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
.
3
,
J
u
n
e
2
0
2
6
,
p
p
.
2736
~
2
7
4
9
I
SS
N:
2
2
5
2
-
8
9
3
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijai.v
15
.i
3
.
p
p
2
7
3
6
-
2
7
4
9
2736
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
i
.
ia
esco
r
e.
co
m
H
y
brid d
eep l
ea
rning
f
o
r sen
timen
t
a
na
ly
sis
of onlin
e studen
t
ex
peri
ences
Ra
j
a
O
ua
da
d,
H
icha
m
M
o
un
cif
La
b
o
r
a
t
o
r
y
o
f
LI
M
A
TI
,
D
e
p
a
r
t
me
n
t
o
f
M
a
t
h
e
ma
t
i
c
s
a
n
d
I
n
f
o
r
ma
t
i
c
s,
F
a
c
u
l
t
y
o
f
P
o
l
y
d
i
sc
i
p
l
i
n
a
r
y
,
S
u
l
t
a
n
M
o
u
l
a
y
S
l
i
ma
n
e
U
n
i
v
e
r
si
t
y
,
B
e
n
i
M
e
l
l
a
l
,
M
o
r
o
c
c
o
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Feb
6
,
2
0
2
5
R
ev
is
ed
Feb
1
2
,
2
0
2
6
Acc
ep
ted
Ap
r
2
2
,
2
0
2
6
Th
e
C
OV
ID
-
1
9
p
a
n
d
e
m
ic
d
isru
p
ted
m
il
li
o
n
s
o
f
li
v
e
s
wo
rl
d
wid
e
,
a
n
d
s
o
c
ial
m
e
d
ia
p
latfo
rm
s
b
e
c
a
m
e
a
sig
n
ifi
c
a
n
t
o
u
tl
e
t
f
o
r
p
e
o
p
le
to
s
h
a
re
th
e
ir
e
m
o
ti
o
n
s
a
n
d
e
x
p
e
rien
c
e
s,
p
r
o
v
i
d
in
g
v
a
lu
a
b
le
in
si
g
h
ts
i
n
to
t
h
e
c
h
a
ll
e
n
g
e
s
a
n
d
o
p
p
o
rtu
n
it
ies
o
f
re
m
o
te
e
d
u
c
a
ti
o
n
.
Th
is
p
a
p
e
r
a
n
a
ly
z
e
s
stu
d
e
n
t
se
n
ti
m
e
n
ts
a
b
o
u
t
o
n
li
n
e
lea
rn
in
g
d
u
ri
n
g
t
h
e
p
a
n
d
e
m
ic
u
sin
g
Twit
ter
d
a
ta.
An
e
x
p
e
rime
n
tal
a
p
p
r
o
a
c
h
is
d
e
v
e
l
o
p
e
d
to
a
n
a
ly
z
e
p
u
b
li
c
c
o
m
m
e
n
ts
,
fo
c
u
sin
g
o
n
th
e
se
n
t
ime
n
t
e
x
p
re
ss
e
d
i
n
t
we
e
ts
re
late
d
to
o
n
l
in
e
e
d
u
c
a
ti
o
n
.
A
h
y
b
rid
d
e
e
p
lea
rn
i
n
g
m
o
d
e
l,
b
a
se
d
o
n
t
h
e
lo
g
isti
c
re
g
re
ss
io
n
(LR)
se
n
ti
m
e
n
t
m
o
d
e
l,
is
u
se
d
to
p
re
d
ict
se
n
ti
m
e
n
t
fro
m
a
larg
e
d
a
tas
e
t
o
f
o
n
li
n
e
lea
rn
in
g
-
re
late
d
twe
e
ts.
Afte
r
p
e
rfo
rm
in
g
n
-
g
ra
m
a
n
a
ly
sis
to
e
x
trac
t
k
e
y
t
o
p
ics
,
twe
e
ts
a
re
c
las
sified
in
to
se
n
ti
m
e
n
t
c
las
se
s.
Th
e
p
ro
p
o
se
d
c
o
n
v
o
l
u
ti
o
n
a
l
l
o
n
g
sh
o
rt
-
term
m
e
m
o
ry
(Co
n
v
-
L
S
TM
)
a
n
d
c
o
n
v
o
lu
ti
o
n
a
l
b
i
d
irec
ti
o
n
a
l
lo
n
g
s
h
o
rt
-
term
m
e
m
o
ry
(Co
n
v
-
BiL
S
TM
)
m
o
d
e
ls
a
re
train
e
d
o
n
twe
e
ts
a
n
n
o
t
a
ted
with
g
ra
n
u
lar
se
n
ti
m
e
n
t
c
las
sifica
ti
o
n
s,
a
c
h
iev
in
g
v
a
li
d
a
ti
o
n
a
c
c
u
ra
c
ies
o
f
9
3
%
a
n
d
9
5
%
,
re
sp
e
c
ti
v
e
ly
.
Th
is
wo
rk
p
r
o
v
i
d
e
s
m
e
a
n
in
g
fu
l
in
si
g
h
ts
in
to
t
h
e
e
m
o
ti
o
n
a
l
e
ffe
c
ts
o
f
o
n
li
n
e
lea
rn
i
n
g
d
u
ri
n
g
th
e
p
a
n
d
e
m
ic,
c
o
n
tri
b
u
ti
n
g
to
th
e
u
n
d
e
rsta
n
d
i
n
g
o
f
st
u
d
e
n
ts'
e
x
p
e
ri
e
n
c
e
s a
n
d
c
h
a
ll
e
n
g
e
s in
re
m
o
te ed
u
c
a
ti
o
n
.
K
ey
w
o
r
d
s
:
C
OVI
D
-
1
9
p
an
d
em
ic
Hy
b
r
id
d
ee
p
lear
n
i
n
g
m
o
d
el
L
o
g
is
tic
r
eg
r
ess
io
n
On
lin
e
lear
n
in
g
Sen
tim
en
t a
n
aly
s
is
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
:
R
aja
Ou
ad
ad
L
ab
o
r
ato
r
y
o
f
L
I
MA
T
I
,
De
p
ar
tm
en
t o
f
Ma
th
em
atics a
n
d
I
n
f
o
r
m
atics,
Facu
lty
o
f
Po
ly
d
is
cip
lin
ar
y
Su
ltan
Mo
u
lay
Sli
m
an
e
Un
iv
e
r
s
ity
B
en
i M
ellal,
Mo
r
o
cc
o
E
m
ail:
o
u
ad
ad
r
aja2
@
g
m
ail.
co
m
1.
I
NT
RO
D
UCT
I
O
N
Or
ig
in
atin
g
in
W
u
h
an
,
C
h
in
a
,
in
Dec
em
b
er
2
0
1
9
,
C
OVI
D
-
1
9
s
wif
tly
ex
p
an
d
e
d
b
ey
o
n
d
n
atio
n
al
b
o
r
d
er
s
,
cu
lm
in
atin
g
in
its
r
ec
o
g
n
itio
n
as
a
p
a
n
d
em
ic
b
y
t
h
e
W
o
r
ld
Hea
lth
O
r
g
an
izatio
n
(
W
HO)
o
n
Ma
r
c
h
1
1
,
2
0
2
0
[
1
]
.
T
h
is
u
n
p
r
ec
e
d
en
ted
ev
en
t
s
ev
er
ely
i
m
p
ac
te
d
co
u
n
tr
ies
wo
r
ld
wid
e
,
r
esu
lt
in
g
in
m
illi
o
n
s
o
f
f
atalities
an
d
n
ec
ess
itatin
g
s
t
r
ict
lo
ck
d
o
wn
m
ea
s
u
r
es
to
c
u
r
b
th
e
v
ir
u
s
'
s
s
p
r
ea
d
.
E
v
en
with
th
e
ad
v
en
t
o
f
v
ac
cin
es,
m
ain
tain
in
g
s
o
cial
d
is
tan
cin
g
r
em
ain
ed
cr
u
cial
i
n
m
itig
atin
g
tr
an
s
m
is
s
io
n
[
2
]
.
T
h
ese
r
estrictio
n
s
b
r
o
u
g
h
t
s
ig
n
i
f
ican
t
c
h
an
g
es
t
o
ed
u
ca
tio
n
,
with
tr
ad
itio
n
al
in
-
p
er
s
o
n
lea
r
n
in
g
tr
an
s
itio
n
in
g
t
o
o
n
lin
e
p
latf
o
r
m
s
to
en
s
u
r
e
co
n
tin
u
ity
.
T
h
is
s
h
if
t
u
n
d
er
s
co
r
e
d
th
e
cr
itical
r
o
l
e
o
f
e
-
lear
n
in
g
as
th
e
p
r
im
ar
y
alter
n
ativ
e
wh
e
n
p
h
y
s
ical
atten
d
a
n
ce
was
n
o
t
f
ea
s
ib
le
[
3
]
.
Stu
d
en
ts
f
ac
ed
c
o
n
s
id
er
ab
le
a
d
ju
s
tm
en
ts
,
ad
a
p
tin
g
to
an
en
tire
l
y
n
ew
ed
u
ca
tio
n
al
en
v
ir
o
n
m
en
t
[
4
]
.
T
h
eir
f
ee
d
b
ac
k
d
u
r
in
g
th
i
s
tr
an
s
itio
n
h
as
p
r
o
v
id
ed
m
ea
n
in
g
f
u
l
in
s
ig
h
ts
in
to
th
e
ch
allen
g
es
an
d
p
o
ten
tial
o
f
o
n
lin
e
e
d
u
ca
tio
n
,
o
f
f
er
in
g
a
b
asis
f
o
r
r
ef
in
i
n
g
lear
n
in
g
s
y
s
tem
s
.
So
cial
m
ed
ia
p
latf
o
r
m
s
,
s
u
ch
as
T
witter
,
h
av
e
f
u
r
th
er
e
n
ab
led
r
esear
ch
e
r
s
to
ac
ce
s
s
an
d
an
aly
ze
s
tu
d
en
t
f
ee
d
b
ac
k
m
o
r
e
ef
f
icien
tly
[
5
]
.
W
h
i
le
o
p
in
io
n
s
w
er
e
tr
a
d
iti
o
n
a
ll
y
g
at
h
e
r
e
d
t
h
r
o
u
g
h
q
u
esti
o
n
n
ai
r
es
,
t
h
e
i
n
cr
e
asin
g
p
r
e
v
al
e
n
ce
o
f
o
n
li
n
e
p
lat
f
o
r
m
s
h
as
f
ac
i
lit
ate
d
t
h
e
c
o
l
lec
ti
o
n
o
f
d
iv
er
s
e
an
d
s
p
o
n
ta
n
e
o
u
s
i
n
s
i
g
h
ts
,
en
r
i
ch
e
d
o
u
r
u
n
d
e
r
s
t
an
d
i
n
g
o
f
s
t
u
d
e
n
ts
'
ex
p
e
r
i
e
n
ce
s
,
a
n
d
i
n
f
o
r
m
e
d
t
h
e
co
n
t
in
u
o
u
s
i
m
p
r
o
v
e
m
e
n
t
o
f
e
-
l
ea
r
n
in
g
s
y
s
te
m
s
[
6
]
,
[
7
]
.
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
Hyb
r
id
d
ee
p
lea
r
n
in
g
fo
r
s
en
timen
t a
n
a
lysi
s
o
f o
n
lin
e
s
tu
d
e
n
t e
xp
erien
ce
s
(
R
a
ja
Ou
a
d
a
d
)
2737
Nu
m
er
o
u
s
s
tu
d
ies
h
av
e
ap
p
lied
s
en
tim
en
t
an
aly
s
is
tech
n
iq
u
es
to
an
aly
ze
s
tu
d
en
t
f
ee
d
b
a
ck
[
8
]
.
I
n
th
ese
s
tu
d
ies,
f
ee
d
b
ac
k
is
o
f
te
n
class
if
ied
b
ased
o
n
th
e
s
en
ti
m
en
t
p
o
lar
ity
it
c
o
n
v
e
y
s
,
ca
teg
o
r
izin
g
r
esp
o
n
s
es
as
eith
er
p
o
s
itiv
e
o
r
n
e
g
ativ
e
[
9
]
.
B
o
th
o
p
i
n
io
n
m
in
in
g
an
d
s
en
tim
en
t
an
aly
s
is
s
h
ar
e
s
im
ilar
o
b
jectiv
es,
f
o
cu
s
in
g
o
n
id
en
tify
i
n
g
s
en
ti
m
en
ts
,
p
er
s
p
ec
tiv
es
,
a
n
d
o
p
in
io
n
s
em
b
ed
d
ed
in
te
x
ts
p
r
o
d
u
ce
d
b
y
h
u
m
an
s
o
n
a
r
an
g
e
o
f
to
p
ics
o
r
en
titi
es
[
1
0
]
.
T
h
e
an
aly
s
is
r
elies
o
n
n
atu
r
al
lan
g
u
ag
e
p
r
o
ce
s
s
in
g
(
NL
P)
,
tex
t
m
in
in
g
,
an
d
f
ea
tu
r
e
ex
tr
ac
tio
n
m
eth
o
d
s
to
in
f
er
an
d
la
b
el
s
en
tim
en
t
o
r
ien
tatio
n
[
1
1
]
.
W
h
ile
clo
s
ely
r
elate
d
,
s
en
tim
en
t
an
aly
s
is
s
p
ec
if
ically
tar
g
ets
e
m
o
tio
n
al
co
n
ten
t,
id
e
n
tify
in
g
wo
r
d
s
o
r
p
h
r
ases
th
at
r
ef
lect
em
o
tio
n
s
,
wh
er
ea
s
o
p
in
io
n
m
i
n
in
g
b
r
o
ad
l
y
ex
tr
a
cts
o
p
in
io
n
s
ab
o
u
t
a
p
a
r
ticu
lar
s
u
b
ject
o
r
en
tity
[
7
]
.
Ov
er
th
e
y
ea
r
s
,
r
esear
ch
er
s
h
av
e
ex
p
lo
r
ed
v
ar
i
o
u
s
m
eth
o
d
s
to
ac
cu
r
ately
in
ter
p
r
et
s
en
tim
en
ts
in
tex
tu
al
d
ata.
T
h
ese
ap
p
r
o
ac
h
es
h
a
v
e
p
r
o
g
r
ess
ed
f
r
o
m
lex
ico
n
-
b
ased
tech
n
iq
u
es
to
m
ac
h
in
e
lear
n
in
g
alg
o
r
ith
m
s
,
an
d
m
o
r
e
r
e
ce
n
tly
,
to
ad
v
an
ce
d
h
y
b
r
id
d
ee
p
lear
n
in
g
m
o
d
els
th
at
lev
er
ag
e
th
e
co
m
b
in
ed
s
tr
en
g
th
s
o
f
m
u
ltip
le
ar
c
h
itectu
r
es
to
ac
h
iev
e
im
p
r
o
v
e
d
p
er
f
o
r
m
a
n
ce
in
s
en
t
im
en
t p
r
ed
ictio
n
[
1
2
]
.
I
n
th
e
co
n
tex
t
o
f
o
u
r
s
tu
d
y
,
r
ec
en
t
p
r
o
g
r
ess
in
h
y
b
r
i
d
d
ee
p
lear
n
in
g
m
o
d
els
h
as
d
em
o
n
s
tr
ated
s
ig
n
if
ican
t
p
o
ten
tial
f
o
r
a
n
aly
zin
g
s
tu
d
en
t
f
ee
d
b
ac
k
an
d
d
e
r
iv
in
g
m
ea
n
i
n
g
f
u
l
in
s
ig
h
ts
in
to
o
n
lin
e
lear
n
in
g
ex
p
er
ien
ce
s
.
T
h
ese
ap
p
r
o
ac
h
e
s
co
m
b
in
e
th
e
s
tr
en
g
th
s
o
f
m
u
ltip
le
n
eu
r
al
n
etwo
r
k
a
r
ch
ite
ctu
r
es
to
ef
f
ec
tiv
ely
m
an
ag
e
t
h
e
co
m
p
lex
ity
an
d
l
ar
g
e
v
o
lu
m
e
o
f
ed
u
ca
tio
n
al
d
ata.
Fo
r
e
x
am
p
le,
Ma
r
y
[
1
3
]
in
tr
o
d
u
ce
d
a
h
y
b
r
id
m
o
d
el
in
teg
r
atin
g
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
s
(
C
NN)
with
lo
n
g
s
h
o
r
t
-
ter
m
m
e
m
o
r
y
(
L
STM
)
n
etwo
r
k
s
,
ac
h
iev
in
g
h
i
g
h
ac
cu
r
ac
y
in
e
v
alu
atin
g
s
tu
d
en
t
r
ev
iews
f
r
o
m
o
n
lin
e
lear
n
i
n
g
p
latf
o
r
m
s
.
T
h
is
h
ig
h
lig
h
ts
th
e
p
r
o
m
is
e
o
f
d
ee
p
lear
n
in
g
tec
h
n
iq
u
es f
o
r
p
r
o
ce
s
s
in
g
u
n
s
tr
u
ct
u
r
ed
f
ee
d
b
ac
k
an
d
s
u
p
p
o
r
tin
g
th
e
im
p
r
o
v
e
m
en
t o
f
teac
h
in
g
a
n
d
lear
n
in
g
p
r
ac
t
ices.
Similar
ly
,
Alawi
an
d
B
o
zk
u
r
t
[
1
4
]
in
teg
r
ated
b
id
ir
ec
tio
n
al
en
co
d
e
r
r
ep
r
esen
tatio
n
s
f
r
o
m
tr
an
s
f
o
r
m
er
s
(
B
E
R
T
)
,
b
id
ir
ec
tio
n
al
lo
n
g
s
h
o
r
t
-
ter
m
m
em
o
r
y
(
B
iLST
M)
,
an
d
C
NN
in
a
h
y
b
r
id
m
o
d
el
to
an
al
y
ze
T
u
r
k
i
s
h
twee
ts
ab
o
u
t
u
n
iv
e
r
s
ity
s
atis
f
ac
tio
n
,
ac
h
iev
i
n
g
an
ac
c
u
r
ac
y
o
f
o
v
er
9
1
%
a
n
d
ef
f
ec
tiv
ely
h
an
d
lin
g
th
e
lin
g
u
i
s
tic
co
m
p
lex
ities
o
f
T
u
r
k
is
h
.
E
x
p
an
d
i
n
g
th
is
r
esear
ch
in
to
m
ass
iv
e
o
p
en
o
n
lin
e
co
u
r
s
es
(
MO
OC
s
)
,
Mr
h
ar
et
a
l.
[
1
5
]
u
tili
ze
d
a
B
ay
esian
C
NN
-
L
STM
m
o
d
el
to
an
al
y
z
e
1
4
0
,
3
2
0
C
o
u
r
s
er
a
co
u
r
s
e
r
ev
iews,
ac
h
iev
in
g
9
1
%
ac
cu
r
ac
y
wh
ile
lin
k
in
g
f
o
r
u
m
p
o
s
t
s
en
tim
en
ts
to
d
r
o
p
o
u
t
r
is
k
s
an
d
co
u
r
s
e
s
u
cc
ess
.
C
o
m
p
lem
en
tin
g
th
is
,
L
i
et
a
l.
[
1
6
]
p
r
esen
ted
a
s
h
a
llo
w
h
y
b
r
id
B
E
R
T
–
C
NN
ap
p
r
o
ac
h
f
o
r
s
en
tim
en
t
an
aly
s
is
b
ased
o
n
1
9
,
1
4
8
c
o
m
m
en
ts
ex
tr
ac
ted
f
r
o
m
a
C
h
i
n
ese
MO
OC
p
latf
o
r
m
,
ac
h
iev
in
g
8
1
.
3
%
ac
c
u
r
ac
y
an
d
9
2
.
8
%
F1
-
s
co
r
e,
all
wh
ile
s
ig
n
if
ican
tly
r
ed
u
cin
g
c
o
m
p
u
tatio
n
al
co
m
p
lex
ity
.
Similar
ly
,
Z
h
en
g
et
a
l.
[
1
7
]
in
tr
o
d
u
ce
d
a
m
o
d
el
co
m
b
in
in
g
B
E
R
T
an
d
a
b
id
ir
ec
tio
n
al
g
ated
r
ec
u
r
r
e
n
t
u
n
it
(
B
iGR
U)
m
o
d
el
with
in
t
h
e
f
ee
d
b
ac
k
s
y
s
tem
s
tr
u
ctu
r
e
(
FS
S)
f
o
r
in
tellig
en
t
teac
h
er
-
s
tu
d
e
n
t
in
ter
ac
tio
n
an
d
cu
r
r
icu
lu
m
en
h
an
ce
m
e
n
t.
On
a
s
elf
-
co
n
s
tr
u
cted
d
ataset
o
f
o
n
l
in
e
ed
u
ca
tio
n
p
latf
o
r
m
r
e
v
iews,
th
eir
m
o
d
el
ac
h
iev
ed
a
r
em
ar
k
ab
le
ac
cu
r
ac
y
o
f
9
8
.
8
2
%,
s
u
r
p
ass
in
g
tr
ad
itio
n
al
m
eth
o
d
s
lik
e
n
aiv
e
B
ay
es
b
y
2
1
.
5
4
%.
A
d
d
itio
n
ally
,
th
eir
m
o
d
el
d
em
o
n
s
tr
ated
s
tr
o
n
g
s
en
tim
en
t
an
aly
s
is
ca
p
ab
ilit
ies,
esp
ec
ially
in
h
a
n
d
li
n
g
tex
ts
c
o
n
tain
in
g
C
h
in
ese
i
n
ter
n
et
b
u
zz
wo
r
d
s
,
h
ig
h
lig
h
tin
g
its
ef
f
ec
tiv
en
ess
in
in
tellig
en
t f
ee
d
b
ac
k
s
y
s
tem
s
f
o
r
ed
u
ca
tio
n
al
co
n
tex
ts
.
I
n
al
ig
n
m
en
t w
ith
th
ese
ad
v
an
ce
m
e
n
ts
,
J
eb
b
ar
i
et
a
l.
[
1
8
]
p
r
esen
ted
a
h
y
b
r
id
ap
p
r
o
a
ch
f
o
r
s
en
tim
en
t
a
n
aly
s
is
in
MO
OC
f
o
r
u
m
s
th
at
m
er
g
es
wo
r
d
-
lev
el
r
e
p
r
esen
ta
tio
n
s
f
r
o
m
B
iLST
M
lay
er
s
w
ith
ch
ar
ac
ter
-
lev
el
f
ea
tu
r
es
f
r
o
m
C
NN
lay
er
s
to
ef
f
ec
tiv
ely
en
co
d
e
u
n
i
q
u
e
li
n
g
u
is
tic
p
atter
n
s
.
W
ith
an
ac
cu
r
ac
y
o
f
9
3
.
1
1
%,
th
e
m
o
d
el
h
ig
h
lig
h
ts
its
ef
f
ec
tiv
en
ess
an
d
co
n
tr
i
b
u
tes
s
ig
n
if
ican
tly
to
s
en
tim
en
t
an
a
ly
s
is
in
o
n
lin
e
ed
u
ca
tio
n
p
latf
o
r
m
s
.
C
o
llectiv
ely
,
th
ese
s
tu
d
ies
illu
s
tr
ate
th
e
tr
an
s
f
o
r
m
ativ
e
p
o
ten
tial
o
f
h
y
b
r
id
d
ee
p
lear
n
in
g
ap
p
r
o
ac
h
es
in
ed
u
ca
tio
n
al
s
en
tim
en
t a
n
aly
s
is
ac
r
o
s
s
d
iv
er
s
e
co
n
tex
ts
an
d
d
atasets
.
B
u
ild
in
g
o
n
th
e
in
s
ig
h
ts
g
ain
ed
f
r
o
m
an
aly
zin
g
p
u
b
lic
s
en
tim
en
t
r
elate
d
to
th
e
s
h
if
t
t
o
war
d
o
n
lin
e
ed
u
ca
tio
n
p
r
o
m
p
ted
b
y
th
e
C
OVI
D
-
1
9
p
a
n
d
em
ic,
th
is
s
tu
d
y
in
tr
o
d
u
ce
s
s
ev
er
al
k
ey
co
n
tr
ib
u
tio
n
s
to
ad
v
a
n
ce
s
en
tim
en
t
an
aly
s
is
tech
n
iq
u
e
s
.
T
h
ese
co
n
tr
ib
u
tio
n
s
in
clu
d
e
lev
er
ag
in
g
le
x
ical
n
-
g
r
a
m
m
o
d
els
to
id
en
tify
p
r
o
m
in
e
n
t
tr
en
d
s
in
twee
t
-
s
p
ec
if
ic
v
o
ca
b
u
lar
y
an
d
em
p
lo
y
in
g
d
o
m
ain
-
s
p
ec
if
ic
wo
r
d
an
al
y
s
is
to
m
ea
s
u
r
e
th
e
im
p
ac
t
o
f
o
n
lin
e
lear
n
in
g
-
r
elate
d
ter
m
s
.
T
h
e
s
en
tim
en
t
p
o
lar
ity
r
atin
g
s
wer
e
f
in
e
-
tu
n
ed
u
s
in
g
a
l
o
g
is
tic
r
eg
r
ess
io
n
(
L
R
)
m
o
d
el,
f
ac
ilit
atin
g
th
e
class
if
icatio
n
o
f
twe
ets
in
to
p
o
s
itiv
e,
n
eg
ativ
e,
an
d
n
eu
tr
al
ca
teg
o
r
ies.
Fu
r
th
er
m
o
r
e
,
a
h
y
b
r
id
co
n
v
o
lu
tio
n
al
lo
n
g
s
h
o
r
t
-
ter
m
m
em
o
r
y
(
C
o
n
v
-
L
STM
)
m
o
d
el
was
d
ev
elo
p
ed
,
lev
er
ag
in
g
th
ese
r
ef
in
e
d
r
atin
g
s
to
p
r
ed
ict
s
en
tim
en
t
tr
en
d
s
m
o
r
e
ac
cu
r
ately
.
T
o
g
eth
er
,
th
ese
m
eth
o
d
s
p
r
o
v
id
e
a
r
o
b
u
s
t
f
r
am
ewo
r
k
f
o
r
u
n
d
er
s
tan
d
in
g
an
d
ad
d
r
ess
in
g
th
e
ev
o
lv
in
g
ch
allen
g
es
o
f
o
n
lin
e
l
ea
r
n
in
g
in
tim
es
o
f
g
lo
b
al
d
is
r
u
p
tio
n
.
T
h
e
p
ap
e
r
is
s
tr
u
ctu
r
ed
as
f
o
l
lo
ws:
s
ec
tio
n
2
in
tr
o
d
u
ce
s
th
e
s
tu
d
y
’
s
m
eth
o
d
s
,
co
m
m
en
ci
n
g
with
a
d
escr
ip
tio
n
o
f
th
e
p
r
o
p
o
s
ed
m
eth
o
d
o
lo
g
y
an
d
th
e
o
r
g
a
n
izatio
n
o
f
th
e
d
ataset
,
p
r
e
p
r
o
ce
s
s
in
g
tech
n
iq
u
es,
an
d
th
e
id
en
tific
atio
n
o
f
wo
r
d
tr
e
n
d
s
u
s
in
g
n
-
g
r
am
m
o
d
els.
I
t
als
o
co
v
e
r
s
th
e
d
etec
tio
n
o
f
o
n
lin
e
lear
n
in
g
-
s
p
ec
if
ic
ter
m
s
d
u
r
in
g
th
e
C
OVI
D
-
1
9
p
an
d
em
ic,
s
en
tim
en
t
an
al
y
s
is
an
d
class
if
icatio
n
u
s
in
g
n
atu
r
al
lan
g
u
a
g
e
to
o
lk
it
(
NL
T
K)
,
th
e
d
e
v
elo
p
m
e
n
t
p
r
o
ce
s
s
o
f
th
e
s
en
tim
en
t
class
if
icatio
n
m
o
d
el
u
s
in
g
LR
,
an
aly
s
is
o
f
g
lo
b
al
s
en
tim
en
t
tr
en
d
s
,
an
d
th
e
ar
c
h
itectu
r
e
an
d
h
y
p
er
p
ar
am
eter
s
o
f
th
e
h
y
b
r
id
C
o
n
v
-
L
STM
m
o
d
el.
Sectio
n
3
d
is
cu
s
s
es
th
e
m
o
d
el'
s
p
er
f
o
r
m
an
ce
an
d
co
m
p
ar
es
it
wit
h
o
th
er
s
tate
o
f
ar
t
m
o
d
els.
T
h
e
f
in
al
s
ec
tio
n
s
u
m
m
ar
izes th
e
f
in
d
in
g
s
an
d
e
x
p
lo
r
es p
o
te
n
tial d
ir
ec
tio
n
s
f
o
r
f
u
tu
r
e
r
esear
ch
.
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
.
3
,
J
u
n
e
2
0
2
6
:
2
7
3
6
-
2
7
4
9
2738
2.
M
E
T
H
O
D
2
.
1
.
O
v
er
v
iew
T
h
e
C
OVI
D
-
1
9
p
an
d
em
ic
h
as
p
r
o
f
o
u
n
d
ly
tr
a
n
s
f
o
r
m
e
d
ed
u
ca
tio
n
o
n
a
g
lo
b
al
s
ca
le,
s
h
if
tin
g
co
n
v
en
tio
n
al
in
-
p
er
s
o
n
lear
n
i
n
g
to
o
n
lin
e
lear
n
i
n
g
p
latf
o
r
m
s
b
ec
au
s
e
o
f
lo
ck
d
o
wn
s
a
n
d
s
o
cial
d
is
tan
cin
g
m
ea
s
u
r
es.
Alth
o
u
g
h
th
is
s
u
d
d
en
s
h
if
t
p
o
s
ed
c
h
allen
g
es,
s
u
c
h
as
ad
ap
tin
g
t
o
n
ew
tech
n
o
lo
g
ies
an
d
m
an
a
g
in
g
s
elf
-
d
ir
ec
ted
lear
n
in
g
,
it
also
h
ig
h
lig
h
ted
th
e
p
o
ten
tial
o
f
o
n
lin
e
ed
u
ca
tio
n
as
a
m
o
r
e
f
le
x
ib
le
an
d
ac
ce
s
s
ib
le
alter
n
ativ
e.
T
o
ass
ess
its
e
f
f
ec
t
iv
e
n
ess
d
u
r
i
n
g
t
h
is
p
er
i
o
d
,
o
u
r
r
es
ea
r
c
h
e
x
am
i
n
ed
le
ar
n
e
r
f
e
e
d
b
ac
k
b
y
a
n
al
y
zi
n
g
a
d
atas
et
o
f
3
1
0
,
5
4
7
twe
ets
r
el
ate
d
t
o
o
n
li
n
e
l
ea
r
n
i
n
g
an
d
co
u
r
s
es,
p
o
s
t
ed
b
e
twe
e
n
J
a
n
u
a
r
y
a
n
d
O
ct
o
b
er
2
0
2
0
.
T
h
ese
t
wee
ts
u
n
d
e
r
w
en
t
a
t
h
o
r
o
u
g
h
cle
a
n
i
n
g
a
n
d
p
r
e
p
r
o
c
ess
in
g
p
r
o
ce
s
s
u
s
i
n
g
t
h
e
N
L
T
K
p
ac
k
a
g
e.
T
h
is
in
c
lu
d
ed
r
em
o
v
i
n
g
n
o
is
y
d
at
a,
t
o
k
e
n
i
zi
n
g
te
x
t
i
n
t
o
li
n
g
u
i
s
tic
u
n
its
,
a
n
d
a
p
p
l
y
i
n
g
l
em
m
at
iza
ti
o
n
t
o
r
e
d
u
ce
wo
r
d
s
to
t
h
ei
r
f
u
n
d
a
m
e
n
t
al
l
e
x
ical
f
o
r
m
s
f
o
r
m
o
r
e
i
n
-
d
ep
th
a
n
al
y
s
is
.
A
f
t
er
th
e
c
lea
n
i
n
g
a
n
d
p
r
e
p
r
o
c
ess
i
n
g
s
te
p
,
we
i
d
e
n
ti
f
i
ed
t
h
e
m
o
s
t
f
r
eq
u
e
n
tl
y
o
cc
u
r
r
i
n
g
u
n
i
g
r
am
s
,
b
i
g
r
am
s
,
a
n
d
t
r
i
g
r
a
m
s
a
cr
o
s
s
t
h
e
en
t
ir
e
tw
ee
t
d
a
tase
t.
Su
b
s
e
q
u
e
n
tl
y
,
we
i
d
en
ti
f
ie
d
t
e
r
m
s
s
p
e
ci
f
ic
t
o
t
h
e
e
f
f
e
cts
o
f
t
h
e
C
O
VI
D
-
1
9
p
a
n
d
em
ic
o
n
o
n
li
n
e
e
d
u
ca
ti
o
n
,
as
well
as
t
h
ei
r
f
r
e
q
u
e
n
cy
i
n
th
e
r
ev
i
ews
,
to
b
et
te
r
u
n
d
e
r
s
ta
n
d
th
eir
i
n
f
l
u
e
n
c
e
o
n
lea
r
n
e
r
s
'
o
p
in
i
o
n
s
.
Ad
d
itio
n
ally
,
f
o
r
ea
ch
clea
n
e
d
an
d
p
r
ep
r
o
ce
s
s
ed
twee
t,
th
e
p
o
lar
ity
was
ass
ess
ed
u
s
in
g
a
s
en
tim
en
t
an
aly
s
is
to
o
l
f
r
o
m
th
e
NL
T
K
lib
r
ar
y
.
Utilizin
g
an
ad
ap
tiv
e
class
if
icatio
n
ap
p
r
o
ac
h
,
twee
ts
wer
e
ca
teg
o
r
ized
in
to
th
r
ee
s
en
tim
en
t
class
es:
p
o
s
itiv
e,
n
eg
ativ
e,
an
d
n
eu
tr
al.
An
LR
m
o
d
el
was
t
h
en
d
ev
elo
p
ed
,
le
v
er
ag
in
g
t
h
e
m
o
s
t
f
r
eq
u
en
t
n
-
g
r
am
s
to
ac
h
iev
e
f
in
e
-
g
r
ain
ed
s
en
tim
en
t
class
if
icatio
n
o
f
twee
ts
.
T
h
i
s
m
o
d
el
ass
ig
n
s
a
d
etailed
s
en
tim
en
t
s
co
r
e
to
ea
ch
r
ev
iew
b
y
c
o
m
p
u
tin
g
p
r
o
b
a
b
ilit
y
s
co
r
es,
en
s
u
r
in
g
ac
cu
r
ate
class
if
icatio
n
in
to
th
e
ap
p
r
o
p
r
iate
s
en
tim
en
t
ca
t
eg
o
r
ies.
W
e
th
en
ass
ig
n
ed
s
e
n
tim
en
t
s
co
r
es
to
r
ev
iews
in
f
iv
e
d
is
tin
ct
class
es,
r
an
g
in
g
f
r
o
m
th
e
h
ig
h
ly
p
o
s
itiv
e
(
1
.
0
)
to
th
e
h
ig
h
ly
n
eg
ativ
e
(
-
1
.
0
)
,
b
ased
o
n
th
e
o
b
tain
ed
p
r
o
b
a
b
ilit
y
s
co
r
es.
Neu
tr
al
r
ev
iews
wer
e
g
r
o
u
p
e
d
in
to
a
s
ep
ar
ate
class
(
0
.
0
)
.
Fi
n
ally
,
we
ex
am
i
n
ed
th
e
d
aily
v
ar
iatio
n
s
in
o
v
er
all
s
en
tim
en
t
tr
en
d
s
am
o
n
g
lear
n
er
s
ab
o
u
t
o
n
lin
e
co
u
r
s
es
d
u
r
in
g
t
h
e
C
OVI
D
-
1
9
p
e
r
io
d
b
y
ca
lcu
latin
g
th
e
av
er
ag
e
s
en
tim
en
t
f
o
r
ea
ch
d
ay
.
T
h
e
twee
ts
,
lab
eled
o
n
a
f
in
e
-
g
r
ain
e
d
s
ca
le
f
r
o
m
0
to
4
,
wer
e
tr
an
s
f
o
r
m
e
d
in
to
wo
r
d
em
b
e
d
d
in
g
s
to
f
a
cilitate
f
u
r
th
er
an
aly
s
is
.
T
h
e
d
ataset
was
p
ar
titi
o
n
ed
,
allo
ca
tin
g
8
0
%
o
f
t
h
e
s
am
p
les
f
o
r
tr
ain
in
g
a
n
d
2
0
%
f
o
r
v
alid
atio
n
o
f
th
e
h
y
b
r
id
m
o
d
els.
B
o
th
s
eq
u
en
tial
a
n
d
B
iLST
M
n
eu
r
al
n
etwo
r
k
s
wer
e
tr
ain
e
d
o
n
th
e
v
ec
to
r
ized
twee
ts
an
d
th
eir
c
o
r
r
esp
o
n
d
i
n
g
lab
els
to
p
r
e
d
ict
s
en
tim
en
t
p
o
lar
ity
.
Fo
llo
win
g
th
e
tr
ain
in
g
p
r
o
ce
s
s
,
th
e
m
o
d
els’
p
er
f
o
r
m
an
ce
was
ev
alu
ated
o
n
test
d
atas
ets
to
as
s
es
s
th
eir
ac
cu
r
ac
y
.
T
h
e
o
v
er
all
s
y
s
tem
ar
ch
itectu
r
e
ad
o
p
ted
in
th
is
s
tu
d
y
is
s
u
m
m
ar
ized
in
Fig
u
r
e
1
.
Fig
u
r
e
1
.
Pro
p
o
s
ed
s
y
s
tem
a
r
c
h
itectu
r
e
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
Hyb
r
id
d
ee
p
lea
r
n
in
g
fo
r
s
en
timen
t a
n
a
lysi
s
o
f o
n
lin
e
s
tu
d
e
n
t e
xp
erien
ce
s
(
R
a
ja
Ou
a
d
a
d
)
2739
2
.
2
.
Da
t
a
s
et
des
cr
iptio
n
T
h
e
d
ataset
em
p
lo
y
e
d
in
t
h
is
s
tu
d
y
was
co
llected
th
r
o
u
g
h
th
e
T
witter
API
,
wh
ich
n
ec
ess
itate
s
th
e
u
s
e
o
f
a
T
witter
d
e
v
elo
p
er
ac
co
u
n
t.
A
cu
s
to
m
Py
th
o
n
s
cr
ip
t,
g
et_
twee
ts
.
p
y
,
was
d
ev
elo
p
ed
to
co
llect
E
n
g
lis
h
-
lan
g
u
a
g
e
twee
ts
r
elate
d
to
d
is
tan
ce
lear
n
in
g
.
T
h
e
s
cr
ip
t
s
ea
r
ch
ed
f
o
r
twee
ts
co
n
t
ain
in
g
th
e
f
o
llo
win
g
h
ash
tag
s
:
#
d
is
tan
ce
lear
n
in
g
,
#
o
n
lin
esch
o
o
l,
#
o
n
lin
et
ea
ch
in
g
,
#
v
ir
tu
allea
r
n
in
g
,
#
o
n
lin
ed
u
ca
tio
n
,
#
d
is
tan
ce
ed
u
ca
tio
n
,
#
On
lin
e
C
las
s
es,
#
Dig
italLe
ar
n
in
g
,
#
elea
r
n
in
g
,
an
d
#
o
n
lin
elea
r
n
in
g
.
Ad
d
itio
n
ally
,
th
e
s
cr
ip
t
tar
g
eted
k
ey
wo
r
d
s
s
u
ch
as
“d
is
tan
ce
lear
n
in
g
,
”
“o
n
lin
e
teac
h
in
g
,
”
“o
n
lin
e
ed
u
ca
tio
n
,
”
“o
n
lin
e
co
u
r
s
e,
”
“o
n
lin
e
s
em
ester
,
”
“d
is
tan
ce
co
u
r
s
e,
”
“d
is
tan
ce
ed
u
ca
tio
n
,
”
“o
n
lin
e
class
,
”
an
d
“e
-
lear
n
in
g
.
”
T
o
m
ain
tai
n
d
ata
q
u
ality
an
d
p
r
e
v
en
t r
e
d
u
n
d
an
cy
,
r
etwe
ets we
r
e
r
em
o
v
e
d
f
r
o
m
t
h
e
d
ataset.
T
h
e
g
et_
twee
ts
f
u
n
ctio
n
tem
p
o
r
ar
ily
s
to
r
ed
th
e
r
etr
iev
ed
tw
ee
ts
in
a
p
an
d
a
Data
Fra
m
e
an
d
s
av
ed
th
e
d
ata
as
C
SV
f
iles
in
an
o
u
tp
u
t
d
ir
ec
to
r
y
.
T
h
e
d
ata
c
o
llectio
n
p
r
o
ce
s
s
to
o
k
a
p
p
r
o
x
im
ately
4
5
h
o
u
r
s
to
g
ath
er
3
2
6
,
1
4
0
twee
ts
.
Su
b
s
eq
u
en
tly
,
th
e
co
n
ca
ten
ate.
p
y
s
cr
ip
t
was
em
p
lo
y
ed
t
o
m
er
g
e
all
i
n
d
iv
i
d
u
al
C
SV
f
iles
in
to
a
u
n
if
ie
d
d
ataset.
T
h
e
f
in
al
d
ataset,
r
ef
er
r
e
d
to
as
twee
ts
_
r
aw.
csv
,
c
o
n
tain
s
3
1
0
,
5
4
7
t
wee
ts
,
with
v
ar
io
u
s
f
ea
tu
r
es,
as o
u
tlin
ed
i
n
T
ab
le
1
.
2
.
3
.
Da
t
a
p
re
pro
ce
s
s
ing
E
m
p
i
r
ic
al
d
ata
is
i
n
h
e
r
e
n
t
ly
i
m
p
er
f
ec
t
,
o
f
te
n
co
n
t
ai
n
i
n
g
i
n
c
o
n
s
is
te
n
cies
o
r
i
r
r
el
e
v
a
n
t in
f
o
r
m
at
io
n
.
T
h
e
p
r
e
p
r
o
c
ess
i
n
g
s
t
a
g
e
is
v
i
tal
f
o
r
a
d
d
r
ess
i
n
g
th
ese
is
s
u
es
b
y
r
em
o
v
i
n
g
n
o
is
e
a
n
d
ir
r
e
le
v
a
n
t
el
em
en
ts
,
t
h
e
r
e
b
y
im
p
r
o
v
i
n
g
d
at
a
q
u
ali
ty
a
n
d
e
n
s
u
r
i
n
g
o
p
ti
m
al
p
e
r
f
o
r
m
an
ce
o
f
m
ac
h
i
n
e
le
ar
n
i
n
g
m
o
d
els
[
1
6
]
.
I
n
NL
P
,
t
e
x
t
p
r
e
p
r
o
c
ess
i
n
g
p
la
y
s
a
v
it
al
r
o
l
e
i
n
o
r
g
a
n
i
zi
n
g
a
n
d
r
e
f
i
n
i
n
g
te
x
t
u
al
d
at
a,
wh
ic
h
i
n
t
u
r
n
i
m
p
r
o
v
es
d
ata
q
u
a
lit
y
an
d
th
e
e
f
f
ec
ti
v
e
n
ess
o
f
l
at
er
a
n
al
y
s
es
.
T
h
e
c
u
r
r
e
n
t
s
tu
d
y
f
o
c
u
s
es
o
n
twe
ets
r
elat
e
d
t
o
o
n
li
n
e
lea
r
n
i
n
g
d
u
r
i
n
g
t
h
e
C
OVI
D
-
1
9
p
a
n
d
e
m
ic
,
w
h
ic
h
wer
e
c
o
ll
ec
t
ed
u
s
i
n
g
t
h
e
T
wi
tt
er
AP
I
.
T
h
e
r
a
w
d
a
ta
,
wh
ic
h
c
o
n
tai
n
ed
r
e
d
u
n
d
an
t
in
f
o
r
m
at
io
n
an
d
d
u
p
li
ca
t
e
e
n
t
r
ies,
was
p
r
o
ce
s
s
ed
a
n
d
cl
ea
n
e
d
t
o
e
n
s
u
r
e
i
ts
s
u
ita
b
ilit
y
f
o
r
a
n
al
y
s
is
.
D
u
r
i
n
g
t
h
e
d
at
a
c
o
l
le
cti
o
n
p
h
ase
,
w
e
i
n
iti
a
lly
g
at
h
e
r
e
d
a
p
p
r
o
x
i
m
a
tel
y
3
2
6
,
1
4
0
tw
ee
ts
.
T
o
e
n
s
u
r
e
d
at
a
q
u
al
it
y
,
all
d
u
p
li
ca
t
e
twee
ts
w
e
r
e
r
em
o
v
e
d
f
r
o
m
t
h
e
d
at
ase
t,
r
es
u
lti
n
g
i
n
a
f
i
n
a
l
d
at
ase
t
o
f
3
1
0
,
5
4
7
u
n
iq
u
e
t
we
ets.
T
h
e
d
a
tase
t
was
th
e
n
p
r
ep
r
o
c
ess
e
d
u
s
i
n
g
a
u
s
er
-
d
e
f
i
n
e
d
f
u
n
cti
o
n
b
u
il
t
wit
h
N
L
T
K
,
a
wi
d
el
y
u
s
e
d
P
y
t
h
o
n
t
o
o
l
k
it
f
o
r
NL
P
.
T
h
e
p
r
e
p
r
o
ce
s
s
in
g
p
r
o
ce
d
u
r
e
in
v
o
lv
e
d
s
tan
d
ar
d
izin
g
t
h
e
t
ex
tu
al
d
ata
b
y
c
o
n
v
e
r
tin
g
all
twee
ts
to
lo
wer
ca
s
e.
Ad
d
itio
n
al
clea
n
in
g
s
tep
s
in
clu
d
ed
r
em
o
v
in
g
ex
tr
a
s
p
ac
es,
n
u
m
er
ic
v
alu
es,
s
p
ec
ial
s
y
m
b
o
ls
,
UR
L
s
,
p
u
n
ctu
atio
n
,
an
d
co
m
m
o
n
ly
u
s
ed
s
to
p
w
o
r
d
s
.
Me
n
tio
n
s
an
d
h
ash
tag
s
wer
e
also
elim
in
ated
to
a
v
o
id
in
tr
o
d
u
cin
g
b
ias
d
u
r
in
g
f
ea
tu
r
e
ex
tr
ac
tio
n
.
T
h
e
twee
ts
wer
e
th
en
to
k
en
ized
,
s
p
litt
in
g
th
e
t
ex
t
in
to
in
d
iv
id
u
al
wo
r
d
s
o
r
m
ea
n
in
g
f
u
l
p
h
r
ases
.
L
em
m
atiza
tio
n
was
ap
p
lied
to
co
n
v
er
t
wo
r
d
s
in
to
th
eir
b
a
s
e
f
o
r
m
s
(
lem
m
as),
en
s
u
r
in
g
t
h
at
d
if
f
er
en
t
f
o
r
m
s
o
f
a
w
o
r
d
(
e.
g
.
,
“
s
tu
d
y
in
g
,
”
“
s
tu
d
ied
,
”
an
d
“
s
tu
d
ies
”
)
wer
e
tr
ea
ted
as
a
s
in
g
le
r
ep
r
esen
tatio
n
,
wh
ich
h
elp
ed
n
o
r
m
alize
th
e
d
ataset.
T
h
ese
p
r
ep
r
o
ce
s
s
in
g
s
tep
s
ef
f
ec
tiv
ely
p
r
ep
ar
e
d
th
e
twee
ts
f
o
r
s
u
b
s
eq
u
en
t
a
n
aly
s
is
.
Fig
u
r
e
2
illu
s
tr
ates
th
e
d
aily
d
is
tr
ib
u
tio
n
o
f
twee
ts
,
h
ig
h
lig
h
tin
g
s
u
b
m
is
s
io
n
f
r
eq
u
e
n
cies
o
v
er
tim
e.
T
h
is
v
is
u
aliza
tio
n
p
r
o
v
id
es
in
s
ig
h
ts
in
to
tem
p
o
r
al
p
atter
n
s
,
r
e
v
ea
lin
g
p
er
io
d
s
o
f
h
eig
h
ten
ed
o
r
r
ed
u
ce
d
ac
tiv
ity
in
th
e
d
ataset.
T
ab
le
1
.
Featu
r
es a
n
d
d
ata
ty
p
es o
f
th
e
co
llected
twee
t d
atas
et
F
e
a
t
u
r
e
D
e
scri
p
t
i
o
n
D
a
t
a
t
y
p
e
Id
U
n
i
q
u
e
i
d
e
n
t
i
f
i
e
r
f
o
r
e
a
c
h
t
w
e
e
t
N
u
mb
e
r
C
o
n
t
e
n
t
Te
x
t
o
f
t
h
e
t
w
e
e
t
Te
x
t
Lo
c
a
t
i
o
n
Lo
c
a
t
i
o
n
o
f
t
h
e
u
ser (
i
f
a
v
a
i
l
a
b
l
e
)
Te
x
t
U
sern
a
m
e
Tw
i
t
t
e
r
u
ser
n
a
m
e
o
f
t
h
e
t
w
e
e
t
a
u
t
h
o
r
Te
x
t
R
e
t
w
e
e
t
c
o
u
n
t
N
u
mb
e
r
o
f
r
e
t
w
e
e
t
s
N
u
mb
e
r
F
a
v
o
r
i
t
e
N
u
mb
e
r
o
f
l
i
k
e
s
o
r
f
a
v
o
r
i
t
e
s
N
u
mb
e
r
C
r
e
a
t
e
d
a
t
Ti
mes
t
a
m
p
w
h
e
n
t
h
e
t
w
e
e
t
w
a
s
p
o
s
t
e
d
D
a
t
e
T
i
me
Fig
u
r
e
2
.
T
e
m
p
o
r
al
d
is
tr
ib
u
ti
o
n
o
f
C
OVI
D
-
19
t
wee
ts
in
th
e
c
o
r
p
u
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
.
3
,
J
u
n
e
2
0
2
6
:
2
7
3
6
-
2
7
4
9
2740
2
.
4
.
I
dentif
y
ing
wo
rd
t
re
nd
s
wit
h
n
-
g
ra
m
s
I
n
th
e
f
ield
o
f
NL
P,
n
-
g
r
am
m
o
d
els
ar
e
f
r
e
q
u
en
tly
ap
p
lied
t
o
an
aly
ze
s
tatis
tical
p
atter
n
s
an
d
ca
p
tu
r
e
s
y
n
tactic
s
tr
u
ctu
r
es
[
1
9
]
.
T
o
e
x
p
lo
r
e
th
e
to
k
en
ize
d
co
r
p
u
s
,
we
co
m
p
u
ted
th
e
f
r
e
q
u
en
c
y
o
f
s
in
g
le
wo
r
d
s
an
d
ad
jace
n
t
wo
r
d
co
m
b
in
atio
n
s
t
h
r
o
u
g
h
a
n
n
-
g
r
a
m
ap
p
r
o
ac
h
.
W
e
th
en
ap
p
lie
d
th
e
ch
ain
r
u
le
o
f
p
r
o
b
ab
ilit
y
t
o
esti
m
ate
th
e
lik
elih
o
o
d
o
f
s
p
ec
if
ic
wo
r
d
s
eq
u
e
n
ce
s
ap
p
ea
r
in
g
,
as d
ef
in
ed
i
n
(
1
)
.
(
1
,
2
,
3
,
…
,
ₙ
)
=
(
1
)
(
2
|
1
)
(
3
|
1
,
2
)
…
(
ₙ
|
1
,
2
,
3
,
…
,
ₙ
−
1
)
=
∏
=
1
(
ᵢ
|
1
−
1
)
(
1
)
F
o
r
e
x
a
m
p
l
e
,
a
s
e
n
t
e
n
ce
c
a
n
b
e
c
o
n
s
i
d
e
r
e
d
as
“
o
n
li
n
e
l
e
a
r
n
i
n
g
i
s
e
x
p
a
n
d
i
n
g
”
.
B
as
e
d
o
n
t
h
e
c
h
a
i
n
r
u
l
e
o
f
p
r
o
b
a
b
i
l
i
t
y
,
t
h
e
p
r
o
b
a
b
i
l
i
t
y
o
f
t
h
e
s
e
n
t
e
n
c
e
c
a
n
b
e
e
x
p
r
e
s
s
e
d
a
s
:
P
(
“
o
n
l
i
n
e
l
e
a
r
n
i
n
g
i
s
e
x
p
a
n
d
i
n
g
”
)
=
P
(
“
o
n
l
i
n
e
”
)
×
P
(
“
l
e
a
r
n
i
n
g
”
|
“
o
n
l
i
n
e
”
)
×
P
(
“i
s
”
|
“
o
n
l
i
n
e
l
ea
r
n
i
n
g
”
)
×
P
(
“
e
x
p
a
n
d
i
n
g
”
|
“
o
n
l
i
n
e
l
e
a
r
n
i
n
g
i
s
”
)
.
In
(
2
)
,
t
h
e
c
h
a
i
n
r
u
l
e
o
f
p
r
o
b
a
b
i
l
i
t
y
is
u
s
e
d
t
o
d
et
e
r
m
i
n
e
t
h
e
p
r
o
b
a
b
i
l
i
t
i
es
o
f
w
o
r
d
s
i
n
e
a
c
h
s
en
t
e
n
c
e
.
(
1
)
=
∏
ₗ
(
ⱼ
|
₁
,
₂
,
₃
,
…
,
ⱼ
₋₁
)
=
∏
=
1
(
ⱼ
|
1
−
1
)
(
2
)
T
h
e
p
r
in
cip
le
o
f
th
e
Ma
r
k
o
v
ass
u
m
p
tio
n
ass
er
ts
th
at
a
wo
r
d
’
s
o
cc
u
r
r
en
ce
is
co
n
d
itio
n
ed
s
o
lely
o
n
th
e
wo
r
d
d
ir
ec
tly
b
e
f
o
r
e
it.
T
h
is
m
o
d
el
allo
ws
th
e
p
r
e
d
ictio
n
o
f
a
f
u
t
u
r
e
wo
r
d
b
ased
o
n
th
e
im
m
ed
iate
p
ast,
r
ath
er
th
an
r
ely
in
g
o
n
th
e
e
n
tire
s
eq
u
en
ce
o
f
p
r
ev
io
u
s
wo
r
d
s
.
I
n
th
e
b
ig
r
am
m
o
d
el,
th
e
p
r
o
b
ab
ilit
y
o
f
a
wo
r
d
is
co
m
p
u
ted
u
s
in
g
th
e
co
n
d
iti
o
n
al
p
r
o
b
a
b
ilit
y
(
|
−
1
)
,
wh
ich
tak
es
in
to
ac
co
u
n
t
o
n
ly
th
e
p
r
e
ce
d
in
g
wo
r
d
,
with
o
u
t c
o
n
s
id
er
in
g
an
y
o
th
er
p
r
io
r
wo
r
d
s
[
2
0
]
as in
(
3
)
.
(
1
,
2
)
=
∏
(
=
2
)
(
2
|
1
)
(
3
)
T
h
e
f
o
r
m
u
la
f
o
r
ca
lcu
latin
g
th
e
b
ig
r
am
p
r
o
b
ab
ilit
y
is
(
4
)
.
(
|
−
1
)
=
(
(
−
1
)
,
)
(
(
−
1
)
)
(
4
)
T
h
e
n
-
g
r
am
m
o
d
el
was
u
s
ed
t
o
ex
tr
ac
t
th
e
m
o
s
t
co
m
m
o
n
u
n
ig
r
am
s
,
b
ig
r
am
s
,
an
d
tr
ig
r
am
s
f
r
o
m
th
e
d
ataset.
T
h
is
an
aly
s
is
s
h
ed
s
li
g
h
t
o
n
p
r
o
m
in
en
t
to
p
ics,
u
n
co
v
er
in
g
th
e
k
ey
d
r
iv
er
s
o
f
s
en
tim
en
t
an
d
p
r
o
v
id
in
g
d
ee
p
er
in
s
ig
h
ts
in
t
o
h
u
m
an
em
o
tio
n
s
ex
p
r
ess
ed
in
twee
ts
ab
o
u
t
o
n
lin
e
co
u
r
s
es
an
d
l
ea
r
n
in
g
d
u
r
in
g
th
e
C
OVI
D
-
1
9
p
an
d
em
ic.
Am
o
n
g
th
ese,
u
n
ig
r
am
s
ap
p
ea
r
e
d
m
o
s
t
f
r
eq
u
e
n
tly
,
f
o
llo
wed
b
y
b
ig
r
am
s
,
with
tr
ig
r
a
m
s
b
ein
g
th
e
least
co
m
m
o
n
.
Fig
u
r
e
3
s
h
o
ws
th
e
to
p
3
0
u
n
i
g
r
am
s
,
b
ig
r
am
s
,
an
d
tr
ig
r
am
s
,
alo
n
g
with
th
eir
r
esp
ec
tiv
e
o
cc
u
r
r
en
ce
co
u
n
ts
.
2
.
5
.
O
nli
ne
lea
rning
-
s
pecif
ic
wo
rd
det
ec
t
io
n durin
g
CO
V
I
D
-
19
Fo
llo
win
g
th
e
c
o
m
p
letio
n
o
f
t
h
e
p
r
e
p
r
o
ce
s
s
in
g
s
tag
e,
a
b
ag
-
of
-
wo
r
d
s
(B
o
W
)
m
o
d
el
was
d
ev
elo
p
e
d
u
s
in
g
th
e
m
o
s
t
co
m
m
o
n
ly
o
cc
u
r
r
in
g
ter
m
s
f
r
o
m
th
e
d
ataset,
h
ig
h
lig
h
tin
g
k
ey
wo
r
d
s
r
elate
d
to
o
n
lin
e
lear
n
in
g
an
d
ed
u
ca
tio
n
d
u
r
i
n
g
th
e
C
OVI
D
-
1
9
p
an
d
em
ic.
Ad
d
itio
n
ally
,
a
wo
r
d
clo
u
d
was
g
en
er
ated
to
p
r
o
v
id
e
a
v
is
u
ally
ap
p
ea
lin
g
r
e
p
r
esen
tat
io
n
o
f
th
e
ex
tr
ac
ted
tex
t
d
ata
,
as
s
h
o
wn
i
n
Fig
u
r
e
4
.
Sev
er
al
wo
r
d
s
h
a
v
e
b
ee
n
r
ec
o
g
n
ized
f
r
o
m
th
e
cu
r
ate
d
lex
ico
n
,
o
cc
u
r
r
in
g
at
d
if
f
er
e
n
t
p
o
in
ts
an
d
in
d
iv
er
s
e
co
n
tex
ts
th
r
o
u
g
h
o
u
t
th
e
twee
ts
.
An
aly
zin
g
th
e
h
ig
h
est
f
r
eq
u
e
n
cy
o
cc
u
r
r
in
g
wo
r
d
s
in
th
e
d
ata
s
et
p
r
o
v
i
d
es
v
alu
ab
le
in
s
ig
h
ts
in
to
th
eir
in
f
lu
en
ce
o
n
t
h
e
o
v
er
all
s
en
tim
en
t
o
f
th
e
twee
ts
.
Ap
p
r
o
x
im
ately
9
,
8
7
1
m
illi
o
n
wo
r
d
s
wer
e
id
en
tifie
d
f
r
o
m
th
e
d
ataset.
T
o
p
in
p
o
in
t
t
h
e
m
o
s
t
p
r
o
m
in
e
n
t
wo
r
d
s
co
n
ce
r
n
in
g
o
n
lin
e
c
o
u
r
s
es
an
d
lear
n
in
g
d
u
r
in
g
th
e
C
OVI
D
-
1
9
p
an
d
em
ic,
th
e
f
r
e
q
u
en
c
y
o
f
e
ac
h
to
k
en
was
ass
es
s
ed
.
A
p
o
p
u
lar
ity
s
co
r
e
f
o
r
ea
ch
wo
r
d
was
th
en
co
m
p
u
ted
,
o
f
f
er
in
g
a
q
u
an
titativ
e
m
ea
s
u
r
e
o
f
its
p
r
o
m
i
n
en
ce
with
in
th
e
d
ataset.
Af
ter
id
en
tify
in
g
th
e
wo
r
d
f
r
eq
u
en
cies,
th
e
lik
elih
o
o
d
o
f
o
cc
u
r
r
en
ce
f
o
r
ea
ch
wo
r
d
was
ca
lcu
lated
b
ased
o
n
th
e
to
tal
o
f
wo
r
d
s
in
th
e
d
ataset.
T
ab
le
2
s
h
o
ws
th
e
p
r
o
m
in
e
n
ce
an
d
ass
o
ciate
d
p
r
o
b
ab
ilit
y
m
ea
s
u
r
es
o
f
th
e
wo
r
d
s
ap
p
ea
r
in
g
m
o
s
t
f
r
eq
u
e
n
tly
in
th
e
d
ataset.
T
h
e
p
r
o
b
a
b
ilit
y
o
f
ea
ch
wo
r
d
in
th
e
d
ataset
ca
n
b
e
ca
lcu
lated
u
s
in
g
(
5
)
,
wh
ich
n
o
r
m
alize
s
th
e
f
r
eq
u
en
cy
o
f
ea
ch
wo
r
d
b
y
d
iv
id
i
n
g
its
co
u
n
t
b
y
th
e
to
tal
wo
r
d
co
u
n
t in
t
h
e
d
ataset.
(
)
=
(
)
∑
(
=
0
)
=
0
(
5
)
T
h
is
n
o
r
m
aliza
tio
n
m
et
h
o
d
p
r
o
v
id
es
a
clea
r
er
p
e
r
s
p
ec
tiv
e
o
n
th
e
r
elativ
e
f
r
e
q
u
en
c
y
o
f
w
o
r
d
s
in
th
e
d
ataset,
h
elp
in
g
to
i
d
en
tify
k
ey
th
em
es.
Fo
r
i
n
s
tan
ce
,
t
h
e
wo
r
d
“
o
n
lin
e
”
s
tan
d
s
o
u
t
with
th
e
h
ig
h
est
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
Hyb
r
id
d
ee
p
lea
r
n
in
g
fo
r
s
en
timen
t a
n
a
lysi
s
o
f o
n
lin
e
s
tu
d
e
n
t e
xp
erien
ce
s
(
R
a
ja
Ou
a
d
a
d
)
2741
f
r
eq
u
e
n
cy
—
1
3
4
,
1
1
2
o
cc
u
r
r
en
ce
s
—
r
ep
r
esen
tin
g
a
p
r
o
b
ab
il
ity
o
f
0
.
0
2
7
(
2
.
7
1
%).
Simil
ar
ly
,
“
co
u
r
s
e
”
an
d
“
lear
n
in
g
”
f
o
llo
w
with
1
1
2
,
2
2
6
o
cc
u
r
r
e
n
ce
s
(
2
.
2
6
%)
an
d
6
4
,
3
6
7
o
cc
u
r
r
e
n
ce
s
(
1
.
3
0
%),
r
esp
ec
tiv
ely
.
T
h
ese
p
r
o
m
in
e
n
t
f
ig
u
r
es
u
n
d
er
lin
e
t
h
e
s
tr
o
n
g
em
p
h
asis
o
n
o
n
lin
e
p
latf
o
r
m
s
a
n
d
ed
u
ca
tio
n
al
co
n
ten
t
with
in
th
e
d
ataset.
Oth
er
ter
m
s
,
s
u
ch
a
s
“
class
”
(
6
1
,
1
0
3
o
cc
u
r
r
e
n
ce
s
,
1
.
2
3
%)
an
d
“
s
tu
d
en
t
”
(
4
1
,
4
7
5
o
cc
u
r
r
e
n
ce
s
,
0
.
8
4
%),
f
u
r
th
er
r
ein
f
o
r
ce
th
e
lear
n
er
-
ce
n
t
r
ic
n
atu
r
e
o
f
th
e
d
i
s
cu
s
s
io
n
s
.
C
o
llect
iv
ely
,
th
ese
in
s
ig
h
ts
r
ev
ea
l
th
e
ce
n
tr
al
f
o
cu
s
o
n
o
n
lin
e
lear
n
in
g
an
d
c
o
u
r
s
e
ex
p
er
ien
ce
s
in
th
e
an
aly
ze
d
c
o
n
ten
t.
Fig
u
r
e
3
.
T
o
p
3
0
u
n
ig
r
am
s
,
b
i
g
r
am
s
,
an
d
t
r
ig
r
am
s
with
th
eir
o
cc
u
r
r
e
n
ce
co
u
n
ts
Fig
u
r
e
4
.
W
o
r
d
clo
u
d
s
h
ig
h
lig
h
tin
g
k
e
y
ter
m
s
in
o
n
lin
e
lear
n
in
g
d
u
r
in
g
th
e
p
an
d
em
ic
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
.
3
,
J
u
n
e
2
0
2
6
:
2
7
3
6
-
2
7
4
9
2742
T
ab
le
2
.
Fre
q
u
en
cy
an
d
p
r
o
b
a
b
ilit
y
o
f
h
ig
h
ly
o
cc
u
r
r
in
g
wo
r
d
s
W
o
r
d
P
o
p
u
l
a
r
i
t
y
P
r
o
b
a
b
i
l
i
t
y
O
n
l
i
n
e
1
3
4
,
1
1
2
0
.
0
2
7
0
6
0
C
o
u
r
se
1
1
2
,
2
2
6
0
.
0
2
2
6
4
4
Le
a
r
n
i
n
g
64
,
367
0
.
0
1
2
9
8
7
C
l
a
s
s
61
,
103
0
.
0
1
2
3
2
9
S
t
u
d
e
n
t
41
,
475
0
.
0
0
8
3
6
9
2
.
6
.
Sentim
ent
a
na
ly
s
is
a
nd
cla
s
s
if
ica
t
io
n us
ing
na
t
ura
l l
a
ng
ua
g
e
t
o
o
lk
it
I
n
r
ec
e
n
t
y
ea
r
s
,
d
ee
p
l
ea
r
n
in
g
-
b
as
ed
s
e
n
ti
m
e
n
t
an
al
y
s
is
h
as
g
ain
e
d
s
i
g
n
if
ica
n
t
p
o
p
u
la
r
it
y
f
o
r
d
e
r
i
v
i
n
g
in
s
i
g
h
ts
a
n
d
p
r
e
d
i
cti
n
g
te
x
t
t
o
e
v
al
u
ate
p
u
b
lic
r
es
p
o
n
s
es
t
o
d
if
f
e
r
e
n
t
ev
e
n
ts
[
2
1
]
.
T
h
is
m
et
h
o
d
is
m
a
in
ly
em
p
l
o
y
e
d
t
o
m
o
n
it
o
r
s
h
i
f
ts
i
n
p
u
b
lic
o
p
i
n
i
o
n
an
d
s
e
n
t
im
en
t.
I
t
b
el
o
n
g
s
to
t
h
e
f
i
el
d
o
f
d
at
a
m
i
n
i
n
g
an
d
u
ti
liz
es
co
m
p
u
tati
o
n
a
l
l
in
g
u
is
t
ics
a
n
d
s
tr
a
te
g
ies
f
o
r
th
e
ex
am
in
ati
o
n
an
d
a
n
al
y
s
is
o
f
te
x
t
.
B
y
ex
am
i
n
i
n
g
u
s
e
r
-
g
en
er
at
ed
s
u
b
je
cti
v
e
i
n
f
o
r
m
ati
o
n
o
n
s
o
ci
al
n
e
tw
o
r
k
i
n
g
p
l
at
f
o
r
m
s
,
s
e
n
t
i
m
e
n
t
a
n
al
y
s
is
a
ll
o
ws
f
o
r
t
h
e
c
lass
i
f
ic
ati
o
n
o
f
te
x
t
in
t
o
d
if
f
e
r
e
n
t
s
e
n
t
im
e
n
t
ca
t
e
g
o
r
ies
,
o
f
f
e
r
i
n
g
v
al
u
ab
le
i
n
s
i
g
h
ts
in
t
o
p
u
b
li
c
f
e
eli
n
g
s
a
n
d
a
ttit
u
d
es
[
2
2
]
.
T
h
e
NL
T
K
s
en
tim
en
t
an
aly
ze
r
is
a
wid
ely
u
s
ed
to
o
l
f
o
r
ev
al
u
atin
g
s
en
tim
en
t
p
o
la
r
ity
.
R
en
o
wn
ed
f
o
r
p
r
o
v
id
i
n
g
r
eliab
le
an
d
ac
cu
r
at
e
r
esu
lts
in
tex
t
an
al
y
s
is
,
th
e
s
en
tim
en
t
an
aly
s
is
f
u
n
ctio
n
ali
ty
f
r
o
m
th
e
NL
T
K
lib
r
ar
y
was
em
p
lo
y
ed
i
n
th
is
s
tu
d
y
to
e
v
alu
ate
th
e
s
en
tim
en
t
p
o
lar
ities
o
f
p
r
ep
r
o
ce
s
s
ed
twee
ts
.
T
h
is
to
o
l
ap
p
lies
p
r
ed
ef
i
n
ed
r
u
les
an
d
h
eu
r
is
tics
to
p
er
f
o
r
m
a
d
etail
ed
s
en
tim
en
t
ass
ess
m
en
t.
Giv
en
th
at
th
e
d
ataset
co
n
s
is
ts
o
f
r
aw
twee
ts
with
o
u
t
lab
eled
s
en
tim
en
t
class
es,
s
en
tim
en
t
an
aly
s
is
is
ess
en
tial
f
o
r
u
n
d
er
s
tan
d
in
g
th
e
d
is
tr
ib
u
tio
n
o
f
twee
ts
ac
r
o
s
s
v
ar
io
u
s
s
en
tim
en
t
ca
teg
o
r
ies.
E
ac
h
twee
t’
s
s
en
tim
en
t
p
o
lar
ity
is
co
m
p
u
ted
u
s
in
g
NL
T
K’
s
s
en
tim
en
t
an
aly
ze
r
,
wh
ich
g
en
e
r
ates
s
co
r
es
f
o
r
p
o
s
itiv
e,
n
eg
ativ
e,
a
n
d
n
eu
tr
al
s
en
tim
en
ts
,
as
well
as
an
o
v
er
all
c
o
m
p
o
u
n
d
s
en
tim
en
t sco
r
e.
T
h
e
o
v
er
all
s
en
tim
en
t
(
co
m
p
o
u
n
d
s
co
r
e)
is
o
b
tain
e
d
b
y
s
u
m
m
in
g
th
e
v
alen
ce
s
co
r
es
o
f
th
e
wo
r
d
s
p
r
esen
t
in
t
h
e
lex
ico
n
,
w
h
ich
ar
e
th
e
n
ad
j
u
s
ted
ac
co
r
d
in
g
to
a
s
et
o
f
p
r
ed
e
f
in
ed
r
u
les.
T
h
ese
s
co
r
es
ar
e
n
o
r
m
alize
d
to
a
r
a
n
g
e
b
etwe
en
-
1
(
ex
tr
em
ely
n
eg
ati
v
e)
a
n
d
+1
(
ex
t
r
em
ely
p
o
s
itiv
e)
,
p
r
o
v
id
i
n
g
a
s
in
g
le,
u
n
id
im
en
s
io
n
al
m
ea
s
u
r
e
o
f
s
e
n
tim
en
t f
o
r
th
e
e
n
tire
s
en
ten
ce
.
A
co
m
p
o
u
n
d
s
co
r
e
cl
o
s
er
to
+1
in
d
icate
s
a
m
o
r
e
p
o
s
itiv
e
s
en
tim
en
t,
w
h
ile
a
s
c
o
r
e
clo
s
er
to
-
1
r
ef
lects
a
m
o
r
e
n
e
g
ativ
e
s
en
tim
en
t.
T
wee
ts
ar
e
class
if
ied
in
t
o
th
r
ee
s
en
tim
en
t
ca
teg
o
r
ies
—
p
o
s
itiv
e,
n
eg
ativ
e,
o
r
n
eu
tr
al
—
b
ased
o
n
th
eir
c
o
m
p
o
u
n
d
p
o
l
ar
ity
s
co
r
es.
I
f
th
e
p
o
lar
ity
s
co
r
e
is
g
r
ea
ter
t
h
an
0
,
th
e
twee
t
is
co
n
s
id
er
ed
p
o
s
itiv
e;
if
less
th
an
0
,
it'
s
n
eg
ati
v
e;
an
d
if
th
e
s
co
r
e
eq
u
als
0
,
th
e
twee
t
is
d
ee
m
ed
n
eu
tr
al.
T
h
e
s
en
tim
en
t
class
if
icatio
n
o
f
o
u
r
d
ataset
is
v
is
u
ally
r
ep
r
esen
ted
in
Fig
u
r
e
5
,
w
h
er
e
twee
ts
ar
e
ca
t
eg
o
r
ized
i
n
to
p
o
s
itiv
e,
n
eu
tr
al
,
an
d
n
eg
ativ
e
s
en
tim
en
ts
.
T
h
e
f
ig
u
r
e
s
h
o
ws
th
at
5
9
.
3
1
%
o
f
th
e
twee
ts
ex
p
r
ess
a
p
o
s
itiv
e
s
en
tim
en
t,
w
h
ile
2
3
.
0
8
%
ar
e
n
eu
tr
al,
an
d
1
7
.
6
1
%
r
ef
lect
n
eg
ativ
e
s
en
tim
en
t.
Po
s
itiv
e
an
d
n
eg
ativ
e
twee
ts
ar
e
f
u
r
th
er
an
aly
ze
d
f
o
r
a
m
o
r
e
g
r
an
u
lar
s
en
tim
en
t
ass
es
s
m
en
t,
p
r
o
v
id
i
n
g
a
d
ee
p
er
u
n
d
e
r
s
tan
d
in
g
o
f
t
h
e
o
v
er
all
s
en
tim
en
t
d
is
tr
ib
u
tio
n
in
r
elatio
n
to
o
n
li
n
e
lear
n
in
g
d
u
r
in
g
th
e
C
OVI
D
-
1
9
p
an
d
em
ic.
Fig
u
r
e
5
.
Sen
tim
en
t
d
is
tr
ib
u
tio
n
o
f
t
wee
ts
u
s
in
g
NL
T
K
2
.
7
.
Dev
el
o
pm
ent
o
f
t
he
lo
g
i
s
t
ic
re
g
re
s
s
io
n sentim
ent
m
o
del
LR
is
a
wid
ely
u
s
ed
s
tati
s
tical
m
o
d
el
f
o
r
b
in
ar
y
class
if
icatio
n
task
s
,
m
ak
in
g
it
a
s
u
itab
le
ch
o
ice
f
o
r
s
en
tim
en
t
an
aly
s
is
[
2
3
]
.
I
t
is
p
ar
ticu
lar
ly
ef
f
ec
tiv
e
i
n
h
ig
h
-
d
im
en
s
io
n
al
s
p
ac
es,
s
u
ch
as
tex
t
d
ata,
an
d
o
f
f
er
s
in
ter
p
r
etab
ilit
y
th
r
o
u
g
h
its
p
r
o
b
ab
ilis
tic
p
r
ed
ictio
n
s
.
Un
lik
e
lin
ea
r
r
eg
r
ess
io
n
,
wh
ich
p
r
o
d
u
ce
s
co
n
tin
u
o
u
s
o
u
tp
u
ts
,
LR
ap
p
lies
th
e
s
ig
m
o
id
f
u
n
ctio
n
to
m
a
p
th
e
weig
h
ted
s
u
m
o
f
i
n
p
u
t
f
ea
tu
r
es
to
a
p
r
o
b
ab
ilit
y
s
co
r
e
b
etwe
en
0
an
d
1
[
2
4
]
.
T
h
e
p
r
o
b
ab
ilit
y
o
f
a
n
in
s
tan
ce
b
elo
n
g
in
g
to
th
e
p
o
s
itiv
e
s
en
tim
en
t
class
(
Y
=1
)
is
co
m
p
u
ted
as in
(
6
)
.
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
Hyb
r
id
d
ee
p
lea
r
n
in
g
fo
r
s
en
timen
t a
n
a
lysi
s
o
f o
n
lin
e
s
tu
d
e
n
t e
xp
erien
ce
s
(
R
a
ja
Ou
a
d
a
d
)
2743
(
=
1
∣
)
=
1
1
+
−
(
+
)
(
6
)
W
h
er
e
d
en
o
tes
th
e
f
ea
tu
r
e
v
e
cto
r
d
er
i
v
ed
f
r
o
m
th
e
i
n
p
u
t
tex
t,
r
ep
r
esen
ts
th
e
weig
h
t
co
e
f
f
icien
ts
lear
n
ed
d
u
r
in
g
th
e
tr
ain
in
g
p
r
o
ce
s
s
,
is
th
e
b
ias ter
m
,
an
d
co
r
r
esp
o
n
d
s
to
E
u
ler
’
s
n
u
m
b
er
.
I
n
th
is
s
tu
d
y
,
we
em
p
lo
y
a
s
en
tim
en
t
class
if
icatio
n
m
o
d
el
b
ased
o
n
LR
to
r
ef
in
e
th
e
p
o
lar
ity
r
atin
g
s
o
f
twee
ts
.
LR
,
a
d
is
cr
im
in
ativ
e
m
o
d
el,
is
c
h
o
s
en
f
o
r
its
co
m
p
u
tatio
n
al
ef
f
icien
cy
,
i
n
ter
p
r
et
ab
ilit
y
,
an
d
ab
ilit
y
to
h
an
d
le
lar
g
e
-
s
ca
le
tex
tu
al
d
ata
[
2
5
]
.
T
h
e
m
o
d
el
is
o
p
tim
iz
ed
u
s
in
g
m
a
x
im
u
m
lik
elih
o
o
d
esti
m
atio
n
(
ML
E
)
,
en
s
u
r
in
g
an
ef
f
ec
tiv
e
d
ec
is
io
n
b
o
u
n
d
ar
y
b
etwe
en
s
en
tim
e
n
t
class
es.
Du
r
in
g
f
ea
tu
r
e
ex
tr
ac
tio
n
,
we
ap
p
l
y
n
-
g
r
am
m
o
d
elin
g
to
ca
p
tu
r
e
u
n
ig
r
am
s
,
b
ig
r
am
s
,
an
d
tr
ig
r
am
s
,
alo
n
g
s
id
e
d
o
m
ain
-
s
p
ec
if
ic
wo
r
d
s
f
r
o
m
th
e
B
o
W
m
o
d
el.
T
h
e
ex
tr
ac
ted
f
ea
tu
r
es
ar
e
t
r
an
s
f
o
r
m
e
d
u
s
in
g
ter
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
(
TF
-
I
DF
)
to
n
o
r
m
alize
ter
m
im
p
o
r
tan
ce
,
r
ed
u
cin
g
th
e
i
n
f
l
u
en
ce
o
f
h
ig
h
-
f
r
eq
u
e
n
cy
wo
r
d
s
.
T
h
e
m
o
d
el
is
tr
ain
ed
o
n
a
d
ataset
co
m
p
r
is
in
g
3
2
0
,
0
0
0
tex
tu
al
twee
ts
,
u
s
in
g
th
e
lb
f
g
s
s
o
lv
e
r
with
1
,
0
0
0
iter
atio
n
s
to
e
n
s
u
r
e
co
n
v
er
g
en
ce
in
h
ig
h
-
d
im
en
s
i
o
n
al
s
p
ac
es.
E
ac
h
n
ew
twee
t
is
tr
an
s
f
o
r
m
ed
in
to
a
T
F
-
I
DF
f
ea
tu
r
e
v
ec
to
r
an
d
p
ass
ed
th
r
o
u
g
h
th
e
lo
g
is
tic
f
u
n
ctio
n
to
co
m
p
u
te
t
h
e
p
r
ed
ict
ed
s
en
tim
en
t
class
an
d
its
ass
o
ciate
d
p
r
o
b
ab
ilit
y
s
co
r
e,
p
r
o
v
id
in
g
p
r
o
b
ab
ilis
tic
co
n
f
id
en
ce
lev
els ess
en
tial f
o
r
s
en
tim
en
t c
lass
if
icatio
n
ap
p
licatio
n
s
.
T
h
e
class
if
ier
p
r
ed
icts
th
e
p
r
o
b
ab
ilit
y
o
f
ea
ch
twee
t
b
elo
n
g
i
n
g
to
th
e
p
o
s
itiv
e
o
r
n
eg
ativ
e
s
en
tim
en
t
class
,
with
s
co
r
es
b
elo
w
0
.
5
r
ef
lectin
g
n
e
g
ativ
e
s
en
tim
en
t
a
n
d
s
co
r
es
a
b
o
v
e
0
.
5
r
e
f
lectin
g
p
o
s
itiv
e
s
en
tim
en
t.
Usi
n
g
th
e
class
if
icatio
n
alg
o
r
ith
m
p
r
esen
ted
in
Alg
o
r
ith
m
1
,
th
e
twee
ts
ar
e
f
u
r
th
er
ca
te
g
o
r
ized
in
to
h
ig
h
l
y
p
o
s
itiv
e
(
≥
0
.
7
5
)
,
p
o
s
itiv
e
(
<
0
.
7
5
)
,
n
eg
ativ
e
(
>
0
.
2
5
)
,
a
n
d
h
ig
h
ly
n
e
g
ativ
e
(
≤
0
.
2
5
)
.
T
h
e
s
en
t
im
en
t
class
if
icatio
n
r
esu
lts
ar
e
f
u
r
th
er
r
ef
in
ed
b
y
in
co
r
p
o
r
atin
g
n
e
u
tr
al
twee
ts
with
s
en
tim
en
t
s
co
r
es
d
er
iv
ed
f
r
o
m
an
alter
n
ativ
e
class
if
icatio
n
ap
p
r
o
ac
h
.
As
illu
s
tr
ated
in
Fig
u
r
e
6
,
th
e
r
e
f
in
e
d
s
en
tim
en
t
d
is
tr
ib
u
tio
n
s
h
o
w
s
5
2
.
5
7
%
o
f
twee
ts
class
if
ied
as
h
ig
h
ly
p
o
s
itiv
e,
9
.
9
3
%
as
p
o
s
itiv
e,
2
2
.
8
0
%
as
n
eu
tr
al,
1
0
.
1
8
%
as
n
eg
ativ
e,
an
d
4
.
5
2
%
as
h
ig
h
l
y
n
eg
ativ
e
.
Giv
en
its
s
ca
lab
ilit
y
,
in
ter
p
r
etab
ilit
y
,
an
d
p
r
o
b
a
b
ilis
tic
o
u
tp
u
t,
LR
i
s
wel
l
-
s
u
ited
f
o
r
lar
g
e
-
s
ca
le
s
en
tim
en
t a
n
aly
s
is
task
s
,
o
f
f
er
in
g
r
o
b
u
s
t a
n
d
e
f
f
icien
t sen
tim
en
t p
r
ed
ictio
n
ac
r
o
s
s
d
iv
er
s
e
t
ex
tu
al
d
atasets
.
Alg
o
r
ith
m
1
.
Fin
e
-
g
r
ain
e
d
s
en
tim
en
t c
lass
if
icatio
n
Input:
d
ataset of tweets with:
–
class (positive, neutral, or negative).
–
LR_proba (probability score).
Output: refined_sentiment (fine
-
grained sentiment):
–
1.0:
highly positive
–
0.5:
positive
–
0.0:
neutral
–
-
0.5:
neg
ative
–
-
1.0:
highly negative
1
:
For each tweet i in the dataset:
2
:
If the tweet class is positive and score (LR_proba) >=0.75:
3
:
Assign refined_sentiment =1.0
4
:
Else if
the tweet's class is
positive and
s
core
(
LR
_proba
) <0.75:
5
:
Assign refined_sentiment =0.5
6
:
Else if
the tweet's class is
neutral
:
7
:
Assign refined_sentiment =0.0
8
:
Else if
the tweet's class is
negative and
s
core (
L
R
_proba)>0.25:
9
:
Assign refined_sentiment =
-
0.5
10
:
Else
:
11
:
Assign refined_sentiment =
-
1.0
12
:
End loop
.
Fig
u
r
e
6
.
Sen
tim
en
t
d
is
tr
ib
u
tio
n
o
f
t
wee
ts
u
s
in
g
th
e
LR
s
en
tim
en
t m
o
d
el
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
.
3
,
J
u
n
e
2
0
2
6
:
2
7
3
6
-
2
7
4
9
2744
2
.
8
.
Ana
ly
s
is
o
f
g
lo
ba
l sent
im
ent
t
re
nd
s
I
n
th
is
an
aly
s
is
,
s
en
tim
en
t
-
r
ate
d
ev
alu
atio
n
s
with
a
h
ig
h
d
eg
r
ee
o
f
p
r
ec
is
io
n
,
alo
n
g
with
th
e
ir
r
ef
in
ed
s
en
tim
en
t
s
co
r
es,
wer
e
u
s
ed
t
o
ex
am
i
n
e
s
en
tim
en
t
t
r
en
d
s
in
o
n
lin
e
co
u
r
s
es.
R
ep
r
esen
tativ
e
r
ev
iews
f
o
r
ea
ch
class
o
f
s
en
tim
en
t,
as
we
ll
as
th
eir
ass
o
ciate
d
LR
p
r
o
b
ab
ilit
y
s
co
r
es,
ar
e
s
h
o
wn
in
T
ab
le
3
.
T
h
ese
r
ev
iews
ca
p
tu
r
e
a
r
a
n
g
e
o
f
em
o
tio
n
s
,
f
r
o
m
h
ig
h
ly
p
o
s
itiv
e
to
h
ig
h
ly
n
eg
ativ
e
,
wh
ich
co
n
tr
i
b
u
te
to
f
lu
ctu
atio
n
s
in
th
e
g
lo
b
al
s
en
tim
en
t
tr
en
d
.
B
y
ca
lcu
latin
g
d
ate
-
wis
e
av
er
ag
e
s
e
n
tim
en
t
s
co
r
es,
n
o
tab
le
v
ar
iatio
n
s
o
v
er
tim
e
wer
e
id
en
tifie
d
.
T
h
ese
v
ar
iatio
n
s
p
r
o
v
id
e
v
al
u
ab
le
in
s
ig
h
ts
in
to
t
h
e
d
y
n
am
ic
em
o
ti
o
n
al
lan
d
s
ca
p
e
o
f
lear
n
er
s
.
T
h
e
p
o
s
itiv
e,
n
eu
tr
al,
an
d
n
eg
ati
v
e
s
en
tim
en
ts
r
ev
ea
led
in
th
e
r
ev
iews
h
ig
h
lig
h
t
k
e
y
m
o
m
en
ts
o
f
em
o
tio
n
a
l
ch
an
g
e,
o
f
f
er
in
g
a
d
ee
p
er
u
n
d
er
s
tan
d
in
g
o
f
h
o
w
lear
n
er
s
'
ex
p
er
ien
ce
s
ev
o
lv
e
in
r
esp
o
n
s
e
to
th
e
o
n
lin
e
lear
n
in
g
en
v
ir
o
n
m
en
t.
T
ab
le
3
.
O
n
lin
e
c
o
u
r
s
es
t
wee
ts
f
r
o
m
d
if
f
er
en
t sen
tim
en
t c
las
s
es
D
a
t
e
O
r
i
g
i
n
a
l
_
Tw
e
e
t
C
l
a
s
s
P
r
o
b
R
e
f
i
n
e
d
_
se
n
t
i
me
n
t
R
e
f
i
n
e
d
_
r
a
t
i
n
g
2
0
2
0
-
05
-
22
9
0
%
o
f
g
r
a
d
u
a
t
e
s
h
a
v
e
a
c
h
i
e
v
e
d
t
h
e
i
r
g
o
a
l
s
.
J
o
i
n
t
h
e
m!
pos
0
.
9
4
7
6
5
2
8
3
4
5
9
7
1
1
6
9
H
i
g
h
l
y
p
o
si
t
i
v
e
1
.
0
2
0
2
0
-
04
-
26
D
i
st
a
n
c
e
l
e
a
r
n
i
n
g
f
o
r
c
e
d
p
a
r
e
n
t
s
t
o
p
a
r
t
i
c
i
p
a
t
e
i
n
t
h
e
i
r
k
i
d
s’
e
d
u
c
a
t
i
o
n
.
I
t
h
i
n
k
i
t
w
a
s
a
c
t
u
a
l
l
y
a
g
o
o
d
t
h
i
n
g
.
pos
0
.
7
4
7
2
8
5
1
7
6
5
1
9
5
6
4
9
P
o
si
t
i
v
e
0
.
5
2
0
2
0
-
08
-
03
So
,
I
can
’
t
b
e
l
i
e
v
e
o
u
r
o
n
l
i
n
e
c
l
a
ss
w
a
s
r
e
c
o
r
d
e
d
t
h
a
t
mea
n
s
my
p
a
r
e
n
t
s
w
i
l
l
s
e
e
h
o
w
i
a
c
t
i
n
c
l
a
ss
.
n
e
u
1
.
0
0
0
0
0
0
N
e
u
t
r
a
l
0
.
0
2
0
2
0
-
09
-
01
V
e
r
y
l
o
w
-
l
e
v
e
l
c
o
n
t
e
n
t
.
N
o
st
r
a
t
e
g
i
e
s
o
r
a
c
t
i
o
n
a
b
l
e
l
e
sso
n
s.
n
e
g
0
.
4
7
4
3
7
4
4
6
8
7
4
1
8
4
9
1
3
N
e
g
a
t
i
v
e
-
0
.
5
2
0
2
0
-
10
-
08
Te
a
c
h
e
r
s
n
o
w
h
a
v
e
t
o
c
o
m
p
e
t
e
w
i
t
h
s
o
c
i
a
l
me
d
i
a
a
s
a
s
o
u
r
c
e
o
f
i
n
f
o
r
mat
i
o
n
.
T
h
e
d
i
s
t
a
n
c
e
l
e
a
r
n
i
n
g
set
u
p
m
a
k
e
s
i
t
d
i
f
f
i
c
u
l
t
f
o
r
t
h
e
m t
o
g
u
i
d
e
s
t
u
d
e
n
t
s
i
mm
e
r
se
d
i
n
f
a
l
se
i
n
f
o
r
mat
i
o
n
o
n
l
i
n
e
.
n
e
g
0
.
0
5
3
6
6
9
2
0
5
3
8
0
5
0
8
6
8
5
H
i
g
h
l
y
n
e
g
a
t
i
v
e
-
1
.
0
2
.
9
.
Arc
hite
ct
ure
a
nd
hy
perpa
ra
m
et
er
s
o
f
t
he
hy
brid co
nv
o
lutio
na
l
lo
ng
s
ho
rt
-
t
er
m
m
em
o
ry
m
o
del
T
h
e
L
STM
n
etwo
r
k
h
as
p
r
o
v
e
n
to
b
e
h
ig
h
l
y
ef
f
ec
tiv
e
in
s
en
tim
en
t
an
aly
s
is
,
s
u
r
p
ass
in
g
o
th
er
n
eu
r
al
n
etwo
r
k
m
o
d
els
in
ter
m
s
o
f
p
r
ed
ictio
n
ac
c
u
r
ac
y
.
L
STM
,
a
s
p
ec
ialized
ty
p
e
o
f
r
ec
u
r
r
e
n
t
n
eu
r
al
n
etwo
r
k
(
R
NN)
,
is
ca
p
ab
le
o
f
f
o
r
ec
as
tin
g
f
u
tu
r
e
o
u
tco
m
es
b
ased
o
n
f
ea
tu
r
es
ex
t
r
ac
ted
f
r
o
m
t
h
e
d
ataset
[
2
6
]
.
I
ts
ar
ch
itectu
r
e
co
n
s
is
ts
o
f
m
em
o
r
y
ce
lls
(
)
an
d
th
r
ee
g
ates:
in
p
u
t
(
)
,
f
o
r
g
et
(
)
,
an
d
o
u
tp
u
t
(
)
.
T
h
ese
g
ates
co
n
tr
o
l
th
e
f
lo
w
o
f
in
f
o
r
m
atio
n
b
y
allo
win
g
n
ew
d
ata
in
to
t
h
e
m
em
o
r
y
ce
ll,
f
o
r
g
ettin
g
o
u
td
ated
in
f
o
r
m
atio
n
,
an
d
p
r
o
p
a
g
atin
g
th
e
cu
r
r
e
n
t
m
em
o
r
y
s
tate
th
r
o
u
g
h
th
e
n
etwo
r
k
.
T
h
e
L
STM
ef
f
ec
tiv
el
y
p
r
e
v
en
ts
th
e
v
a
n
is
h
in
g
g
r
ad
ien
t
p
r
o
b
lem
,
en
s
u
r
in
g
ac
cu
r
ate
p
r
ed
ictio
n
s
.
T
h
e
g
a
tes
an
d
m
em
o
r
y
ce
lls
ar
e
g
o
v
er
n
e
d
b
y
weig
h
t
m
atr
ices
(
)
,
th
e
lo
g
is
tic
s
ig
m
o
id
ac
tiv
atio
n
f
u
n
ctio
n
(
σ
)
,
a
n
d
th
e
Had
a
m
ar
d
p
r
o
d
u
ct
(
⊙
)
,
wh
ich
to
g
eth
er
r
eg
u
late
th
e
f
lo
w
an
d
r
ete
n
tio
n
o
f
in
f
o
r
m
atio
n
[
2
7
]
.
T
h
e
m
a
th
em
atica
l
f
o
r
m
u
latio
n
g
o
v
er
n
in
g
th
e
b
e
h
av
io
r
o
f
th
e
L
STM
g
ates,
m
em
o
r
y
ce
ll
s
tate,
an
d
h
id
d
e
n
s
tate
is
d
ef
in
ed
in
(
7
)
.
=
(
+
ℎ
−
1
+
⊙
−
1
+
)
=
(
+
ℎ
ℎ
−
1
+
⊙
−
1
+
)
=
⊙
−
1
+
⊙
ℎ
(
+
ℎ
ℎ
−
1
+
)
=
(
+
ℎ
ℎ
−
1
+
+
)
ℎ
=
⊙
ℎ
(
)
(
7
)
T
h
e
p
r
o
p
o
s
ed
h
y
b
r
i
d
C
o
n
v
-
L
STM
m
o
d
el
ef
f
ec
tiv
ely
a
d
d
r
ess
es
th
e
r
e
d
u
n
d
an
cy
is
s
u
es
o
f
ten
en
co
u
n
ter
e
d
b
y
L
STM
n
etwo
r
k
s
.
W
h
ile
L
STM
s
ar
e
p
r
o
f
icien
t
at
ca
p
tu
r
in
g
lo
ca
l
d
ep
en
d
en
cies
b
etwe
e
n
n
eig
h
b
o
r
in
g
w
o
r
d
s
in
lar
g
e
d
atasets
,
th
ey
f
ac
e
ch
allen
g
e
s
in
o
p
tim
izin
g
p
er
f
o
r
m
an
ce
.
T
h
e
C
o
n
v
-
L
STM
m
o
d
el
co
m
b
in
es
co
n
v
o
lu
tio
n
a
l
lay
er
s
with
L
STM
s
to
o
v
er
co
m
e
th
ese
is
s
u
es,
p
r
o
v
id
in
g
im
p
r
o
v
e
d
ac
cu
r
ac
y
i
n
s
en
tim
en
t c
lass
if
icatio
n
.
T
h
is
s
tu
d
y
twee
ts
wer
e
lab
el
ed
in
to
s
en
tim
en
t
class
es
r
a
n
g
in
g
f
r
o
m
m
o
s
t
n
eg
ativ
e
(
0
)
to
m
o
s
t
p
o
s
itiv
e
(
4
)
,
with
th
e
d
ataset
d
iv
id
ed
in
to
8
0
%
f
o
r
tr
ain
in
g
a
n
d
2
0
%
f
o
r
test
in
g
.
E
ac
h
twee
t
was
p
r
ep
r
o
ce
s
s
ed
an
d
tr
a
n
s
f
o
r
m
e
d
in
t
o
1
0
0
-
d
im
en
s
io
n
al
wo
r
d
em
b
ed
d
i
n
g
s
u
s
in
g
th
e
e
m
b
e
d
d
in
g
lay
e
r
,
w
h
ich
was
co
n
f
ig
u
r
ed
with
v
o
ca
b
u
lar
y
s
ize
o
f
8
7
,
5
8
7
u
n
iq
u
e
wo
r
d
s
an
d
m
a
x
im
u
m
s
eq
u
en
ce
len
g
t
h
o
f
4
6
0
.
Sen
tim
en
t
class
if
icatio
n
was
p
er
f
o
r
m
e
d
u
s
in
g
b
o
th
s
e
q
u
en
tial
an
d
B
iLST
M
m
o
d
els
,
im
p
lem
en
ted
with
T
e
n
s
o
r
Flo
w
an
d
Ker
as.
T
h
e
m
o
d
el
ar
ch
itectu
r
e
in
co
r
p
o
r
ate
d
em
b
ed
d
in
g
,
co
n
v
o
lu
tio
n
al,
m
ax
p
o
o
lin
g
,
L
STM
,
a
n
d
d
en
s
e
lay
er
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
Hyb
r
id
d
ee
p
lea
r
n
in
g
fo
r
s
en
timen
t a
n
a
lysi
s
o
f o
n
lin
e
s
tu
d
e
n
t e
xp
erien
ce
s
(
R
a
ja
Ou
a
d
a
d
)
2745
T
h
e
n
etwo
r
k
ar
c
h
itectu
r
e,
illu
s
tr
ated
in
Fig
u
r
e
7
,
s
tar
ts
with
an
em
b
ed
d
i
n
g
lay
e
r
th
at
tr
an
s
f
o
r
m
s
th
e
in
p
u
t
te
x
t
in
to
d
e
n
s
e
v
ec
to
r
s
o
f
1
0
0
d
im
e
n
s
io
n
s
.
T
h
is
is
f
o
llo
wed
b
y
a
o
n
e
-
d
im
en
s
io
n
al
co
n
v
o
l
u
tio
n
al
lay
er
co
n
tain
in
g
1
2
8
f
ilter
s
with
a
k
er
n
el
s
ize
o
f
3
,
u
s
in
g
th
e
R
e
L
U
ac
tiv
atio
n
f
u
n
ctio
n
.
T
o
s
tab
ilize
th
e
tr
ain
in
g
p
r
o
ce
s
s
,
b
atch
n
o
r
m
aliza
tio
n
was
ap
p
lied
,
an
d
a
Ma
x
Po
o
lin
g
lay
er
with
a
p
o
o
l
s
ize
o
f
2
was
u
s
ed
to
d
o
wn
s
am
p
le
th
e
f
ea
tu
r
e
m
ap
s
.
A
d
r
o
p
o
u
t
r
ate
o
f
0
.
3
was
in
co
r
p
o
r
ated
with
i
n
th
e
c
o
n
v
o
lu
tio
n
al
la
y
er
s
to
p
r
ev
en
t
o
v
er
f
itti
n
g
.
T
h
e
o
u
tp
u
ts
g
en
er
ated
b
y
th
e
co
n
v
o
l
u
tio
n
al
lay
er
s
ar
e
f
e
d
i
n
to
a
L
STM
lay
er
co
m
p
r
is
in
g
1
2
8
u
n
its
,
wh
ich
en
a
b
les
th
e
m
o
d
el
t
o
ca
p
tu
r
e
lo
n
g
-
r
a
n
g
e
d
ep
en
d
e
n
cies
with
in
th
e
in
p
u
t
s
eq
u
en
ce
s
.
T
o
p
r
ev
en
t
o
v
er
f
itti
n
g
,
a
d
r
o
p
o
u
t
r
ate
o
f
0
.
4
was
ap
p
lied
t
o
th
e
L
STM
la
y
er
.
I
n
s
tead
o
f
f
latten
in
g
th
e
s
eq
u
en
ce
o
u
tp
u
ts
,
a
Glo
b
alM
ax
Po
o
lin
g
1
D
lay
e
r
was
u
s
ed
to
id
en
tify
an
d
r
etain
th
e
m
o
s
t
s
alien
t
f
ea
tu
r
es.
T
h
e
r
esu
ltin
g
f
ea
tu
r
e
r
ep
r
esen
tatio
n
is
th
e
n
p
ass
ed
th
r
o
u
g
h
two
f
u
lly
co
n
n
ec
ted
d
en
s
e
lay
er
s
with
6
4
a
n
d
3
2
u
n
its
,
r
esp
ec
tiv
ely
,
ea
ch
f
o
llo
wed
b
y
d
r
o
p
o
u
t r
ate
s
o
f
0
.
3
a
n
d
0
.
2
to
f
u
r
t
h
er
en
h
an
ce
r
eg
u
la
r
izatio
n
.
T
h
e
f
in
al
lay
e
r
co
n
s
is
ts
o
f
f
iv
e
u
n
its
with
s
o
f
tm
ax
ac
tiv
atio
n
,
en
a
b
lin
g
m
u
lticlas
s
s
en
tim
en
t
class
if
icatio
n
.
C
o
m
p
ilatio
n
o
f
th
e
m
o
d
el
in
v
o
lv
ed
th
e
Ad
a
m
o
p
tim
izer
with
a
0
.
0
0
1
lear
n
in
g
r
ate,
w
h
ile
th
e
lo
s
s
was
co
m
p
u
ted
u
s
in
g
ca
te
g
o
r
ical
cr
o
s
s
-
en
tr
o
p
y
.
T
o
d
y
n
am
ically
ad
ju
s
t
th
e
lear
n
in
g
r
ate
wh
en
v
alid
atio
n
lo
s
s
p
latea
u
s
,
th
e
R
ed
u
ce
L
R
On
Plateau
ca
llb
ac
k
was
u
s
ed
,
r
e
d
u
cin
g
th
e
r
ate
b
y
a
f
ac
to
r
o
f
0
.
5
with
a
p
atien
ce
o
f
th
r
ee
ep
o
ch
s
.
T
h
e
co
m
p
lete
m
o
d
el,
wh
ich
co
n
tain
s
1
0
,
8
1
5
,
2
8
1
tr
ain
a
b
le
p
ar
am
eter
s
,
d
e
m
o
n
s
tr
ates
a
s
tr
o
n
g
ca
p
ac
ity
to
lear
n
f
r
o
m
f
i
n
e
-
g
r
ain
ed
tex
tu
al
d
ata,
r
esu
ltin
g
in
en
h
an
ce
d
s
en
tim
en
t
p
r
e
d
ictio
n
p
er
f
o
r
m
an
ce
,
p
ar
ticu
lar
ly
f
o
r
twee
ts
r
elate
d
to
o
n
lin
e
lear
n
in
g
.
Fig
u
r
e
7
.
Ar
c
h
itectu
r
e
o
f
th
e
p
r
o
p
o
s
ed
h
y
b
r
id
C
o
n
v
-
L
STM
m
o
d
el
3.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
T
h
is
s
ec
tio
n
r
ep
o
r
ts
th
e
p
er
f
o
r
m
an
ce
o
u
tco
m
es
o
f
t
h
e
p
r
o
p
o
s
ed
h
y
b
r
id
m
o
d
el
f
o
r
cla
s
s
if
y
in
g
th
e
s
en
tim
en
t
o
f
twee
ts
ass
o
ciate
d
with
o
n
lin
e
lear
n
in
g
an
d
c
o
u
r
s
es.
Mo
d
el
ev
alu
atio
n
wa
s
co
n
d
u
cte
d
u
s
in
g
s
tan
d
ar
d
m
etr
ics,
in
clu
d
in
g
ac
cu
r
ac
y
,
p
r
ec
is
io
n
,
F1
-
s
co
r
e,
a
n
d
s
u
p
p
o
r
t.
T
o
ass
ess
i
ts
ef
f
ec
tiv
en
ess
,
th
e
r
esu
lts
o
f
th
e
p
r
o
p
o
s
ed
ap
p
r
o
ac
h
ar
e
co
m
p
ar
e
d
with
th
o
s
e
o
f
o
t
h
er
s
tate
-
of
-
th
e
-
a
r
t
m
o
d
els
p
r
e
v
io
u
s
ly
ap
p
lied
to
s
tu
d
en
t
r
ev
iews
in
o
n
lin
e
lear
n
in
g
c
o
n
tex
ts
,
h
ig
h
lig
h
ti
n
g
th
e
im
p
r
o
v
em
en
ts
ac
h
ie
v
ed
in
s
en
tim
en
t
class
if
icatio
n
ac
cu
r
ac
y
.
3
.
1
.
M
o
del
p
er
f
o
rma
nce
Du
r
in
g
tr
ain
in
g
,
b
o
th
th
e
h
y
b
r
id
C
o
n
v
-
L
STM
an
d
co
n
v
o
lu
tio
n
al
b
id
ir
ec
tio
n
al
lo
n
g
s
h
o
r
t
-
ter
m
m
em
o
r
y
(
C
o
n
v
-
B
iLST
M)
m
o
d
els
wer
e
tr
ain
ed
with
id
en
tical
co
n
d
itio
n
s
:
a
b
atch
s
ize
o
f
3
2
,
a
lear
n
in
g
r
ate
o
f
0
.
0
0
1
,
an
d
a
v
er
b
o
s
ity
lev
el
o
f
2
.
T
o
en
h
a
n
ce
g
en
er
aliza
ti
o
n
an
d
r
ed
u
ce
o
v
er
f
itti
n
g
,
th
e
d
r
o
p
o
u
t
tech
n
iq
u
e
was
ap
p
lied
.
T
a
b
le
4
r
e
p
o
r
ts
th
e
tr
ai
n
in
g
an
d
v
alid
atio
n
a
cc
u
r
ac
y
a
n
d
lo
s
s
v
alu
es
o
f
t
h
e
p
r
o
p
o
s
ed
h
y
b
r
id
m
o
d
els ac
r
o
s
s
s
ix
tr
ain
in
g
ep
o
ch
s
.
T
h
e
C
o
n
v
-
L
STM
m
o
d
el
ac
h
iev
ed
a
p
ea
k
t
r
ain
in
g
ac
c
u
r
a
cy
o
f
9
5
.
7
8
% with
Evaluation Warning : The document was created with Spire.PDF for Python.