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
.
2
,
A
p
r
il 2
0
2
6
,
p
p
.
1
8
5
1
~
1
8
6
2
I
SS
N:
2
2
5
2
-
8
9
3
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijai.v
15
.i
2
.
p
p
1
8
5
1
-
1
8
6
2
1851
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
i
.
ia
esco
r
e.
co
m
Sentimen
t
-
a
wa
re
user
-
ite
m recom
menda
tion co
mbi
ning
weig
hted
XG
B
o
o
s
t
a
nd op
timi
zed si
mila
rity metri
cs
Sn
eha
l B
ho
g
a
n,
Vij
a
y
S.
Ra
j
pu
ro
hit,
Sa
nje
ev
S.
Sa
nn
a
k
ki
D
e
p
a
r
t
me
n
t
o
f
C
o
mp
u
t
e
r
S
c
i
e
n
c
e
a
n
d
En
g
i
n
e
e
r
i
n
g
,
K
a
r
n
a
t
a
k
L
a
w
S
o
c
i
e
t
y
G
o
g
t
e
I
n
st
i
t
u
t
e
o
f
T
e
c
h
n
o
l
o
g
y
,
a
f
f
i
l
i
a
t
e
d
t
o
V
i
sv
e
sv
a
r
a
y
a
Te
c
h
n
o
l
o
g
i
c
a
l
U
n
i
v
e
r
si
t
y
,
B
e
l
a
g
a
v
i
,
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
Ma
y
2
6
,
2
0
2
5
R
ev
is
ed
J
an
5
,
2
0
2
6
Acc
ep
ted
J
an
2
5
,
2
0
2
6
Us
e
r
-
it
e
m
re
c
o
m
m
e
n
d
a
ti
o
n
sy
ste
m
s
p
lay
a
v
it
a
l
ro
le
i
n
e
n
h
a
n
c
in
g
p
e
rso
n
a
li
z
e
d
d
i
g
it
a
l
e
x
p
e
rien
c
e
s
a
c
ro
ss
e
-
c
o
m
m
e
rc
e
a
n
d
so
c
ial
m
e
d
ia
p
latfo
rm
s.
Tra
d
it
i
o
n
a
l
re
c
o
m
m
e
n
d
a
ti
o
n
a
p
p
r
o
a
c
h
e
s,
su
c
h
a
s
c
o
l
lab
o
ra
ti
v
e
fil
terin
g
(C
F
)
a
n
d
c
o
n
ten
t
-
b
a
se
d
fil
terin
g
(CB
F
)
,
o
fte
n
su
f
fe
r
fro
m
c
h
a
ll
e
n
g
e
s
li
k
e
d
a
ta
sp
a
rsity
,
c
o
ld
-
sta
rt
iss
u
e
s,
a
n
d
li
m
it
e
d
c
o
n
tex
t
u
a
l
u
n
d
e
rsta
n
d
i
n
g
.
S
e
n
ti
m
e
n
t
-
a
wa
re
re
c
o
m
m
e
n
d
a
ti
o
n
s
y
ste
m
s
h
a
v
e
e
m
e
rg
e
d
a
s
a
p
ro
m
isin
g
so
l
u
ti
o
n
b
y
i
n
c
o
r
p
o
ra
ti
n
g
e
m
o
t
io
n
a
l
in
si
g
h
ts
e
x
trac
ted
fro
m
u
se
r
re
v
iew
s,
th
e
re
b
y
imp
r
o
v
in
g
re
c
o
m
m
e
n
d
a
ti
o
n
a
c
c
u
ra
c
y
a
n
d
p
e
rso
n
a
li
z
a
ti
o
n
.
Th
is
stu
d
y
p
r
o
p
o
se
s
a
n
o
v
e
l
se
n
ti
m
e
n
t
-
a
wa
re
u
se
r
-
it
e
m
re
c
o
m
m
e
n
d
a
ti
o
n
sy
ste
m
(S
AU
IRS
)
fra
m
e
wo
rk
th
a
t
in
teg
ra
tes
o
p
t
imiz
e
d
term
f
re
q
u
e
n
c
y
-
in
v
e
rse
d
o
c
u
m
e
n
t
fre
q
u
e
n
c
y
(
O
-
TF
-
IDF),
p
a
ra
m
e
teriz
e
d
b
id
irec
ti
o
n
a
l
e
n
c
o
d
e
r
re
p
re
se
n
tati
o
n
s
fr
o
m
tr
a
n
sfo
rm
e
rs
(P
-
BERT
),
we
ig
h
ted
e
x
trem
e
g
ra
d
ien
t
b
o
o
sti
n
g
(W
XG
Bo
o
st),
a
n
d
a
n
o
p
ti
m
ize
d
sim
il
a
rit
y
m
e
tri
c
s
m
o
d
e
l
.
Th
e
o
p
t
imiz
e
d
TF
-
IDF
e
n
h
a
n
c
e
s
fe
a
tu
re
se
lec
ti
o
n
,
re
d
u
c
in
g
d
ime
n
sio
n
a
li
ty
wh
il
e
p
re
se
rv
i
n
g
re
lev
a
n
t
tex
t
u
a
l
in
f
o
rm
a
t
io
n
.
P
-
B
ERT
,
a
fin
e
-
t
u
n
e
d
BERT
m
o
d
e
l,
imp
r
o
v
e
s
se
n
ti
m
e
n
t
c
las
sifica
ti
o
n
a
c
c
u
ra
c
y
b
y
lev
e
ra
g
in
g
d
e
e
p
c
o
n
tex
t
u
a
l
e
m
b
e
d
d
in
g
s.
WXG
Bo
o
st
f
u
rth
e
r
re
fin
e
d
se
n
ti
m
e
n
t
p
re
d
ictio
n
s,
a
d
d
re
ss
in
g
c
las
s
imb
a
lan
c
e
a
n
d
e
n
h
a
n
c
i
n
g
m
o
d
e
l
r
o
b
u
stn
e
ss
.
T
h
e
e
x
trac
ted
se
n
ti
m
e
n
t
in
f
o
rm
a
ti
o
n
is
i
n
c
o
r
p
o
ra
ted
in
t
o
a
n
o
p
ti
m
ize
d
sim
il
a
rit
y
m
e
tri
c
s
m
o
d
e
l
to
imp
r
o
v
e
re
c
o
m
m
e
n
d
a
ti
o
n
p
re
c
isio
n
b
y
a
li
g
n
in
g
u
se
r
p
r
e
fe
re
n
c
e
s
with
se
n
ti
m
e
n
t
-
d
riv
e
n
i
n
sig
h
ts.
Ex
ten
si
v
e
e
x
p
e
rime
n
ts
c
o
n
d
u
c
ted
o
n
Am
a
z
o
n
b
e
n
c
h
m
a
rk
d
a
tas
e
ts
d
e
m
o
n
stra
te
th
e
su
p
e
rio
r
p
e
rf
o
rm
a
n
c
e
in
term
s
o
f
a
c
c
u
ra
c
y
,
ro
o
t
m
e
a
n
s
q
u
a
re
e
rro
r
(R
M
S
E)
,
a
n
d
m
e
a
n
a
b
so
lu
te
e
rro
r
(M
AE)
o
f
t
h
e
p
ro
p
o
se
d
fr
a
m
e
wo
rk
c
o
m
p
a
re
d
to
sta
te
-
of
-
th
e
-
a
rt
re
c
o
m
m
e
n
d
a
ti
o
n
m
o
d
e
ls.
K
ey
w
o
r
d
s
:
C
o
llab
o
r
ativ
e
f
ilter
in
g
Dee
p
lear
n
in
g
E
co
m
m
er
ce
Par
am
eter
ized
B
E
R
T
Sen
tim
en
t a
n
aly
s
is
Similar
ity
m
etr
ics
W
eig
h
ted
ex
tr
em
e
g
r
ad
ien
t
b
o
o
s
tin
g
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
:
Sn
eh
al
B
h
o
g
an
Dep
ar
tm
en
t o
f
C
o
m
p
u
ter
Scie
n
ce
an
d
E
n
g
in
ee
r
in
g
,
Kar
n
ata
k
L
aw
So
ciety
Go
g
te
I
n
s
titu
te
o
f
T
ec
h
n
o
lo
g
y
af
f
iliated
to
Vis
v
esv
ar
ay
a
T
ec
h
n
o
lo
g
ical
U
n
iv
er
s
ity
,
B
elag
av
i,
I
n
d
ia
B
elag
av
i
-
5
9
0
0
1
8
,
I
n
d
ia
E
m
ail: sn
eh
alr
b
h
o
g
an
@
g
m
ail.
co
m
1.
I
NT
RO
D
UCT
I
O
N
W
ith
th
e
r
ap
id
ex
p
an
s
io
n
o
f
e
-
co
m
m
er
ce
,
s
o
cial
m
ed
ia,
an
d
o
n
lin
e
co
n
ten
t
p
latf
o
r
m
s
,
u
s
er
-
item
r
ec
o
m
m
en
d
atio
n
s
y
s
tem
s
h
av
e
b
ec
o
m
e
a
cr
u
cial
co
m
p
o
n
e
n
t
o
f
p
e
r
s
o
n
alize
d
d
ig
ital
ex
p
er
ien
ce
s
[
1
]
,
[
2
]
.
T
r
ad
itio
n
al
r
ec
o
m
m
en
d
atio
n
alg
o
r
ith
m
s
,
s
u
ch
as
co
llab
o
r
ativ
e
f
ilter
in
g
(
C
F)
an
d
co
n
ten
t
-
b
ased
f
ilter
in
g
(
C
B
F)
,
o
f
ten
s
u
f
f
er
f
r
o
m
ch
a
llen
g
es
lik
e
d
ata
s
p
ar
s
ity
,
co
l
d
-
s
tar
t
p
r
o
b
lem
s
,
an
d
an
in
ab
ilit
y
to
in
co
r
p
o
r
ate
co
n
tex
tu
al
an
d
s
en
tim
en
t
-
d
r
iv
en
in
s
ig
h
ts
[
3
]
,
[
4
]
.
Sen
tim
en
t
-
awa
r
e
r
ec
o
m
m
en
d
atio
n
s
y
s
tem
s
aim
to
b
r
id
g
e
th
is
g
ap
b
y
in
te
g
r
atin
g
u
s
er
s
’
em
o
tio
n
al
in
clin
atio
n
s
ex
tr
ac
t
ed
f
r
o
m
tex
t
u
al
r
ev
iews,
lea
d
in
g
to
m
o
r
e
ac
cu
r
ate
,
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
.
2
,
Ap
r
il 2
0
2
6
:
1
8
5
1
-
1
8
6
2
1852
an
d
p
er
s
o
n
alize
d
r
ec
o
m
m
en
d
atio
n
s
[
5
]
,
[
6
]
.
Natu
r
al
lan
g
u
ag
e
p
r
o
ce
s
s
in
g
(
NL
P)
tech
n
i
q
u
es
h
av
e
p
lay
ed
a
tr
an
s
f
o
r
m
ativ
e
r
o
le
in
im
p
r
o
v
i
n
g
s
en
tim
en
t
a
n
aly
s
is
f
o
r
r
ec
o
m
m
en
d
atio
n
s
y
s
tem
s
[
7
]
,
[
8
]
.
C
o
n
v
en
tio
n
al
ter
m
f
r
eq
u
e
n
cy
-
in
v
er
s
e
d
o
c
u
m
en
t
f
r
eq
u
en
c
y
(
T
F
-
I
DF)
m
et
h
o
d
s
[
9
]
,
[
1
0
]
o
f
ten
f
all
s
h
o
r
t
in
ca
p
tu
r
in
g
co
n
tex
t
u
al
m
ea
n
in
g
s
,
w
h
er
ea
s
ad
v
an
ce
d
d
ee
p
lear
n
in
g
m
o
d
els
[
1
1
]
,
[
1
2
]
s
u
ch
as
b
id
ir
ec
tio
n
al
en
co
d
er
r
e
p
r
esen
tatio
n
s
f
r
o
m
tr
an
s
f
o
r
m
er
s
(
B
E
R
T
)
[
1
2
]
,
[
1
3
]
h
a
v
e
s
ig
n
if
ican
tl
y
en
h
an
ce
d
s
en
tim
en
t
class
if
icatio
n
.
Ho
wev
er
,
o
p
tim
izin
g
th
ese
m
o
d
els
f
o
r
ef
f
ec
tiv
e
u
s
er
-
item
r
ec
o
m
m
en
d
atio
n
[
1
4
]
r
em
ain
s
a
ch
allen
g
e
d
u
e
t
o
co
m
p
u
tatio
n
al
c
o
m
p
lex
ity
an
d
th
e
n
ee
d
f
o
r
e
f
f
icien
t f
ea
tu
r
e
r
ep
r
esen
tatio
n
[
1
5
]
,
[
1
6
]
.
T
h
is
s
tu
d
y
p
r
o
p
o
s
es
a
n
o
v
el
s
en
tim
en
t
-
awa
r
e
u
s
er
-
item
r
ec
o
m
m
en
d
atio
n
s
y
s
tem
(
SAUI
R
S)
th
at
lev
er
ag
es
an
o
p
tim
ized
T
F
-
I
D
F
f
o
r
in
itial
f
ea
tu
r
e
e
x
tr
ac
tio
n
,
a
p
ar
am
eter
ized
B
E
R
T
(
P
-
B
E
R
T
)
f
o
r
en
h
an
ce
d
s
en
tim
en
t
class
if
icatio
n
,
an
d
a
weig
h
ted
ex
tr
em
e
g
r
ad
ien
t
b
o
o
s
tin
g
(
W
XGBo
o
s
t)
m
o
d
el
f
o
r
r
o
b
u
s
t
s
en
tim
en
t
p
r
ed
ictio
n
.
T
h
e
s
en
tim
en
t
in
f
o
r
m
atio
n
is
f
u
r
t
h
er
i
n
teg
r
ated
in
to
a
n
o
p
tim
ized
s
im
ilar
ity
m
etr
ics
m
o
d
el
to
r
ef
in
e
u
s
er
-
item
r
ec
o
m
m
en
d
atio
n
s
.
T
h
e
p
r
o
p
o
s
ed
h
y
b
r
id
m
o
d
el
ef
f
ec
tiv
ely
ca
p
t
u
r
es
u
s
er
p
r
ef
er
en
ce
s
,
en
h
an
ce
s
in
ter
p
r
etab
ilit
y
,
an
d
im
p
r
o
v
es
r
ec
o
m
m
e
n
d
atio
n
a
cc
u
r
ac
y
b
y
in
c
o
r
p
o
r
atin
g
b
o
t
h
s
en
tim
en
t
-
d
r
iv
en
an
d
co
n
ten
t
-
b
ased
f
ea
t
u
r
es.
T
h
e
p
r
im
ar
y
co
n
tr
ib
u
tio
n
s
o
f
th
is
r
esear
ch
in
clu
d
e:
i)
Dev
elo
p
m
en
t
o
f
an
o
p
tim
i
ze
d
T
F
-
I
DF
tech
n
iq
u
e
to
en
h
a
n
ce
f
ea
t
u
r
e
s
elec
tio
n
an
d
r
ed
u
ce
d
im
en
s
io
n
ality
.
ii)
I
m
p
lem
en
tatio
n
o
f
a
P
-
B
E
R
T
m
o
d
el
to
im
p
r
o
v
e
s
en
tim
en
t c
l
ass
if
icatio
n
ac
cu
r
ac
y
.
iii)
I
n
teg
r
atio
n
o
f
a
W
XGBo
o
s
t
m
o
d
el
to
en
h
an
ce
class
if
icatio
n
r
o
b
u
s
tn
ess
an
d
m
itig
ate
class
im
b
alan
ce
.
iv
)
Pro
p
o
s
al
o
f
a
n
o
p
tim
ize
d
s
im
ilar
ity
m
etr
ics
m
o
d
el
to
r
ef
in
e
u
s
er
-
item
r
ec
o
m
m
e
n
d
atio
n
s
u
s
in
g
s
en
tim
en
t
-
awa
r
e
in
s
ig
h
ts
.
v)
E
m
p
ir
ical
v
alid
atio
n
t
h
r
o
u
g
h
ex
ten
s
iv
e
ex
p
er
im
e
n
ts
d
em
o
n
s
tr
atin
g
im
p
r
o
v
e
d
ac
cu
r
ac
y
a
n
d
ef
f
icien
c
y
o
v
er
ex
is
tin
g
r
ec
o
m
m
e
n
d
atio
n
f
r
am
ewo
r
k
s
.
Ma
n
u
s
cr
ip
t
o
r
g
an
izatio
n
:
s
ec
tio
n
2
r
e
v
iews
ex
is
tin
g
s
en
tim
en
t
an
aly
s
is
–
b
ased
r
ec
o
m
m
en
d
atio
n
m
eth
o
d
s
u
s
ed
in
e
-
c
o
m
m
er
ce
p
latf
o
r
m
s
.
Sectio
n
3
p
r
esen
t
s
a
h
y
b
r
id
s
en
tim
en
t
a
n
aly
s
is
an
d
en
h
a
n
ce
d
CF
ap
p
r
o
ac
h
,
wh
ile
s
ec
tio
n
4
d
is
cu
s
s
es
ex
p
er
im
en
tal
r
esu
lts
an
d
co
m
p
ar
ativ
e
an
aly
s
is
,
f
o
llo
wed
b
y
co
n
clu
s
io
n
s
an
d
r
esear
ch
s
ig
n
if
ican
ce
.
2.
L
I
T
E
R
AT
U
RE
SU
RVE
Y
T
h
e
i
n
te
g
r
a
t
i
o
n
o
f
d
e
ep
l
e
ar
n
i
n
g
t
e
ch
n
iq
u
e
s
,
p
a
r
t
ic
u
l
ar
ly
l
a
r
g
e
l
a
n
g
u
ag
e
m
o
d
e
l
s
(
L
L
M
s
)
a
n
d
s
e
n
t
i
m
e
n
t
a
n
a
ly
s
i
s
,
h
a
s
s
i
g
n
i
f
i
c
a
n
t
l
y
a
d
v
a
n
ce
d
in
t
e
l
l
i
g
en
t
p
r
o
d
u
c
t
r
e
c
o
m
m
en
d
a
t
io
n
s
y
s
t
em
s
.
T
h
i
s
l
i
t
er
a
t
u
r
e
s
u
r
v
ey
r
e
v
i
e
w
s
r
e
c
en
t
s
tu
d
i
e
s
b
y
a
n
a
l
y
z
in
g
t
h
e
ir
m
e
t
h
o
d
o
lo
g
i
e
s
,
d
a
t
a
s
e
t
s
,
o
p
t
i
m
i
z
ed
m
et
r
i
c
s
,
co
n
tr
i
b
u
t
i
o
n
s
,
l
i
m
i
t
a
t
i
o
n
s
,
an
d
f
u
tu
r
e
r
e
s
e
ar
c
h
d
ir
e
c
t
io
n
s
.
T
h
o
m
a
s
an
d
J
eb
a
[
1
7
]
p
r
o
p
o
s
ed
a
b
i
g
r
a
m
-
b
a
s
e
d
d
ee
p
le
a
r
n
i
n
g
f
r
a
m
e
w
o
r
k
t
h
a
t
in
co
r
p
o
r
a
t
es
s
e
n
t
i
m
e
n
t
an
a
l
y
s
i
s
t
o
en
h
a
n
ce
p
r
o
d
u
c
t
r
ec
o
m
m
e
n
d
at
i
o
n
a
c
cu
r
a
c
y
.
B
y
e
x
t
r
a
c
t
in
g
s
e
n
t
im
en
t
s
co
r
e
s
f
r
o
m
u
s
er
r
e
v
ie
w
s
,
th
e
s
y
s
t
em
r
e
f
i
n
e
s
r
e
c
o
m
m
en
d
a
t
io
n
p
r
e
c
i
s
i
o
n
an
d
r
ec
a
l
l
.
T
h
e
m
o
d
e
l
w
a
s
tr
a
i
n
ed
a
n
d
v
a
l
i
d
a
te
d
u
s
in
g
e
-
c
o
m
m
er
c
e
d
a
t
a
s
e
t
s
c
o
n
s
i
s
t
i
n
g
o
f
u
s
er
r
ev
i
e
w
s
a
n
d
r
a
t
in
g
s
.
I
n
co
r
p
o
r
a
t
i
n
g
s
e
n
t
im
e
n
t
i
n
f
o
r
m
a
t
i
o
n
en
a
b
l
e
s
t
h
e
s
y
s
t
e
m
t
o
c
a
p
tu
r
e
n
u
a
n
c
e
d
u
s
e
r
p
r
e
f
e
r
e
n
c
e
s
,
r
e
s
u
l
t
in
g
i
n
i
m
p
r
o
v
ed
p
e
r
s
o
n
a
l
i
za
t
i
o
n
.
Ho
w
e
v
er
,
th
e
s
t
u
d
y
h
i
g
h
l
i
g
h
t
s
c
h
a
l
l
en
g
e
s
r
e
l
a
t
ed
to
l
ar
g
e
-
s
ca
l
e
d
a
ta
p
r
o
c
e
s
s
i
n
g
a
n
d
r
ec
o
m
m
e
n
d
s
f
u
tu
r
e
ex
p
l
o
r
a
t
i
o
n
o
f
e
f
f
i
c
i
en
t
a
l
g
o
r
i
t
h
m
s
an
d
r
e
a
l
-
t
im
e
p
r
o
c
e
s
s
i
n
g
m
e
ch
a
n
i
s
m
s
.
A
b
d
a
l
l
a
e
t
a
l.
[
1
8
]
i
n
tr
o
d
u
c
ed
a
h
y
b
r
i
d
r
e
c
o
m
m
en
d
a
t
io
n
m
o
d
e
l
co
m
b
i
n
in
g
s
e
l
f
-
a
t
t
en
t
io
n
m
e
ch
an
i
s
m
s
w
i
th
b
i
d
i
r
e
c
t
i
o
n
a
l
lo
n
g
s
h
o
r
t
-
te
r
m
m
e
m
o
r
y
(
B
i
-
L
S
T
M)
n
e
t
w
o
r
k
s
a
n
d
i
n
ce
n
t
iv
e
l
e
ar
n
in
g
-
b
a
s
e
d
CF
.
T
h
e
m
o
d
e
l
w
a
s
e
v
a
l
u
a
te
d
u
s
in
g
m
e
a
n
a
b
s
o
l
u
te
e
r
r
o
r
(
M
A
E
)
a
n
d
r
o
o
t
m
ea
n
s
q
u
a
r
e
e
r
r
o
r
(
R
M
S
E
)
m
e
t
r
ic
s
o
n
e
-
c
o
m
m
e
r
c
e
d
a
t
a
s
e
t
s
co
n
t
a
in
i
n
g
u
s
e
r
-
i
t
e
m
i
n
t
er
a
c
t
i
o
n
s
an
d
r
e
v
i
e
w
t
e
x
t
s
.
S
e
lf
-
a
t
te
n
t
io
n
e
n
a
b
le
s
e
f
f
e
c
t
i
v
e
c
o
n
t
ex
t
u
a
l
r
e
p
r
e
s
e
n
ta
t
i
o
n
,
i
m
p
r
o
v
i
n
g
r
e
co
m
m
e
n
d
a
t
i
o
n
a
c
cu
r
a
cy
.
N
e
v
e
r
th
e
l
e
s
s
,
th
e
m
o
d
e
l
’
s
a
r
ch
i
t
e
c
tu
r
a
l
c
o
m
p
l
e
x
i
ty
i
n
cr
e
a
s
e
s
co
m
p
u
ta
t
i
o
n
a
l
c
o
s
t
,
s
u
g
g
e
s
t
i
n
g
th
e
n
e
e
d
f
o
r
o
p
t
im
i
z
a
t
i
o
n
an
d
d
o
m
a
i
n
g
e
n
er
a
l
i
z
a
t
io
n
i
n
f
u
tu
r
e
r
e
s
e
a
r
ch
.
I
b
r
ah
i
m
e
t
a
l.
[
1
9
]
p
r
e
s
en
t
e
d
a
h
y
b
r
i
d
n
eu
r
a
l
CF
a
p
p
r
o
a
c
h
th
a
t
i
n
t
eg
r
a
t
e
s
n
e
u
r
a
l
n
e
t
w
o
r
k
s
w
i
t
h
t
r
ad
i
t
i
o
n
a
l
CF
t
e
c
h
n
iq
u
e
s
.
P
e
r
f
o
r
m
a
n
c
e
w
a
s
e
v
a
l
u
a
t
e
d
u
s
i
n
g
p
r
e
c
i
s
i
o
n
,
r
ec
a
l
l
,
an
d
F
1
-
s
c
o
r
e
o
n
p
u
b
l
i
c
ly
av
a
i
l
ab
l
e
d
a
t
a
s
e
ts
s
u
c
h
a
s
M
o
v
i
e
L
e
n
s
.
T
h
e
h
y
b
r
id
f
r
am
e
w
o
r
k
e
f
f
e
c
t
iv
e
l
y
m
o
d
e
l
s
b
o
t
h
l
i
n
ea
r
a
n
d
n
o
n
l
in
e
a
r
u
s
er
-
i
t
em
in
t
e
r
a
c
t
i
o
n
s
,
i
m
p
r
o
v
i
n
g
r
e
c
o
m
m
e
n
d
a
t
io
n
a
c
cu
r
a
cy
.
H
o
w
e
v
e
r
,
p
e
r
f
o
r
m
a
n
c
e
d
e
g
r
a
d
a
t
io
n
u
n
d
e
r
s
p
ar
s
e
d
a
t
a
c
o
n
d
i
t
io
n
s
r
em
a
i
n
s
a
c
h
a
l
l
e
n
g
e
,
m
o
t
i
v
a
t
i
n
g
f
u
tu
r
e
w
o
r
k
o
n
d
a
t
a
a
u
g
m
e
n
t
a
t
io
n
a
n
d
c
o
n
t
e
x
tu
a
l
f
e
a
t
u
r
e
i
n
te
g
r
a
t
i
o
n
.
S
a
m
i
et
a
l
.
[
2
0
]
p
r
o
p
o
s
e
d
a
h
y
b
r
id
r
e
co
m
m
e
n
d
a
t
i
o
n
m
o
d
e
l
t
h
a
t
co
m
b
i
n
e
s
d
e
ep
l
e
a
r
n
i
n
g
w
i
t
h
tr
a
d
i
t
io
n
a
l
r
e
c
o
m
m
en
d
a
t
i
o
n
a
l
g
o
r
i
th
m
s
t
o
en
h
an
c
e
p
e
r
s
o
n
a
l
i
z
a
t
io
n
.
T
h
e
s
y
s
t
e
m
w
a
s
e
v
a
l
u
a
t
e
d
u
s
i
n
g
a
c
cu
r
a
cy
a
n
d
co
v
er
a
g
e
m
e
tr
i
c
s
o
n
d
a
t
a
s
e
t
s
c
o
n
t
a
in
i
n
g
u
s
e
r
i
n
t
er
a
c
t
io
n
l
o
g
s
an
d
p
r
o
f
i
l
e
in
f
o
r
m
a
t
i
o
n
f
r
o
m
o
n
l
i
n
e
p
l
a
t
f
o
r
m
s
.
W
h
i
l
e
th
e
h
y
b
r
id
a
p
p
r
o
a
ch
i
m
p
r
o
v
e
s
r
e
c
o
m
m
en
d
a
t
io
n
q
u
a
l
i
t
y
,
t
h
e
s
t
u
d
y
e
m
p
h
a
s
i
z
e
s
t
h
e
n
ee
d
f
o
r
s
c
a
l
ab
l
e
ar
c
h
it
e
c
t
u
r
e
s
an
d
r
e
a
l
-
t
i
m
e
d
a
t
a
p
r
o
c
e
s
s
i
n
g
f
o
r
p
r
a
c
t
i
ca
l
d
e
p
lo
y
m
e
n
t
.
Gu
o
a
n
d
Z
h
an
g
[
2
1
]
em
p
lo
y
ed
d
e
ep
l
e
a
r
n
i
n
g
-
b
as
e
d
em
b
ed
d
in
g
s
t
o
p
r
ed
i
c
t
c
u
s
to
m
er
s
a
t
i
s
f
a
c
t
io
n
i
n
c
r
o
s
s
-
b
o
r
d
er
e
-
c
o
m
m
e
r
c
e
en
v
ir
o
n
m
en
t
s
.
T
h
e
m
o
d
el
o
p
t
im
i
z
ed
c
l
a
s
s
i
f
i
c
a
t
i
o
n
a
c
cu
r
a
c
y
a
n
d
F1
-
s
c
o
r
e
u
s
i
n
g
d
a
ta
s
e
t
s
c
o
m
p
r
i
s
i
n
g
cu
s
t
o
m
e
r
r
e
v
ie
w
s
a
n
d
s
a
t
i
s
f
a
c
t
i
o
n
r
a
t
i
n
g
s
.
A
c
c
u
r
a
te
s
a
t
i
s
f
a
c
t
i
o
n
p
r
ed
i
c
ti
o
n
s
u
p
p
o
r
t
s
s
t
r
a
t
eg
i
c
d
e
c
i
s
i
o
n
-
m
a
k
in
g
a
n
d
c
u
s
t
o
m
e
r
r
e
te
n
t
i
o
n
.
H
o
we
v
er
,
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
S
en
timen
t
-
a
w
a
r
e
u
s
er
-
item
r
e
co
mme
n
d
a
tio
n
c
o
mb
in
in
g
w
eig
h
ted
X
GB
o
o
s
t a
n
d
…
(
S
n
e
h
a
l
B
h
o
g
a
n
)
1853
p
o
t
en
t
i
a
l
b
i
a
s
e
s
i
n
cu
s
t
o
m
e
r
f
e
ed
b
ac
k
l
i
m
i
t
m
o
d
e
l
g
e
n
er
a
l
i
z
ab
i
l
i
t
y
,
a
n
d
t
h
e
au
t
h
o
r
s
r
e
co
m
m
e
n
d
i
n
c
o
r
p
o
r
a
t
in
g
d
iv
e
r
s
e
d
a
t
a
s
o
u
r
c
e
s
in
f
u
tu
r
e
s
tu
d
i
e
s
.
Y
a
o
an
d
Z
h
en
g
[
2
2
]
p
r
o
p
o
s
ed
an
en
h
an
c
ed
s
e
n
t
im
e
n
t
a
n
a
l
y
s
i
s
m
o
d
e
l
co
m
b
in
i
n
g
tr
a
n
s
f
o
r
m
e
r
s
w
i
th
c
o
n
d
i
t
io
n
al
r
a
n
d
o
m
f
ie
l
d
s
(
C
R
F
s
)
t
o
c
a
p
t
u
r
e
co
n
te
x
tu
a
l
d
e
p
en
d
e
n
c
i
e
s
in
t
ex
t
.
E
v
a
lu
a
te
d
o
n
Y
e
lp
a
n
d
A
m
a
z
o
n
r
e
v
ie
w
d
a
t
a
s
e
t
s
,
th
e
m
o
d
e
l
a
ch
i
ev
e
d
an
a
cc
u
r
a
c
y
o
f
9
3
.
4
%
.
W
h
i
l
e
t
h
e
tr
a
n
s
f
o
r
m
e
r
-
C
R
F
in
t
e
g
r
a
t
i
o
n
i
m
p
r
o
v
e
s
l
in
g
u
i
s
t
i
c
u
n
d
e
r
s
t
an
d
in
g
,
i
t
s
c
o
m
p
u
t
a
t
i
o
n
al
c
o
m
p
l
e
x
i
ty
m
o
t
i
v
a
t
e
s
f
u
t
u
r
e
r
e
s
e
a
r
c
h
o
n
m
o
d
e
l
o
p
t
i
m
i
za
t
i
o
n
a
n
d
r
e
a
l
-
t
i
m
e
a
p
p
l
i
c
a
b
i
l
i
ty
.
R
o
y
e
t
a
l.
[
2
3
]
in
t
r
o
d
u
c
ed
a
h
y
b
r
id
r
e
co
m
m
e
n
d
a
t
i
o
n
s
y
s
t
e
m
i
n
t
eg
r
a
t
in
g
co
n
te
n
t
-
b
a
s
e
d
a
n
d
i
t
e
m
-
b
a
s
e
d
CF
w
i
t
h
c
a
s
c
ad
ed
L
L
M
s
.
T
h
e
m
o
d
e
l
p
r
o
c
e
s
s
e
s
p
r
o
d
u
c
t
m
e
t
ad
a
t
a
an
d
u
s
e
r
r
e
v
i
e
w
s
t
o
en
h
an
c
e
r
e
c
o
m
m
en
d
a
t
io
n
ac
c
u
r
a
c
y
,
p
r
e
c
i
s
i
o
n
,
an
d
r
e
c
a
l
l.
E
x
p
e
r
im
en
t
s
c
o
n
d
u
c
te
d
o
n
o
n
l
i
n
e
a
p
p
ar
e
l
r
e
t
a
i
l
d
a
t
a
s
e
t
s
,
p
a
r
t
i
cu
l
a
r
ly
m
e
n
’
s
s
h
i
r
t
ca
t
eg
o
r
ie
s
,
d
e
m
o
n
s
t
r
a
t
e
i
m
p
r
o
v
ed
p
er
s
o
n
a
l
i
z
a
t
io
n
th
r
o
u
g
h
n
u
a
n
ce
d
p
r
ef
e
r
en
c
e
m
o
d
e
l
i
n
g
.
H
o
we
v
er
,
s
c
a
l
ab
i
l
i
t
y
a
n
d
r
e
a
l
-
t
im
e
p
r
o
ce
s
s
i
n
g
r
e
m
a
in
o
p
en
c
h
a
l
l
e
n
g
e
s
.
G
a
r
ap
a
t
i
an
d
C
h
a
k
r
ab
o
r
t
y
[
2
4
]
p
r
o
p
o
s
ed
th
e
r
e
v
i
e
w
t
e
x
t
g
r
a
n
u
l
a
r
i
ty
(
R
T
G
)
m
o
d
e
l
t
o
im
p
r
o
v
e
s
en
t
i
m
e
n
t
a
n
a
ly
s
i
s
a
n
d
r
a
t
i
n
g
p
r
e
d
i
c
t
io
n
b
y
a
n
a
ly
z
in
g
r
ev
i
e
w
t
ex
t
s
a
t
v
ar
y
in
g
le
v
el
s
o
f
d
e
ta
i
l
.
E
v
a
lu
a
t
e
d
u
s
i
n
g
M
A
E
a
n
d
R
M
S
E
o
n
e
-
c
o
m
m
er
c
e
d
a
t
a
s
e
t
s
,
t
h
e
m
o
d
e
l
ca
p
tu
r
e
s
s
u
b
t
l
e
o
p
i
n
io
n
n
u
a
n
ce
s
.
D
e
s
p
it
e
i
t
s
e
f
f
e
c
t
i
v
e
n
e
s
s
,
c
o
m
p
u
t
a
t
i
o
n
a
l
c
o
m
p
l
e
x
i
ty
r
e
m
a
i
n
s
a
l
i
m
i
t
a
ti
o
n
,
h
i
g
h
l
i
g
h
t
i
n
g
t
h
e
n
e
e
d
f
o
r
e
f
f
i
c
ie
n
t
ar
c
h
i
t
ec
t
u
r
e
s
a
n
d
c
o
n
t
ex
t
u
a
l
f
e
a
tu
r
e
i
n
t
e
g
r
a
t
i
o
n
.
Y
an
g
e
t
a
l.
[
2
5
]
d
e
v
e
lo
p
e
d
an
a
t
t
en
t
i
v
e
a
s
p
ec
t
-
b
a
s
ed
r
e
co
m
m
e
n
d
a
t
i
o
n
n
e
tw
o
r
k
(
A
A
R
N
)
th
a
t
e
m
p
lo
y
s
a
t
t
e
n
t
io
n
m
e
c
h
an
i
s
m
s
t
o
c
ap
t
u
r
e
a
s
p
e
ct
-
l
ev
e
l
u
s
er
p
r
ef
e
r
e
n
c
e
s
f
r
o
m
r
ev
i
e
w
s
.
T
h
e
m
o
d
e
l
o
p
t
i
m
iz
e
d
p
r
e
c
i
s
i
o
n
,
r
e
c
a
l
l
,
a
n
d
F
1
-
s
co
r
e
u
s
i
n
g
an
n
o
t
at
e
d
e
-
co
m
m
e
r
c
e
r
e
v
i
e
w
d
a
ta
s
e
t
s
.
A
s
p
e
ct
-
l
ev
e
l
m
o
d
e
l
i
n
g
e
n
ab
l
e
s
m
o
r
e
p
e
r
s
o
n
a
l
i
z
ed
r
e
co
m
m
e
n
d
a
t
i
o
n
s
,
t
h
o
u
g
h
ac
c
u
r
a
t
e
ly
e
x
tr
a
c
t
in
g
an
d
w
e
ig
h
t
in
g
a
s
p
e
c
t
s
r
em
a
in
s
c
h
a
l
l
e
n
g
i
n
g
.
F
u
t
u
r
e
wo
r
k
s
u
g
g
e
s
t
s
r
e
f
in
i
n
g
a
s
p
e
c
t
e
x
tr
a
c
t
io
n
a
n
d
in
t
e
g
r
a
t
in
g
e
x
te
r
n
a
l
k
n
o
w
le
d
g
e
s
o
u
r
ce
s
.
I
n
s
u
m
m
ar
y
,
th
ese
s
tu
d
ies
d
e
m
o
n
s
tr
ate
th
at
in
co
r
p
o
r
atin
g
a
d
v
an
ce
d
NL
P
tech
n
iq
u
es
s
u
ch
as
L
L
Ms,
atten
tio
n
m
ec
h
an
is
m
s
,
an
d
asp
ec
t
-
lev
el
s
en
tim
en
t
m
o
d
elin
g
s
ig
n
if
ican
tly
im
p
r
o
v
es
r
ec
o
m
m
en
d
atio
n
ac
c
u
r
ac
y
an
d
p
er
s
o
n
aliza
tio
n
.
Ho
we
v
er
,
ch
allen
g
es
r
elate
d
to
d
ata
im
b
alan
ce
,
m
u
lti
-
d
o
m
ain
e
-
co
m
m
e
r
ce
en
v
ir
o
n
m
en
ts
,
co
m
p
u
tatio
n
al
co
m
p
lex
ity
,
a
n
d
r
ea
l
-
tim
e
p
r
o
ce
s
s
in
g
p
er
s
is
t.
T
h
ese
lim
it
atio
n
s
m
o
tiv
ate
th
e
d
ev
elo
p
m
e
n
t o
f
t
h
e
p
r
o
p
o
s
ed
SAUR
I
S f
r
am
ewo
r
k
d
is
cu
s
s
ed
in
th
e
s
u
b
s
eq
u
e
n
t sectio
n
.
3.
P
RO
P
O
SE
D
M
E
T
H
O
D
T
h
e
SAUI
R
S
p
r
esen
ted
in
Fig
u
r
e
1
is
d
esig
n
ed
co
n
s
id
er
in
g
r
ec
o
m
m
en
d
i
n
g
T
o
p
N
item
s
t
o
u
s
er
s
an
d
id
en
tify
in
g
u
s
er
s
with
s
im
ilar
in
ter
ests
.
T
h
e
SAUI
R
S
f
ir
s
t
co
llects
r
ea
lis
tic
d
ata
f
r
o
m
th
e
Am
az
o
n
web
s
ite,
co
n
s
id
er
in
g
m
u
ltip
le
p
r
o
d
u
cts
o
f
d
if
f
er
en
t
d
o
m
ain
s
,
wh
ich
in
clu
d
e
co
s
m
etics,
f
o
o
d
,
a
n
d
elec
tr
o
n
ics.
Mo
r
e
d
etails
o
f
th
e
d
atasets
an
d
th
eir
attr
ib
u
te
ar
e
p
r
o
v
id
e
d
i
n
s
u
b
-
s
ec
tio
n
3
.
1
.
T
h
en
,
th
e
SAUI
R
S
p
er
f
o
r
m
s
s
en
tim
en
t
an
aly
s
is
u
s
in
g
o
p
ti
m
ized
TF
-
I
DF
an
d
a
tr
an
s
f
o
r
m
er
m
o
d
el
em
p
lo
y
in
g
a
P
-
B
E
R
T
m
o
d
el.
Mo
r
e
d
etails
o
f
th
e
p
r
o
ce
s
s
ar
e
d
is
c
u
s
s
ed
in
s
u
b
-
s
ec
tio
n
s
3
.
2
an
d
3
.
3
.
T
h
en
,
in
tr
o
d
u
ce
s
W
XGBo
o
s
t
to
p
er
f
o
r
m
t
h
e
s
en
tim
en
t
class
if
icatio
n
task
an
d
o
p
tim
ized
s
im
ilar
ity
m
etr
ics
.
A
m
o
d
el
is
in
tr
o
d
u
ce
d
to
p
er
f
o
r
m
T
o
p
N
r
ec
o
m
m
en
d
atio
n
s
.
Mo
r
e
d
etails o
f
th
is
ar
e
p
r
esen
ted
in
s
u
b
-
s
ec
tio
n
s
3
.
4
an
d
3
.
5
.
Fig
u
r
e
1
.
Ar
c
h
itectu
r
e
o
f
SAUI
R
S
3
.
1
.
Da
t
a
s
et
a
nd
prepro
ce
s
s
i
ng
T
h
e
s
tu
d
y
u
s
e
s
t
h
e
A
m
az
o
n
d
a
t
a
s
e
t
w
h
i
ch
s
p
a
n
s
m
u
l
t
ip
l
e
p
r
o
d
u
c
t
c
a
t
eg
o
r
i
e
s
l
i
k
e
p
r
o
d
u
c
t
r
ev
i
e
w
s
,
l
a
p
t
o
p
r
e
v
i
e
w
s
,
an
d
t
a
b
l
e
t
r
e
v
i
e
w
s
[
2
4
]
,
[
2
5
]
,
m
a
i
n
t
a
in
ed
b
y
th
e
Un
i
v
e
r
s
i
t
y
o
f
C
a
l
if
o
r
n
i
a
S
an
D
ie
g
o
.
D
u
e
t
o
t
h
e
d
a
ta
s
e
t
’
s
l
a
r
g
e
s
i
ze
[
2
5
]
,
a
s
u
b
s
e
t
o
f
3
0
,
0
0
0
r
o
w
s
h
a
v
in
g
m
u
l
t
ip
l
e
ca
t
e
g
o
r
i
es
w
e
r
e
ex
t
r
a
c
te
d
t
o
t
h
e
m
en
’
s
s
h
i
r
t
c
a
t
e
g
o
r
y
u
n
d
e
r
c
l
o
th
i
n
g
,
s
h
o
e
s
,
an
d
j
e
w
e
lr
y
w
e
r
e
s
e
l
e
c
t
ed
r
an
d
o
m
l
y
f
o
r
c
r
e
a
t
i
n
g
a
n
o
v
e
l
d
a
t
a
s
e
t
f
o
r
tr
a
i
n
i
n
g
,
te
s
t
i
n
g
,
an
d
v
a
l
id
a
t
io
n
.
F
u
r
t
h
er
,
t
h
e
Am
a
z
o
n
la
p
to
p
a
n
d
ta
b
l
e
t
d
a
t
as
e
t
s
[
2
5
]
w
e
r
e
a
l
s
o
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
.
2
,
Ap
r
il 2
0
2
6
:
1
8
5
1
-
1
8
6
2
1854
c
o
l
l
e
c
t
ed
f
r
o
m
th
e
p
u
b
l
i
c
l
y
r
e
l
e
a
s
ed
A
m
a
z
o
n
p
r
o
d
u
c
t
r
e
v
i
e
w
s
2
(
5
-
c
o
r
e
)
c
o
l
l
ec
t
i
o
n
,
wh
i
c
h
e
n
s
u
r
e
s
t
h
a
t
ea
c
h
u
s
er
an
d
e
a
ch
i
t
em
h
a
v
e
a
t
l
e
a
s
t
f
i
v
e
a
s
s
o
c
i
a
t
ed
r
e
v
i
e
w
s
.
F
r
o
m
th
e
e
l
e
c
tr
o
n
i
cs
c
a
t
e
g
o
r
y
,
l
a
p
to
p
-
a
n
d
t
a
b
le
t
-
r
e
l
a
t
ed
en
t
r
i
e
s
w
e
r
e
s
e
l
e
c
t
ed
,
p
r
e
p
r
o
c
e
s
s
ed
,
a
n
d
s
u
b
s
eq
u
en
t
l
y
d
iv
i
d
e
d
i
n
to
tr
a
i
n
i
n
g
,
v
a
l
id
a
t
i
o
n
,
a
n
d
t
e
s
t
p
a
r
t
i
t
io
n
s
f
o
l
l
o
w
in
g
a
n
8
0
:2
0
s
p
l
i
t.
O
n
l
y
r
el
e
v
a
n
t
f
i
e
l
d
s
n
e
c
e
s
s
a
r
y
f
o
r
c
o
n
s
t
r
u
c
t
i
n
g
t
h
e
s
e
n
t
i
m
e
n
t
-
b
a
s
e
d
r
ec
o
m
m
en
d
a
t
i
o
n
s
y
s
t
em
w
e
r
e
r
e
t
a
in
e
d
f
r
o
m
th
e
o
r
i
g
in
a
l
d
a
t
a
s
e
t
.
T
h
e
s
e
f
i
e
l
d
s
i
n
c
l
u
d
e
:
I
d
(
a
s
in
)
:
a
u
n
i
q
u
e
i
d
e
n
t
if
i
e
r
f
o
r
ea
c
h
p
r
o
d
u
c
t
,
B
r
a
n
d
(
d
e
ta
i
l
s
)
:
d
e
t
a
i
l
s
o
f
t
h
e
p
r
o
d
u
c
t
’
s
b
r
a
n
d
.
C
a
t
eg
o
r
ie
s
(
m
a
i
n
_
c
a
t
eg
o
r
y
)
:
th
e
p
r
i
m
a
r
y
c
a
t
e
g
o
r
y
o
f
th
e
p
r
o
d
u
c
t
.
M
an
u
f
a
c
tu
r
er
n
a
m
e
(
M
an
u
f
a
c
tu
r
er
)
:
th
e
n
a
m
e
o
f
th
e
m
a
n
u
f
a
c
tu
r
er
.
R
ev
i
e
w
s
_
d
a
t
e
(
t
i
m
e
s
t
a
m
p
)
:
t
h
e
d
a
t
e
wh
en
t
h
e
r
e
v
i
e
w
w
a
s
p
o
s
t
e
d
.
R
e
v
i
e
w
s
_
d
id
P
u
r
c
h
a
s
e
(
v
e
r
i
f
i
ed
_
p
u
r
c
h
a
s
e)
:
in
d
i
c
at
e
s
w
h
e
t
h
e
r
th
e
p
r
o
d
u
c
t
w
a
s
p
u
r
ch
a
s
ed
b
y
t
h
e
r
ev
i
e
w
e
r
.
R
e
v
i
e
w
s
_
d
o
R
e
c
o
m
m
e
n
d
(
b
o
u
g
h
t_
to
g
e
th
e
r
)
:
r
ec
o
m
m
e
n
d
ed
b
u
n
d
l
e
s
f
r
o
m
t
h
e
we
b
s
i
t
es
.
R
e
v
i
e
w
s
_
r
a
t
in
g
(
r
a
t
in
g
)
:
n
u
m
e
r
i
c
a
l
r
a
t
i
n
g
s
g
iv
e
n
b
y
u
s
e
r
s
.
R
e
v
i
e
w
s
_
te
x
t
(
t
e
x
t
)
:
t
h
e
t
ex
t
u
a
l
r
e
v
i
e
w
c
o
n
te
n
t
.
R
e
v
i
e
w
s
_
t
i
t
l
e
(
t
i
t
l
e
)
:
t
i
t
l
e
s
o
f
t
h
e
r
ev
i
ew
s
.
R
e
v
i
e
w
s
_
u
s
e
r
n
a
m
e
(
r
ev
i
e
w
e
r
N
am
e
)
:
th
e
u
s
er
n
am
e
o
f
th
e
r
ev
i
e
w
er
.
U
s
e
r
_
s
e
n
t
im
e
n
t
:
t
h
e
m
an
u
a
ll
y
an
n
o
t
a
t
e
d
s
e
n
t
i
m
e
n
t
la
b
el
f
o
r
e
ac
h
r
e
v
ie
w
(
p
o
s
i
t
iv
e
o
r
n
eg
a
t
i
v
e
)
.
T
h
e
p
r
ep
r
o
c
e
s
s
i
n
g
s
t
e
p
e
n
s
u
r
e
d
th
e
d
a
t
a
s
e
t
i
s
c
le
a
n
a
n
d
s
t
r
u
c
t
u
r
ed
f
o
r
a
n
a
l
y
s
i
s
.
K
ey
t
a
s
k
s
i
n
c
lu
d
ed
ar
e
a
s
f
o
l
l
o
w
s
:
h
an
d
l
in
g
m
i
s
s
i
n
g
d
a
ta
:
n
u
l
l
v
a
lu
e
s
in
t
h
e
d
a
ta
s
e
t
w
e
r
e
ad
d
r
e
s
s
e
d
b
y
r
e
m
o
v
in
g
t
h
a
t
p
a
r
t
i
cu
l
a
r
r
o
w.
D
a
t
a
c
l
e
a
n
in
g
(
r
em
o
v
i
n
g
u
n
n
e
c
e
s
s
a
r
y
co
l
u
m
n
s
)
:
i
r
r
e
l
ev
a
n
t
d
a
t
a
c
o
lu
m
n
s
w
e
r
e
r
e
m
o
v
ed
.
O
n
l
y
t
h
e
e
s
s
e
n
t
i
a
l
c
o
l
u
m
n
s
f
r
o
m
t
h
e
d
a
t
a
s
e
t
w
e
r
e
r
e
t
a
in
e
d
,
b
a
s
ed
o
n
ex
p
l
o
r
a
t
o
r
y
d
a
ta
a
n
a
ly
s
i
s
(
E
DA
)
.
M
o
r
e
o
v
e
r
,
th
e
t
e
x
t
-
b
a
s
e
d
r
ev
i
e
w
d
a
t
a
u
n
d
er
w
e
n
t
th
e
f
o
l
lo
w
i
n
g
s
t
e
p
s
:
p
u
n
c
t
u
a
t
io
n
r
em
o
v
a
l
:
n
o
n
-
a
lp
h
an
u
m
e
r
i
c
c
h
ar
a
c
t
er
s
w
e
r
e
e
l
im
i
n
a
te
d
.
S
t
o
p
w
o
r
d
r
e
m
o
v
a
l
:
c
o
m
m
o
n
wo
r
d
s
(
e.
g
.
,
“
an
d
”
a
n
d
“
th
e
”
)
th
a
t
d
i
d
n
o
t
c
o
n
t
r
ib
u
t
e
t
o
th
e
s
e
n
t
im
e
n
t
w
er
e
f
i
l
t
er
ed
o
u
t.
L
e
m
m
a
t
i
z
a
t
io
n
:
w
o
r
d
s
w
e
r
e
r
e
d
u
c
ed
t
o
t
h
e
i
r
r
o
o
t
f
o
r
m
to
s
t
an
d
ar
d
iz
e
t
e
x
t
u
a
l
c
o
n
t
en
t
.
3
.
2
.
F
e
a
t
ure
ex
t
r
a
ct
io
n
T
o
m
ak
e
th
e
clea
n
ed
d
ataset
s
u
itab
le
f
o
r
m
ac
h
in
e
lear
n
i
n
g
,
f
ea
tu
r
e
ex
tr
ac
tio
n
was c
ar
r
ied
o
u
t u
s
in
g
a
n
ewly
in
tr
o
d
u
ce
d
o
p
tim
ized
t
er
m
f
r
e
q
u
en
c
y
-
in
v
e
r
s
e
d
o
cu
m
en
t
f
r
eq
u
en
cy
(O
-
TF
-
I
DF)
m
eth
o
d
.
T
F
-
I
DF
is
a
s
tatis
t
ical
tech
n
iq
u
e
th
at
r
ef
le
cts
th
e
r
elev
an
ce
o
f
a
te
r
m
with
in
a
s
p
ec
if
ic
d
o
cu
m
en
t
r
elativ
e
to
an
en
tir
e
co
llectio
n
.
I
t
is
co
m
m
o
n
l
y
a
p
p
lied
in
in
f
o
r
m
atio
n
r
etr
ie
v
al
an
d
te
x
t
an
al
y
tics
to
tr
a
n
s
f
o
r
m
tex
tu
al
in
p
u
t
in
to
m
ea
n
in
g
f
u
l
n
u
m
er
ical
r
e
p
r
esen
tatio
n
s
th
at
ca
n
b
e
u
tili
ze
d
b
y
lear
n
in
g
alg
o
r
ith
m
s
.
T
e
r
m
f
r
e
q
u
en
c
y
(
T
F
)
r
ef
lects
h
o
w
o
f
ten
a
p
a
r
ticu
lar
wo
r
d
o
cc
u
r
s
with
in
a
d
o
cu
m
en
t
co
m
p
ar
e
d
with
th
e
to
tal
n
u
m
b
er
o
f
wo
r
d
s
in
th
at
d
o
cu
m
e
n
t a
s
o
b
tain
ed
in
(
1
)
.
(
)
=
ℎ
(
1
)
I
n
co
n
tr
ast,
in
v
er
s
e
d
o
cu
m
en
t
f
r
eq
u
e
n
cy
(
I
DF)
in
d
icate
s
th
e
s
ig
n
if
ican
ce
o
f
a
ter
m
b
y
lo
wer
in
g
th
e
in
f
lu
en
c
e
o
f
wo
r
d
s
th
at
a
p
p
ea
r
f
r
eq
u
en
tl
y
ac
r
o
s
s
m
an
y
d
o
cu
m
en
ts
,
s
u
c
h
as c
o
m
m
o
n
s
to
p
wo
r
d
s
as in
(
2
)
.
(
)
=
l
og
(
)
(
2
)
T
h
e
T
F
-
I
DF v
alu
e
is
o
b
tain
e
d
b
y
m
u
ltip
ly
i
n
g
th
ese
two
co
m
p
o
n
en
ts
as in
(
3
)
.
−
(
)
=
(
)
×
(
)
(
3
)
T
h
is
co
m
b
in
ed
m
ea
s
u
r
e
em
p
h
asizes
ter
m
s
th
at
ar
e
m
o
r
e
d
is
tin
ctiv
e
o
r
m
ea
n
in
g
f
u
l
with
in
a
s
p
ec
if
ic
d
o
cu
m
e
n
t
wh
ile
d
o
wn
p
lay
i
n
g
wo
r
d
s
th
at
ar
e
co
m
m
o
n
ly
f
o
u
n
d
th
r
o
u
g
h
o
u
t
th
e
en
tir
e
co
r
p
u
s
.
Alth
o
u
g
h
TF
-
I
DF
r
em
ain
s
a
wid
ely
u
s
ed
an
d
ef
f
ec
tiv
e
m
eth
o
d
f
o
r
tr
an
s
f
o
r
m
in
g
tex
t
i
n
to
n
u
m
er
i
ca
l
f
ea
tu
r
es,
it
s
till
p
r
esen
ts
ce
r
tain
s
h
o
r
tco
m
in
g
s
th
at
m
ay
lim
it
it
s
ef
f
ec
tiv
en
ess
f
o
r
s
p
ec
if
ic
ap
p
licatio
n
s
.
T
h
e
m
ain
p
r
o
b
lem
id
en
tifie
d
d
u
r
in
g
r
ec
o
m
m
e
n
d
a
tio
n
s
y
s
tem
d
esig
n
u
s
in
g
s
en
ti
m
en
t a
n
aly
s
is
ar
e
as f
o
llo
ws:
i)
TF
-
I
DF
tr
ea
ts
wo
r
d
s
o
r
n
-
g
r
am
s
as
in
d
ep
en
d
en
t
f
ea
tu
r
es
an
d
d
o
es
n
o
t
co
n
s
id
er
th
e
s
em
an
tic
o
r
s
y
n
tactic
r
elatio
n
s
h
ip
s
b
etwe
en
th
em
.
Hen
ce
,
d
u
e
to
th
is
it
ca
n
n
o
t
u
n
d
er
s
tan
d
th
e
m
ea
n
in
g
o
f
wo
r
d
s
o
r
p
h
r
ases
,
lead
in
g
to
a
lack
o
f
co
n
tex
t
in
f
ea
tu
r
e
r
ep
r
esen
tatio
n
.
Fo
r
ex
am
p
le,
“
n
o
t
b
ad
”
an
d
“
b
ad
”
m
ay
b
e
tr
ea
ted
s
im
ilar
ly
d
esp
ite
h
av
in
g
o
p
p
o
s
ite
s
en
tim
en
ts
.
ii)
T
h
e
f
ea
tu
r
e
s
p
ac
e
g
r
o
ws
r
ap
id
ly
with
th
e
s
ize
o
f
th
e
v
o
ca
b
u
l
ar
y
an
d
th
e
u
s
e
o
f
n
-
g
r
am
s
(
e.
g
.
,
b
ig
r
a
m
s
o
r
tr
ig
r
am
s
)
.
Hen
ce
,
r
esu
ltin
g
i
n
lar
g
e,
s
p
ar
s
e
m
atr
ices,
in
cr
ea
s
in
g
co
m
p
u
tatio
n
al
co
s
ts
an
d
m
em
o
r
y
r
eq
u
ir
em
e
n
ts
.
I
t a
ls
o
m
ak
es m
o
d
els p
r
o
n
e
to
o
v
er
f
itti
n
g
,
esp
ec
ially
with
s
m
all
d
atasets
.
iii)
W
o
r
d
s
th
at
ar
e
v
er
y
co
m
m
o
n
ac
r
o
s
s
d
o
cu
m
e
n
ts
(
e.
g
.
,
“
p
r
o
d
u
ct
”
an
d
“
r
ev
iew
”
)
m
ay
s
till
h
av
e
n
o
n
-
ze
r
o
weig
h
ts
d
u
e
to
ter
m
f
r
eq
u
en
cy
,
ev
en
if
t
h
ey
ad
d
litt
le
v
alu
e.
T
h
is
ca
n
in
tr
o
d
u
ce
n
o
is
e
in
to
t
h
e
f
ea
tu
r
e
s
et.
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
S
en
timen
t
-
a
w
a
r
e
u
s
er
-
item
r
e
co
mme
n
d
a
tio
n
c
o
mb
in
in
g
w
eig
h
ted
X
GB
o
o
s
t a
n
d
…
(
S
n
e
h
a
l
B
h
o
g
a
n
)
1855
iv
)
R
ar
e
b
u
t
m
ea
n
in
g
f
u
l
ter
m
s
(
e.
g
.
,
“
awe
s
o
m
e
”
an
d
“
d
is
astro
u
s
”
)
m
ay
b
e
d
is
ca
r
d
e
d
if
th
e
m
i
n
ed
th
r
esh
o
l
d
is
to
o
h
ig
h
.
T
h
is
ca
n
lead
to
a
lo
s
s
o
f
cr
itical
in
f
o
r
m
atio
n
.
v)
T
h
e
p
e
r
f
o
r
m
an
ce
o
f
T
F
-
I
DF
h
ea
v
ily
d
ep
en
d
s
o
n
th
e
s
elec
tio
n
o
f
p
ar
am
eter
s
lik
e
m
in
d
s
,
m
ax
_
d
f
,
an
d
n
g
r
am
_
r
an
g
e
.
Po
o
r
p
ar
am
ete
r
ch
o
ices
ca
n
r
esu
lt
in
ir
r
ele
v
an
t
f
ea
tu
r
es
b
ei
n
g
in
cl
u
d
ed
o
r
s
ig
n
if
ican
t
f
ea
tu
r
es b
ein
g
e
x
clu
d
e
d
.
Fo
r
s
o
lv
in
g
th
e
ab
o
v
e
is
s
u
es,
t
h
e
O
-
TF
-
I
DF
h
as
b
ee
n
en
h
an
ce
d
b
y
s
elec
tin
g
th
e
m
in
_
d
f
,
m
ax
_
d
f
an
d
n
_
g
r
am
r
an
g
e.
As
T
F
-
I
DF
v
ec
to
r
izatio
n
is
u
s
ed
to
co
n
v
er
t
tex
t
d
ata
in
t
o
n
u
m
er
ical
f
ea
tu
r
es
f
o
r
m
o
d
el
tr
ain
in
g
.
T
h
e
f
o
llo
win
g
p
ar
a
m
eter
o
p
tim
ized
m
etr
ics
b
ee
n
u
s
ed
f
o
r
en
h
a
n
cin
g
th
e
T
F
-
I
DF.
T
h
at
is
,
th
e
co
n
f
ig
u
r
atio
n
o
f
p
ar
a
m
eter
s
ar
e
tf
id
f
=T
f
i
d
f
Vec
to
r
izer
(
m
in
_
d
f
=
5
,
m
ax
_
d
f
=
0
.
9
5
,
n
g
r
am
_
r
a
n
g
e
=(
1
,
2
)
)
,
m
in
_
d
f
=5
:
ig
n
o
r
es
ter
m
s
ap
p
ea
r
in
g
in
f
ewe
r
th
an
5
d
o
c
u
m
en
ts
,
m
ax
_
d
f
=0
.
9
5
:
ex
cl
u
d
es
ter
m
s
th
at
ap
p
ea
r
in
m
o
r
e
th
an
9
5
% o
f
d
o
c
u
m
en
ts
,
an
d
n
g
r
am
_
r
an
g
e
=
(
1
,
2
)
: th
e
a
d
v
an
tag
e
o
f
o
p
tim
ized
T
F
-
I
D
F a
r
e
as f
o
llo
ws
:
i)
B
y
in
clu
d
in
g
b
ig
r
am
s
(
n
g
r
am
_
r
an
g
e
=
(
1
,
2
)
)
,
t
h
is
wo
r
k
s
ig
n
if
ican
tly
in
cr
ea
s
es
th
e
n
u
m
b
er
o
f
f
ea
tu
r
es,
lead
in
g
to
a
h
ig
h
-
d
im
en
s
io
n
al
s
p
ar
s
e
m
atr
ix
.
T
h
is
p
r
o
v
id
es b
etter
f
ea
tu
r
es.
ii)
T
h
e
m
in
_
d
f
=
5
a
n
d
m
a
x
_
d
f
=0
.
9
5
th
r
esh
o
ld
s
a
r
e
ch
o
s
en
ac
co
r
d
in
g
to
t
h
e
d
ataset.
T
h
e
s
e
p
ar
am
eter
s
p
r
o
v
id
e
a
b
alan
ce
b
etwe
en
r
et
ain
in
g
im
p
o
r
tan
t te
r
m
s
an
d
r
e
m
o
v
in
g
n
o
is
e.
iii)
B
y
o
p
tim
izin
g
th
e
T
F
-
I
DF
ap
p
r
o
ac
h
,
b
etter
p
er
f
o
r
m
a
n
c
e
is
ac
h
iev
ed
p
r
o
v
id
in
g
b
ett
er
r
esu
lts
f
o
r
s
en
tim
en
t
-
b
ased
p
r
o
d
u
ct
r
ec
o
m
m
en
d
atio
n
.
T
h
is
o
p
tim
ized
T
F
-
I
DF
m
eth
o
d
ass
ig
n
ed
im
p
o
r
tan
ce
to
w
o
r
d
s
co
n
s
id
er
in
g
u
n
ig
r
am
s
an
d
b
ig
r
am
s
b
ased
o
n
th
eir
f
r
eq
u
en
cy
ac
r
o
s
s
r
ev
iew
s
,
en
s
u
r
in
g
b
etter
r
e
p
r
esen
tatio
n
o
f
tex
tu
al
c
o
n
ten
t
f
o
r
th
e
s
u
p
er
v
is
ed
m
ac
h
in
e
lear
n
in
g
class
if
ier
.
3
.
3
.
Sentim
ent
a
na
ly
s
is
wit
h t
ra
ns
f
o
rm
er
-
ba
s
ed
mo
del
T
h
e
ex
tr
ac
ted
f
ea
tu
r
e
s
et
was
s
u
b
s
eq
u
en
tly
f
ed
in
to
a
B
E
R
T
-
b
ased
m
o
d
el
[
1
5
]
to
o
b
tain
s
en
tim
en
t
s
co
r
es.
Am
o
n
g
tr
an
s
f
o
r
m
er
-
d
r
iv
en
tech
n
iq
u
es,
B
E
R
T
[
1
5
]
r
em
ain
s
o
n
e
o
f
th
e
m
o
s
t
wid
el
y
ad
o
p
ted
d
u
e
to
its
s
tr
o
n
g
co
n
te
x
tu
al
u
n
d
er
s
tan
d
i
n
g
.
T
r
a
n
s
f
o
r
m
er
s
,
a
class
o
f
d
ee
p
lear
n
in
g
m
o
d
els,
r
ely
o
n
a
s
elf
-
atten
tio
n
m
ec
h
an
is
m
th
at
id
e
n
tifie
s
th
e
r
elev
an
ce
o
f
ea
c
h
f
ea
tu
r
e
an
d
its
co
n
tex
tu
al
m
ea
n
in
g
with
in
a
r
ev
iew.
Un
lik
e
tr
ad
itio
n
al
n
eu
r
al
n
etwo
r
k
ar
ch
itectu
r
es,
tr
an
s
f
o
r
m
er
m
o
d
els
s
u
p
p
o
r
t
p
ar
allel
co
m
p
u
tatio
n
,
en
ab
led
b
y
s
elf
-
atten
tio
n
an
d
p
o
s
itio
n
al
em
b
ed
d
in
g
s
,
wh
ich
ca
p
tu
r
e
lo
n
g
-
r
an
g
e
d
ep
e
n
d
en
cies
ef
f
icien
tly
.
Fo
r
th
is
r
ea
s
o
n
,
B
E
R
T
is
s
elec
ted
f
o
r
s
en
tim
en
t
ev
alu
atio
n
.
Ho
wev
er
,
g
iv
en
th
e
s
ize
o
f
t
h
e
d
ataset
u
s
ed
in
th
is
s
tu
d
y
—
ap
p
r
o
x
im
ately
3
0
,
0
0
0
r
e
v
iews
s
tan
d
ar
d
B
E
R
T
m
o
d
els
an
d
t
h
eir
co
m
m
o
n
v
ar
ian
ts
s
tr
u
g
g
l
e
to
d
eliv
er
o
p
tim
al
p
er
f
o
r
m
an
ce
.
T
o
a
d
d
r
ess
th
is
l
im
itatio
n
,
th
is
wo
r
k
p
r
o
p
o
s
es a
p
ar
am
eter
-
o
p
tim
ized
v
er
s
io
n
o
f
B
E
R
T
,
r
ef
e
r
r
ed
to
as P
-
B
E
R
T
.
T
h
e
ar
ch
itectu
r
e
o
f
th
e
p
r
o
p
o
s
ed
P
-
B
E
R
T
m
o
d
el
is
illu
s
tr
ated
in
Fig
u
r
e
2
.
Fig
u
r
e
2
.
Sen
tim
en
t
p
r
ed
ictiv
e
s
co
r
e,
p
o
lar
ity
lab
elin
g
,
a
n
d
p
o
lar
ity
class
if
icatio
n
I
n
ex
tr
ac
t
in
g
p
o
la
r
it
ies
a
t
a
g
iv
en
r
e
v
i
ew
c
o
n
s
id
e
r
i
n
g
t
h
e
c
o
n
t
ex
t
u
al
f
ea
t
u
r
e
l
e
v
el
,
le
t
th
e
c
o
l
lect
io
n
o
f
r
e
v
i
ew
-
te
x
t
i
n
th
e
d
a
tas
et
b
e
a
n
d
d
e
n
o
te
w
o
r
d
s
i
n
ea
c
h
r
e
v
ie
w
-
t
ex
t.
T
h
is
is
m
at
h
e
m
at
ic
all
y
d
e
n
o
te
d
as
(
4
)
.
=
{
}
(
4
)
B
y
u
s
in
g
,
o
v
e
r
all
p
o
lar
ity
o
f
r
ev
iew
-
tex
t is id
en
tifie
d
u
s
in
g
(
5
)
,
in
a
t
u
p
le
-
s
et
f
o
r
m
at
r
ep
r
es
en
ted
as
.
=
{
(
,
,
)
}
(
5
)
I
n
(
5
)
,
d
en
o
tes
co
n
tex
tu
r
al
f
e
atu
r
es,
d
en
o
tes
p
o
lar
ity
,
an
d
d
en
o
tes
r
ev
iew
-
p
o
lar
ity
f
o
r
g
iv
en
.
T
h
e
P
-
B
E
R
T
ap
p
r
o
ac
h
co
n
s
id
er
s
p
o
s
itio
n
-
tag
g
in
g
tech
n
iq
u
e
f
o
r
r
e
p
r
esen
tin
g
co
n
tex
tu
al
f
ea
tu
r
es;
h
en
ce
,
tag
g
in
g
is
d
o
n
e
o
n
c
o
m
p
lete
r
ev
iew
-
tex
t
wh
er
e
th
e
c
o
n
tex
t
u
al
-
tag
s
f
o
r
s
tar
tin
g
-
p
o
in
t
an
d
en
d
in
g
-
p
o
in
t
was
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
.
2
,
Ap
r
il 2
0
2
6
:
1
8
5
1
-
1
8
6
2
1856
d
en
o
ted
as
an
d
an
d
its
r
esp
ec
tiv
e
p
o
lar
ities
wer
e
d
en
o
ted
as
an
d
.
Mo
r
e
o
v
er
,
p
o
s
itio
n
-
en
co
d
in
g
was in
co
r
p
o
r
ated
in
P
-
B
E
R
T
f
o
r
r
etr
iev
in
g
co
n
tex
tu
r
al
f
ea
tu
r
e
p
o
s
itio
n
an
d
its
r
esp
ec
tiv
e
p
o
lar
ity
as to
k
en
s
.
I
n
(
6
)
,
th
e
d
en
o
tes
o
v
er
all
to
k
en
s
p
r
esen
t
in
r
ev
iew
-
tex
t
,
d
en
o
tes
s
in
g
le
to
k
en
s
in
th
e
r
ev
iew
-
tex
t.
Usi
n
g
(
6
)
,
t
h
e
co
n
tex
tu
r
al
f
ea
tu
r
e
p
o
lar
ity
w
as
ex
tr
ac
ted
.
T
h
e
B
E
R
T
ap
p
r
o
ac
h
was
m
o
d
if
ie
d
d
u
r
in
g
f
in
e
-
t
u
n
in
g
p
r
o
ce
s
s
in
th
is
wo
r
k
.
Fo
r
m
o
d
if
y
in
g
B
E
R
T
,
a
So
f
tMa
x
f
u
n
ctio
n
was
u
s
ed
f
o
r
ex
tr
ac
tin
g
r
ev
iew
-
lev
el
p
o
lar
ity
an
d
class
if
y
in
g
p
o
lar
ity
,
to
g
et
th
e
class
if
icatio
n
r
esu
lt
s
ac
h
iev
ed
b
y
th
e
P
-
B
E
R
T
.
T
h
e
o
u
tp
u
t
(
to
k
e
n
s
)
f
r
o
m
B
E
R
T
wer
e
u
s
ed
as
in
p
u
t
f
o
r
So
f
tMa
x
f
u
n
ctio
n
.
T
h
e
m
ain
ai
m
o
f
u
s
in
g
So
f
tMa
x
f
u
n
ctio
n
was
to
u
s
e
p
o
lar
ity
g
iv
en
b
y
B
E
R
T
ap
p
r
o
ac
h
,
an
d
c
o
n
v
er
t
it
to
p
r
o
b
ab
ilit
ie
s
.
T
h
e
p
r
o
b
ab
ilit
ies
p
r
o
v
id
e
d
p
o
lar
ity
lik
elih
o
o
d
f
o
r
ev
er
y
class
(
n
eu
tr
al,
n
eg
ativ
e,
an
d
p
o
s
itiv
e)
,
u
s
in
g
wh
ich
r
ev
iew
-
lev
el
p
o
lar
ity
was e
x
tr
ac
ted
.
T
h
e
So
f
tMa
x
p
r
o
b
ab
ilit
y
was e
v
alu
at
ed
u
s
in
g
(
7
)
.
=
{
1
,
2
,
…
,
}
(
6
)
=
(
1
1
+
1
)
(
7
)
I
n
(
7
)
,
1
d
en
o
tes
a
weig
h
t
m
atr
ix
,
wh
ich
was
u
tili
ze
d
f
o
r
tr
an
s
f
o
r
m
in
g
c
o
n
tex
tu
alize
d
t
o
k
en
s
to
f
o
r
m
at
wh
ich
was u
s
ed
f
o
r
class
if
icatio
n
an
d
1
d
en
o
tes
a
b
ias
v
ec
to
r
wh
ich
was a
d
d
ed
to
lin
ea
r
tr
an
s
f
o
r
m
ed
o
u
tp
u
t
f
o
r
c
h
an
g
in
g
th
e
o
u
tco
m
e
an
d
e
n
h
an
cin
g
B
E
R
T
f
lex
ib
ilit
y
.
I
n
th
e
P
-
B
E
R
T
f
r
am
e
wo
r
k
,
C
LS
d
en
o
tes
th
e
class
if
icatio
n
to
k
en
,
TOK
r
ef
er
s
to
in
d
iv
id
u
al
to
k
en
s
,
a
n
d
S
E
P
m
ar
k
s
th
e
s
ep
ar
ato
r
.
E
ac
h
r
ev
iew
is
f
ir
s
t
p
r
o
ce
s
s
ed
th
r
o
u
g
h
to
k
en
an
d
p
o
s
itio
n
al
em
b
ed
d
i
n
g
lay
er
s
,
wh
er
e
ev
er
y
w
o
r
d
is
ass
ig
n
ed
a
to
k
en
id
en
tifie
r
an
d
a
co
r
r
esp
o
n
d
in
g
p
o
s
itio
n
in
d
ex
.
T
h
is
s
tu
d
y
em
p
lo
y
s
o
n
ly
th
e
e
n
co
d
er
co
m
p
o
n
en
t
o
f
B
E
R
T
,
as
th
e
en
co
d
er
al
o
n
e
is
s
u
f
f
icien
t
to
g
en
er
ate
r
ich
co
n
tex
t
u
al
em
b
e
d
d
in
g
s
f
o
r
th
e
i
n
p
u
t
tex
t.
T
h
e
en
co
d
er
c
o
n
tain
s
a
s
elf
-
atten
tio
n
m
ec
h
an
is
m
co
m
b
in
ed
with
a
f
ee
d
-
f
o
r
war
d
n
etwo
r
k
th
at
tr
an
s
f
o
r
m
s
th
e
r
ev
iew
s
eq
u
en
ce
in
to
co
n
tex
tu
al
r
e
p
r
esen
tatio
n
s
.
U
n
lik
e
th
e
o
r
i
g
in
al
B
E
R
T
m
o
d
el,
wh
ich
r
elies
o
n
s
tatic
m
ask
in
g
,
P
-
B
E
R
T
in
co
r
p
o
r
ates
d
y
n
am
ic
m
ask
in
g
to
p
r
o
d
u
ce
m
o
r
e
d
iv
er
s
e
a
n
d
co
n
tex
t
-
awa
r
e
r
ep
r
esen
tati
o
n
s
.
Ad
d
itio
n
all
y
,
in
s
tead
o
f
u
s
in
g
ch
ar
ac
ter
-
lev
el
b
y
te
-
p
air
en
co
d
in
g
,
t
h
e
p
r
o
p
o
s
ed
m
o
d
el
ad
o
p
ts
b
y
te
-
lev
e
l
b
y
te
p
air
en
co
d
in
g
(
B
PE
)
to
ac
ce
ler
ate
co
m
p
u
tat
io
n
.
T
h
ese
en
h
a
n
ce
m
en
ts
allo
w
P
-
B
E
R
T
to
h
an
d
le
la
r
g
er
d
atasets
with
m
o
r
e
b
atch
es
an
d
lo
n
g
e
r
s
eq
u
en
ce
s
th
an
s
tan
d
ar
d
B
E
R
T
.
B
ef
o
r
e
co
n
s
tr
u
ctin
g
th
e
atten
tio
n
m
ask
s
an
d
in
p
u
t
I
Ds,
ea
ch
r
ev
iew
is
to
k
en
ized
.
Du
r
in
g
co
n
ca
ten
atio
n
,
th
e
atten
tio
n
m
ask
id
en
tifie
s
th
e
r
elativ
e
i
m
p
o
r
tan
ce
o
f
ea
ch
to
k
en
,
wh
e
r
ea
s
in
p
u
t
I
Ds
co
n
v
er
t
th
e
tex
t
i
n
to
a
s
eq
u
e
n
tial
n
u
m
er
ic
f
o
r
m
at.
B
o
th
co
m
p
o
n
en
ts
s
er
v
e
as
in
p
u
ts
to
th
e
P
-
B
E
R
T
ar
ch
itectu
r
e.
T
h
e
f
in
al
P
-
B
E
R
T
co
n
f
ig
u
r
atio
n
in
clu
d
es
7
6
8
h
i
d
d
en
u
n
it
s
an
d
1
2
e
n
co
d
er
lay
er
s
.
I
ts
o
u
t
p
u
t
lay
e
r
ca
teg
o
r
izes
ea
ch
r
e
v
iew
as
p
o
s
itiv
e,
n
eu
tr
al,
o
r
n
eg
ati
v
e.
T
h
e
r
es
u
ltin
g
class
if
icatio
n
s
co
r
es a
r
e
s
u
b
s
eq
u
en
tly
p
r
o
ce
s
s
ed
u
s
in
g
a
W
XGBo
o
s
t
m
o
d
el,
wh
ich
is
d
escr
ib
ed
in
t
h
e
f
o
l
lo
win
g
s
ec
tio
n
.
3
.
4
.
Weig
hte
d e
x
t
re
m
e
g
ra
d
ient
bo
o
s
t
ing
f
o
r
s
ent
im
ent
c
la
s
s
if
ica
t
io
n
T
h
is
m
o
d
el
class
if
ied
u
s
er
r
e
v
iews
as
eith
er
p
o
s
itiv
e
o
r
n
eg
ativ
e
,
u
s
in
g
ac
c
u
r
ac
y
,
p
r
ec
is
io
n
,
an
d
F
-
s
co
r
e
p
er
f
o
r
m
an
ce
m
etr
ics.
T
h
e
s
en
tim
en
t
an
aly
s
is
s
tep
f
o
r
m
ed
th
e
f
o
u
n
d
atio
n
f
o
r
g
en
er
atin
g
u
s
er
an
d
item
-
b
ased
r
ec
o
m
m
e
n
d
atio
n
s
.
Fo
r
p
o
lar
ity
u
s
in
g
t
h
e
p
r
o
p
o
s
ed
m
o
d
el,
co
n
s
id
er
t
h
e
e
-
co
m
m
er
ce
d
ataset
with
ℎ
n
u
m
b
e
r
o
f
s
am
p
les
r
ated
b
y
th
e
P
-
B
E
R
T
m
o
d
el
is
d
ef
i
n
ed
in
(
8
)
.
=
{
(
1
,
1
)
,
(
2
,
2
)
,
…
(
,
)
}
(
8
)
I
n
(
8
)
,
d
en
o
tes
a
f
ea
tu
r
e
v
ec
to
r
f
o
r
ev
e
r
y
ℎ
o
b
s
er
v
atio
n
an
d
∈
{
0
,
1
}
wh
er
e
=
1
d
en
o
tes
p
o
s
itiv
e
s
am
p
le
a
n
d
=
0
d
e
n
o
tes
a
n
eg
ativ
e
s
am
p
le.
T
h
e
s
u
p
er
v
i
s
ed
class
if
ier
m
o
d
el
m
ap
s
in
p
u
t
f
ea
tu
r
e
s
f
o
r
g
ettin
g
o
u
tp
u
t
,
as p
r
esen
ted
in
(
9
)
.
=
(
)
=
∑
∙
ℎ
(
)
=
1
(
9
)
I
n
(
9
)
,
d
en
o
tes
th
e
to
tal
n
u
m
b
er
o
f
t
r
ee
s
p
r
esen
t,
d
en
o
t
es
weig
h
ts
ass
ig
n
ed
to
ℎ
tr
ee
an
d
ℎ
(
)
d
en
o
tes
o
u
tp
u
t
o
f
ℎ
f
o
r
in
p
u
t
.
T
h
e
W
XGB
o
o
s
t
m
ain
aim
is
to
m
in
im
ize
lo
g
-
lo
s
s
wh
ile
in
co
r
p
o
r
atin
g
a
r
eg
u
lar
izatio
n
ter
m
f
o
r
p
r
ev
e
n
tin
g
o
v
er
f
itti
n
g
.
T
h
e
lo
s
s
f
u
n
ctio
n
is
ev
alu
ate
d
u
s
in
g
(
1
0
)
.
=
∑
[
l
og
(
)
+
(
1
−
)
l
og
(
1
−
)
]
+
∑
(
ℎ
)
=
1
=
1
(
1
0
)
I
n
(
1
0
)
,
=
(
(
)
)
is
p
r
ed
icted
p
o
s
itiv
e
p
o
lar
ity
p
r
o
b
a
b
ilit
y
,
wh
er
e
s
ig
m
o
id
-
ac
tiv
atio
n
is
ev
alu
ated
as p
r
esen
ted
i
n
(
1
1
)
.
Als
o
,
(
ℎ
)
in
(
1
0
)
is
ev
alu
ated
u
s
in
g
(
1
2
)
.
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
S
en
timen
t
-
a
w
a
r
e
u
s
er
-
item
r
e
co
mme
n
d
a
tio
n
c
o
mb
in
in
g
w
eig
h
ted
X
GB
o
o
s
t a
n
d
…
(
S
n
e
h
a
l
B
h
o
g
a
n
)
1857
(
)
=
1
1
+
−
(
1
1
)
(
ℎ
)
=
+
1
2
|
|
|
|
2
(
1
2
)
I
n
(
1
2
)
,
r
ep
r
esen
ts
th
e
to
tal
n
u
m
b
er
o
f
leav
es
in
ea
ch
d
ec
is
io
n
tr
ee
,
wh
ile
co
r
r
esp
o
n
d
s
t
o
th
e
weig
h
t
ass
ig
n
ed
to
t
h
e
k
-
th
lea
f
.
T
h
e
p
ar
am
eter
s
an
d
s
er
v
e
as
r
eg
u
lar
izatio
n
ter
m
s
th
at
m
an
ag
e
th
e
o
v
er
all
co
m
p
lex
ity
o
f
th
e
tr
ee
.
Du
r
in
g
tr
ain
i
n
g
,
th
e
W
XGB
o
o
s
t
m
o
d
el
p
r
o
ce
s
s
es
th
e
e
-
c
o
m
m
er
ce
d
ataset
b
y
m
in
im
izin
g
th
e
lo
s
s
f
u
n
ctio
n
,
id
en
tify
in
g
th
e
m
o
s
t
ef
f
ec
tiv
e
tr
ee
s
tr
u
ctu
r
e
th
r
o
u
g
h
o
p
tim
al
n
o
d
e
s
p
litt
in
g
,
d
eter
m
in
in
g
th
e
i
d
ea
l
leaf
wei
g
h
ts
,
an
d
e
n
s
u
r
in
g
an
a
p
p
r
o
p
r
iate
b
alan
ce
b
etwe
en
p
r
ed
icti
v
e
ac
cu
r
ac
y
an
d
g
en
er
aliza
tio
n
b
y
r
e
g
u
latin
g
(
ℎ
)
.
3
.
5
.
Rec
o
mm
enda
t
io
n us
ing
s
im
ila
rit
y
m
et
rics a
nd
CF
CF
co
m
es
in
to
p
lay
,
wh
ich
in
v
o
lv
es
th
e
c
r
ea
tio
n
o
f
a
cu
m
u
l
ativ
e
p
r
o
d
u
ct
s
et
b
ased
o
n
o
th
er
p
r
o
d
u
cts
d
ee
m
ed
lik
e
th
e
o
n
es
s
elec
ted
th
r
o
u
g
h
co
n
te
n
t
f
ilter
in
g
,
an
d
u
s
in
g
th
is
ap
p
r
o
ac
h
,
p
r
o
d
u
cts
th
at
s
h
ar
e
co
m
m
o
n
alities
with
u
s
er
p
r
ef
er
en
ce
s
ar
e
id
en
tifie
d
.
Su
b
s
eq
u
en
tly
,
th
e
s
im
ilar
ity
b
etwe
en
th
e
em
b
ed
d
in
g
s
o
f
th
e
u
s
er
-
en
ter
e
d
cu
s
to
m
p
h
r
a
s
e
an
d
th
e
p
r
o
d
u
cts
’
d
escr
ip
t
io
n
an
d
r
ev
iew
in
th
e
co
r
p
u
s
u
s
in
g
th
e
co
s
in
e
f
o
r
m
u
la
as
in
(
1
3
)
.
=
〈
,
〉
‖
‖
∙
‖
‖
(
1
3
)
W
h
er
e
,
an
d
r
ep
r
esen
t
th
e
ca
s
ca
d
ed
em
b
ed
d
in
g
s
o
f
cu
s
to
m
u
s
er
p
h
r
ases
an
d
c
o
n
ca
t
en
ated
p
r
o
d
u
ct
d
escr
ip
tio
n
a
n
d
r
ev
iew
i
n
f
o
r
m
atio
n
in
th
e
d
ataset
r
esp
ec
ti
v
ely
.
〈
,
〉
r
ep
r
esen
ts
th
e
d
o
t
p
r
o
d
u
ct
o
f
v
ec
to
r
s
.
T
h
e
v
ec
to
r
n
o
r
m
f
o
r
is
s
h
o
wn
in
(
1
4
)
.
‖
‖
∙
‖
‖
=
√
1
2
+
2
2
+
⋯
+
2
(
1
4
)
I
n
th
is
s
tu
d
y
,
th
e
m
o
d
el
ev
alu
ates
th
e
s
im
ilar
ity
b
etwe
en
an
ac
tiv
e
u
s
er
an
d
ea
ch
n
eig
h
b
o
r
i
n
g
u
s
er
u
s
in
g
th
e
c
o
s
in
e
s
im
ilar
ity
m
e
tr
ic
s
h
o
wn
in
(
1
3
)
.
Her
e,
th
e
n
eig
h
b
o
r
s
o
f
ar
e
d
e
f
in
ed
as
u
s
er
s
wh
o
h
av
e
p
r
o
v
id
e
d
r
atin
g
s
f
o
r
t
h
e
s
am
e
item
s
an
d
wh
o
s
e
r
atin
g
p
atter
n
s
clo
s
ely
r
esem
b
le
th
o
s
e
o
f
th
e
ac
tiv
e
u
s
er
.
T
o
esti
m
ate
th
e
f
in
al
p
r
e
d
icted
r
atin
g
s
,
th
e
s
y
s
tem
co
m
b
in
es
th
e
cr
iter
io
n
s
co
r
es
f
o
r
ea
ch
r
ec
o
m
m
en
d
ed
item
u
s
in
g
a
weig
h
ted
ag
g
r
eg
atio
n
s
tr
ateg
y
.
T
h
e
a
g
g
r
e
g
ated
s
co
r
es
ar
e
u
s
ed
to
c
o
m
p
u
te
th
e
ex
p
ec
ted
r
atin
g
f
o
r
ea
ch
item
,
af
ter
wh
ich
th
e
ite
m
s
ar
e
s
o
r
ted
ac
co
r
d
i
n
g
to
th
e
ir
p
r
ed
icted
v
al
u
es.
T
h
e
h
ig
h
e
s
t
-
r
an
k
ed
item
s
ar
e
u
ltima
tely
r
ec
o
m
m
e
n
d
ed
t
o
th
e
u
s
er
.
(
,
)
=
∑
=
1
√
∑
2
=
1
√
∑
2
=
1
(
1
5
)
L
et
th
e
s
tr
in
g
r
ep
r
esen
tatio
n
o
f
th
e
ac
tiv
e
u
s
er
b
e
d
en
o
ted
as
an
d
th
at
o
f
th
e
n
eig
h
b
o
r
in
g
u
s
er
as
.
T
h
eir
v
ec
to
r
ized
f
o
r
m
s
ar
e
r
ep
r
esen
ted
b
y
an
d
,
r
esp
ec
tiv
ely
.
Fo
r
ea
ch
item
ass
o
ciate
d
with
th
e
ac
tiv
e
u
s
er
,
th
e
co
r
r
esp
o
n
d
i
n
g
g
en
r
es
ar
e
ex
tr
ac
ted
,
co
n
ca
ten
ated
in
to
a
tex
tu
al
s
eq
u
en
c
e
,
an
d
th
en
tr
an
s
f
o
r
m
ed
in
t
o
a
v
ec
to
r
.
A
s
im
ilar
p
r
o
ce
s
s
is
ap
p
lied
f
o
r
t
h
e
n
eig
h
b
o
r
in
g
u
s
er
,
wh
er
e
th
e
ca
teg
o
r
ies
f
o
r
ea
ch
item
a
r
e
co
m
b
in
ed
in
to
an
d
c
o
n
v
e
r
ted
in
to
t
h
e
v
ec
to
r
f
o
r
m
.
T
h
ese
r
ep
r
ese
n
tatio
n
s
ar
e
g
en
er
ated
u
s
in
g
th
e
B
E
R
T
-
XGB
o
o
s
t
m
o
d
el.
B
ec
au
s
e
a
u
s
er
’
s
f
u
ll
r
atin
g
h
is
to
r
y
d
ef
in
e
s
th
eir
b
eh
av
io
r
al
p
r
o
f
ile,
th
e
B
E
R
T
-
XGB
o
o
s
t
m
o
d
el
p
r
o
ce
s
s
es a
ll item
s
r
ate
d
b
y
a
u
s
er
as th
e
co
n
te
x
tu
al
i
n
p
u
t f
o
r
ea
ch
.
T
h
e
s
im
ilar
ity
s
ig
n
if
ican
ce
b
etwe
en
an
d
is
f
u
r
th
er
r
ef
in
e
d
u
s
in
g
a
n
o
r
m
alize
d
E
u
clid
ea
n
d
is
tan
ce
m
ea
s
u
r
e.
I
n
(
1
6
)
in
teg
r
ates
b
o
th
th
e
s
im
ilar
ity
s
co
r
e
(
,
)
an
d
th
e
co
r
r
esp
o
n
d
i
n
g
weig
h
t
(
,
)
to
co
m
p
u
te
t
h
e
p
r
ed
icted
r
atin
g
f
o
r
an
u
n
r
ated
item
.
T
h
e
p
r
e
f
e
r
en
ce
o
f
u
s
er
f
o
r
item
is
in
f
e
r
r
ed
f
r
o
m
th
e
r
atin
g
s
o
f
t
h
e
m
o
s
t similar
n
eig
h
b
o
r
s
.
=
ˉ
+
∑
(
,
)
⋅
(
,
)
⋅
(
−
ˉ
)
=
1
∑
∣
(
,
)
⋅
(
,
)
∣
=
1
(
1
6
)
Her
e,
d
en
o
tes th
e
r
atin
g
ass
ig
n
ed
b
y
to
th
e
item
,
ˉ
is
th
e
av
er
ag
e
r
atin
g
o
f
u
s
er
,
r
ep
r
esen
ts
th
e
n
eig
h
b
o
r
in
g
u
s
er
,
(
,
)
is
th
e
s
i
m
ilar
ity
m
ea
s
u
r
e,
an
d
(
,
)
is
th
e
weig
h
t
ass
o
ciate
d
with
th
at
s
im
ilar
ity
.
T
h
e
o
u
tp
u
ts
o
b
tain
ed
f
r
o
m
CF
an
d
s
en
tim
en
t
a
n
aly
s
is
ar
e
th
en
c
o
m
b
i
n
ed
to
g
en
er
ate
a
u
n
if
ied
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
.
2
,
Ap
r
il 2
0
2
6
:
1
8
5
1
-
1
8
6
2
1858
r
ec
o
m
m
en
d
atio
n
lis
t.
Giv
en
a
r
atin
g
m
atr
ix
×
,
wh
er
e
d
en
o
tes
th
e
n
u
m
b
er
o
f
u
s
er
s
an
d
d
en
o
tes
th
e
n
u
m
b
er
o
f
item
s
,
ea
ch
r
atin
g
∈
×
r
ep
r
esen
ts
th
e
s
co
r
e
ass
ig
n
ed
b
y
u
s
er
to
item
.
T
h
e
f
in
al
p
r
ed
icted
r
atin
g
f
o
r
item
b
y
u
s
er
is
o
b
tain
ed
as
in
(
1
7
)
.
=
∙
(
1
7
)
Usi
n
g
th
e
m
o
d
elled
s
im
ilar
it
y
m
atr
ix
in
th
is
wo
r
k
,
u
s
er
-
item
s
im
ilar
ity
m
atr
ix
is
co
n
s
tr
u
cted
to
id
en
tify
s
im
ilar
in
te
r
est
u
s
er
s
;
th
en
,
u
s
er
r
atin
g
n
o
r
m
alize
d
to
p
er
f
o
r
m
p
r
e
d
ictio
n
o
f
wh
at
item
s
a
u
s
er
m
ay
p
u
r
ch
ase.
T
h
en
,
item
-
b
ased
r
ec
o
m
m
en
d
atio
n
is
p
er
f
o
r
m
e
d
b
y
i
d
en
tify
in
g
to
p
-
1
0
K,
T
o
p
-
2
0
K
item
s
f
o
r
th
e
u
s
er
s
.
T
h
is
s
y
s
tem
g
en
er
ates
r
ec
o
m
m
en
d
atio
n
s
f
o
r
p
r
o
d
u
ct
s
th
at
ar
e
s
im
ilar
to
th
o
s
e
p
r
ev
io
u
s
ly
b
o
u
g
h
t
o
r
r
ated
h
ig
h
ly
b
y
th
e
u
s
er
.
T
h
e
m
o
d
el
s
im
ilar
ity
m
etr
ics
is
d
esig
n
ed
in
s
u
ch
a
way
wh
en
n
o
r
atin
g
is
av
ailab
le
f
o
r
p
a
r
ticu
lar
p
r
o
d
u
cts
,
it c
an
s
till
m
ak
e
T
op
-
N
r
ec
o
m
m
e
n
d
at
io
n
s
to
u
s
er
s
.
4.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
T
h
is
s
ec
tio
n
s
tu
d
ies
th
e
p
er
f
o
r
m
an
ce
attain
ed
b
y
p
r
o
p
o
s
ed
m
o
d
el
an
d
v
ar
io
u
s
o
th
er
b
asel
in
e
m
o
d
els
u
s
in
g
m
u
lti
-
d
o
m
ain
ec
o
m
m
er
ce
Am
az
o
n
d
ataset.
T
h
e
s
tu
d
y
in
clu
d
es
s
tu
d
y
in
g
p
er
f
o
r
m
a
n
ce
f
o
r
s
en
tim
en
t
class
if
icatio
n
an
d
T
o
p
-
N
r
e
co
m
m
en
d
atio
n
u
s
in
g
n
o
v
el
h
y
b
r
id
m
o
d
el
c
o
m
b
in
e
d
with
n
ew
s
im
ilar
ity
m
etr
ics
-
b
ased
co
llab
o
r
ativ
e
m
etr
ics.
T
h
e
f
in
d
in
g
s
o
f
p
o
l
ar
ity
class
if
icatio
n
ar
e
ev
alu
ated
u
s
in
g
ac
cu
r
ac
y
d
ef
in
ed
in
(
1
8
)
.
=
+
+
+
+
(
1
8
)
T
o
ass
es
s
th
e
ef
f
ec
tiv
en
ess
o
f
th
e
p
r
o
p
o
s
ed
m
o
d
el,
two
co
m
m
o
n
ly
ad
o
p
ted
ev
alu
atio
n
m
et
r
ics:
MA
E
in
(
1
9
)
a
n
d
R
MSE
in
(
2
0
)
wer
e
em
p
lo
y
ed
f
o
r
r
atin
g
p
r
ed
ictio
n
.
T
h
ese
m
ea
s
u
r
es
a
r
e
wid
ely
u
s
ed
in
r
ec
o
m
m
en
d
atio
n
s
y
s
tem
s
to
q
u
an
tify
p
r
e
d
ictio
n
ac
cu
r
ac
y
.
R
MSE
em
p
h
asizes
lar
g
er
d
ev
iatio
n
s
b
y
s
q
u
ar
in
g
th
e
er
r
o
r
s
,
th
er
eb
y
p
en
alizin
g
s
ig
n
if
ican
t
m
is
p
r
ed
ictio
n
s
m
o
r
e
h
ea
v
ily
th
an
MA
E
.
Fo
r
th
is
r
ea
s
o
n
,
R
MSE
is
f
r
eq
u
e
n
tly
p
r
ef
er
r
e
d
wh
e
n
lar
g
e
er
r
o
r
s
ar
e
p
ar
ticu
lar
ly
u
n
d
esira
b
le.
Nev
e
r
th
eless
,
b
ec
au
s
e
b
o
th
m
etr
ics
ca
n
b
e
in
f
lu
en
ce
d
b
y
o
u
tlier
s
,
MA
E
m
ay
b
e
m
o
r
e
s
u
itab
le
f
o
r
d
ata
s
ets with
ir
r
eg
u
lar
o
r
s
k
ewe
d
r
atin
g
d
is
tr
ib
u
tio
n
s
.
=
1
∑
|
,
−
̂
,
|
=
1
(
1
9
)
=
√
1
∑
(
,
−
̂
,
)
2
=
1
(
2
0
)
Her
e,
d
en
o
tes
th
e
to
tal
n
u
m
b
er
o
f
s
am
p
les
in
th
e
tes
t
s
et,
wh
ile
,
an
d
̂
,
co
r
r
esp
o
n
d
to
th
e
ac
tu
al
an
d
p
r
ed
icted
r
atin
g
s
,
r
esp
ec
tiv
ely
.
Sin
ce
b
o
th
m
etr
ics
m
ea
s
u
r
e
p
r
ed
ictio
n
er
r
o
r
,
l
o
wer
v
al
u
es
in
d
icate
im
p
r
o
v
e
d
ac
cu
r
ac
y
,
a
n
d
s
tr
o
n
g
er
p
r
ed
ict
iv
e
ca
p
ab
ilit
y
o
f
th
e
m
o
d
el
.
4
.
1
.
Sentim
ent
a
na
ly
s
is
perf
o
rm
a
nce
ev
a
lua
t
io
n
T
h
is
s
ec
tio
n
s
tu
d
ies
ev
alu
ated
th
e
s
en
tim
en
t
class
if
icatio
n
p
er
f
o
r
m
a
n
ce
u
s
in
g
p
r
o
p
o
s
ed
W
XGB
o
o
s
t
b
ased
m
o
d
el
with
o
th
er
ex
is
tin
g
m
ac
h
in
e
lear
n
in
g
-
b
ased
class
if
ier
s
[
2
3
]
.
T
h
e
s
tu
d
y
s
h
o
wed
in
[
2
3
]
,
th
at
em
p
lo
y
in
g
T
F
-
I
DF
with
m
ac
h
in
e
lear
n
in
g
class
if
ier
aid
ed
th
e
p
er
f
o
r
m
a
n
ce
i
n
co
m
p
ar
is
o
n
with
C
o
u
n
tv
ec
to
r
izer
-
b
ased
m
ac
h
i
n
e
lear
n
in
g
m
o
d
el.
T
h
e
wo
r
k
in
tr
o
d
u
ce
s
a
n
o
v
el
XGBo
o
s
t+o
p
tim
ized
T
F
-
I
DF
m
o
d
el
en
s
u
r
i
n
g
an
ac
cu
r
ac
y
o
f
8
8
%;
th
en
,
th
e
wo
r
k
in
tr
o
d
u
ce
d
a
n
o
v
el
W
XGBo
o
s
t
co
m
b
in
ed
with
o
p
tim
ized
TF
-
I
DF
,
en
s
u
r
in
g
an
ac
c
u
r
ac
y
o
f
9
4
%
as sh
o
wn
in
T
ab
le
1
.
T
h
u
s
,
th
e
r
esu
lt sh
o
ws th
e
p
r
o
p
o
s
ed
m
o
d
el
attain
s
s
ig
n
if
ican
tly
h
ig
h
er
ac
c
u
r
ac
y
b
y
lear
n
in
g
th
r
o
u
g
h
B
E
R
T
-
b
ased
s
co
r
in
g
co
m
b
in
ed
with
XGB
o
o
s
t
-
b
ased
weig
h
ted
class
if
ier
wh
en
co
m
p
ar
ed
with
ex
is
tin
g
m
eth
o
d
s
.
T
ab
le
1
.
C
o
m
p
a
r
ativ
e
s
tu
d
y
w
ith
m
ac
h
in
e
lear
n
in
g
-
b
ased
s
en
tim
en
t
an
aly
s
is
m
o
d
el
M
o
d
e
l
A
c
c
u
r
a
c
y
(
%)
Lo
g
i
s
t
i
c
r
e
g
r
e
ssi
o
n
+
TF
-
I
D
F
[
2
3
]
,
2
0
2
4
8
1
.
3
2
N
a
ï
v
e
B
a
y
e
s
+
TF
-
I
D
F
[
2
3
]
,
2
0
2
4
7
9
.
8
2
X
G
B
o
o
st
+
TF
-
I
D
F
[
2
3
]
,
2
0
2
4
8
1
.
6
7
R
a
n
d
o
m
f
o
r
e
s
t
+
TF
-
I
D
F
[
2
3
]
,
2
0
2
4
8
3
.
0
1
C
R
F
[
2
2
]
,
2
0
2
4
9
3
.
4
X
G
B
o
o
st
+
O
-
TF
-
I
D
F
[
p
r
o
p
o
s
e
d
]
88
W
X
G
B
o
o
st
+
O
-
TF
-
I
D
F
[
p
r
o
p
o
s
e
d
]
9
4
.
5
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
S
en
timen
t
-
a
w
a
r
e
u
s
er
-
item
r
e
co
mme
n
d
a
tio
n
c
o
mb
in
in
g
w
eig
h
ted
X
GB
o
o
s
t a
n
d
…
(
S
n
e
h
a
l
B
h
o
g
a
n
)
1859
4
.
2
.
Rec
o
mm
enda
t
io
n
perf
o
rm
a
nce
ev
a
lua
t
i
o
n
T
h
is
s
ec
tio
n
p
r
o
v
id
es
a
c
o
m
p
ar
ativ
e
ev
al
u
atio
n
o
f
th
e
SAUI
R
S
r
ec
o
m
m
en
d
atio
n
f
r
a
m
ewo
r
k
a
n
d
s
ev
er
al
well
-
estab
lis
h
ed
b
asel
in
e
m
o
d
els.
T
o
v
er
if
y
th
e
ef
f
ec
tiv
en
ess
o
f
th
e
SAUI
R
S
m
o
d
el,
th
e
b
aselin
es
wer
e
g
r
o
u
p
ed
in
to
f
o
u
r
m
ain
ca
teg
o
r
ies:
i
)
m
eth
o
d
s
th
at
r
ely
ex
clu
s
iv
ely
o
n
n
u
m
e
r
ical
r
atin
g
s
(
CF
an
d
p
r
o
b
a
b
ilis
tic
m
atr
ix
f
ac
to
r
izatio
n
(
PMF
)
)
,
ii
)
ap
p
r
o
ac
h
es
th
at
in
teg
r
ate
r
ev
iew
tex
t
in
to
th
e
p
r
ed
ictio
n
p
r
o
ce
s
s
(
h
id
d
en
f
ac
to
r
s
an
d
h
i
d
d
en
to
p
ics
(
HFT
)
an
d
r
et
r
iev
al
-
au
g
m
en
ted
r
etr
iev
al
(
R
AR
)
V2
)
,
i
ii
)
atten
tio
n
-
d
r
iv
e
n
m
o
d
els
th
at
ex
p
lo
it
n
e
u
r
al
a
tten
tio
n
m
ec
h
an
is
m
s
(
n
e
u
r
al
atten
tio
n
al
r
eg
r
ess
io
n
m
o
d
e
l
with
r
ev
iew
-
lev
el
ex
p
lan
atio
n
s
(
NARR
E
)
an
d
r
e
v
iew
s
em
an
tics
b
ased
m
o
d
el
(
R
S
B
M
)
)
,
an
d
iv
)
asp
ec
t
-
awa
r
e
r
ec
o
m
m
en
d
atio
n
m
o
d
els
(
asp
ec
t
atten
ti
o
n
-
b
ase
d
n
e
u
r
al
c
o
llab
o
r
ativ
e
f
ilter
in
g
(
A3
NC
F
)
an
d
u
s
er
an
d
c
o
n
tex
t
awa
r
e
m
o
d
el
(
UC
AM
)
)
.
A
co
n
cise o
v
er
v
iew
o
f
th
e
b
aselin
e
tech
n
iq
u
es is
s
u
m
m
ar
ized
as f
o
llo
ws
:
−
C
F
[
2
6
]
:
on
e
o
f
th
e
ea
r
lies
t
a
n
d
m
o
s
t
in
f
lu
en
tial
r
ec
o
m
m
e
n
d
atio
n
s
tr
ateg
ies,
CF
id
en
tifie
s
s
im
ilar
u
s
er
s
o
r
item
s
b
y
a
n
aly
zin
g
h
is
to
r
ical
r
atin
g
p
atter
n
s
,
a
n
d
r
ec
o
m
m
en
d
s
item
s
f
r
o
m
th
e
clo
s
est n
eig
h
b
o
r
s
.
−
PMF
[
2
7
]
:
PMF
ex
ten
d
s
th
e
m
atr
ix
f
ac
to
r
izatio
n
p
a
r
ad
ig
m
b
y
in
tr
o
d
u
cin
g
Gau
s
s
ian
p
r
io
r
s
,
allo
win
g
it
to
r
esp
o
n
d
ef
f
ec
tiv
ely
to
h
ig
h
l
y
s
p
ar
s
e
an
d
im
b
alan
ce
d
r
atin
g
m
atr
ices.
−
HFT
[
2
8
]
:
th
e
HFT
m
o
d
el
e
m
p
lo
y
s
laten
t
Dir
ich
let
allo
ca
tio
n
(
L
DA)
to
d
er
iv
e
u
n
d
er
ly
i
n
g
to
p
ics
f
r
o
m
u
s
er
an
d
item
r
e
v
iews a
n
d
f
u
s
es th
ese
to
p
ics with
laten
t f
ac
to
r
s
o
b
tain
ed
th
r
o
u
g
h
m
atr
ix
f
a
cto
r
izatio
n
.
−
NAR
R
E
[
2
9
]
:
th
e
NARR
E
m
ec
h
an
is
m
th
at
h
ig
h
lig
h
ts
im
p
ac
tf
u
l
r
ev
iews.
User
an
d
item
e
m
b
ed
d
in
g
s
ar
e
lear
n
ed
u
s
in
g
m
atr
ix
f
ac
to
r
i
za
tio
n
,
wh
ile
a
C
NN
id
en
tifie
s
in
f
o
r
m
ativ
e
r
ev
iew
s
eg
m
en
ts
f
o
r
th
e
f
in
al
p
r
ed
ictio
n
.
−
R
AR
V2
[
3
0
]
:
th
is
m
eth
o
d
en
r
ich
es
r
ev
iew
-
tex
t
r
ep
r
esen
tatio
n
b
y
co
m
b
in
in
g
m
u
ltip
le
em
b
ed
d
in
g
ty
p
es.
I
t
u
tili
ze
s
B
E
R
T
an
d
r
o
b
u
s
tly
o
p
tim
ized
B
E
R
T
p
r
etr
ain
in
g
a
p
p
r
o
ac
h
(
R
o
B
E
R
T
a
)
as
au
x
ili
ar
y
s
o
u
r
ce
s
o
f
s
em
an
tic
in
f
o
r
m
atio
n
with
in
a
d
ee
p
m
atr
ix
f
ac
to
r
izatio
n
f
r
am
ewo
r
k
.
−
R
S
B
M
[
3
1
]
:
t
h
e
R
SB
M
ex
tr
ac
ts
s
em
an
tic
f
ea
tu
r
es
f
r
o
m
r
e
v
iews
u
s
in
g
a
C
NN
an
d
a
p
p
li
es
an
atten
tio
n
n
etwo
r
k
to
ca
p
tu
r
e
asp
ec
t
-
s
p
e
cif
ic
ev
alu
atio
n
s
f
o
r
r
atin
g
p
r
e
d
ictio
n
.
−
A3
NC
F
[
3
2
]
:
an
asp
ec
t
-
awa
r
e
n
eu
r
al
m
o
d
el
th
at
ad
ap
ts
to
u
s
er
-
s
p
ec
if
ic
p
r
ef
er
en
ce
s
ac
r
o
s
s
v
ar
io
u
s
item
asp
ec
ts
.
T
o
p
ic
m
o
d
elin
g
is
u
s
ed
to
i
d
en
tify
u
s
er
in
ter
ests
a
n
d
item
attr
i
b
u
tes,
wh
ich
ar
e
t
h
en
in
teg
r
ated
in
to
th
e
r
ec
o
m
m
en
d
ati
o
n
p
r
o
c
ess
.
−
UC
AM
[
3
3
]
:
a
d
ee
p
-
lear
n
in
g
-
b
ased
,
co
n
te
x
t
-
awa
r
e
m
o
d
el
th
at
m
er
g
es
u
s
er
-
item
in
ter
ac
tio
n
s
with
r
ev
iew
r
ep
r
esen
tatio
n
s
.
R
ev
i
ew
f
ea
tu
r
es
ar
e
o
b
tain
ed
u
s
in
g
a
B
E
R
T
-
b
ased
asp
ec
t
-
b
a
s
ed
s
en
tim
en
t
an
aly
s
is
(
AB
SA
)
m
o
d
u
le
to
ex
tr
ac
t a
s
p
ec
t
-
o
r
ien
ted
s
en
tim
en
t
s
.
T
h
e
R
MSE
an
d
MA
E
s
co
r
es
f
o
r
th
e
b
aselin
e
m
o
d
els
ar
e
r
ep
o
r
te
d
in
T
ab
le
2
.
T
h
e
r
esu
lts
d
em
o
n
s
tr
ate
th
at
th
e
p
r
o
p
o
s
e
d
SAUI
R
S
f
r
am
ewo
r
k
,
lev
e
r
ag
in
g
h
y
b
r
id
o
p
tim
izatio
n
,
e
n
h
an
ce
d
s
im
ilar
ity
co
m
p
u
tatio
n
,
an
d
a
s
en
tim
e
n
t
-
awa
r
e
class
if
icatio
n
m
o
d
u
le
,
a
ch
iev
es a
s
u
b
s
tan
tial r
ed
u
ctio
n
in
p
r
ed
ictiv
e
er
r
o
r
co
m
p
ar
ed
with
all
co
m
p
etin
g
ap
p
r
o
ac
h
es.
T
h
e
MA
E
p
e
r
f
o
r
m
an
ce
o
f
d
if
f
er
e
n
t
tr
an
s
f
o
r
m
er
an
d
atten
tio
n
-
b
ased
m
o
d
els
is
tab
u
lated
in
T
ab
le
3
.
T
h
e
r
esu
lts
ca
n
b
e
s
ee
n
in
th
e
p
r
o
p
o
s
ed
SAUI
R
S
m
o
d
el
,
em
p
l
o
y
in
g
h
y
b
r
id
o
p
tim
izatio
n
with
o
p
ti
m
ized
s
im
ilar
ity
m
etr
ics
co
m
b
in
ed
with
s
en
tim
en
t
class
if
ier
en
ab
le
d
th
e
m
o
d
el
to
s
ig
n
if
ican
tly
r
ed
u
ce
p
r
ed
ictiv
e
er
r
o
r
in
co
m
p
ar
is
o
n
with
d
i
f
f
er
en
t b
aselin
e
m
o
d
els.
T
ab
le
2
.
C
o
m
p
a
r
ativ
e
s
tu
d
y
w
ith
b
aselin
e
m
o
d
els
M
o
d
e
l
R
M
S
E
M
A
E
C
F
[
2
6
]
,
2
0
0
1
1
.
2
9
1
1
.
6
0
2
P
M
F
[
2
7
]
,
2
0
0
7
1
.
0
0
3
1
.
0
0
8
H
F
T
[
2
8
]
,
2
0
1
3
0
.
9
0
1
1
.
2
9
7
R
A
R
V
2
[
2
9
]
,
2
0
1
8
0
.
6
5
5
0
.
8
0
9
N
A
R
R
E
[
3
0
]
,
2
0
2
3
1
.
0
2
3
1
.
3
7
3
R
S
B
M
[
3
1
]
,
2
0
2
0
0
.
5
7
8
0
.
7
9
9
A
3
N
C
F
[
3
2
]
,
2
0
1
8
0
.
9
9
2
1
.
3
0
7
U
C
A
M
[
3
3
]
2
0
2
0
0
.
7
6
7
0
.
9
9
7
A
A
R
N
[
2
5
]
,
2
0
2
4
0
.
5
0
9
0
.
7
6
4
R
TG
[
2
4
]
,
2
0
2
5
0
.
2
6
7
0
.
2
1
7
S
A
U
I
R
S
[
p
r
o
p
o
se
d]
0
.
2
4
5
0
.
3
1
1
T
ab
le
3
.
C
o
m
p
a
r
ativ
e
s
tu
d
y
w
ith
tr
an
s
f
o
r
m
er
s
a
n
d
atten
tio
n
-
b
ased
m
o
d
els
M
o
d
e
l
M
A
E
D
o
c
2
V
e
c
0
.
6
9
2
B
ER
T
0
.
5
8
5
R
o
B
E
R
Ta
0
.
5
7
6
A
A
R
N
[
2
5
]
0
.
5
0
9
S
A
U
I
R
S
[
p
r
o
p
o
se
d
]
0
.
2
4
5
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
.
2
,
Ap
r
il 2
0
2
6
:
1
8
5
1
-
1
8
6
2
1860
T
h
e
r
esu
lts
in
d
icate
s
ig
n
if
i
ca
n
t
im
p
r
o
v
em
en
ts
in
r
ec
o
m
m
en
d
atio
n
ac
cu
r
ac
y
.
v
alid
atin
g
th
e
ef
f
ec
tiv
en
ess
o
f
s
en
tim
en
t
-
awa
r
e
f
ea
tu
r
es in
r
ef
in
in
g
u
s
er
-
item
r
ec
o
m
m
en
d
atio
n
s
.
T
h
e
p
r
o
p
o
s
ed
m
eth
o
d
o
lo
g
y
o
f
f
er
s
a
s
ca
lab
le
a
n
d
i
n
ter
p
r
etab
le
a
p
p
r
o
ac
h
,
b
r
id
g
i
n
g
th
e
g
ap
b
etwe
en
s
en
ti
m
en
t
an
aly
s
is
an
d
r
ec
o
m
m
en
d
atio
n
s
y
s
tem
s
.
T
h
is
s
tu
d
y
co
n
tr
i
b
u
tes
to
ad
v
an
cin
g
i
n
tellig
en
t
r
ec
o
m
m
en
d
er
s
y
s
tem
s
b
y
in
teg
r
atin
g
d
ee
p
lear
n
i
n
g
-
b
ase
d
s
en
tim
en
t
class
if
icatio
n
with
o
p
tim
ized
s
im
ilar
ity
m
ea
s
u
r
es,
p
av
in
g
th
e
way
f
o
r
m
o
r
e
ac
cu
r
ate
an
d
u
s
er
-
ce
n
tr
ic
r
ec
o
m
m
e
n
d
atio
n
s
.
5.
CO
NCLU
SI
O
N
T
h
is
s
tu
d
y
in
t
r
o
d
u
ce
d
a
n
o
v
el
SAUI
R
S
f
r
am
ewo
r
k
in
teg
r
ati
n
g
O
-
TF
-
I
DF,
P
-
B
E
R
T
-
b
ased
s
en
tim
en
t
class
if
icatio
n
,
W
XG
B
o
o
s
t
,
an
d
an
o
p
tim
ized
s
im
ilar
ity
m
et
r
ics
m
o
d
el.
T
h
e
ap
p
r
o
ac
h
was
ev
alu
ated
u
s
in
g
th
e
Am
az
o
n
m
u
lti
-
d
o
m
ai
n
d
ataset
co
m
p
r
is
in
g
3
0
,
0
0
0
r
ev
iews
ac
r
o
s
s
d
iv
er
s
e
p
r
o
d
u
ct
ca
teg
o
r
ies.
Per
f
o
r
m
an
c
e
was
ass
e
s
s
ed
u
s
in
g
ac
cu
r
ac
y
,
MA
E
,
an
d
R
MSE
,
en
s
u
r
i
n
g
a
co
m
p
r
eh
e
n
s
iv
e
ev
alu
atio
n
o
f
e
f
f
ec
tiv
en
ess
.
E
x
p
er
im
en
tal
r
esu
lts
d
em
o
n
s
tr
ated
th
at
th
e
p
r
o
p
o
s
ed
f
r
a
m
ewo
r
k
o
u
tp
e
r
f
o
r
m
s
ex
is
tin
g
r
ec
o
m
m
en
d
atio
n
m
o
d
els
b
y
ac
h
iev
in
g
h
ig
h
er
ac
cu
r
ac
y
wh
ile
r
e
d
u
cin
g
er
r
o
r
m
etr
ics.
T
h
e
s
en
tim
en
t
clas
s
if
icatio
n
m
o
d
u
le,
p
o
wer
ed
b
y
P
-
B
E
R
T
an
d
W
XGBo
o
s
t
,
s
ig
n
if
ican
tly
im
p
r
o
v
e
d
p
r
e
d
ictio
n
p
r
ec
is
io
n
,
lead
i
n
g
to
m
o
r
e
r
elev
an
t
an
d
u
s
er
-
ce
n
tr
ic
r
ec
o
m
m
en
d
atio
n
s
.
O
-
TF
-
I
DF
ef
f
ec
tiv
ely
ca
p
tu
r
ed
ess
en
tial
f
ea
tu
r
es,
m
itig
atin
g
s
p
ar
s
ity
is
s
u
es
an
d
en
h
an
cin
g
co
n
tex
t
u
al
u
n
d
e
r
s
tan
d
in
g
o
f
r
e
v
iews.
Fu
r
th
er
m
o
r
e,
s
en
tim
en
t
-
awa
r
e
s
im
ilar
ity
m
etr
ics
r
ef
in
ed
th
e
r
ec
o
m
m
en
d
atio
n
p
r
o
ce
s
s
b
y
alig
n
in
g
s
u
g
g
esti
o
n
s
with
u
s
er
s
’
em
o
tio
n
al
in
clin
atio
n
s
.
Qu
an
titativ
e
an
aly
s
is
co
n
f
ir
m
e
d
ac
c
u
r
ac
y
im
p
r
o
v
em
e
n
ts
o
v
e
r
tr
a
d
itio
n
a
l
CF
m
eth
o
d
s
,
w
h
ile
r
ed
u
ce
d
MA
E
an
d
R
MSE
v
alid
ated
r
o
b
u
s
tn
ess
an
d
r
eliab
ilit
y
.
Ov
er
all,
th
e
p
r
o
p
o
s
ed
SAUI
R
S
m
o
d
el
o
f
f
er
s
a
s
ca
lab
le,
ef
f
icien
t,
an
d
s
en
tim
en
t
-
awa
r
e
s
o
lu
tio
n
.
Fu
tu
r
e
wo
r
k
will
f
o
cu
s
o
n
r
ea
l
-
tim
e
d
ep
l
o
y
m
en
t,
lar
g
e
-
s
ca
le
o
p
tim
izatio
n
,
an
d
r
ein
f
o
r
ce
m
e
n
t le
ar
n
in
g
–
b
ased
ad
ap
tiv
e
r
ef
i
n
em
en
t.
F
UNDING
I
NF
O
R
M
A
T
I
O
N
T
h
is
r
esear
ch
r
ec
eiv
ed
n
o
e
x
te
r
n
al
f
u
n
d
in
g
.
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
t
r
ib
u
to
r
R
o
les
T
a
x
o
n
o
m
y
(
C
R
ed
iT)
to
r
ec
o
g
n
ize
in
d
iv
i
d
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
.
Na
m
e
o
f
Aut
ho
r
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
Sn
eh
al
B
h
o
g
an
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
Vijay
S.
R
ajp
u
r
o
h
it
✓
✓
✓
✓
✓
✓
✓
San
jeev
S.
San
n
ak
k
i
✓
✓
✓
✓
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
T
h
e
au
t
h
o
r
s
d
ec
lar
e
th
at
n
o
f
i
n
an
cial
o
r
co
m
m
er
cial
r
elatio
n
s
h
ip
s
ex
is
t
th
at
c
o
u
ld
b
e
co
n
s
tr
u
ed
as
a
p
o
ten
tial c
o
n
f
lict
o
f
in
ter
est.
T
h
e
co
r
r
esp
o
n
d
in
g
au
th
o
r
c
o
n
f
i
r
m
s
th
is
d
ec
lar
atio
n
o
n
b
eh
alf
o
f
all
co
-
a
u
th
o
r
s
.
E
T
H
I
CAL AP
P
RO
V
AL
No
t
ap
p
licab
le.
T
h
is
s
tu
d
y
d
id
n
o
t
in
v
o
l
v
e
h
u
m
an
p
ar
tici
p
an
ts
,
h
u
m
an
d
ata,
o
r
a
n
y
p
er
s
o
n
ally
id
en
tifia
b
le
in
f
o
r
m
atio
n
.
All
d
ata
u
s
ed
wer
e
eith
er
p
u
b
licl
y
av
ailab
le,
f
u
lly
an
o
n
y
m
ize
d
,
o
r
d
er
i
v
ed
f
r
o
m
n
o
n
‑
h
u
m
an
s
o
u
r
ce
s
,
an
d
th
er
e
f
o
r
e
n
o
in
f
o
r
m
ed
co
n
s
en
t w
as
r
eq
u
ir
ed
f
r
o
m
in
d
iv
i
d
u
als.
DATA AV
AI
L
AB
I
L
I
T
Y
T
h
e
d
ata
an
d
to
o
ls
u
s
ed
in
t
h
is
s
tu
d
y
a
r
e
a
v
ailab
le
f
r
o
m
th
e
c
o
r
r
esp
o
n
d
in
g
a
u
th
o
r
,
[
SB
],
u
p
o
n
r
ea
s
o
n
ab
le
r
eq
u
est.
An
y
u
s
e
o
f
th
e
s
h
ar
ed
r
eso
u
r
ce
s
m
u
s
t b
e
ap
p
r
o
p
r
iately
cited
.
Evaluation Warning : The document was created with Spire.PDF for Python.