)
A
sp
e
c
t
Tr
a
n
s
a
c
t
i
o
n
(
Ti
)
R
e
v
i
e
w
U
t
i
l
i
t
y
(
U
i
)
S
e
n
t
i
me
n
t
p
r
o
d
u
c
t
v
a
l
u
e
X
u
t
i
l
i
t
y
s
c
o
r
e
Ex
t
e
r
n
a
l
(
U
i
)
U
t
i
l
i
t
y
s
c
o
r
e
I
n
t
e
r
n
a
l
(
U
i
)
S
e
n
t
i
me
n
t
p
r
o
d
u
c
t
r
e
v
i
e
w
Pro
p
o
s
ed
m
o
d
el
p
e
r
f
o
r
m
s
two
p
r
o
ce
d
u
r
es
o
n
th
e
ex
tr
ac
ted
asp
ec
t
to
co
m
p
lete
th
e
tr
an
s
f
o
r
m
atio
n
.
T
r
ip
les
ar
e
ca
teg
o
r
ized
b
y
r
e
v
iew
id
in
th
e
f
ir
s
t
o
n
e
.
A
r
e
v
iew
d
ep
en
d
in
g
u
p
o
n
is
s
u
e
a
d
d
r
ess
ed
is
lik
ely
to
h
av
e
m
o
r
e
th
an
o
n
e
tr
ip
le
w
ith
th
e
s
am
e
asp
ec
t
.
T
h
e
s
en
tim
en
t
s
co
r
e
o
f
th
e
elem
e
n
ts
th
at
h
av
e
s
ev
er
al
o
cc
u
r
r
e
n
ce
s
ar
e
r
eso
lv
ed
b
y
ad
d
in
g
u
p
ea
ch
s
en
tim
en
t
s
co
r
e
s
ep
ar
ately
o
r
ca
lcu
latin
g
th
eir
av
e
r
ag
e.
I
n
th
e
f
o
r
m
er
ca
s
e,
it
is
ass
u
m
ed
th
at
r
ep
ea
ted
m
en
tio
n
s
o
f
a
n
elem
en
t
in
cr
ea
s
e
its
in
f
lu
en
ce
o
r
p
r
ef
er
en
ce
(
i.e
.
,
p
o
s
itiv
e
o
r
n
eg
ativ
e)
.
T
h
e
ar
r
a
y
o
f
tr
ip
les
with
th
e
id
en
tical
r
ev
iew
I
D
is
ar
r
an
g
e
d
in
to
a
tr
an
s
ac
tio
n
d
u
r
in
g
th
e
n
ex
t
o
p
er
atio
n
o
f
th
e
tr
an
s
f
o
r
m
atio
n
p
r
o
ce
d
u
r
e.
An
ex
am
p
le
tr
an
s
f
o
r
m
atio
n
f
o
r
th
e
r
ev
iews
m
en
tio
n
ed
i
n
F
ig
u
r
e
6
is
s
h
o
wn
in
F
ig
u
r
e
8
[
5
9
]
,
[
6
0
]
.
Fig
u
r
e
8
.
Sco
r
e
esti
m
ates f
o
r
t
h
e
u
tili
ty
p
atter
n
5.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
T
h
e
wo
r
k
co
n
s
id
er
ed
o
n
ly
p
o
s
itiv
e
u
tili
ty
v
alu
es
f
o
r
asp
ec
t
g
r
o
u
p
s
h
av
i
n
g
th
e
h
ig
h
e
s
t
p
o
s
itiv
e
r
atin
g
s
f
o
r
s
en
tim
en
t,
m
ap
p
in
g
ar
ticles
with
h
ig
h
s
ale
im
p
ac
t
o
n
ly
;
esti
m
ates
wer
e
ca
lcu
l
ated
b
y
ex
tr
ac
tin
g
tr
ip
les.
T
h
e
p
r
o
p
o
s
ed
wo
r
k
em
p
lo
y
ed
th
e
f
aster
h
ig
h
-
u
tili
ty
item
s
et
m
in
er
(
FHN
)
alg
o
r
it
h
m
[
5
1
]
,
[
5
2
]
,
u
s
in
g
th
e
FHN
tech
n
iq
u
e,
h
ig
h
u
ti
lity
asp
ec
t
(
HUA)
g
r
o
u
p
s
ar
e
r
etr
iev
ed
f
r
o
m
tr
an
s
ac
tio
n
d
atab
ase.
Patter
n
em
p
lo
y
ed
is
a
co
llectio
n
o
f
u
s
ef
u
l c
h
ar
ac
ter
is
tics
th
at
ar
e
in
d
icato
r
s
o
f
h
ig
h
p
r
o
f
it f
ea
tu
r
es o
r
asp
ec
ts
,
with
th
e
h
ig
h
est p
o
s
itiv
e
s
en
tim
en
t r
ati
n
g
s
an
d
,
th
u
s
,
th
e
lar
g
est lev
el
s
o
f
s
atis
f
ac
tio
n
ar
e
in
d
icate
d
b
y
th
e
p
atter
n
s
with
th
e
h
ig
h
est
p
o
s
itiv
e
u
tili
ty
.
Fig
u
r
e
9
d
ep
icts
c
o
m
p
ar
is
o
n
o
f
ac
cu
m
u
lated
u
tili
ty
f
o
r
FIM
an
d
p
r
o
p
o
s
ed
tech
n
iq
u
e
u
n
d
er
p
o
s
itiv
e
u
tili
ty
p
atter
n
s
.
Fig
u
r
e
9
.
C
o
m
p
a
r
is
o
n
o
f
ac
cu
m
u
lated
u
tili
ty
f
o
r
FIM
an
d
p
r
o
p
o
s
ed
tech
n
iq
u
e
u
n
d
e
r
p
o
s
itiv
e
u
tili
ty
p
atter
n
s
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
5
,
No
.
4
,
Au
g
u
s
t 2
0
2
6
:
3
2
4
0
-
3
2
5
1
3248
Hy
b
r
id
VNN
+
FHN,
C
NN,
L
STM
,
an
d
tr
an
s
f
o
r
m
er
m
o
d
el
wer
e
u
s
ed
f
o
r
ac
ce
s
s
in
g
th
e
u
t
ilit
y
d
r
iv
en
p
r
o
d
u
ct
an
aly
s
is
.
C
NN
ex
ce
l
s
in
ex
tr
ac
tin
g
r
elev
an
t
f
ea
tu
r
es
f
r
o
m
s
h
o
r
t
an
d
s
tr
u
ctu
r
e
d
tex
ts
,
th
e
L
STM
n
etwo
r
k
f
o
cu
s
es
o
n
lo
n
g
-
ter
m
co
n
tex
t
d
e
p
en
d
e
n
cies
in
p
r
o
d
u
ct
r
e
v
iew
tex
ts
,
an
d
T
r
a
n
s
f
o
r
m
er
s
s
h
in
e
with
r
o
b
u
s
t
co
n
tex
t
em
b
ed
d
in
g
s
th
at
ar
e
in
s
en
s
itiv
e
to
n
o
is
es
.
Hy
b
r
id
VNN
+
FHN
co
m
b
in
es
d
ee
p
co
n
tex
t
in
tellig
en
ce
with
ef
f
icien
t
u
tili
za
tio
n
-
p
atter
n
d
is
co
v
er
y
.
B
y
co
m
p
ar
in
g
th
e
f
o
u
r
alg
o
r
ith
m
s
,
th
e
m
o
s
t
r
eliab
le
o
n
e
will
b
e
p
ick
ed
th
at
b
alan
c
es
th
e
ef
f
icien
cy
o
f
p
r
o
d
u
ct
-
p
atter
n
d
is
co
v
er
y
,
c
o
n
tex
t
in
tel
lig
en
ce
,
p
r
o
ce
s
s
in
g
co
m
p
lex
ity
,
a
n
d
n
o
is
e
r
o
b
u
s
tn
ess
.
T
h
e
f
o
llo
win
g
m
etr
ics we
r
e
u
s
ed
to
ass
ess
th
e
alg
o
r
ith
m
s
:
−
E
s
tim
atio
n
o
f
n
u
m
b
er
o
f
p
atte
r
n
s
(
N)
.
−
Utilit
y
o
f
p
r
o
d
u
ct
an
d
av
er
a
g
e
u
tili
ty
u
(
p
i)
.
−
R
u
n
tim
e
o
f
an
al
g
o
r
ith
m
(
T
)
.
T
ab
le
3
d
ep
icts
c
o
m
p
ar
is
o
n
m
etr
ics
o
f
t
h
e
p
r
o
p
o
s
ed
h
y
b
r
id
alg
o
r
ith
m
with
C
NN,
L
STM
,
an
d
t
r
an
s
f
o
r
m
er
-
b
ased
m
o
d
els.
I
t
s
h
o
ws
th
at
th
e
h
y
b
r
i
d
m
et
h
o
d
d
e
m
o
n
s
tr
ated
th
e
s
tr
o
n
g
est
o
v
er
all
b
alan
ce
b
etwe
en
u
tili
ty
,
p
atter
n
ex
tr
ac
t
io
n
,
an
d
r
u
n
tim
e
. T
h
e
wea
k
est p
er
f
o
r
m
an
ce
is
o
f
C
NN
f
o
r
1
4
3
p
atter
n
s
.
T
ab
le
3
.
C
o
m
p
a
r
is
o
n
m
etr
ics
o
f
p
r
o
p
o
s
ed
h
y
b
r
i
d
VNN+
FHN
with
C
NN,
L
STM
,
an
d
tr
an
s
f
o
r
m
er
m
o
d
el
A
l
g
o
r
i
t
h
m
P
a
t
t
e
r
n
c
o
u
nt
(N)
A
v
e
r
a
g
e
u
t
i
l
i
t
y
(
u
n
i
t
s
)
R
u
n
t
i
m
e
(
µs
)
H
y
b
r
i
d
(
F
H
N
+
VNN)
1
5
8
1
6
2
.
8
8
70
C
N
N
1
4
3
1
5
5
.
4
6
60
LSTM
1
7
1
1
6
7
.
3
3
80
T
r
a
n
sf
o
r
mer
-
b
a
se
d
p
r
e
d
i
c
t
i
o
n
m
o
d
e
l
(
TPM
)
1
8
9
1
7
3
.
5
2
90
L
STM
p
er
f
o
r
m
s
b
etter
with
1
7
1
p
atter
n
s
id
en
tifie
d
an
d
h
ig
h
er
u
tili
ty
o
f
1
6
7
.
3
3
with
r
u
n
tim
e
o
f
8
0
µs.
T
PM
d
eliv
er
s
b
est
d
ee
p
-
lear
n
in
g
r
esu
lts
,
ac
h
iev
in
g
t
h
e
h
ig
h
est
p
atter
n
co
u
n
t
o
f
1
8
9
with
u
tili
ty
v
alu
e
1
7
3
.
5
2
,
h
o
wev
e
r
it
h
as
s
lo
w
est
r
u
n
tim
e
o
f
9
0
µs.
T
h
e
h
y
b
r
id
VNN
+
FHN
tech
n
i
q
u
e
b
ala
n
ce
s
th
ese
tr
ad
e
-
o
f
f
s
b
y
ex
t
r
ac
tin
g
1
5
8
p
atter
n
s
with
u
tili
ty
o
f
1
6
2
.
8
8
an
d
a
f
ast
r
u
n
tim
e
o
f
7
0
µs,
o
u
tp
er
f
o
r
m
in
g
C
NN
in
u
tili
ty
,
L
STM
an
d
T
PM
in
s
p
ee
d
,
an
d
C
NN
in
p
atter
n
c
o
u
n
t.
I
n
e
x
ten
d
ed
ev
alu
atio
n
,
h
y
b
r
id
VNN
-
FHN
ac
h
iev
es h
ig
h
est o
v
er
all
u
tili
ty
o
f
1
8
2
.
4
u
n
its
b
u
t r
eq
u
ir
ed
th
e
lo
n
g
est r
u
n
tim
e
o
f
0
.
0
1
8
s
,
in
d
icatin
g
in
cr
ea
s
ed
co
m
p
u
tatio
n
al
c
o
s
t
with
h
ig
h
e
r
ac
cu
r
ac
y
.
C
NN
an
d
L
STM
p
r
o
d
u
ce
d
m
o
d
er
ate
u
tili
ties
o
f
1
5
0
.
6
–
1
6
1
.
2
u
n
its
an
d
o
cc
asio
n
ally
m
is
in
ter
p
r
eted
s
h
o
r
t
o
r
n
o
is
y
tex
t,
wh
ile
th
e
T
r
an
s
f
o
r
m
er
m
o
d
el
m
ain
ta
in
ed
s
tab
le
u
tili
ty
o
f
1
7
6
.
9
u
n
its
b
u
t
s
tr
u
g
g
led
with
am
b
ig
u
o
u
s
o
r
i
n
co
m
p
lete
r
ev
iews.
All
m
o
d
els
s
ca
led
w
ell
f
o
r
1
0
0
–
1
,
0
0
0
e
-
co
m
m
er
ce
p
r
o
d
u
cts
with
r
u
n
tim
es
u
n
d
er
0
.
0
2
s
;
C
NN
an
d
L
STM
s
h
o
wed
th
e
m
o
s
t
s
tab
le
r
u
n
tim
es
o
f
0
.
0
1
0
–
0
.
0
1
3
s
d
u
e
to
GPU
-
f
r
i
en
d
ly
ar
ch
itectu
r
es,
wh
er
ea
s
h
y
b
r
id
VNN
-
FHN
s
lo
wed
s
li
g
h
tly
as
p
atter
n
s
ize
in
cr
ea
s
ed
.
I
n
cr
o
s
s
-
d
o
m
ain
s
c
en
ar
io
s
f
in
an
ce
,
h
ea
lth
ca
r
e,
a
n
d
r
etail,
all
m
o
d
els
ca
n
g
en
e
r
a
lize
with
r
ed
ef
in
ed
u
tili
ty
f
u
n
ctio
n
s
,
with
h
y
b
r
i
d
VNN
+
FHN
tech
n
iq
u
e
s
h
o
win
g
th
e
b
est ad
ap
tab
ilit
y
.
6.
CO
NCLU
SI
O
N
A
tar
g
eted
R
S
d
esig
n
h
elp
s
i
n
o
p
tim
izatio
n
o
f
r
eso
u
r
ce
s
in
ter
m
s
o
f
r
aw
m
ater
ial,
tr
an
s
p
o
r
tatio
n
,
m
an
u
f
ac
tu
r
in
g
co
s
t,
h
u
m
a
n
r
eso
u
r
ce
s
,
p
ac
k
ag
in
g
,
m
ar
k
eti
n
g
,
an
d
s
ale.
Pro
p
o
s
ed
wo
r
k
em
p
lo
y
s
a
h
y
b
r
id
ap
p
r
o
ac
h
o
f
class
if
icatio
n
an
d
h
ig
h
u
tili
ty
p
atter
n
to
id
en
tif
y
th
e
d
o
m
i
n
an
t
asp
ec
ts
o
f
ar
ti
cles
with
h
ig
h
s
ale
p
r
o
b
a
b
ilit
y
.
Pro
p
o
s
ed
tech
n
iq
u
e
h
y
b
r
id
VNN
+
FHN
was
ass
es
s
ed
an
d
co
m
p
a
r
ed
with
C
NN,
L
STM
,
an
d
T
r
an
s
f
o
r
m
e
r
m
o
d
el.
L
STM
p
er
f
o
r
m
s
b
etter
with
1
7
1
p
atter
n
s
id
en
tifie
d
an
d
h
i
g
h
er
u
tili
ty
o
f
1
6
7
.
3
3
with
r
u
n
tim
e
o
f
8
0
µs.
T
PM
d
eliv
er
s
b
est
d
ee
p
-
lear
n
in
g
r
esu
lts
,
ac
h
iev
in
g
th
e
h
ig
h
est
p
atter
n
co
u
n
t
o
f
1
8
9
with
u
tili
ty
v
alu
e
1
7
3
.
5
2
,
h
o
wev
er
it
h
as
s
lo
west
r
u
n
tim
e
o
f
9
0
µs.
T
h
e
h
y
b
r
id
VNN
+
FHN
tech
n
iq
u
e
b
alan
ce
s
th
ese
tr
ad
e
-
o
f
f
s
b
y
e
x
tr
ac
tin
g
1
5
8
p
atter
n
s
with
u
tili
ty
o
f
1
6
2
.
8
8
an
d
a
f
ast
r
u
n
tim
e
o
f
7
0
µs,
o
u
tp
er
f
o
r
m
in
g
C
NN
in
u
tili
ty
,
L
STM
a
n
d
T
PM
in
s
p
ee
d
,
an
d
C
NN
in
p
atter
n
co
u
n
t.
I
n
cr
o
s
s
-
d
o
m
ai
n
s
ce
n
ar
io
s
f
in
an
ce
,
h
ea
lth
ca
r
e,
an
d
r
etail,
all
m
o
d
els
ca
n
g
en
er
alize
with
r
ed
e
f
in
ed
u
tili
ty
f
u
n
ctio
n
s
,
with
th
e
h
y
b
r
id
VNN
+
FHN
tech
n
iq
u
e
d
is
p
lay
in
g
th
e
b
est ad
ap
tab
ilit
y
.
F
UNDING
I
NF
O
R
M
A
T
I
O
N
Au
th
o
r
s
s
tate
n
o
f
u
n
d
in
g
in
v
o
lv
ed
.
AUTHO
R
CO
NT
RI
B
UT
I
O
NS ST
A
T
E
M
E
N
T
T
h
is
jo
u
r
n
al
u
s
es
th
e
C
o
n
tr
ib
u
to
r
R
o
les
T
ax
o
n
o
m
y
(
C
R
ed
iT)
to
r
ec
o
g
n
ize
in
d
iv
id
u
al
au
th
o
r
co
n
tr
ib
u
tio
n
s
,
r
ed
u
ce
au
th
o
r
s
h
ip
d
is
p
u
tes,
an
d
f
ac
ilit
ate
co
llab
o
r
atio
n
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
R
eso
u
r
ce
o
p
timiz
a
tio
n
via
id
e
n
tifi
ca
tio
n
o
f a
r
ticles w
ith
a
s
p
ec
ts
o
f p
r
io
r
ity
…
(
Ga
r
ima
S
r
i
va
s
ta
va
)
3249
Na
m
e
o
f
Aut
ho
r
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
Gar
im
a
Sriv
astav
a
✓
✓
✓
✓
✓
✓
✓
✓
✓
Vaish
ali
Sin
g
h
✓
✓
✓
✓
✓
✓
Sach
in
Ku
m
ar
✓
✓
✓
✓
✓
✓
✓
✓
✓
C
:
C
o
n
c
e
p
t
u
a
l
i
z
a
t
i
o
n
M
:
M
e
t
h
o
d
o
l
o
g
y
So
:
So
f
t
w
a
r
e
Va
:
Va
l
i
d
a
t
i
o
n
Fo
:
Fo
r
mal
a
n
a
l
y
s
i
s
I
:
I
n
v
e
s
t
i
g
a
t
i
o
n
R
:
R
e
so
u
r
c
e
s
D
:
D
a
t
a
C
u
r
a
t
i
o
n
O
:
W
r
i
t
i
n
g
-
O
r
i
g
i
n
a
l
D
r
a
f
t
E
:
W
r
i
t
i
n
g
-
R
e
v
i
e
w
&
E
d
i
t
i
n
g
Vi
:
Vi
su
a
l
i
z
a
t
i
o
n
Su
:
Su
p
e
r
v
i
s
i
o
n
P
:
P
r
o
j
e
c
t
a
d
mi
n
i
st
r
a
t
i
o
n
Fu
:
Fu
n
d
i
n
g
a
c
q
u
i
si
t
i
o
n
CO
NF
L
I
C
T
O
F
I
N
T
E
R
E
S
T
ST
A
T
E
M
E
NT
Au
th
o
r
s
s
tate
n
o
co
n
f
lict o
f
in
t
er
est.
DATA AV
AI
L
AB
I
L
I
T
Y
T
h
e
d
ataset
g
en
er
ate
d
is
av
ailab
le
f
r
o
m
th
e
co
r
r
esp
o
n
d
in
g
a
u
th
o
r
,
[
GS]
,
u
p
o
n
r
ea
s
o
n
a
b
le
r
e
q
u
est.
RE
F
E
R
E
NC
E
S
[
1
]
H
.
J.
A
l
e
q
a
b
i
e
,
M
.
S
.
S
f
o
q
,
R
.
A
.
A
l
b
e
e
r
,
a
n
d
E
.
H
.
A
b
d
,
“
A
r
e
v
i
e
w
o
f
t
e
x
t
mi
n
i
n
g
t
e
c
h
n
i
q
u
e
s
:
t
r
e
n
d
s,
a
n
d
a
p
p
l
i
c
a
t
i
o
n
s
i
n
v
a
r
i
o
u
s
dom
a
i
n
s,
”
I
ra
q
i
J
o
u
r
n
a
l
f
o
r
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
a
n
d
Ma
t
h
e
m
a
t
i
c
s
,
v
o
l
.
5
,
n
o
.
1
,
p
p
.
1
2
5
–
1
4
1
,
2
0
2
4
,
d
o
i
:
1
0
.
5
2
8
6
6
/
i
j
c
s
m.
2
0
2
4
.
0
5
.
0
1
.
0
0
9
.
[
2
]
S
.
Li
,
B
.
Zh
u
,
Y
.
Z
h
a
n
g
,
F
.
L
i
u
,
a
n
d
Z.
Y
u
,
“
A
t
w
o
-
st
a
g
e
n
o
n
l
i
n
e
a
r
u
se
r
sat
i
sf
a
c
t
i
o
n
d
e
c
i
s
i
o
n
m
o
d
e
l
b
a
se
d
o
n
o
n
l
i
n
e
r
e
v
i
e
w
mi
n
i
n
g
:
c
o
n
s
i
d
e
r
i
n
g
n
o
n
-
c
o
m
p
e
n
sa
t
o
r
y
a
n
d
c
o
m
p
e
n
sa
t
o
r
y
st
a
g
e
s,
”
J
o
u
r
n
a
l
o
f
T
h
e
o
re
t
i
c
a
l
a
n
d
Ap
p
l
i
e
d
E
l
e
c
t
ro
n
i
c
C
o
m
m
e
r
c
e
Re
se
a
rc
h
,
v
o
l
.
1
9
,
n
o
.
1
,
p
p
.
2
7
2
–
2
9
6
,
2
0
2
4
,
d
o
i
:
1
0
.
3
3
9
0
/
j
t
a
e
r
1
9
0
1
0
0
1
5
.
[
3
]
J.
D
i
n
g
,
M
.
X
u
,
Y
.
K
.
T
se,
K
.
Y
.
Li
n
,
a
n
d
M
.
Z
h
a
n
g
,
“
C
u
s
t
o
m
e
r
o
p
i
n
i
o
n
s
m
i
n
i
n
g
t
h
r
o
u
g
h
s
o
c
i
a
l
m
e
d
i
a
:
i
n
si
g
h
t
s
f
r
o
m
su
st
a
i
n
a
b
i
l
i
t
y
f
r
a
u
d
c
r
i
s
i
s
-
V
o
l
k
s
w
a
g
e
n
e
mi
ss
i
o
n
s
sc
a
n
d
a
l
,
”
E
n
t
e
rp
ri
se
I
n
f
o
rm
a
t
i
o
n
S
y
st
e
m
s
,
v
o
l
.
1
7
,
n
o
.
8
,
2
0
2
3
,
d
o
i
:
1
0
.
1
0
8
0
/
1
7
5
1
7
5
7
5
.
2
0
2
2
.
2
1
3
0
0
1
2
.
[
4
]
H
.
Y
a
o
,
H
.
J.
H
a
mi
l
t
o
n
,
a
n
d
C
.
J.
B
u
t
z
,
“
A
f
o
u
n
d
a
t
i
o
n
a
l
a
p
p
r
o
a
c
h
t
o
m
i
n
i
n
g
i
t
e
ms
e
t
u
t
i
l
i
t
i
e
s
f
r
o
m
d
a
t
a
b
a
ses
,
”
i
n
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
2
0
0
4
S
I
AM
I
n
t
e
rn
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
D
a
t
a
Mi
n
i
n
g
,
A
p
r
.
2
0
0
4
,
p
p
.
4
8
2
–
4
8
6
,
d
o
i
:
1
0
.
1
1
3
7
/
1
.
9
7
8
1
6
1
1
9
7
2
7
4
0
.
5
1
.
[
5
]
H
.
Y
a
o
,
H
.
J
.
H
a
m
i
l
t
o
n
,
a
n
d
L.
G
e
n
g
,
“
A
u
n
i
f
i
e
d
f
r
a
m
e
w
o
r
k
f
o
r
u
t
i
l
i
t
y
-
b
a
sed
m
e
a
s
u
r
e
s
f
o
r
mi
n
i
n
g
i
t
e
mse
t
s,”
i
n
A
C
M
S
I
G
K
D
D
2
n
d
Wo
rks
h
o
p
o
n
U
t
i
l
i
t
y
B
a
se
d
D
a
t
a
Mi
n
i
n
g
,
P
h
i
l
a
d
e
l
p
h
i
a
,
2
0
0
6
,
p
p
.
2
8
–
3
7
.
[
6
]
V
.
R
a
mi
n
e
n
i
a
n
d
L.
S
u
d
u
l
a
,
“
M
i
n
i
n
g
f
r
e
q
u
e
n
t
h
i
g
h
u
t
i
l
i
t
y
i
t
e
mse
t
s
o
n
d
y
n
a
mi
c
d
a
t
a
b
a
s
e
s
w
i
t
h
o
u
t
c
a
n
d
i
d
a
t
e
g
e
n
e
r
a
t
i
o
n
,
”
i
n
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
o
n
Em
e
rg
i
n
g
T
e
c
h
n
o
l
o
g
i
e
s
i
n
En
g
i
n
e
e
ri
n
g
a
n
d
S
c
i
e
n
c
e
:
I
C
ETE
S
2
0
2
3
,
2
0
2
5
,
d
o
i
:
1
0
.
1
0
6
3
/
5
.
0
2
4
2
3
2
1
.
[
7
]
C
.
F
.
A
h
m
e
d
,
S
.
K
.
T
a
n
b
e
e
r
,
B
.
S
.
J
e
o
n
g
,
a
n
d
Y
.
K
.
Le
e
,
“
Ef
f
i
c
i
e
n
t
t
r
e
e
s
t
r
u
c
t
u
r
e
s
f
o
r
h
i
g
h
u
t
i
l
i
t
y
p
a
t
t
e
r
n
m
i
n
i
n
g
i
n
i
n
c
r
e
m
e
n
t
a
l
d
a
t
a
b
a
se
s,”
I
EE
E
T
ra
n
s
a
c
t
i
o
n
s
o
n
K
n
o
w
l
e
d
g
e
a
n
d
D
a
t
a
E
n
g
i
n
e
e
ri
n
g
,
v
o
l
.
2
1
,
n
o
.
1
2
,
p
p
.
1
7
0
8
–
1
7
2
1
,
2
0
0
9
,
d
o
i
:
1
0
.
1
1
0
9
/
T
K
D
E.
2
0
0
9
.
4
6
.
[
8
]
V
.
S
.
T
s
e
n
g
,
B
.
E
.
S
h
i
e
,
C
.
W
.
W
u
,
a
n
d
P
.
S
.
Y
u
,
“
E
f
f
i
c
i
e
n
t
a
l
g
o
r
i
t
h
m
s
f
o
r
m
i
n
i
n
g
h
i
g
h
u
t
i
l
i
t
y
i
t
e
m
s
e
t
s
f
r
o
m
t
r
a
n
s
a
c
t
i
o
n
a
l
d
a
t
a
b
a
s
e
s
,
”
I
E
E
E
T
r
a
n
s
a
c
t
i
o
n
s
o
n
K
n
o
w
l
e
d
g
e
a
n
d
D
a
t
a
E
n
g
i
n
e
e
r
i
n
g
,
v
o
l
.
2
5
,
n
o
.
8
,
p
p
.
1
7
7
2
–
1
7
8
6
,
2
0
1
3
,
d
o
i
:
1
0
.
1
1
0
9
/
T
K
D
E
.
2
0
1
2
.
5
9
.
[
9
]
M
.
M
.
A
z
i
z
,
A
.
A
.
B
a
k
a
r
,
a
n
d
M
.
R
.
Y
a
a
k
u
b
,
“
C
o
r
e
N
LP
d
e
p
e
n
d
e
n
c
y
p
a
r
si
n
g
a
n
d
p
a
t
t
e
r
n
i
d
e
n
t
i
f
i
c
a
t
i
o
n
f
o
r
e
n
h
a
n
c
e
d
o
p
i
n
i
o
n
mi
n
i
n
g
i
n
a
s
p
e
c
t
-
b
a
se
d
se
n
t
i
me
n
t
a
n
a
l
y
si
s
,
”
J
o
u
rn
a
l
o
f
K
i
n
g
S
a
u
d
U
n
i
v
e
rs
i
t
y
-
C
o
m
p
u
t
e
r
a
n
d
I
n
f
o
rm
a
t
i
o
n
S
c
i
e
n
c
e
s
,
v
o
l
.
3
6
,
n
o
.
4
,
2
0
2
4
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
j
k
su
c
i
.
2
0
2
4
.
1
0
2
0
3
5
.
[
1
0
]
K
.
K
.
Y
u
su
f
,
E.
O
g
b
u
j
u
,
T.
A
b
i
o
d
u
n
,
a
n
d
F
.
O
l
a
d
i
p
o
,
“
A
t
e
c
h
n
i
c
a
l
r
e
v
i
e
w
o
f
t
h
e
st
a
t
e
-
of
-
t
h
e
-
a
r
t
me
t
h
o
d
s
i
n
a
s
p
e
c
t
-
b
a
s
e
d
sen
t
i
m
e
n
t
a
n
a
l
y
si
s
,
”
J
o
u
r
n
a
l
o
f
C
o
m
p
u
t
i
n
g
T
h
e
o
ri
e
s
a
n
d
A
p
p
l
i
c
a
t
i
o
n
s
,
v
o
l
.
1
,
n
o
.
3
,
p
p
.
2
8
7
–
2
9
8
,
2
0
2
4
,
d
o
i
:
1
0
.
6
2
4
1
1
/
j
c
t
a
.
9
9
9
9
.
[
1
1
]
H
.
H
.
D
o
,
P
.
P
r
a
s
a
d
,
A
.
M
a
a
g
,
a
n
d
A
.
A
l
s
a
d
o
o
n
,
“
D
e
e
p
l
e
a
r
n
i
n
g
f
o
r
a
s
p
e
c
t
-
b
a
s
e
d
se
n
t
i
m
e
n
t
a
n
a
l
y
s
i
s:
a
c
o
m
p
a
r
a
t
i
v
e
r
e
v
i
e
w
,
”
Ex
p
e
rt
S
y
st
e
m
s w
i
t
h
Ap
p
l
i
c
a
t
i
o
n
s
,
v
o
l
.
1
1
8
,
p
p
.
2
7
2
–
2
9
9
,
M
a
r
.
2
0
1
9
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
e
sw
a
.
2
0
1
8
.
1
0
.
0
0
3
.
[
1
2
]
A
.
Q
a
z
i
,
R
.
G
.
R
a
j
,
G
.
H
a
r
d
a
k
e
r
,
a
n
d
C
.
S
t
a
n
d
i
n
g
,
“
A
s
y
s
t
e
m
a
t
i
c
l
i
t
e
r
a
t
u
r
e
r
e
v
i
e
w
o
n
o
p
i
n
i
o
n
t
y
p
e
s
a
n
d
se
n
t
i
me
n
t
a
n
a
l
y
si
s
t
e
c
h
n
i
q
u
e
s,
”
I
n
t
e
r
n
e
t
Re
se
a
r
c
h
,
v
o
l
.
2
7
,
n
o
.
3
,
p
p
.
6
0
8
–
6
3
0
,
J
u
n
.
2
0
1
7
,
d
o
i
:
1
0
.
1
1
0
8
/
I
n
t
R
-
04
-
2
0
1
6
-
0
0
8
6
.
[
1
3
]
P
.
H
a
j
e
k
,
L.
H
i
k
k
e
r
o
v
a
,
a
n
d
J.
M
.
S
a
h
u
t
,
“
F
a
k
e
r
e
v
i
e
w
d
e
t
e
c
t
i
o
n
i
n
e
-
c
o
mm
e
r
c
e
p
l
a
t
f
o
r
ms u
s
i
n
g
a
s
p
e
c
t
-
b
a
s
e
d
se
n
t
i
m
e
n
t
a
n
a
l
y
s
i
s,
”
J
o
u
rn
a
l
o
f
Bu
s
i
n
e
ss
Re
s
e
a
r
c
h
,
v
o
l
.
1
6
7
,
2
0
2
3
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
j
b
u
sr
e
s.
2
0
2
3
.
1
1
4
1
4
3
.
[
1
4
]
Z.
L
i
n
a
n
d
B
.
Li
,
“
A
s
p
e
c
t
-
b
a
s
e
d
se
n
t
i
men
t
a
n
a
l
y
si
s
b
a
se
d
o
n
l
o
c
a
l
c
o
n
t
e
x
t
f
o
c
u
s
m
e
c
h
a
n
i
sm
a
n
d
t
a
l
k
i
n
g
-
h
e
a
d
a
t
t
e
n
t
i
o
n
,
”
i
n
2
0
2
2
4
t
h
I
n
t
e
rn
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
F
ro
n
t
i
e
rs
T
e
c
h
n
o
l
o
g
y
o
f
I
n
f
o
rm
a
t
i
o
n
a
n
d
C
o
m
p
u
t
e
r,
I
C
FTI
C
2
0
2
2
,
2
0
2
2
,
p
p
.
1
1
2
–
118
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
F
TI
C
5
7
6
9
6
.
2
0
2
2
.
1
0
0
7
5
3
2
1
.
[
1
5
]
W
.
A
n
,
F
.
T
i
a
n
,
P
.
C
h
e
n
,
a
n
d
Q
.
Z
h
e
n
g
,
“
A
s
p
e
c
t
-
b
a
se
d
se
n
t
i
me
n
t
a
n
a
l
y
si
s
w
i
t
h
h
e
t
e
r
o
g
e
n
e
o
u
s
g
r
a
p
h
n
e
u
r
a
l
n
e
t
w
o
r
k
,
”
I
EE
E
T
ra
n
s
a
c
t
i
o
n
s
o
n
C
o
m
p
u
t
a
t
i
o
n
a
l
S
o
c
i
a
l
S
y
st
e
m
s
,
v
o
l
.
1
0
,
n
o
.
1
,
p
p
.
4
0
3
–
4
1
2
,
2
0
2
3
,
d
o
i
:
1
0
.
1
1
0
9
/
T
C
S
S
.
2
0
2
2
.
3
1
4
8
8
6
6
.
[
1
6
]
H
.
X
u
,
S
.
Li
u
,
W
.
W
a
n
g
,
a
n
d
L
.
D
e
n
g
,
“
R
A
G
-
TC
G
C
N
:
a
s
p
e
c
t
s
e
n
t
i
m
e
n
t
a
n
a
l
y
s
i
s
b
a
s
e
d
o
n
r
e
si
d
u
a
l
a
t
t
e
n
t
i
o
n
g
a
t
i
n
g
a
n
d
t
h
r
e
e
-
c
h
a
n
n
e
l
g
r
a
p
h
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
t
w
o
r
k
s
,
”
A
p
p
l
i
e
d
S
c
i
e
n
c
e
s
,
v
o
l
.
1
2
,
n
o
.
2
3
,
2
0
2
2
,
d
o
i
:
1
0
.
3
3
9
0
/
a
p
p
1
2
2
3
1
2
1
0
8
.
[
1
7
]
Y
.
W
a
n
g
,
C
.
Li
u
,
J
.
X
i
e
,
S
.
Y
a
n
g
,
Y
.
Ji
a
,
a
n
d
H
.
Za
n
,
“
A
sp
e
c
t
-
b
a
s
e
d
sen
t
i
me
n
t
a
n
a
l
y
s
i
s
w
i
t
h
d
e
p
e
n
d
e
n
c
y
r
e
l
a
t
i
o
n
g
r
a
p
h
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
t
w
o
r
k
,
”
i
n
2
0
2
2
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
As
i
a
n
L
a
n
g
u
a
g
e
Pro
c
e
ss
i
n
g
,
I
ALP
2
0
2
2
,
2
0
2
2
,
p
p
.
6
3
–
6
8
.
d
o
i
:
1
0
.
1
1
0
9
/
I
A
LP5
7
1
5
9
.
2
0
2
2
.
9
9
6
1
3
2
1
.
[
1
8
]
A
.
K
u
mar
a
n
d
G
.
G
a
r
g
,
“
S
y
st
e
mat
i
c
l
i
t
e
r
a
t
u
r
e
r
e
v
i
e
w
o
n
c
o
n
t
e
x
t
-
b
a
s
e
d
s
e
n
t
i
m
e
n
t
a
n
a
l
y
s
i
s
i
n
s
o
c
i
a
l
m
u
l
t
i
m
e
d
i
a
,
”
M
u
l
t
i
m
e
d
i
a
T
o
o
l
s
a
n
d
Ap
p
l
i
c
a
t
i
o
n
s
,
v
o
l
.
7
9
,
n
o
.
2
1
–
2
2
,
p
p
.
1
5
3
4
9
–
1
5
3
8
0
,
2
0
2
0
,
d
o
i
:
1
0
.
1
0
0
7
/
s1
1
0
4
2
-
019
-
7
3
4
6
-
5.
[
1
9
]
H
.
H
a
n
,
X
.
B
a
i
,
a
n
d
P
.
Li
,
“
A
u
g
m
e
n
t
e
d
se
n
t
i
m
e
n
t
r
e
p
r
e
s
e
n
t
a
t
i
o
n
b
y
l
e
a
r
n
i
n
g
c
o
n
t
e
x
t
i
n
f
o
r
ma
t
i
o
n
,
”
N
e
u
r
a
l
C
o
m
p
u
t
i
n
g
a
n
d
Ap
p
l
i
c
a
t
i
o
n
s
,
v
o
l
.
3
1
,
n
o
.
1
2
,
p
p
.
8
4
7
5
–
8
4
8
2
,
2
0
1
9
,
d
o
i
:
1
0
.
1
0
0
7
/
s0
0
5
2
1
-
0
1
8
-
3
6
9
8
-
4.
[
2
0
]
S
.
S
.
H
a
r
i
,
S
.
P
o
r
k
o
d
i
,
R
.
S
a
r
a
n
y
a
,
a
n
d
N
.
V
i
j
a
y
a
k
u
mar,
“
I
n
t
e
l
l
i
g
e
n
t
mo
d
e
l
t
o
i
m
p
r
o
v
e
t
h
e
e
f
f
i
c
a
c
y
o
f
h
e
a
l
t
h
c
a
r
e
c
o
n
t
e
n
t
mark
e
t
i
n
g
b
y
a
u
t
o
-
t
a
g
g
i
n
g
a
n
d
e
x
p
l
o
r
i
n
g
t
h
e
v
e
r
a
c
i
t
y
o
f
c
o
n
t
e
n
t
u
si
n
g
o
p
i
n
i
o
n
m
i
n
i
n
g
,
”
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
r
n
a
l
o
f
El
e
c
t
ro
n
i
c
Ma
r
k
e
t
i
n
g
a
n
d
R
e
t
a
i
l
i
n
g
,
v
o
l
.
1
5
,
n
o
.
2
,
p
p
.
2
4
0
–
2
6
0
,
2
0
2
4
,
d
o
i
:
1
0
.
1
5
0
4
/
I
JEM
R
.
2
0
2
4
.
1
3
6
9
7
8
.
Evaluation Warning : The document was created with Spire.PDF for Python.