I
nte
rna
t
io
na
l J
o
urna
l o
f
E
lect
rica
l a
nd
Co
m
pu
t
er
E
ng
ineering
(
I
J
E
CE
)
Vo
l.
15
,
No
.
2
,
A
p
r
il
20
25
,
p
p
.
1
9
3
3
~
1
9
4
8
I
SS
N:
2088
-
8
7
0
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijece.
v
15
i
2
.
pp
1
9
3
3
-
1
9
4
8
1933
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ec
e.
ia
esco
r
e.
co
m
Exploring
topic
mo
delling
:
a
com
pa
ra
tive a
na
ly
sis
of t
ra
ditiona
l
a
nd t
ra
nsfo
rmer
-
ba
sed a
ppro
a
ches with
empha
sis
on
coherence
a
nd div
ersity
Ay
esh
a
Ria
z
1
,
O
m
a
r
Abdu
lka
der
2
,
M
uh
a
m
m
a
d J
a
wa
d Ik
ra
m
2
,
Sa
da
qa
t
J
a
n
1
1
D
e
p
a
r
t
me
n
t
o
f
C
o
m
p
u
t
e
r
S
o
f
t
w
a
r
e
E
n
g
i
n
e
e
r
i
n
g
,
U
n
i
v
e
r
si
t
y
o
f
E
n
g
i
n
e
e
r
i
n
g
a
n
d
Te
c
h
n
o
l
o
g
y
,
M
a
r
d
a
n
,
P
a
k
i
s
t
a
n
2
F
a
c
u
l
t
y
o
f
C
o
mp
u
t
e
r
S
t
u
d
i
e
s,
A
r
a
b
O
p
e
n
U
n
i
v
e
r
si
t
y
,
J
e
d
d
a
h
,
S
a
u
d
i
A
r
a
b
i
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Ma
y
1
2
,
2
0
2
4
R
ev
is
ed
Oct
3
0
,
2
0
2
4
Acc
ep
ted
No
v
2
0
,
2
0
2
4
To
p
ic
m
o
d
e
li
n
g
(TM
)
is
a
n
u
n
su
p
e
rv
ise
d
tec
h
n
iq
u
e
u
se
d
to
re
c
o
g
n
i
z
e
h
id
d
e
n
o
r
a
b
stra
c
t
to
p
ics
in
larg
e
c
o
rp
o
r
a
,
e
x
trac
ti
n
g
m
e
a
n
in
g
fu
l
p
a
tt
e
rn
s
o
f
wo
rd
s
(se
m
a
n
ti
c
s).
Th
is
p
a
p
e
r
e
x
p
lo
re
s
TM
wit
h
i
n
d
a
ta
m
in
i
n
g
(DM)
,
f
o
c
u
sin
g
o
n
c
h
a
ll
e
n
g
e
s
a
n
d
a
d
v
a
n
c
e
m
e
n
ts
in
e
x
trac
ti
n
g
in
si
g
h
ts
fro
m
d
a
tas
e
ts,
e
sp
e
c
ially
fro
m
so
c
ial
m
e
d
ia
p
latfo
rm
s
(
S
M
P
s).
Trad
it
io
n
a
l
tec
h
n
iq
u
e
s
li
k
e
late
n
t
Dirich
let
a
ll
o
c
a
ti
o
n
(LDA),
a
l
o
n
g
si
d
e
n
e
we
r
m
e
th
o
d
o
lo
g
ies
su
c
h
a
s
b
id
irec
ti
o
n
a
l
e
n
c
o
d
e
r
re
p
re
se
n
tatio
n
s
fr
o
m
tran
sfo
rm
e
rs
(BERT
)
,
g
e
n
e
ra
ti
v
e
p
re
-
train
e
d
tran
sfo
rm
e
rs
(G
P
T),
a
n
d
e
x
tra
l
o
n
g
-
term
m
e
m
o
ry
n
e
two
r
k
s
(XLNe
t)
a
re
e
x
a
m
in
e
d
.
T
h
is
p
a
p
e
r
h
i
g
h
l
ig
h
ts
t
h
e
l
imitatio
n
s
o
f
LDA,
p
ro
m
p
ti
n
g
th
e
a
d
o
p
ti
o
n
o
f
e
m
b
e
d
d
i
n
g
-
b
a
se
d
m
o
d
e
ls
li
k
e
BERT
a
n
d
G
P
T,
ro
o
ted
i
n
t
ra
n
sfo
rm
e
r
a
rc
h
it
e
c
tu
re
,
o
ffe
ri
n
g
e
n
h
a
n
c
e
d
c
o
n
tex
t
-
a
wa
re
n
e
ss
a
n
d
se
m
a
n
ti
c
u
n
d
e
rsta
n
d
i
n
g
.
Th
e
p
a
p
e
r
e
m
p
h
a
siz
e
s
lev
e
ra
g
in
g
p
re
-
train
e
d
tran
sfo
rm
e
r
-
b
a
se
d
lan
g
u
a
g
e
m
o
d
e
ls
to
g
e
n
e
ra
te
d
o
c
u
m
e
n
t
e
m
b
e
d
d
i
n
g
,
re
fin
in
g
TM
a
n
d
imp
ro
v
i
n
g
a
c
c
u
ra
c
y
.
No
tab
ly
,
in
te
g
ra
ti
n
g
B
ERT
with
XLNe
t
su
m
m
a
ries
e
m
e
rg
e
s
a
s
a
p
ro
m
isin
g
a
p
p
ro
a
c
h
.
By
sy
n
th
e
siz
in
g
in
sig
h
ts,
th
e
p
a
p
e
r
a
ims
to
in
f
o
rm
re
se
a
r
c
h
e
rs
o
n
o
p
ti
m
izi
n
g
TM
te
c
h
n
iq
u
e
s,
p
o
ten
ti
a
ll
y
s
h
ift
i
n
g
h
o
w i
n
si
g
h
ts a
re
e
x
trac
ted
fro
m
tex
t
u
a
l
d
a
ta.
K
ey
w
o
r
d
s
:
B
i
d
i
r
e
ct
i
o
n
a
l
e
n
c
o
d
e
r
r
e
p
r
e
s
e
n
t
a
ti
o
n
s
f
r
o
m
t
r
a
n
s
f
o
r
m
e
r
s
E
x
tr
a
lo
n
g
-
ter
m
m
e
m
o
r
y
n
etwo
r
k
s
Gen
er
ativ
e
p
r
e
-
t
r
ain
ed
tr
an
s
f
o
r
m
er
s
L
aten
t D
ir
ich
let
allo
ca
tio
n
So
cial
m
ed
ia
p
latf
o
r
m
s
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
:
Om
ar
Ab
d
u
lk
a
d
er
Facu
lty
o
f
C
o
m
p
u
ter
Stu
d
ies,
Ar
ab
Op
en
Un
i
v
er
s
ity
J
ed
d
ah
,
Sau
d
i A
r
a
b
ia
E
m
ail
:
o
.
ab
d
u
l
k
ad
er
@
ar
ab
o
u
.
ed
u
.
s
a
1.
I
NT
RO
D
UCT
I
O
N
Ar
tific
ial
in
tellig
en
ce
(
AI
)
is
a
co
m
p
r
e
h
en
s
iv
e
f
ield
in
c
o
m
p
u
ter
s
cien
ce
th
at
e
n
co
m
p
ass
es
v
ar
io
u
s
tech
n
o
lo
g
ies
an
d
m
eth
o
d
o
lo
g
ies
to
en
ab
le
m
ac
h
in
es
to
p
er
f
o
r
m
task
s
th
at
tr
ad
itio
n
ally
r
eq
u
ir
ed
h
u
m
an
in
tellig
en
ce
.
Ma
ch
i
n
e
lear
n
in
g
(
ML
)
is
a
s
u
b
s
et
o
f
AI
,
f
o
cu
s
in
g
o
n
t
h
e
d
ev
elo
p
m
en
t
o
f
alg
o
r
ith
m
s
th
at
allo
w
m
ac
h
in
es
to
lear
n
p
atter
n
s
an
d
m
ak
e
p
r
ed
ictio
n
s
o
r
d
ec
is
io
n
s
f
r
o
m
d
ata
with
o
u
t
ex
p
licit
p
r
o
g
r
a
m
m
in
g
.
ML
en
h
an
ce
s
p
e
r
f
o
r
m
an
ce
t
h
r
o
u
g
h
au
to
n
o
m
o
u
s
lear
n
in
g
.
On
th
e
o
th
er
h
an
d
,
n
atu
r
al
lan
g
u
ag
e
p
r
o
ce
s
s
in
g
(
NL
P),
an
o
th
er
c
r
u
cial
co
m
p
o
n
en
t
o
f
AI
,
in
v
o
lv
es
th
e
a
p
p
licatio
n
o
f
tech
n
i
q
u
es
to
e
n
ab
le
m
ac
h
in
es
to
u
n
d
er
s
tan
d
an
d
p
r
o
ce
s
s
h
u
m
an
lan
g
u
a
g
e
in
s
p
o
k
e
n
o
r
wr
itten
f
o
r
m
.
NL
P
f
ac
ilit
ates
s
p
ee
ch
-
b
ased
in
ter
ac
tio
n
s
b
etwe
e
n
h
u
m
an
s
an
d
co
m
p
u
ter
s
,
a
d
d
r
e
s
s
in
g
th
e
g
ap
b
etwe
en
h
u
m
an
co
m
m
u
n
icatio
n
an
d
co
m
p
u
ter
co
m
p
r
e
h
en
s
io
n
.
I
t
f
in
d
s
ap
p
licatio
n
s
in
s
en
tim
en
t
an
aly
s
is
,
tex
t
cla
s
s
if
icatio
n
,
s
em
an
tic
r
elate
d
n
ess
,
o
r
to
p
ic
m
o
d
elin
g
.
Du
e
t
o
th
e
r
ap
id
g
r
o
wth
o
f
o
n
lin
e
te
x
tu
al
in
f
o
r
m
atio
n
s
u
ch
as
web
p
ag
es,
em
ail,
n
ews
ar
ticles,
an
d
p
er
s
o
n
al
b
lo
g
s
,
th
er
e
is
a
n
ee
d
to
s
ea
r
ch
,
an
a
ly
ze
,
an
d
u
n
d
er
s
tan
d
th
e
m
as
s
iv
e
am
o
u
n
t
o
f
d
ata
to
d
eter
m
in
e
th
e
s
u
b
jectiv
e
in
f
o
r
m
atio
n
in
te
x
t.
B
ec
au
s
e
o
f
th
e
tr
em
e
n
d
o
u
s
am
o
u
n
t
o
f
d
ata
o
n
o
n
lin
e
p
latf
o
r
m
s
,
it
is
a
d
if
f
icu
lt
an
d
tim
e
-
co
n
s
u
m
in
g
task
f
o
r
in
d
iv
id
u
a
ls
to
ex
tr
ac
t
in
f
o
r
m
atio
n
ab
o
u
t
th
e
to
p
ic
o
f
th
eir
in
ter
est.
Dif
f
er
en
t
m
ac
h
in
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
2
,
Ap
r
il
20
25
:
1
9
3
3
-
1
9
4
8
1934
lear
n
in
g
alg
o
r
ith
m
s
s
u
ch
as
tex
t
s
u
m
m
ar
izatio
n
a
n
d
t
o
p
ic
m
o
d
elin
g
a
p
p
r
o
ac
h
es
ar
e
u
s
ed
f
o
r
an
aly
s
is
o
f
s
u
c
h
d
ata.
T
o
an
aly
ze
d
o
c
u
m
en
ts
t
o
lear
n
m
ea
n
in
g
f
u
l
p
atter
n
s
o
f
wo
r
d
s
,
to
p
ic
m
o
d
elin
g
is
u
s
ed
.
E
x
tr
ac
tin
g
to
p
ics
f
r
o
m
tex
t
is
an
im
p
o
r
ta
n
t
ap
p
l
icatio
n
.
T
o
p
ic
m
o
d
elin
g
is
a
m
eth
o
d
u
s
ed
f
o
r
f
in
d
i
n
g
th
e
u
n
iq
u
e
to
p
ics
th
at
ar
e
p
r
esen
t
in
a
d
ataset.
I
t
i
s
a
tex
t
-
m
in
in
g
ap
p
licatio
n
[
1
]
.
As
d
ep
icted
in
Fig
u
r
e
1
,
th
e
p
r
o
ce
s
s
o
f
to
p
ic
m
o
d
elin
g
in
itiates
with
th
e
ac
q
u
is
itio
n
o
f
r
aw
in
p
u
t
te
x
t
d
ata.
T
h
is
d
ata
th
en
p
r
o
ce
ed
s
th
r
o
u
g
h
a
cr
u
cial
d
ata
p
r
ep
r
o
ce
s
s
in
g
p
h
ase
aim
ed
at
clea
n
in
g
an
d
o
r
g
an
izin
g
it
f
o
r
s
u
b
s
eq
u
en
t
an
al
y
s
is
.
On
ce
th
e
p
r
ep
r
o
ce
s
s
in
g
s
tag
e
is
co
m
p
lete,
v
ar
i
o
u
s
to
p
i
c
m
o
d
elin
g
tech
n
iq
u
es
ar
e
em
p
lo
y
ed
t
o
ex
tr
ac
t
co
h
er
e
n
t
an
d
m
ea
n
in
g
f
u
l
to
p
ics
f
r
o
m
th
e
r
ef
in
ed
tex
t
c
o
r
p
u
s
.
Su
b
s
eq
u
en
tly
,
th
ese
d
e
r
iv
ed
to
p
ics
ar
e
v
is
u
alize
d
,
f
ac
ilit
atin
g
e
x
p
lo
r
atio
n
a
n
d
in
ter
p
r
etatio
n
an
d
o
f
f
er
in
g
v
alu
ab
le
in
s
ig
h
ts
in
to
th
e
in
h
er
en
t
s
tr
u
ctu
r
es
an
d
th
em
es
with
in
th
e
tex
t.
T
h
e
p
r
o
b
a
b
ilis
tic
to
p
ic
m
o
d
els
ar
e
u
s
ed
t
o
d
is
co
v
er
t
h
e
s
ec
r
et
t
h
em
atic
s
tr
u
ctu
r
e
in
a
co
r
p
u
s
o
f
d
o
c
u
m
en
ts
a
n
d
h
elp
in
r
ec
o
g
n
izin
g
th
e
m
as
s
iv
e
am
o
u
n
t
o
f
u
n
lab
eled
d
a
ta
[
2
]
.
T
h
is
tech
n
iq
u
e
is
u
s
ed
to
g
et
wo
r
d
co
-
o
cc
u
r
r
e
n
ce
p
atter
n
s
ac
r
o
s
s
a
co
llectio
n
o
f
d
o
c
u
m
en
ts
;
th
ese
p
atter
n
s
ar
e
th
en
co
n
s
id
er
ed
as
h
id
d
en
“
to
p
ics
”
p
r
esen
t in
th
e
c
o
r
p
u
s
.
Fig
u
r
e
1
.
T
o
p
ic
m
o
d
elin
g
T
o
p
ic
m
o
d
els,
lik
e
laten
t
s
e
m
an
tic
an
aly
s
is
(
L
SA)
wh
ic
h
u
tili
ze
s
d
im
en
s
io
n
ality
r
e
d
u
ctio
n
f
o
r
u
n
co
v
e
r
in
g
laten
t
s
tr
u
ctu
r
es
in
tex
t
co
r
p
o
r
a
b
ased
o
n
ter
m
-
d
o
cu
m
e
n
t
r
elatio
n
s
h
ip
s
.
On
th
e
o
th
er
h
a
n
d
,
laten
t
Dir
ich
let
allo
ca
tio
n
(
L
DA)
is
a
p
r
o
b
a
b
ilis
tic
m
o
d
el
r
ep
r
ese
n
tin
g
d
o
c
u
m
en
ts
as
m
ix
tu
r
es
o
f
to
p
ics.
W
h
ile
it
in
co
r
p
o
r
ates
p
r
o
b
ab
ilis
tic
m
o
d
elin
g
.
T
r
ad
itio
n
al
to
p
ic
m
o
d
els,
s
u
ch
as
laten
t
Dir
ich
let
allo
ca
tio
n
[
3
]
an
d
p
r
o
b
a
b
ilis
tic
laten
t
s
em
an
tic
an
aly
s
is
[
3
]
,
h
a
v
e
s
h
o
wn
t
o
b
e
an
e
f
f
ec
tiv
e
u
n
s
u
p
er
v
is
ed
m
eth
o
d
f
o
r
th
e
s
tatis
t
ical
an
aly
s
is
o
f
d
o
cu
m
e
n
t
co
llectio
n
s
.
T
h
ese
a
p
p
r
o
ac
h
es
[
5
]
,
[
6
]
m
o
d
el
ea
ch
d
o
cu
m
en
t
as
a
m
ix
tu
r
e
o
f
laten
t
to
p
ics,
wh
ich
ar
e
m
u
lti
n
o
m
ial
d
is
tr
ib
u
tio
n
s
o
v
er
wo
r
d
s
,
an
d
ad
h
e
r
e
to
th
e
b
ag
-
of
-
wo
r
d
s
ass
u
m
p
tio
n
.
No
n
-
n
eg
ativ
e
m
atr
ix
f
ac
to
r
izatio
n
(
NFM)
a
n
alter
n
ativ
e
t
o
p
r
o
b
ab
ilis
tic
m
o
d
els,
d
ec
o
m
p
o
s
in
g
d
o
cu
m
en
ts
in
to
to
p
ics with
n
o
n
-
n
e
g
ativ
e
co
ef
f
icien
ts
.
A
n
o
v
el
ap
p
r
o
ac
h
to
u
n
s
u
p
er
v
is
ed
wo
r
d
e
m
b
ed
d
in
g
h
as
r
ec
en
tly
b
ee
n
in
tr
o
d
u
ce
d
,
o
f
f
er
i
n
g
a
u
n
iq
u
e
r
ep
r
esen
tatio
n
f
o
r
in
d
iv
id
u
al
wo
r
d
ty
p
es
as
v
ec
to
r
s
.
T
h
is
m
eth
o
d
f
ac
ilit
ates
th
e
clu
s
ter
in
g
o
f
wo
r
d
s
b
ased
o
n
th
eir
d
is
tan
ce
s
with
in
a
h
ig
h
-
d
im
en
s
io
n
al
s
p
ac
e.
B
y
p
r
o
v
i
d
in
g
a
v
ec
to
r
r
e
p
r
esen
tatio
n
f
o
r
e
ac
h
wo
r
d
t
y
p
e,
th
is
ap
p
r
o
ac
h
en
h
a
n
ce
s
th
e
u
n
d
er
s
tan
d
in
g
o
f
s
em
an
tic
r
elatio
n
s
h
ip
s
b
etwe
en
wo
r
d
s
an
d
e
n
a
b
les
m
o
r
e
ef
f
ec
tiv
e
an
aly
s
is
an
d
in
ter
p
r
etatio
n
o
f
tex
tu
al
d
ata.
A
p
o
p
u
lar
wo
r
d
r
ep
r
esen
tatio
n
tech
n
iq
u
e
t
h
at
ca
n
b
e
u
s
ed
to
id
en
tify
th
em
es
with
s
em
an
ti
c
m
ea
n
in
g
is
wo
r
d
em
b
e
d
d
in
g
.
I
t
p
r
o
d
u
ce
s
wo
r
d
v
ec
to
r
r
ep
r
esen
tatio
n
s
,
an
d
r
esear
ch
h
as d
em
o
n
s
tr
ated
th
a
t it
is
an
ef
f
ec
tiv
e
in
d
icato
r
o
f
s
em
an
tic
r
elate
d
n
ess
[
7
]
.
W
o
r
d
em
b
ed
d
i
n
g
[
8
]
,
a
tech
n
o
lo
g
y
f
o
r
n
at
u
r
al
lan
g
u
a
g
e
p
r
o
ce
s
s
in
g
th
at
is
n
o
w
u
n
d
er
g
o
in
g
r
ap
id
d
ev
elo
p
m
e
n
t,
allo
w
u
s
to
m
o
d
el
th
em
es
an
d
to
p
ic
c
o
n
n
ec
tio
n
s
in
co
n
tin
u
o
u
s
s
em
an
tic
s
p
ac
e.
W
o
r
d
em
b
ed
d
in
g
ar
e
r
ea
l
-
v
alu
ed
co
n
tin
u
o
u
s
v
ec
to
r
s
f
o
r
wo
r
d
s
th
a
t
ar
e
s
u
cc
ess
f
u
l
at
ca
p
tu
r
in
g
s
em
an
tic
r
eg
u
lar
ities
in
lan
g
u
a
g
e.
T
h
e
y
ar
e
also
r
ef
er
r
ed
to
as
wo
r
d
v
ec
to
r
s
a
n
d
d
is
tr
ib
u
ted
r
ep
r
esen
tatio
n
s
o
f
wo
r
d
s
.
Similar
s
em
an
tic
an
d
s
y
n
tactic
ch
a
r
ac
ter
is
tics
o
f
wo
r
d
s
ten
d
to
p
r
o
ject
in
to
s
u
r
r
o
u
n
d
i
n
g
r
e
g
io
n
s
i
n
th
e
v
ec
t
o
r
s
p
ac
e.
E
m
b
ed
d
in
g
-
b
ased
m
o
d
els
in
t
o
p
ic
m
o
d
elin
g
,
s
u
ch
as
Do
c
2
Vec
,
r
ep
r
esen
t
a
p
ar
a
d
ig
m
s
h
if
t
f
r
o
m
tr
ad
itio
n
al
p
r
o
b
a
b
ilis
tic
ap
p
r
o
ac
h
es.
Do
c2
Vec
in
tr
o
d
u
ce
d
in
2
0
1
4
,
ex
t
en
d
s
W
o
r
d
2
Vec
,
en
ab
lin
g
th
e
r
ep
r
esen
tatio
n
o
f
en
tire
d
o
cu
m
en
ts
in
co
n
tin
u
o
u
s
v
ec
to
r
s
p
ac
es.
Ov
er
tim
e,
th
ese
m
o
d
els
h
av
e
ev
o
lv
ed
with
r
ef
in
em
e
n
ts
in
t
r
ain
in
g
m
eth
o
d
o
l
o
g
ies,
in
co
r
p
o
r
atio
n
o
f
ad
d
itio
n
al
c
o
n
tex
tu
al
in
f
o
r
m
atio
n
,
an
d
o
p
tim
izatio
n
f
o
r
p
er
f
o
r
m
an
ce
an
d
s
ca
lab
ilit
y
.
R
esear
ch
er
s
h
av
e
ex
p
lo
r
ed
ad
v
a
n
ce
m
en
ts
in
e
m
b
ed
d
in
g
tech
n
i
q
u
es,
en
h
an
c
in
g
th
e
m
o
d
elin
g
o
f
s
em
an
tic
r
elatio
n
s
h
ip
s
an
d
ca
p
tu
r
in
g
m
o
r
e
n
u
an
ce
d
d
o
c
u
m
en
t
s
tr
u
ctu
r
es.
T
h
e
ev
o
lu
tio
n
o
f
em
b
ed
d
in
g
-
b
ased
m
o
d
els
co
n
tin
u
es
to
co
n
tr
ib
u
te
t
o
th
e
f
lex
ib
ilit
y
an
d
e
f
f
icien
cy
o
f
to
p
ic
m
o
d
elin
g
,
o
f
f
e
r
in
g
n
o
v
el
in
s
ig
h
ts
in
to
d
o
cu
m
e
n
t
r
ep
r
esen
tatio
n
s
in
v
ec
to
r
s
p
ac
es.
T
h
e
tr
an
s
itio
n
f
r
o
m
em
b
ed
d
in
g
-
b
ased
m
o
d
els
led
to
th
e
r
is
e
o
f
tr
an
s
f
o
r
m
er
-
b
ased
m
o
d
els,
with
b
i
d
ir
ec
tio
n
al
en
co
d
er
r
ep
r
esen
tatio
n
s
f
r
o
m
tr
an
s
f
o
r
m
e
r
s
(
B
E
R
T
)
in
2
0
1
9
as a
lan
d
m
a
r
k
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
E
xp
lo
r
in
g
to
p
ic
mo
d
ellin
g
:
a
co
mp
a
r
a
tive
a
n
a
lysi
s
o
f tra
d
itio
n
a
l a
n
d
tr
a
n
s
fo
r
mer
-
b
a
s
ed
…
(
A
ye
s
h
a
R
ia
z
)
1935
B
E
R
T
r
ev
o
lu
tio
n
ized
n
atu
r
al
lan
g
u
ag
e
u
n
d
er
s
tan
d
in
g
b
y
ca
p
tu
r
in
g
c
o
n
tex
t
an
d
r
elatio
n
s
h
ip
s
b
i
-
d
ir
ec
tio
n
ally
.
On
g
o
in
g
wo
r
k
i
n
v
o
lv
es f
in
e
-
tu
n
in
g
B
E
R
T
f
o
r
s
p
ec
if
ic
task
s
,
s
u
ch
as to
p
ic
m
o
d
elin
g
,
le
v
er
ag
in
g
its
co
n
tex
tu
al
u
n
d
e
r
s
tan
d
in
g
ca
p
ab
ilit
ies.
T
h
is
p
r
o
g
r
ess
io
n
s
ig
n
if
ies
a
s
h
if
t
to
war
d
s
m
o
r
e
p
o
wer
f
u
l
an
d
co
n
tex
t
-
awa
r
e
m
o
d
els,
as
tr
an
s
f
o
r
m
er
s
,
with
t
h
eir
atten
tio
n
m
ec
h
a
n
is
m
s
,
ex
ce
l
at
ca
p
tu
r
in
g
in
tr
icate
r
elatio
n
s
h
ip
s
in
d
ata.
T
h
e
in
t
eg
r
atio
n
o
f
tr
an
s
f
o
r
m
er
-
b
ased
m
o
d
els,
in
clu
d
i
n
g
B
E
R
T
,
in
to
to
p
ic
m
o
d
elin
g
f
r
am
ewo
r
k
s
r
ep
r
esen
ts
a
co
n
tem
p
o
r
ar
y
ap
p
r
o
ac
h
t
o
en
h
a
n
cin
g
t
h
e
e
f
f
icien
cy
an
d
p
e
r
f
o
r
m
an
ce
o
f
th
ese
m
o
d
els
in
ca
p
tu
r
in
g
co
m
p
lex
s
em
an
tic
s
tr
u
ctu
r
es.
I
n
th
is
r
e
s
ea
r
ch
,
we
ex
p
lo
r
e
s
ev
er
al
k
e
y
m
eth
o
d
s
th
at
a
r
e
u
s
ed
f
o
r
to
p
ic
m
o
d
elin
g
s
u
ch
as
L
DA,
B
E
R
T
,
g
en
er
ativ
e
p
r
e
-
tr
ain
ed
t
r
an
s
f
o
r
m
er
s
(
GPT)
,
an
d
e
x
tr
a
l
o
n
g
-
te
r
m
m
em
o
r
y
n
etwo
r
k
s
(
XL
Net)
.
A
k
ey
d
is
co
v
er
y
i
n
th
is
r
ev
ie
w
is
th
at
wh
en
we
a
p
p
ly
B
E
R
T
to
th
e
s
u
m
m
ar
ies
cr
ea
ted
b
y
XL
Net,
we
co
n
s
is
ten
tly
g
et
v
er
y
clea
r
an
d
m
ea
n
in
g
f
u
l
to
p
ics.
T
h
is
d
is
co
v
er
y
co
u
ld
ch
an
g
e
h
o
w
we
ap
p
r
o
ac
h
to
p
ic
m
o
d
elin
g
a
n
d
m
ak
e
it m
o
r
e
u
s
ef
u
l in
d
i
f
f
er
en
t a
r
ea
s
.
T
h
is
p
ap
er
ca
r
ef
u
lly
ex
am
i
n
e
s
an
d
o
f
f
er
s
in
s
ig
h
ts
in
t
o
h
o
w
to
m
ak
e
th
e
m
o
s
t
o
f
th
ese
m
eth
o
d
s
to
u
n
co
v
e
r
clea
r
an
d
m
ea
n
in
g
f
u
l
to
p
ics
in
v
a
r
io
u
s
a
p
p
licatio
n
s
.
T
h
e
m
ain
o
b
jectiv
es
o
f
th
is
s
tu
d
y
ar
e
g
iv
e
n
as
f
o
llo
ws.
Firstl
y
,
it
p
r
o
v
id
es
an
o
v
er
v
iew
o
f
th
e
f
ield
o
f
to
p
ic
m
o
d
elin
g
(
T
M
)
,
e
x
p
lain
in
g
w
h
at
it
is
an
d
wh
y
it
is
im
p
o
r
tan
t.
T
h
en
,
it
d
iv
es
d
ee
p
in
to
d
if
f
er
e
n
t
asp
ec
ts
o
f
to
p
ic
m
o
d
elin
g
lik
e
ch
allen
g
es,
way
s
o
f
d
o
in
g
it
,
wh
er
e
it
is
u
s
ed
,
an
d
h
o
w
we
m
ea
s
u
r
e
its
s
u
cc
es
s
.
T
h
e
s
tu
d
y
also
g
ath
er
s
an
d
p
r
esen
ts
im
p
o
r
tan
t
in
f
o
r
m
atio
n
,
o
r
g
an
izin
g
an
d
s
u
m
m
a
r
izin
g
wh
at
we
k
n
o
w
ab
o
u
t
to
p
ic
m
o
d
elin
g
.
I
t
also
d
is
cu
s
s
ed
th
e
to
o
ls
u
s
ed
in
to
p
ic
m
o
d
elin
g
.
L
astl
y
,
it
talk
s
ab
o
u
t
th
e
way
s
we
m
ea
s
u
r
e
h
o
w
well
to
p
ic
m
o
d
elin
g
s
y
s
tem
s
wo
r
k
.
T
h
e
r
e
m
ain
d
e
r
o
f
th
is
p
ap
er
is
s
tr
u
ctu
r
e
d
s
o
th
at
s
ec
tio
n
2
s
h
o
ws
th
e
b
ac
k
g
r
o
u
n
d
o
f
to
p
ic
m
o
d
elin
g
,
s
ec
tio
n
3
p
r
esen
ts
t
h
e
liter
atu
r
e
r
ev
iew,
s
ec
tio
n
4
p
r
esen
ts
th
e
m
eth
o
d
,
s
ec
tio
n
5
p
r
esen
ts
th
e
r
esu
lts
(
co
m
p
ar
is
o
n
o
f
m
o
d
els),
an
d
s
ec
tio
n
6
co
n
clu
d
e
th
e
wh
o
le
wo
r
k
.
2.
B
ACK
G
RO
UND
2
.
1
.
T
ra
ditio
na
l
m
o
delin
g
m
et
ho
ds
2
.
1
.
1
.
L
a
t
ent
Dirichlet
a
llo
ca
t
io
n
I
n
th
e
r
ea
lm
o
f
tex
t
m
in
in
g
a
n
d
d
ata
d
is
co
v
er
y
,
to
p
ic
m
o
d
elin
g
s
tan
d
s
as
a
p
o
wer
f
u
l
tec
h
n
iq
u
e
f
o
r
u
n
v
eilin
g
laten
t
r
elatio
n
s
h
ip
s
with
in
d
ata
an
d
tex
t
d
o
cu
m
e
n
ts
.
Ov
er
th
e
y
ea
r
s
,
r
esear
ch
er
s
h
av
e
d
elv
ed
in
to
th
is
f
ield
,
ap
p
ly
in
g
to
p
ic
m
o
d
elin
g
to
d
i
v
er
s
e
d
o
m
ain
s
,
i
n
clu
d
in
g
s
o
f
twar
e
en
g
in
ee
r
in
g
,
p
o
liti
ca
l
s
cien
ce
,
m
ed
icin
e,
an
d
lin
g
u
is
tics
.
Am
o
n
g
th
e
v
ar
i
o
u
s
m
eth
o
d
s
av
ail
ab
le
,
L
DA
em
er
g
es
as
a
co
r
n
er
s
to
n
e.
T
h
is
p
ap
er
co
n
tr
ib
u
tes
to
th
e
e
x
is
tin
g
b
o
d
y
o
f
k
n
o
wled
g
e
b
y
p
r
o
v
id
in
g
a
co
m
p
r
eh
en
s
iv
e
e
x
p
lo
r
atio
n
o
f
s
ch
o
lar
l
y
ar
ticles
with
in
th
e
y
ea
r
s
2
0
0
3
to
2
0
1
6
,
ce
n
ter
ed
o
n
L
DA
-
b
ased
to
p
ic
m
o
d
elin
g
.
I
t
n
o
t
o
n
ly
s
h
ed
s
lig
h
t
o
n
th
e
ev
o
lu
tio
n
an
d
cu
r
r
e
n
t
tr
en
d
s
in
th
is
ar
ea
b
u
t
also
d
elin
ea
tes
th
e
in
tellectu
al
lan
d
s
ca
p
e.
Fu
r
th
er
m
o
r
e,
th
is
p
ap
e
r
o
u
tlin
es th
e
ch
allen
g
es e
n
co
u
n
ter
ed
an
d
in
tr
o
d
u
ce
s
p
r
o
m
in
en
t to
o
ls
an
d
d
atasets
ess
en
tia
l t
o
L
DA
-
b
ased
to
p
ic
m
o
d
elin
g
,
m
ak
in
g
it a
v
alu
a
b
le
r
eso
u
r
ce
f
o
r
r
esear
c
h
er
s
an
d
en
th
u
s
iast
s
in
th
is
d
o
m
ain
[
9
]
.
2
.
1
.
2
.
No
n
-
neg
a
t
iv
e
ma
t
rix
f
a
ct
o
riza
t
i
o
n
No
n
n
e
g
ativ
e
m
atr
i
x
f
ac
to
r
izatio
n
(
NM
F)
[
1
0
]
h
as
e
m
er
g
ed
as
a
p
r
o
m
in
e
n
t
tec
h
n
iq
u
e
f
o
r
d
im
en
s
io
n
ality
r
ed
u
ctio
n
,
d
e
m
o
n
s
tr
atin
g
its
p
r
o
m
i
n
en
ce
s
in
ce
its
in
ce
p
tio
n
.
W
ith
t
h
e
i
n
co
r
p
o
r
atio
n
o
f
th
e
n
o
n
n
e
g
ativ
ity
c
o
n
s
tr
ain
t,
NM
F
ac
h
iev
es
a
p
ar
ts
-
b
ased
r
ep
r
esen
tatio
n
,
wh
ich
en
h
a
n
ce
s
th
e
in
ter
p
r
eta
b
ilit
y
o
f
th
e
ex
tr
ac
ted
f
ea
tu
r
es.
T
h
is
s
u
r
v
ey
p
ap
er
aim
s
to
p
r
o
v
i
d
e
a
co
m
p
r
e
h
en
s
iv
e
o
v
er
v
iew
o
f
th
eo
r
etica
l
r
esear
ch
o
n
NM
F
o
v
er
th
e
p
ast
f
iv
e
y
ea
r
s
.
I
t
s
u
m
m
ar
izes
th
e
p
r
in
ci
p
les,
b
asic
m
o
d
els,
p
r
o
p
er
ties
,
an
d
alg
o
r
ith
m
s
o
f
NM
F,
as
well
as
it
s
m
o
d
if
icatio
n
s
,
ex
ten
s
io
n
s
,
an
d
g
en
er
aliza
tio
n
s
.
T
h
e
ex
is
tin
g
N
MF
alg
o
r
ith
m
s
ar
e
class
if
ied
in
to
f
o
u
r
m
ain
g
r
o
u
p
s
:
b
asic
NM
F
(
B
NM
F),
co
n
s
tr
ain
ed
NM
F
(
C
NM
F),
s
tr
u
ctu
r
ed
NM
F
(
SNMF)
,
an
d
g
e
n
er
alize
d
NM
F
(
GN
MF)
,
with
d
etailed
an
aly
s
e
s
o
f
th
eir
d
esig
n
p
r
in
cip
le
s
,
ch
ar
ac
ter
is
tics
,
r
elatio
n
s
h
ip
s
,
an
d
ev
o
lu
tio
n
.
Ad
d
itio
n
ally
,
th
e
s
u
r
v
e
y
ex
p
lo
r
es
r
elate
d
wo
r
k
o
u
ts
id
e
o
f
NM
F
th
at
o
f
f
e
r
s
in
s
ig
h
ts
o
r
co
n
n
ec
tio
n
s
to
N
MF.
Op
en
r
esear
ch
ch
allen
g
e
s
ar
e
d
is
cu
s
s
ed
,
an
d
s
ev
er
al
ap
p
licatio
n
ar
ea
s
o
f
NM
F
ar
e
b
r
ief
l
y
o
u
tlin
ed
.
T
h
is
s
u
r
v
ey
aim
s
to
p
r
o
v
id
e
an
i
n
teg
r
ated
an
d
u
p
-
to
-
d
ate
f
r
am
ewo
r
k
f
o
r
u
n
d
er
s
tan
d
i
n
g
NM
F,
with
th
e
g
o
al
o
f
g
u
id
in
g
f
u
tu
r
e
r
esear
c
h
en
d
ea
v
o
r
s
in
th
is
f
ield
.
2
.
1
.
3
.
L
a
t
ent
s
em
a
ntic
a
na
l
y
s
is
T
h
e
p
a
p
er
r
ef
er
e
n
ce
d
as
[
1
1
]
in
v
esti
g
ates
L
SA
as
a
m
eth
o
d
f
o
r
an
aly
zi
n
g
tex
tu
al
d
ata,
wi
th
a
f
o
cu
s
o
n
its
ef
f
ec
tiv
en
ess
in
u
n
d
er
s
tan
d
in
g
r
elatio
n
s
h
i
p
s
am
o
n
g
ter
m
s
with
in
d
o
cu
m
en
ts
an
d
ac
r
o
s
s
th
e
co
r
p
u
s
.
L
SA
’
s
ap
p
licatio
n
s
p
an
s
v
ar
io
u
s
d
o
m
ain
s
,
in
clu
d
in
g
in
tellig
en
t
in
f
o
r
m
atio
n
r
etr
iev
al,
s
ea
r
ch
en
g
in
es,
an
d
o
n
lin
e
n
ews
p
latf
o
r
m
s
,
wh
er
e
it
p
lay
s
a
cr
u
cial
r
o
le
in
clas
s
if
icatio
n
,
clu
s
ter
in
g
,
s
u
m
m
ar
izatio
n
,
an
d
s
ea
r
ch
task
s
.
T
h
r
o
u
g
h
th
e
u
tili
za
tio
n
o
f
s
in
g
u
lar
v
alu
e
d
ec
o
m
p
o
s
itio
n
(
SVD)
,
L
SA
en
d
ea
v
o
r
s
t
o
r
ev
ea
l
th
e
o
v
er
al
l
s
tr
u
ctu
r
e
o
f
d
o
cu
m
en
ts
an
d
u
n
co
v
er
laten
t
co
n
n
ec
ti
o
n
s
with
in
th
e
d
ataset.
R
e
s
ea
r
ch
er
s
em
p
lo
y
L
SA
to
g
ain
d
ee
p
er
i
n
s
ig
h
ts
in
to
ter
m
-
d
o
c
u
m
en
t
r
elatio
n
s
h
ip
s
,
p
ar
ticu
la
r
ly
in
t
h
e
co
n
tex
t
o
f
an
al
y
zin
g
r
esear
ch
p
ap
er
s
i
n
n
atu
r
al
lan
g
u
ag
e
p
r
o
ce
s
s
in
g
.
T
h
e
s
tu
d
y
d
em
o
n
s
tr
ates
L
SA
’
s
ca
p
ac
ity
to
m
er
g
e
ter
m
s
wit
h
s
im
ilar
m
ea
n
in
g
s
an
d
id
en
tify
s
u
b
tle
n
u
an
ce
s
,
th
er
eb
y
s
tr
ea
m
lin
in
g
d
o
cu
m
e
n
t
r
ep
r
esen
tatio
n
in
a
lo
wer
-
d
im
en
s
io
n
al
s
p
ac
e.
T
h
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
2
,
Ap
r
il
20
25
:
1
9
3
3
-
1
9
4
8
1936
p
ap
er
r
ef
e
r
en
ce
d
as
[
1
2
]
d
el
v
es
in
to
th
e
ex
p
lo
r
atio
n
o
f
w
o
r
d
em
b
e
d
d
in
g
in
th
e
co
n
te
x
t
o
f
s
m
all
co
r
p
o
r
a,
p
ar
ticu
lar
ly
p
er
tin
en
t
in
th
e
s
t
u
d
y
o
f
in
d
iv
id
u
al
te
x
t
p
r
o
d
u
cti
o
n
.
A
c
o
m
p
ar
ativ
e
an
aly
s
is
is
co
n
d
u
cte
d
b
etwe
en
s
k
ip
-
g
r
am
an
d
L
SA
with
in
th
is
f
r
am
ewo
r
k
,
f
o
c
u
s
in
g
o
n
t
h
eir
ef
f
ec
tiv
en
ess
in
ex
tr
ac
tin
g
s
em
an
tic
p
atter
n
s
f
r
o
m
s
in
g
le
-
s
er
ies
d
r
ea
m
r
e
p
o
r
ts
.
T
h
e
f
in
d
in
g
s
r
e
v
ea
l
th
at
L
SA
o
u
tp
er
f
o
r
m
s
Sk
ip
-
g
r
am
w
h
en
ap
p
lied
to
s
m
all
tr
ain
in
g
co
r
p
o
r
a
in
two
d
is
tin
ct
s
em
an
tic
tes
ts
.
A
s
a
co
m
p
ellin
g
ca
s
e
s
tu
d
y
,
th
e
p
ap
er
d
em
o
n
s
tr
ates
L
SA
’
s
ca
p
ac
ity
to
ca
p
tu
r
e
m
ea
n
in
g
f
u
l
wo
r
d
ass
o
ciatio
n
s
with
in
d
r
ea
m
r
ep
o
r
t
s
er
ies,
ev
en
wh
en
d
ea
lin
g
with
a
lim
ited
n
u
m
b
er
o
f
d
r
ea
m
s
o
r
in
f
r
eq
u
e
n
t
wo
r
d
s
.
T
h
is
h
ig
h
li
g
h
ts
th
e
p
o
ten
tial
o
f
L
SA
as
a
v
alu
ab
le
to
o
l
f
o
r
u
n
r
av
elin
g
wo
r
d
ass
o
ciatio
n
s
in
th
e
r
ea
lm
o
f
d
r
ea
m
r
e
p
o
r
t
s
,
th
er
eb
y
co
n
tr
i
b
u
tin
g
f
r
esh
p
er
s
p
ec
tiv
es
to
th
e
tim
eless
f
ield
o
f
p
s
y
ch
o
lo
g
y
r
esear
ch
.
2
.
1
.
4
.
P
ro
ba
bil
is
t
ic
la
t
ent
s
em
a
ntic
a
na
ly
s
is
Pro
b
ab
ilis
tic
laten
t
s
em
an
tic
an
aly
s
is
(
PLSA
)
tr
ea
ts
to
p
ics
a
s
h
id
d
en
v
ar
ia
b
les
an
d
aim
s
to
u
n
co
v
e
r
a
p
r
o
b
a
b
ilis
tic
m
o
d
el
t
h
at
ca
n
ef
f
ec
tiv
ely
e
x
p
lain
th
e
o
b
s
er
v
ed
d
ata
p
r
esen
t
in
th
e
d
o
c
u
m
en
t
-
ter
m
m
atr
ix
.
E
s
s
en
tially
,
PLSA
s
ee
k
s
a
m
o
d
el
(
P
(
D,
W
)
)
wh
er
e,
f
o
r
a
n
y
g
i
v
en
d
o
cu
m
e
n
t
an
d
wo
r
d
in
th
e
c
o
r
p
u
s
,
th
e
p
r
o
b
a
b
ilit
y
(
P
(
D,
W
)
)
co
r
r
esp
o
n
d
s
with
th
e
r
elev
an
t
en
tr
y
in
th
e
d
o
cu
m
en
t
-
ter
m
m
atr
ix
.
T
h
is
m
eth
o
d
m
er
g
es
th
e
p
r
in
cip
les
o
f
L
SA
with
p
r
o
b
ab
ilis
tic
p
r
in
cip
les,
o
f
f
e
r
in
g
a
co
m
p
r
e
h
en
s
iv
e
f
r
a
m
ewo
r
k
f
o
r
to
p
ic
m
o
d
elin
g
in
NL
P
.
PLSA
h
o
ld
s
s
ig
n
if
ic
an
t
im
p
licatio
n
s
f
o
r
u
n
co
v
er
i
n
g
th
e
u
n
d
e
r
ly
in
g
s
tr
u
ctu
r
es
with
in
tex
tu
al
d
ata,
th
u
s
s
er
v
in
g
as
a
v
alu
a
b
le
t
o
o
l
in
NL
P
r
esear
ch
a
n
d
a
p
p
licatio
n
s
.
I
n
p
ap
er
[
1
3
]
,
PLSA
is
a
s
tatis
t
ical
tech
n
iq
u
e
wid
ely
e
m
p
lo
y
e
d
i
n
in
f
o
r
m
atio
n
r
etr
ie
v
al
an
d
v
ar
io
u
s
o
th
er
f
ield
s
,
k
n
o
wn
f
o
r
its
in
tu
itiv
e
an
d
r
o
b
u
s
t
r
esu
lts
.
Desp
ite
its
ef
f
ec
tiv
en
ess
,
PLSA
is
o
f
ten
cr
iticized
f
o
r
th
e
r
ig
i
d
ity
o
f
its
a
s
s
u
m
p
tio
n
s
an
d
th
e
iter
ativ
e
n
atu
r
e
o
f
t
h
e
ex
p
ec
ta
tio
n
-
m
ax
im
izatio
n
alg
o
r
ith
m
,
wh
ich
h
as
led
to
d
i
v
is
io
n
s
am
o
n
g
r
esear
ch
er
s
.
I
n
th
is
m
an
u
s
cr
ip
t,
we
p
r
o
v
id
e
an
o
v
er
v
iew
o
f
PLSA,
f
o
llo
wed
b
y
a
d
is
cu
s
s
io
n
o
n
r
ef
o
r
m
u
latio
n
s
aim
ed
at
ad
d
r
ess
in
g
its
in
h
e
r
en
t
p
r
o
b
l
em
s
.
Sp
ec
ial
atten
tio
n
is
g
iv
en
to
a
p
p
r
o
ac
h
es
th
at
co
n
n
e
ct
PLSA
with
th
e
s
in
g
u
lar
v
alu
e
d
ec
o
m
p
o
s
itio
n
th
eo
r
em
,
s
ee
k
in
g
to
e
n
h
an
ce
i
ts
ef
f
icien
cy
an
d
ef
f
ec
tiv
en
ess
.
Fu
r
th
er
m
o
r
e
,
we
ex
p
lo
r
e
h
o
w
PLSA
s
er
v
es
as
th
e
f
o
u
n
d
atio
n
f
o
r
o
th
er
tec
h
n
iq
u
es,
s
u
c
h
as
k
e
r
n
eliza
tio
n
an
d
p
r
o
b
ab
ilis
tic
tr
an
s
f
er
lear
n
i
n
g
.
T
h
ese
ex
te
n
s
io
n
s
tr
an
s
f
o
r
m
PLSA
f
r
o
m
a
d
escr
ip
tiv
e
t
o
o
l
in
t
o
an
in
f
e
r
en
tial
o
n
e,
th
er
e
b
y
o
p
en
in
g
u
p
n
ew
av
en
u
es f
o
r
r
esear
ch
an
d
a
p
p
licatio
n
s
.
2
.
2
.
Dee
p
lea
rning
a
pp
ro
a
ches f
o
r
t
o
pic m
o
delin
g
I
n
th
eir
s
tu
d
y
,
C
h
ai
an
d
L
i
[
1
4
]
ad
d
r
ess
th
e
is
s
u
e
o
f
in
te
r
p
r
etab
ilit
y
in
d
ee
p
lear
n
in
g
-
b
ased
tex
t
m
in
in
g
,
a
ch
allen
g
e
th
at
h
as
co
n
s
tr
ain
ed
th
e
ad
o
p
tio
n
an
d
im
p
r
o
v
em
en
t
o
f
s
u
ch
m
o
d
els
in
th
e
f
ield
o
f
in
f
o
r
m
atio
n
s
y
s
tem
s
(
I
S).
T
h
e
y
in
tr
o
d
u
ce
a
n
o
v
el
r
esear
ch
f
r
am
ewo
r
k
c
alled
“n
e
u
r
al
to
p
i
c
em
b
ed
d
in
g
”
t
h
at
co
m
b
in
es
d
ee
p
n
eu
r
al
n
etw
o
r
k
s
with
to
p
ic
m
o
d
elin
g
t
o
cr
ea
te
i
n
ter
p
r
etab
le
r
e
p
r
esen
tatio
n
s
o
f
tex
tu
al
d
o
cu
m
e
n
ts
.
T
h
r
o
u
g
h
a
p
r
elim
i
n
ar
y
ex
p
er
im
en
t in
f
ak
e
r
e
v
ie
w
d
etec
tio
n
,
th
eir
ap
p
r
o
ac
h
s
ig
n
if
ican
tly
en
h
an
ce
s
th
e
s
tate
-
of
-
th
e
-
ar
t p
er
f
o
r
m
a
n
c
e,
as in
d
icate
d
b
y
an
8
p
er
ce
n
t
im
p
r
o
v
em
e
n
t in
F1
s
co
r
e.
T
h
is
r
esear
ch
m
ar
k
s
a
s
ig
n
if
ican
t
co
n
tr
ib
u
tio
n
to
th
e
I
S
co
m
m
u
n
ity
,
o
f
f
er
in
g
a
p
r
o
m
is
in
g
p
ath
way
f
o
r
th
e
f
u
tu
r
e
in
teg
r
atio
n
o
f
s
tate
-
of
-
th
e
-
a
r
t d
ee
p
lear
n
in
g
m
eth
o
d
s
wh
ile
en
s
u
r
in
g
in
te
r
p
r
etab
il
ity
in
tex
t m
in
in
g
ap
p
licatio
n
s
.
2
.
2
.
1
.
B
idi
re
ct
io
na
l
enco
der
re
presenta
t
io
ns
f
ro
m
t
ra
ns
f
o
rm
er
s
B
id
ir
ec
tio
n
al
en
co
d
er
r
ep
r
ese
n
tatio
n
s
f
r
o
m
tr
a
n
s
f
o
r
m
e
r
s
(
B
E
R
T
o
p
ic)
[
1
5
]
is
an
ad
v
an
ce
d
to
p
ic
m
o
d
elin
g
ap
p
r
o
ac
h
th
at
h
ar
n
e
s
s
es
th
e
p
o
wer
o
f
tr
an
s
f
o
r
m
e
r
-
b
ased
d
ee
p
lear
n
in
g
m
o
d
els
to
id
en
tify
to
p
ics
in
ex
ten
s
iv
e
tex
t
d
atasets
.
I
ts
ad
v
an
tag
es
in
clu
d
e
th
e
g
en
er
a
tio
n
o
f
m
o
r
e
c
o
h
er
e
n
t
an
d
in
ter
p
r
etab
le
to
p
ics
co
m
p
ar
ed
to
tr
ad
itio
n
al
m
eth
o
d
s
lik
e
laten
t
Dir
ic
h
let
allo
ca
tio
n
.
B
E
R
T
o
p
ic
ex
ce
ls
in
h
an
d
lin
g
lar
g
e
tex
t
co
llectio
n
s
ef
f
ec
tiv
ely
an
d
ca
p
tu
r
es
s
em
an
tic
r
elatio
n
s
h
ip
s
b
etwe
en
wo
r
d
s
,
r
esu
ltin
g
i
n
h
ig
h
ly
ac
cu
r
ate
an
d
m
ea
n
in
g
f
u
l
to
p
ics.
Mo
r
eo
v
er
,
it
o
f
f
er
s
f
in
e
-
g
r
ain
ed
c
o
n
tr
o
l
o
v
er
th
e
n
u
m
b
e
r
o
f
to
p
ics
to
b
e
ex
tr
ac
ted
an
d
is
am
en
ab
le
to
f
in
e
-
t
u
n
in
g
s
p
ec
if
ic
tex
t d
atasets
f
o
r
en
h
an
ce
d
p
er
f
o
r
m
a
n
ce
.
I
n
s
u
m
m
a
r
y
,
B
E
R
T
o
p
ic
r
ep
r
esen
ts
a
s
ig
n
if
ican
t a
d
v
an
ce
m
en
t in
to
p
ic
m
o
d
elin
g
with
a
wid
e
r
a
n
g
e
o
f
b
e
n
ef
its
f
o
r
r
esear
ch
er
s
a
n
d
p
r
ac
titi
o
n
e
r
s
.
2
.
2
.
2
.
G
ener
a
t
iv
e
pre
-
t
ra
ined t
ra
ns
f
o
rm
er
Gen
er
ativ
e
p
r
e
-
tr
ain
e
d
tr
an
s
f
o
r
m
er
s
,
o
r
GPT
m
o
d
els
[
1
6
]
,
b
u
ilt
o
n
th
e
tr
an
s
f
o
r
m
er
ar
ch
itectu
r
e,
r
ep
r
esen
t
a
p
iv
o
tal
ad
v
an
ce
m
en
t
in
ar
tific
ial
in
tellig
en
ce
(
AI
)
.
T
h
e
y
em
p
o
wer
ap
p
licatio
n
s
lik
e
C
h
atGPT
to
g
en
er
ate
h
u
m
an
-
lik
e
te
x
t
an
d
m
u
ltime
d
ia
co
n
ten
t,
a
n
s
wer
q
u
esti
o
n
s
co
n
v
er
s
atio
n
ally
,
a
n
d
f
in
d
ap
p
licatio
n
s
ac
r
o
s
s
v
ar
io
u
s
in
d
u
s
tr
ies.
T
h
e
s
ig
n
if
ican
ce
o
f
GPT
lies
in
its
tr
an
s
f
o
r
m
ativ
e
im
p
ac
t
o
n
AI
r
esear
ch
,
in
tr
o
d
u
cin
g
a
n
ew
er
a
o
f
m
ac
h
in
e
lear
n
in
g
ad
o
p
tio
n
.
GPT
m
o
d
els,
d
u
e
to
th
eir
s
p
ee
d
an
d
s
ca
lab
ilit
y
,
ca
n
au
to
m
ate
an
d
en
h
an
ce
a
n
ar
r
a
y
o
f
task
s
,
f
r
o
m
lan
g
u
ag
e
tr
a
n
s
latio
n
an
d
d
o
cu
m
en
t
s
u
m
m
ar
izatio
n
to
c
o
n
ten
t
cr
ea
tio
n
,
web
s
ite
d
esig
n
,
co
d
e
g
en
er
atio
n
,
a
n
d
ev
e
n
p
o
etr
y
co
m
p
o
s
itio
n
.
T
h
e
y
ar
e
a
d
r
i
v
in
g
f
o
r
ce
b
eh
in
d
th
e
p
u
r
s
u
it
o
f
ar
tific
ial
g
en
e
r
al
in
tellig
en
ce
,
o
f
f
e
r
in
g
th
e
p
o
te
n
tial
to
elev
ate
o
r
g
an
izatio
n
a
l
p
r
o
d
u
ctiv
ity
a
n
d
tr
an
s
f
o
r
m
a
p
p
licatio
n
s
an
d
u
s
er
ex
p
er
ie
n
ce
s
.
T
h
is
r
ev
iew
p
a
p
er
ex
p
lo
r
es
th
e
p
r
o
f
o
u
n
d
im
p
ac
t
o
f
GPT
m
o
d
els
in
AI
r
esear
ch
a
n
d
th
eir
wid
e
-
r
an
g
in
g
ap
p
licatio
n
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
E
xp
lo
r
in
g
to
p
ic
mo
d
ellin
g
:
a
co
mp
a
r
a
tive
a
n
a
lysi
s
o
f tra
d
itio
n
a
l a
n
d
tr
a
n
s
fo
r
mer
-
b
a
s
ed
…
(
A
ye
s
h
a
R
ia
z
)
1937
2
.
2
.
3
.
E
x
t
ra
lo
ng
-
t
er
m m
e
m
o
ry
net
wo
rk
s
XL
Net
[
1
7
]
,
wh
ich
s
tan
d
s
f
o
r
“e
x
tr
a
lo
n
g
-
ter
m
m
em
o
r
y
n
et
wo
r
k
s
,
”
is
an
ad
v
an
ce
d
lan
g
u
ag
e
m
o
d
el
u
s
ed
in
NL
P
.
Un
lik
e
o
ld
er
m
o
d
els
lik
e
GPT,
XL
Net
in
tr
o
d
u
ce
s
a
n
ew
way
o
f
tr
ai
n
in
g
ca
lled
“
p
er
m
u
tatio
n
lan
g
u
ag
e
m
o
d
elin
g
.
”
T
h
is
h
el
p
s
it
u
n
d
er
s
tan
d
t
h
e
o
r
d
er
o
f
wo
r
d
s
in
a
s
en
ten
ce
b
etter
,
c
ap
tu
r
in
g
c
o
n
tex
t
in
b
o
th
d
i
r
ec
tio
n
s
.
I
t
c
o
m
b
in
es
t
wo
ty
p
es
o
f
m
o
d
els,
au
to
r
e
g
r
ess
iv
e
(
ca
p
tu
r
es
th
e
co
n
tex
t
f
r
o
m
lef
t
to
r
ig
h
t,
as
s
ee
n
in
tr
ad
itio
n
al
lan
g
u
ag
e
m
o
d
els)
an
d
a
u
to
en
co
d
in
g
(
allo
ws
th
e
m
o
d
el
to
lev
er
a
g
e
b
i
d
ir
ec
tio
n
al
co
n
te
x
t)
,
allo
win
g
it
to
u
n
d
e
r
s
tan
d
lan
g
u
ag
e
m
o
r
e
th
o
r
o
u
g
h
ly
.
Simil
ar
to
GPT,
XL
Net
is
v
er
s
atile
,
ex
ce
llin
g
in
task
s
lik
e
tex
t
co
m
p
letio
n
an
d
tr
a
n
s
latio
n
.
I
ts
im
p
ac
t
o
n
AI
r
esear
ch
h
as
b
ee
n
s
ig
n
if
ican
t,
o
v
er
c
o
m
in
g
s
o
m
e
lim
itatio
n
s
o
f
o
ld
er
m
o
d
els
an
d
co
n
tr
ib
u
tin
g
to
ad
v
an
ce
m
e
n
ts
in
n
at
u
r
al
lan
g
u
ag
e
p
r
o
ce
s
s
in
g
task
s
.
I
n
s
im
p
ler
ter
m
s
,
XL
Net
is
a
p
o
wer
f
u
l
to
o
l
th
at
u
n
d
er
s
tan
d
s
la
n
g
u
a
g
e
b
etter
,
th
an
k
s
to
its
in
n
o
v
ativ
e
tr
ain
in
g
ap
p
r
o
ac
h
,
m
ak
in
g
it a
k
ey
p
la
y
er
in
t
h
e
d
ev
elo
p
m
en
t
o
f
ar
tific
ial
in
telli
g
en
ce
.
3.
L
I
T
E
R
AT
U
RE
R
E
VI
E
W
I
n
th
is
s
ec
tio
n
,
we
r
ev
iew
th
e
wo
r
k
r
elate
d
to
m
o
d
els
b
ased
o
n
em
b
ed
d
in
g
an
d
th
e
m
o
d
els
in
wh
ich
to
p
ics
ar
e
g
en
er
ated
f
o
r
s
p
ec
if
ic
p
ar
ts
o
f
s
p
ee
ch
in
a
d
ataset.
E
m
b
ed
d
in
g
-
b
ased
m
o
d
els
h
av
e
g
ain
ed
p
r
o
m
in
e
n
ce
f
o
r
th
eir
ab
ilit
y
to
ca
p
tu
r
e
s
em
an
tic
r
elatio
n
s
h
ip
s
b
etwe
en
wo
r
d
s
an
d
p
h
r
ases
,
p
r
o
v
id
i
n
g
a
r
o
b
u
s
t
f
o
u
n
d
atio
n
f
o
r
v
a
r
io
u
s
n
atu
r
al
lan
g
u
ag
e
p
r
o
ce
s
s
in
g
task
s
.
O
n
th
e
o
th
e
r
h
an
d
,
m
o
d
els
th
at
f
o
cu
s
o
n
g
en
er
atin
g
to
p
ics
f
o
r
s
p
ec
if
ic
p
ar
ts
o
f
s
p
ee
ch
,
s
u
ch
as
n
o
u
n
s
o
r
v
er
b
s
,
o
f
f
er
m
o
r
e
f
in
e
-
g
r
ain
e
d
in
s
ig
h
t
s
in
to
th
e
s
tr
u
ctu
r
e
an
d
th
em
atic
d
is
tr
ib
u
tio
n
o
f
d
atasets
.
T
h
is
d
u
al
f
o
cu
s
o
n
e
m
b
ed
d
in
g
tech
n
iq
u
es
an
d
tar
g
eted
to
p
ic
m
o
d
elin
g
h
ig
h
lig
h
ts
th
e
ev
o
lv
in
g
m
eth
o
d
o
lo
g
ies aim
ed
at
e
n
h
an
ci
n
g
t
ex
tu
al
an
aly
s
is
an
d
u
n
d
er
s
tan
d
in
g
.
3
.
1
.
Sentim
ent
a
na
ly
s
is
a
nd
deep
lea
rning
a
pp
ro
a
ches in un
d
er
s
t
a
nd
ing
t
ex
t
ua
l da
t
a
I
n
r
ec
en
t
y
ea
r
s
a
lo
t
o
f
r
esear
ch
h
as
b
ee
n
ca
r
r
ied
o
u
t
th
at
h
av
e
s
h
o
wn
th
at
w
o
r
d
em
b
ed
d
i
n
g
is
q
u
ite
u
s
ef
u
l
f
o
r
to
p
ic
m
o
d
elin
g
.
T
h
e
r
ev
iew
in
th
is
s
ec
tio
n
is
m
ain
ly
co
n
s
is
ts
o
f
em
b
ed
d
in
g
b
ased
m
o
d
els.
W
ith
th
e
p
r
o
life
r
atio
n
a
n
d
ex
p
a
n
s
io
n
o
f
o
n
lin
e
b
u
s
in
ess
es
h
av
e
s
tr
ea
m
lin
ed
th
e
p
r
o
ce
s
s
o
f
o
b
tain
i
n
g
d
ir
ec
t
cu
s
to
m
er
f
ee
d
b
ac
k
.
T
h
is
ac
ce
s
s
ib
ilit
y
e
n
ab
les
co
m
p
an
ies
to
id
en
tify
t
h
eir
wea
k
n
ess
es
an
d
tak
e
m
ea
s
u
r
es
to
s
tr
en
g
th
en
th
eir
m
ar
k
et
p
o
s
itio
n
.
Ad
d
itio
n
ally
,
s
en
tim
en
t
an
aly
s
is
s
er
v
es
as
a
v
alu
ab
le
to
o
l
f
o
r
co
m
p
r
eh
en
d
in
g
p
eo
p
le
’
s
em
o
tio
n
s
an
d
attitu
d
es
to
war
d
th
eir
p
r
o
d
u
cts
[
1
8
]
.
W
h
ile
s
o
cial
m
ed
ia
p
latf
o
r
m
s
h
av
e
g
r
a
n
ted
in
d
i
v
id
u
als
a
p
latf
o
r
m
f
o
r
f
r
ee
ex
p
r
ess
io
n
,
t
h
ey
h
av
e
also
b
ee
n
ass
o
ciate
d
with
d
is
tu
r
b
i
n
g
in
cid
en
ts
.
Ha
te
s
p
ee
ch
o
n
s
o
cial
m
ed
ia
h
as
em
er
g
ed
as
a
s
ig
n
if
ican
t
co
n
ce
r
n
,
p
o
s
in
g
a
s
ev
er
e
th
r
ea
t
to
p
eo
p
le
’
s
well
-
b
ein
g
.
Sen
tim
en
t
an
aly
s
is
ca
n
p
lay
a
c
r
u
cial
r
o
le
i
n
id
en
t
if
y
in
g
an
d
ad
d
r
ess
in
g
h
ate
s
p
e
ec
h
o
n
t
h
ese
p
latf
o
r
m
s
.
Var
io
u
s
d
ee
p
lear
n
in
g
tech
n
i
q
u
es,
s
u
ch
as
c
o
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
s
(
C
NN)
an
d
r
ec
u
r
r
en
t
n
eu
r
al
n
etwo
r
k
s
(
R
NN)
ca
n
b
e
em
p
lo
y
e
d
f
o
r
th
is
p
u
r
p
o
s
e
[
1
9
]
.
Stan
f
o
r
d
U
n
iv
er
s
ity
co
n
d
u
cte
d
r
esear
ch
u
tili
zin
g
em
o
tio
n
s
y
m
b
o
ls
to
ex
tr
ac
t
twee
ts
,
em
p
lo
y
in
g
T
witter
’
s
API
,
with
ea
ch
twee
t
r
ep
r
esen
tin
g
eith
e
r
a
p
o
s
itiv
e
o
r
n
e
g
ativ
e
s
en
tim
en
t
.
T
o
en
h
an
ce
th
e
ac
c
u
r
ac
y
o
f
class
if
icatio
n
m
o
d
els,
s
ev
er
al
d
ata
p
r
e
p
r
o
ce
s
s
in
g
tech
n
iq
u
es
wer
e
im
p
lem
en
ted
.
T
h
is
r
esear
ch
h
as
th
e
p
o
ten
tia
l
to
au
to
m
ate
th
e
class
if
icatio
n
o
f
twee
ts
with
o
u
t
th
e
n
ee
d
f
o
r
m
an
u
al
in
ter
v
en
tio
n
.
I
t
s
h
o
u
ld
b
e
n
o
ted
th
at
th
eir
s
en
tim
en
t
an
aly
s
is
wa
s
co
n
d
u
cted
co
m
p
r
eh
e
n
s
iv
ely
with
o
u
t sp
ec
if
ic
d
o
m
ain
f
o
cu
s
[
2
0
]
.
T
witter
h
as
b
ec
o
m
e
r
ea
ll
y
i
m
p
o
r
tan
t
f
o
r
f
ig
u
r
i
n
g
o
u
t
wh
at
p
eo
p
le
th
i
n
k
an
d
f
ee
l
b
ec
a
u
s
e
lo
ts
o
f
d
if
f
er
en
t
t
y
p
es
o
f
p
eo
p
le
u
s
e
it.
Peo
p
le
u
s
e
all
s
o
r
ts
o
f
m
eth
o
d
s
to
f
ig
u
r
e
o
u
t
wh
at
p
e
o
p
le
ar
e
s
ay
in
g
o
n
T
witter
,
b
u
t
s
o
m
etim
es
it
is
n
o
t
to
tally
ac
cu
r
ate
b
ec
au
s
e
th
er
e
ar
e
n
o
t
alwa
y
s
g
o
o
d
s
ets
o
f
d
ata
to
u
s
e,
an
d
u
n
d
er
s
tan
d
i
n
g
h
o
w
lan
g
u
a
g
e
wo
r
k
s
ca
n
b
e
t
r
ick
y
[
2
1
]
.
Sen
tim
en
t
an
al
y
s
is
is
u
s
ef
u
l
i
n
m
an
y
a
r
ea
s
,
lik
e
en
ter
tain
m
en
t
an
d
h
ea
lth
ca
r
e.
Peo
p
le
ar
e
s
tu
d
y
in
g
it
a
lo
t,
b
u
t
s
o
m
etim
es
it
is
h
ar
d
to
d
o
r
esear
ch
b
ec
au
s
e
th
er
e
is
n
o
t
en
o
u
g
h
d
ata.
E
v
e
n
wh
en
th
er
e
is
d
ata,
g
ettin
g
p
er
m
is
s
io
n
an
d
p
a
p
er
wo
r
k
ca
n
tak
e
a
lo
n
g
tim
e,
ev
en
m
o
n
t
h
s
[
2
2
]
.
As
s
o
cial
n
etwo
r
k
in
g
s
ites
h
av
e
b
ec
o
m
e
i
n
cr
ea
s
in
g
ly
p
o
p
u
lar
,
p
eo
p
le
n
o
w
ten
d
to
ex
p
r
ess
th
eir
em
o
tio
n
s
o
n
lin
e
r
ath
er
th
an
c
o
n
f
i
d
in
g
in
th
eir
f
am
ily
m
em
b
er
s
.
T
h
is
s
h
if
t
h
as
s
p
ar
k
ed
n
u
m
er
o
u
s
r
esear
ch
in
itiativ
es in
th
e
f
ield
o
f
m
en
tal
h
ea
lth
,
u
tili
zin
g
NL
P
tech
n
iq
u
es
[
2
3
]
.
W
h
ile
d
ata
r
elate
d
to
m
en
ta
l
h
ea
lth
r
em
ain
s
h
i
g
h
ly
c
o
n
f
id
en
tial
an
d
o
f
ten
in
ac
ce
s
s
ib
le
d
u
e
to
ag
r
ee
m
en
ts
with
d
ata
s
o
u
r
ce
s
,
r
ec
en
t
tech
n
o
lo
g
ical
a
d
v
an
c
em
en
ts
h
av
e
in
tr
o
d
u
ce
d
API
s
lik
e
T
wee
p
y
an
d
T
witter
4
j.
T
h
ese
API
s
en
a
b
le
th
e
c
o
llectio
n
o
f
o
n
lin
e
d
ata
f
o
r
s
en
tim
en
t
an
aly
s
is
,
p
a
r
ticu
lar
ly
f
r
o
m
v
ar
io
u
s
s
o
cial
m
ed
ia
p
latf
o
r
m
s
.
B
esid
es
tex
tu
al
p
o
s
ts
,
m
an
y
s
o
cial
m
ed
ia
web
s
ites
,
s
u
ch
as
Fli
ck
r
an
d
I
n
s
tag
r
am
,
f
ac
ilit
ate
th
e
s
h
ar
in
g
a
n
d
h
o
s
tin
g
o
f
im
ag
es.
A
n
aly
zin
g
th
ese
im
ag
es,
wh
ich
r
e
f
lect
u
s
er
s
’
m
o
o
d
s
a
n
d
em
o
tio
n
s
,
ca
n
p
r
o
v
id
e
v
alu
a
b
le
in
s
ig
h
ts
in
to
th
eir
em
o
tio
n
al
s
tates
an
d
s
en
tim
en
ts
[
2
4
]
.
Alth
o
u
g
h
s
o
cial
m
ed
ia
is
o
f
ten
ass
o
ciate
d
with
co
n
tr
ib
u
tin
g
t
o
m
e
n
tal
h
ea
lth
is
s
u
es
lik
e
in
cr
ea
s
ed
an
x
iety
,
i
t
h
as
also
p
r
o
v
id
ed
in
d
iv
id
u
als with
a
p
latf
o
r
m
to
ch
allen
g
e
s
tig
m
as a
n
d
o
u
td
ate
d
b
elief
s
,
allo
win
g
th
em
to
ex
p
r
ess
th
eir
th
o
u
g
h
ts
with
o
u
t
f
ea
r
.
Giv
en
t
h
e
o
p
en
n
ess
an
d
f
r
ee
d
o
m
f
ac
ilit
a
ted
b
y
s
o
cial
m
ed
ia,
th
ese
p
latf
o
r
m
s
h
av
e
a
r
esp
o
n
s
ib
ilit
y
to
ad
d
r
ess
th
e
g
r
o
win
g
m
e
n
tal
h
ea
lth
co
n
ce
r
n
s
.
Ho
wev
er
,
it
is
im
p
o
r
tan
t
to
n
o
te
th
at
n
o
n
e
o
f
th
e
ex
is
tin
g
r
esear
ch
ca
n
g
u
a
r
an
tee
th
e
ac
cu
r
ate
p
r
ed
ictio
n
o
f
all
s
u
icid
al
an
d
n
o
n
-
s
u
icid
al
p
o
s
ts
o
n
s
o
cial
m
ed
ia
b
y
d
i
f
f
er
en
t
u
s
er
s
[
2
5
]
.
T
h
e
ch
allen
g
es
ar
is
e
d
u
e
to
is
s
u
es
in
h
er
en
t
i
n
n
atu
r
al
la
n
g
u
ag
e
p
r
o
ce
s
s
in
g
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
2
,
Ap
r
il
20
25
:
1
9
3
3
-
1
9
4
8
1938
W
h
ile
ac
h
iev
in
g
a
1
0
0
%
ac
cu
r
ate
r
esu
lt
is
d
if
f
icu
lt,
s
en
tim
e
n
t
an
aly
s
is
d
o
es
p
r
o
v
id
e
v
alu
a
b
le
in
s
ig
h
ts
an
d
,
in
s
o
m
e
in
s
tan
ce
s
,
ca
n
h
elp
p
r
ev
en
t m
is
h
ap
s
.
B
ey
o
n
d
a
n
aly
zin
g
s
o
cial
m
e
d
ia
p
o
s
ts
,
th
er
e
ar
e
o
th
e
r
m
eth
o
d
s
f
o
r
ass
ess
in
g
in
d
iv
i
d
u
al
s
en
tim
en
t,
s
u
ch
as
clas
s
if
y
in
g
s
u
icid
e
n
o
tes.
T
h
is
p
r
o
ce
s
s
in
v
o
lv
es
l
in
g
u
is
tic
an
aly
s
is
,
as
in
d
iv
id
u
als
ex
p
r
ess
th
eir
em
o
tio
n
s
u
n
iq
u
ely
in
s
u
icid
e
n
o
tes.
Dif
f
er
en
ce
s
in
lan
g
u
ag
e,
p
r
o
n
o
u
n
u
s
ag
e,
an
d
g
r
am
m
ar
m
ak
e
it
ch
allen
g
in
g
to
ac
c
u
r
ately
class
if
y
th
ese
n
o
tes
[
2
6
]
.
Ad
d
itio
n
al
r
esear
ch
ef
f
o
r
ts
h
av
e
also
b
ee
n
d
ed
icate
d
t
o
d
is
tin
g
u
is
h
in
g
g
en
u
in
e
f
r
o
m
f
ak
e
s
u
icid
e
n
o
tes,
e
m
p
lo
y
i
n
g
a
r
an
g
e
o
f
tec
h
n
iq
u
es
f
r
o
m
tr
a
d
itio
n
al
ap
p
r
o
ac
h
es
to
cu
ttin
g
-
ed
g
e
m
ac
h
i
n
e
lear
n
i
n
g
m
eth
o
d
s
.
On
e
lim
itin
g
f
ac
to
r
in
t
h
is
r
esear
ch
d
o
m
ain
is
th
e
lim
ited
av
ailab
ilit
y
o
f
s
am
p
le
d
ata.
T
h
e
u
tili
za
tio
n
o
f
NL
P
an
d
tex
t
m
in
in
g
in
s
e
n
tim
en
t
an
aly
s
is
is
s
till
ev
o
lv
i
n
g
,
wh
ich
ex
p
lain
s
th
e
ch
allen
g
es
f
ac
ed
b
y
m
ajo
r
s
o
cial
m
ed
ia
p
latf
o
r
m
s
in
ac
cu
r
ately
id
en
tify
i
n
g
p
o
s
ts
ex
p
r
es
s
in
g
n
eg
ativ
e
s
en
tim
en
ts
[
2
7
]
.
I
n
th
e
ea
r
ly
s
tag
es
o
f
s
en
tim
en
t
an
aly
s
is
,
th
e
f
o
cu
s
was
p
r
im
ar
ily
o
n
in
d
iv
id
u
a
l
wo
r
d
s
r
ath
er
th
an
co
m
p
r
e
h
e
n
d
in
g
th
e
co
n
te
x
t
in
wh
ich
th
ese
wo
r
d
s
ap
p
ea
r
ed
[
2
8
]
.
Su
b
s
eq
u
en
tly
,
a
W
o
r
d
N
et
-
b
ased
ap
p
r
o
ac
h
was
in
tr
o
d
u
ce
d
f
o
r
s
en
tim
en
t
an
aly
s
is
,
wh
ich
in
v
o
lv
ed
ca
lc
u
latin
g
th
e
d
is
tan
ce
b
etwe
en
wo
r
d
s
in
a
tex
t a
n
d
th
e
ter
m
s
“
good
”
o
r
“
b
a
d
”
[
2
9
]
.
So
m
e
r
esear
ch
er
s
ev
en
u
tili
ze
d
C
o
s
in
e
d
is
tan
ce
to
en
h
an
ce
ac
cu
r
ac
y
.
T
h
e
ad
o
p
tio
n
o
f
b
id
ir
ec
tio
n
al
lo
n
g
s
h
o
r
t
-
ter
m
m
em
o
r
y
(
Bi
-
L
ST
M
)
in
s
en
tim
en
t
an
aly
s
is
m
ar
k
ed
a
s
h
if
t
to
war
d
s
co
n
s
id
er
in
g
co
n
tex
t,
m
ak
i
n
g
it
a
m
o
r
e
ef
f
ec
tiv
e
m
et
h
o
d
c
o
m
p
a
r
ed
to
o
th
er
s
[
3
0
]
.
Dea
lin
g
wit
h
ex
ten
s
iv
e
tex
t
an
d
d
ec
i
p
h
er
i
n
g
its
m
ea
n
in
g
ca
n
b
e
ch
allen
g
in
g
,
an
d
o
n
e
tec
h
n
iq
u
e
to
b
etter
g
r
asp
t
h
e
s
u
b
jects
with
in
lar
g
e
c
o
llectio
n
s
o
f
tex
t
is
to
p
ic
m
o
d
elin
g
[
3
1
]
.
T
o
p
ic
m
o
d
elin
g
h
as
f
o
u
n
d
ap
p
licatio
n
s
in
d
iv
er
s
e
f
ield
s
s
u
ch
as
p
o
liti
ca
l
s
cien
ce
,
cu
s
to
m
er
r
ev
iews,
s
o
f
twar
e
en
g
in
ee
r
in
g
,
m
ed
icin
e,
an
d
lin
g
u
is
tics
[
3
2
]
.
3
.
2
.
T
o
pic
m
o
delin
g
f
ro
m
t
r
a
ditio
na
l a
pp
ro
a
ches t
o
mo
dern
t
ec
hn
iq
ues
W
h
ile
th
er
e
ar
e
n
u
m
er
o
u
s
to
p
ic
m
o
d
elin
g
alg
o
r
ith
m
s
av
aila
b
le,
o
p
tim
izin
g
an
d
f
in
e
-
tu
n
in
g
th
em
is
cr
u
cial
f
o
r
o
b
tain
i
n
g
r
eliab
le
r
esu
lts
.
A
s
o
lid
u
n
d
er
s
tan
d
in
g
o
f
th
e
u
n
d
e
r
ly
in
g
p
r
o
ce
s
s
es
in
th
ese
alg
o
r
ith
m
s
is
n
ec
ess
ar
y
to
d
eter
m
in
e
wh
ich
o
n
e
b
est
alig
n
s
with
th
e
s
p
ec
if
ic
o
b
jectiv
es
at
h
an
d
[
3
3
]
.
T
o
p
ic
m
o
d
elin
g
co
n
tin
u
es
to
h
o
ld
a
p
i
v
o
tal
r
o
le
in
th
e
d
o
m
ain
s
o
f
m
ac
h
in
e
lear
n
in
g
an
d
n
atu
r
al
la
n
g
u
ag
e
p
r
o
ce
s
s
in
g
,
n
ec
ess
itatin
g
o
n
g
o
in
g
r
esear
c
h
ef
f
o
r
ts
to
en
h
an
ce
its
u
tili
ty
ac
r
o
s
s
v
ar
io
u
s
d
o
m
ain
s
.
T
h
e
lan
d
s
ca
p
e
o
f
to
p
ic
m
o
d
elin
g
h
as
u
n
d
er
g
o
n
e
s
ig
n
i
f
ican
t
tr
an
s
f
o
r
m
atio
n
s
s
in
ce
it
s
in
ce
p
tio
n
.
Am
o
n
g
th
e
ea
r
lie
s
t
an
d
m
o
s
t
wid
ely
ad
o
p
ted
m
o
d
els,
L
DA
,
o
r
ig
i
n
ally
in
tr
o
d
u
ce
d
b
y
B
lei
et
a
l.
i
n
2
0
0
3
[
3
]
,
r
em
ain
s
a
f
o
u
n
d
at
io
n
al
an
d
p
r
e
v
alen
t
ap
p
r
o
ac
h
in
th
e
r
ea
lm
o
f
p
r
o
b
ab
ilis
tic
to
p
ic
m
o
d
elin
g
.
R
eisen
b
ich
ler
an
d
R
eu
tter
er
[
3
4
]
em
p
h
asized
th
e
ap
p
licatio
n
o
f
to
p
ic
m
o
d
els
in
m
ar
k
etin
g
r
esear
ch
,
h
ig
h
li
g
h
t
in
g
th
eir
ef
f
icac
y
i
n
r
ev
ea
lin
g
h
id
d
e
n
p
atter
n
s
in
co
m
p
lex
co
-
o
cc
u
r
r
en
ce
d
ata.
T
h
ey
n
o
ted
th
e
in
c
r
ea
s
in
g
in
t
er
est
am
o
n
g
m
ar
k
etin
g
s
ch
o
lar
s
an
d
p
r
ac
titi
o
n
e
r
s
in
u
tili
zin
g
th
ese
v
er
s
atile
clu
s
ter
in
g
tech
n
iq
u
es
d
u
e
to
th
e
av
ailab
ilit
y
o
f
ex
ten
s
iv
e
d
atasets
.
Ho
wev
er
,
th
ey
id
en
tifie
d
a
s
ig
n
if
ican
t g
ap
in
th
e
liter
atu
r
e.
An
ab
s
en
ce
o
f
a
co
m
p
r
eh
e
n
s
iv
e
o
v
er
v
iew
o
f
th
is
r
ap
id
ly
ev
o
lv
in
g
f
ield
.
T
o
a
d
d
r
ess
th
is
g
ap
,
th
e
au
th
o
r
s
c
o
n
d
u
cted
a
s
y
s
tem
a
tic
r
ev
iew
o
f
6
1
r
elev
an
t
p
ap
er
s
an
d
co
n
ce
p
tu
al
co
n
tr
ib
u
tio
n
s
,
ex
p
lo
r
in
g
v
ar
io
u
s
d
im
en
s
io
n
s
in
clu
d
in
g
d
ata
s
tr
u
ctu
r
es,
m
o
d
el
im
p
lem
en
tat
io
n
,
ex
ten
s
io
n
s
,
an
d
p
er
f
o
r
m
an
ce
ev
alu
atio
n
.
T
h
e
ir
o
b
jectiv
e
was
to
ca
teg
o
r
i
ze
an
d
illu
m
in
ate
th
e
c
o
n
tr
ib
u
tio
n
s
m
a
d
e
in
m
ar
k
etin
g
r
esear
ch
t
h
r
o
u
g
h
to
p
ic
m
o
d
els,
s
h
o
wca
s
in
g
a
d
v
an
ce
m
en
ts
ac
r
o
s
s
m
ar
k
etin
g
s
u
b
-
d
o
m
ain
s
.
T
h
e
au
th
o
r
s
r
ev
iew
n
o
t
o
n
ly
s
y
n
t
h
esized
ex
is
tin
g
liter
atu
r
e
b
u
t
also
p
r
o
v
id
ed
v
al
u
ab
le
in
s
ig
h
ts
in
to
p
o
ten
tial
f
u
tu
r
e
r
esear
c
h
d
ir
ec
tio
n
s
in
m
ar
k
etin
g
,
with
a
f
o
cu
s
o
n
th
e
u
tili
za
tio
n
o
f
to
p
ic
m
o
d
elin
g
tech
n
iq
u
es.
Sh
o
r
t
tex
ts
ex
h
i
b
it
less
wo
r
d
co
-
o
cc
u
r
r
en
ce
s
t
h
an
lo
n
g
er
tex
ts
d
o
,
an
d
wo
r
d
c
o
-
o
cc
u
r
r
en
ce
s
ar
e
cr
u
cial
f
o
r
t
o
p
ic
m
o
d
ellin
g
b
e
ca
u
s
e
ex
is
tin
g
m
o
d
els
ar
e
b
ased
o
n
wo
r
d
c
o
-
o
cc
u
r
r
en
ce
s
an
d
f
ail
wh
en
a
p
p
lied
to
s
h
o
r
t
tex
ts
[
3
5
]
.
Me
an
wh
il
e,
Siv
an
an
d
h
am
et
a
l.
[
3
6
]
,
in
th
eir
w
o
r
k
in
2
0
2
1
,
em
p
lo
y
e
d
a
co
m
b
i
n
atio
n
o
f
tech
n
iq
u
es,
in
clu
d
in
g
L
DA
to
p
ic
m
o
d
elin
g
,
NL
P
,
tim
e
s
er
ie
s
an
aly
s
is
,
an
d
v
ec
to
r
au
to
-
r
eg
r
ess
io
n
,
to
an
aly
ze
r
esear
ch
tr
en
d
s
a
n
d
m
a
k
e
p
r
e
d
ictio
n
s
u
s
in
g
to
p
ic
m
o
d
elin
g
.
I
t
is
wo
r
th
n
o
tin
g
th
at
alth
o
u
g
h
th
ei
r
m
eth
o
d
s
wer
e
well
-
d
etailed
,
th
eir
d
ataset
o
n
ly
co
v
er
ed
ar
ticles
s
p
an
n
in
g
ap
p
r
o
x
im
ately
n
in
e
y
e
ar
s
,
wh
ich
m
ay
n
o
t
p
r
o
v
id
e
en
tire
ly
ac
c
u
r
ate
r
esu
l
ts
.
Gr
is
ales
et
-
al
.
[
3
7
]
d
is
cu
s
s
ed
th
e
ev
o
lu
tio
n
an
d
ap
p
licatio
n
s
o
f
to
p
ic
m
o
d
elin
g
,
a
p
r
o
m
in
e
n
t
n
atu
r
al
lan
g
u
ag
e
p
r
o
ce
s
s
in
g
tech
n
iq
u
e,
b
y
em
p
lo
y
in
g
b
ib
lio
m
etr
ic
an
aly
s
is
.
I
t
r
ev
ea
ls
th
at
th
e
USA
an
d
C
h
in
a
ar
e
am
o
n
g
th
e
m
o
s
t
p
r
o
d
u
ctiv
e
c
o
u
n
tr
ies
in
th
is
f
ield
,
with
ap
p
licatio
n
s
p
r
im
ar
ily
f
o
cu
s
ed
o
n
id
en
tify
in
g
s
u
b
-
to
p
ics
in
s
h
o
r
t
tex
ts
lik
e
s
o
cia
l
n
etwo
r
k
s
an
d
b
lo
g
s
.
T
h
e
s
tu
d
y
u
n
d
er
s
co
r
es
th
e
v
er
s
atility
o
f
to
p
ic
m
o
d
elin
g
an
d
its
v
alu
a
b
le
r
o
le
in
s
y
s
tem
atica
lly
r
ev
iewin
g
v
ast
v
o
lu
m
es
o
f
u
n
s
tr
u
ctu
r
e
d
in
f
o
r
m
ati
o
n
,
o
f
f
er
i
n
g
i
n
s
ig
h
ts
b
en
ef
icial
f
o
r
r
esear
c
h
er
s
an
d
ac
ad
em
ics
en
g
ag
ed
in
v
ar
io
u
s
ac
ad
em
ic
an
d
r
esear
ch
c
o
n
te
x
ts
.
T
h
is
s
tu
d
y
u
s
es
th
e
L
DA
m
o
d
el
to
p
er
f
o
r
m
to
p
ic
ex
tr
ac
tio
n
o
n
th
e
ab
s
tr
ac
ts
o
f
d
o
cu
m
en
ts
f
r
o
m
2
5
jo
u
r
n
als
an
d
1
9
co
n
f
er
e
n
ce
s
m
ar
k
ed
.
Yad
av
h
ig
h
lig
h
ted
th
e
s
ig
n
i
f
ican
ce
o
f
s
en
tim
en
t
an
aly
s
is
in
to
d
ay
’
s
co
m
p
etitiv
e
wo
r
ld
,
s
h
o
wca
s
in
g
its
wid
e
-
r
an
g
in
g
a
p
p
licatio
n
s
ac
r
o
s
s
v
a
r
io
u
s
s
ec
t
o
r
s
,
f
r
o
m
h
ea
lth
ca
r
e
to
e
n
ter
ta
in
m
en
t,
c
o
r
p
o
r
ate,
an
d
p
o
liti
cs.
T
h
e
s
tu
d
y
f
o
cu
s
e
s
o
n
an
aly
zin
g
h
ea
d
lin
es
f
r
o
m
I
n
d
ia
’
s
lead
in
g
n
ews
p
o
r
tal,
t
h
e
T
im
es
o
f
I
n
d
ia,
em
p
lo
y
in
g
b
o
t
h
s
u
p
er
v
is
ed
a
n
d
u
n
s
u
p
er
v
is
ed
tech
n
iq
u
es.
I
n
s
u
p
er
v
is
ed
s
en
tim
en
t
an
al
y
s
is
,
th
e
B
i
-
L
STM
tech
n
iq
u
e
is
em
p
l
o
y
ed
,
wh
ile
u
n
s
u
p
er
v
is
ed
to
p
ic
m
o
d
elin
g
is
co
n
d
u
cted
u
s
in
g
L
DA
a
n
d
L
SA
,
with
L
DA
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
E
xp
lo
r
in
g
to
p
ic
mo
d
ellin
g
:
a
co
mp
a
r
a
tive
a
n
a
lysi
s
o
f tra
d
itio
n
a
l a
n
d
tr
a
n
s
fo
r
mer
-
b
a
s
ed
…
(
A
ye
s
h
a
R
ia
z
)
1939
o
u
tp
er
f
o
r
m
in
g
L
SA
in
ter
m
s
o
f
p
r
o
d
u
cin
g
a
m
o
r
e
b
ala
n
ce
d
d
is
tr
ib
u
tio
n
o
f
to
p
ics.
T
h
e
r
esear
ch
em
p
h
asizes
th
e
im
p
o
r
tan
ce
o
f
ch
o
o
s
in
g
an
ap
p
r
o
p
r
iate
n
u
m
b
er
o
f
to
p
ics
b
ased
o
n
d
ataset
ch
ar
ac
ter
is
tics
,
p
r
o
v
id
in
g
v
alu
ab
le
in
s
ig
h
ts
in
to
to
p
ic
s
elec
tio
n
m
eth
o
d
o
l
o
g
ies
b
ey
o
n
d
tr
ial
-
a
n
d
-
e
r
r
o
r
ap
p
r
o
ac
h
e
s
[
3
8
]
.
T
h
is
th
esis
p
r
o
ject
f
o
c
u
s
es
o
n
h
ar
n
ess
in
g
to
p
ic
m
o
d
elin
g
tech
n
iq
u
es
w
ith
in
th
e
r
ea
lm
o
f
n
atu
r
al
lan
g
u
ag
e
p
r
o
ce
s
s
in
g
to
ex
tr
ac
t v
alu
ab
le
in
s
ig
h
ts
f
r
o
m
cu
s
to
m
er
ca
lls
to
a
E
u
r
o
p
ea
n
f
in
an
cial
s
er
v
ice
p
r
o
v
i
d
er
.
I
t a
i
m
s
to
co
m
p
ar
e
two
p
r
o
m
in
e
n
t
to
p
ic
m
o
d
elin
g
alg
o
r
ith
m
s
,
L
DA
an
d
B
E
R
T
o
p
ic,
to
ca
teg
o
r
ize
an
d
an
aly
ze
ca
ll
co
n
ten
t,
u
ltima
tely
en
h
an
cin
g
th
e
co
m
p
an
y
’
s
u
n
d
er
s
tan
d
in
g
o
f
cu
s
to
m
er
n
ee
d
s
an
d
f
ac
ilit
atin
g
m
o
r
e
ef
f
ec
tiv
e
d
ec
is
io
n
-
m
ak
in
g
.
L
DA
o
u
tp
er
f
o
r
m
s
B
E
R
T
o
p
ic
in
ter
m
s
o
f
to
p
ic
q
u
ality
an
d
in
ter
p
r
etab
ilit
y
,
in
d
icatin
g
its
s
u
p
er
io
r
ab
ilit
y
to
ca
p
tu
r
e
m
ea
n
in
g
f
u
l
an
d
co
h
er
en
t
to
p
ics
with
in
th
e
cu
s
to
m
e
r
ca
ll
d
ata.
T
h
e
r
esu
lts
d
em
o
n
s
tr
ate
th
e
p
o
ten
tial
f
o
r
im
p
r
o
v
in
g
cu
s
to
m
er
en
g
a
g
em
en
t,
s
atis
f
ac
tio
n
,
an
d
tailo
r
e
d
s
tr
ateg
ies f
o
r
th
e
c
o
m
p
an
y
’
s
b
en
ef
it
[
3
9
]
.
W
o
r
d
em
b
ed
d
in
g
is
ef
f
ec
tiv
e
in
id
en
tify
in
g
s
y
n
tactica
lly
an
d
s
em
an
tically
co
n
n
ec
t
ed
wo
r
d
s
,
ac
co
r
d
in
g
to
r
esear
c
h
,
an
d
t
h
ey
ca
n
b
e
u
s
ed
to
co
n
n
ec
t
r
elate
d
wo
r
d
s
in
a
co
r
p
u
s
[
4
0
]
.
Dete
r
m
in
in
g
s
em
an
tically
r
elev
an
t
co
n
ten
t
in
s
h
o
r
t
tex
ts
ca
n
b
e
ch
a
llen
g
in
g
s
in
ce
th
ey
o
f
ten
h
av
e
less
wo
r
d
co
-
o
cc
u
r
r
e
n
ce
s
.
No
n
-
n
e
g
ativ
e
m
atr
ix
f
ac
to
r
izatio
n
,
wh
ich
is
b
ased
o
n
p
r
e
-
tr
ai
n
ed
d
is
tr
ib
u
tio
n
al
v
ec
to
r
r
ep
r
esen
tatio
n
,
is
u
s
ed
to
s
o
lv
e
th
is
is
s
u
e
[
4
1
]
.
Patil
et
a
l.
[
4
2
]
d
is
cu
s
s
ed
th
e
u
s
e
o
f
to
p
ic
m
o
d
elin
g
,
s
p
ec
if
ically
L
DA
as
an
u
n
s
u
p
er
v
is
ed
m
ac
h
in
e
lear
n
in
g
tec
h
n
iq
u
e
f
o
r
d
is
co
v
e
r
in
g
to
p
ics
with
in
a
co
llectio
n
o
f
d
o
cu
m
e
n
ts
.
T
h
ey
em
p
h
asize
th
at
wh
ile
L
DA
is
ty
p
ically
ap
p
lied
to
id
en
tif
y
to
p
ics ac
r
o
s
s
m
u
ltip
le
d
o
cu
m
en
ts
,
it
ca
n
also
b
e
v
alu
ab
le
f
o
r
s
in
g
le
-
d
o
cu
m
en
t
a
p
p
licatio
n
s
,
w
h
er
e
it
ef
f
ec
tiv
ely
ex
tr
ac
ts
k
ey
wo
r
d
s
s
u
m
m
ar
izin
g
th
e
co
r
e
id
ea
o
f
th
e
d
o
cu
m
e
n
t
co
n
cisely
.
T
h
is
ap
p
r
o
ac
h
is
p
ar
ticu
lar
ly
u
s
ef
u
l
f
o
r
tex
t
s
u
m
m
ar
izatio
n
,
wh
er
e
th
e
g
o
al
is
to
s
h
o
r
ten
a
tex
t
d
o
cu
m
en
t
wh
ile
r
etain
in
g
its
ess
en
tial
p
o
in
ts
.
T
h
e
au
th
o
r
s
p
r
esen
t
an
L
DA
m
o
d
el
th
at
id
en
tifie
s
d
o
m
in
an
t
to
p
ics
with
in
a
tex
t
d
o
cu
m
en
t
an
d
s
elec
ts
s
en
ten
ce
s
r
ef
lectin
g
th
ese
to
p
ics
to
cr
ea
te
a
h
u
m
an
-
r
ea
d
ab
le
s
u
m
m
ar
y
.
Acc
o
r
d
in
g
to
Sh
i
et
a
l.
[
4
3
]
,
a
n
eu
r
al
wo
r
d
em
b
e
d
d
in
g
is
em
p
lo
y
ed
t
o
d
eter
m
in
e
th
e
co
n
tex
t
o
f
wo
r
d
s
in
o
r
d
er
to
i
d
en
tify
th
e
th
em
es
o
f
s
h
o
r
t
tex
ts
.
Similar
to
th
is
,
W
an
g
et
a
l.
[
4
4
]
em
p
lo
y
wo
r
d
em
b
ed
d
in
g
’
s
to
id
en
tif
y
ad
d
iti
o
n
al
in
f
o
r
m
atio
n
f
o
r
in
teg
r
atin
g
s
h
o
r
t
tex
t
d
o
cu
m
e
n
ts
in
to
s
ec
tio
n
s
o
f
ad
jace
n
t
wo
r
d
s
.
W
o
r
d
em
b
ed
d
in
g
ar
e
cr
u
cial
f
o
r
to
p
ic
m
o
d
ellin
g
to
lo
ca
te
th
e
tar
g
et
ter
m
’
s
im
m
ed
iate
co
n
tex
t
[
4
5
]
.
An
alter
n
ate
m
eth
o
d
o
f
to
p
ic
m
o
d
ellin
g
with
less
r
u
n
-
ti
m
e
co
m
p
le
x
ity
is
o
f
f
e
r
ed
b
y
p
r
e
-
t
r
ain
ed
w
o
r
d
em
b
ed
d
in
g
in
co
m
b
in
atio
n
with
k
-
m
ea
n
s
clu
s
ter
in
g
an
d
T
F
-
b
ased
r
e
-
r
a
n
k
in
g
[
4
6
]
.
Hasan
et
a
l.
[
4
7
]
f
o
c
u
s
ed
o
n
to
p
ic
m
o
d
elin
g
,
a
s
tatis
tic
al
d
ata
m
in
in
g
tech
n
iq
u
e
u
s
e
d
to
ca
teg
o
r
ize
d
iv
er
s
e
d
o
c
u
m
en
ts
in
to
co
h
er
e
n
t
to
p
ics.
T
h
e
s
tu
d
y
ev
alu
ates
th
e
ef
f
ec
tiv
en
ess
o
f
two
p
r
o
m
in
en
t
to
p
ic
m
o
d
elin
g
m
eth
o
d
s
wh
en
ap
p
lied
to
B
an
g
la
n
ews
ar
ticles,
wi
th
th
e
g
o
al
o
f
au
to
m
atin
g
d
o
c
u
m
en
t
ca
teg
o
r
izatio
n
f
o
r
r
ec
o
m
m
en
d
atio
n
s
y
s
tem
s
an
d
s
ea
r
ch
ap
p
licatio
n
s
.
T
h
e
m
et
h
o
d
s
u
n
d
er
e
x
am
in
atio
n
a
r
e
L
DA
an
d
L
DA2
v
ec
,
wh
ich
c
o
m
b
in
es
L
DA
with
W
o
r
d
2
Vec
.
Desp
ite
th
e
in
h
er
en
t
d
if
f
er
e
n
ce
s
in
s
y
n
tax
b
etwe
en
B
an
g
la
an
d
E
n
g
lis
h
,
L
DA
d
em
o
n
s
tr
ates
it
s
ef
f
icac
y
in
e
x
tr
ac
tin
g
m
e
an
in
g
f
u
l
to
p
ics
f
r
o
m
B
an
g
la
n
ew
s
d
o
cu
m
en
ts
.
T
o
f
ac
ilit
ate
test
in
g
an
d
p
r
ac
tical
im
p
lem
en
tatio
n
,
a
n
o
v
el
tech
n
iq
u
e
is
d
ev
is
ed
f
o
r
id
en
tify
i
n
g
to
p
ics
in
n
o
n
-
f
ac
to
r
ized
d
o
c
u
m
en
ts
b
y
L
DA
a
n
d
L
DA2
v
ec
.
R
em
ar
k
a
b
ly
,
t
h
e
r
esu
lts
r
ev
ea
l
th
at
L
DA2
v
ec
o
u
tp
e
r
f
o
r
m
s
L
DA,
ac
h
ie
v
in
g
a
n
im
p
r
ess
iv
e
ac
cu
r
ac
y
r
ate
o
f
8
5
.
6
6
%
in
id
en
tify
in
g
to
p
ics
with
in
test
d
o
cu
m
e
n
ts
,
co
m
p
ar
ed
t
o
L
DA
’
s
ac
cu
r
ac
y
o
f
6
2
.
4
5
%.
As
s
u
ch
,
L
DA2
v
ec
e
m
er
g
es
as
th
e
m
o
r
e
r
eliab
le
c
h
o
ice
f
o
r
r
ea
l
-
wo
r
l
d
a
p
p
licatio
n
s
.
Qi
an
d
He
[
4
8
]
,
p
r
esen
ted
an
o
f
f
-
to
p
ic
d
etec
tio
n
alg
o
r
ith
m
t
h
at
en
h
a
n
ce
s
a
cc
u
r
ac
y
a
n
d
e
f
f
icien
cy
in
E
n
g
lis
h
co
m
p
o
s
itio
n
-
ass
is
ted
r
ev
iew
s
y
s
tem
s
.
I
t
co
m
b
in
es
L
DA
an
d
W
o
r
d
2
Vec
to
m
o
d
el
an
d
tr
ain
d
o
cu
m
en
ts
,
lev
er
ag
in
g
s
em
an
tic
r
elatio
n
s
h
ip
s
b
etwe
e
n
d
o
cu
m
en
t
to
p
ics
a
n
d
w
o
r
d
s
to
ca
lcu
late
p
r
o
b
ab
ilit
y
-
weig
h
ted
s
u
m
s
f
o
r
ea
ch
to
p
ic
an
d
its
f
ea
tu
r
e
wo
r
d
s
.
T
h
e
alg
o
r
ith
m
id
e
n
tifie
s
o
f
f
-
to
p
ic
c
o
m
p
o
s
itio
n
s
th
r
o
u
g
h
a
s
et
th
r
esh
o
ld
,
o
u
tp
er
f
o
r
m
in
g
th
e
v
ec
to
r
s
p
a
ce
m
o
d
el
in
ex
p
er
im
en
tal
r
es
u
lts
,
ac
h
iev
in
g
h
ig
h
er
ac
cu
r
a
cy
an
d
an
F
v
alu
e
ex
ce
ed
in
g
8
8
%.
T
h
is
alg
o
r
it
h
m
h
o
ld
s
p
r
o
m
is
e
f
o
r
in
tellig
en
t
o
f
f
-
to
p
ic
d
etec
tio
n
,
esp
ec
ially
in
E
n
g
lis
h
co
m
p
o
s
itio
n
teac
h
in
g
co
n
tex
ts
.
Gr
o
o
ten
d
o
r
s
t
in
s
tu
d
y
[
4
9
]
d
is
cu
s
s
ed
th
e
B
E
R
T
o
p
ic,
a
to
p
ic
m
o
d
elin
g
ap
p
r
o
ac
h
th
at
g
o
e
s
b
ey
o
n
d
tr
ad
itio
n
al
m
eth
o
d
s
b
y
lev
er
a
g
in
g
s
tate
-
of
-
th
e
-
ar
t
lan
g
u
ag
e
m
o
d
els
an
d
in
tr
o
d
u
cin
g
a
cl
ass
-
b
ased
T
F
-
I
DF
p
r
o
ce
d
u
r
e
f
o
r
cr
ea
tin
g
c
o
h
er
e
n
t
to
p
ic
r
ep
r
esen
tatio
n
s
.
B
E
R
T
o
p
ic
o
p
er
ates
in
s
ev
er
al
s
tep
s
:
it
f
ir
s
t
g
en
er
ates
d
o
cu
m
e
n
t
em
b
ed
d
in
g
u
s
in
g
p
r
e
-
tr
ain
ed
tr
a
n
s
f
o
r
m
e
r
-
b
ased
l
an
g
u
ag
e
m
o
d
els,
th
e
n
clu
s
ter
s
th
ese
em
b
ed
d
in
g
,
an
d
f
in
ally
p
r
o
d
u
ce
s
to
p
ic
r
e
p
r
esen
tatio
n
s
u
s
in
g
th
e
class
-
b
ased
T
F
-
I
DF
ap
p
r
o
ac
h
.
T
h
is
s
ep
ar
atio
n
o
f
task
s
p
r
o
v
id
es
f
lex
ib
ilit
y
an
d
ea
s
e
o
f
u
s
e
in
t
h
e
m
o
d
el.
T
h
e
p
a
p
er
p
r
esen
ts
a
co
m
p
r
eh
en
s
iv
e
an
al
y
s
is
o
f
B
E
R
T
o
p
ic,
in
clu
d
in
g
ev
alu
atio
n
s
with
cl
ass
ical
to
p
ic
co
h
er
e
n
ce
m
ea
s
u
r
es
an
d
co
n
s
id
er
atio
n
s
o
f
r
u
n
n
in
g
tim
es.
T
h
e
r
esu
lts
o
f
ex
p
er
im
en
ts
in
d
i
ca
te
th
at
B
E
R
T
o
p
ic
ef
f
ec
ti
v
ely
ca
p
tu
r
es
co
h
e
r
en
t
lan
g
u
ag
e
p
atter
n
s
an
d
co
n
s
is
ten
tly
p
er
f
o
r
m
s
well
ac
r
o
s
s
v
ar
io
u
s
task
s
,
m
ak
in
g
it
a
co
m
p
etitiv
e
an
d
s
tab
le
o
p
ti
o
n
in
th
e
f
ield
o
f
to
p
ic
m
o
d
elin
g
.
C
y
g
an
[
5
0
]
h
ig
h
li
g
h
ted
th
e
s
ig
n
if
ican
ce
o
f
b
r
o
wsi
n
g
h
is
to
r
y
,
wh
ich
o
f
ten
c
o
n
tain
s
a
wea
lth
o
f
in
tellectu
al
ex
p
lo
r
atio
n
r
ec
o
r
d
ed
as
UR
L
s
,
titl
e
s
,
an
d
tim
estam
p
s
.
Ho
wev
er
,
th
e
tab
u
lar
f
o
r
m
at
o
f
th
is
d
ata
m
ak
es
it
ch
allen
g
in
g
t
o
ex
p
lo
r
e
an
d
e
x
tr
ac
t
m
ea
n
i
n
g
f
u
l
in
s
ig
h
ts
.
T
h
e
au
th
o
r
em
p
lo
y
s
Sen
ten
ce
-
B
E
R
T
(
SB
E
R
T
)
to
cr
ea
te
ex
p
r
ess
iv
e
em
b
ed
d
in
g
’
s
th
at
f
ac
ilit
ate
to
p
ic
m
o
d
elin
g
with
in
b
r
o
wsi
n
g
h
is
to
r
y
d
ata.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
2
,
Ap
r
il
20
25
:
1
9
3
3
-
1
9
4
8
1940
Qu
alitativ
e
an
aly
s
is
r
ev
ea
ls
th
at
to
p
ic
clu
s
ter
s
g
e
n
er
ated
f
r
o
m
Sen
ten
ce
-
B
E
R
T
(
SB
E
R
T
)
web
p
ag
e
em
b
ed
d
in
g
s
o
u
tp
er
f
o
r
m
th
o
s
e
cr
ea
ted
u
s
in
g
Do
c2
Vec
-
b
as
ed
d
o
cu
m
e
n
t
em
b
ed
d
i
n
g
.
De
s
p
ite
a
lo
wer
to
p
ic
co
h
er
en
ce
s
co
r
e,
SB
E
R
T
-
b
ased
em
b
ed
d
in
g
s
ex
ce
l
in
o
r
g
a
n
izin
g
a
d
iv
e
r
s
e
r
an
g
e
o
f
d
o
c
u
m
en
ts
with
o
u
t
th
e
n
ee
d
f
o
r
p
r
io
r
tr
ain
in
g
,
u
n
lik
e
Do
c2
Vec
,
wh
ich
r
eq
u
ir
es
tr
a
in
in
g
o
n
t
h
e
in
p
u
t
d
ataset.
SB
E
R
T
em
b
ed
d
in
g
’
s
ca
p
tu
r
e
s
em
an
tic
co
n
ce
p
ts
ef
f
ec
tiv
ely
,
wh
ile
Do
c2
Vec
ten
d
s
to
p
r
io
r
itize
s
tr
u
ctu
r
al
s
im
il
ar
ity
o
v
er
s
em
an
tic
m
ea
n
in
g
.
Ov
er
all,
t
h
is
s
u
g
g
es
ts
th
at
tr
ain
in
g
o
n
th
e
i
n
p
u
t
d
ataset
f
o
r
d
o
cu
m
e
n
t
to
p
ic
m
o
d
elin
g
m
a
y
n
o
t
b
e
id
ea
l,
as
it
ca
n
lead
t
o
an
o
v
er
em
p
h
asis
o
n
s
tr
u
ct
u
r
al
c
o
m
p
o
s
itio
n
r
ath
er
t
h
an
s
em
a
n
tic
co
n
ten
t.
I
n
Gr
o
o
t
et
al
.
[
5
1
]
,
d
is
cu
s
s
ed
th
e
ap
p
licatio
n
o
f
to
p
ic
m
o
d
elin
g
,
p
ar
ticu
lar
l
y
laten
t
Dir
ich
let
allo
ca
tio
n
,
a
n
d
th
e
c
h
allen
g
es
it
f
ac
es
wh
en
d
ea
lin
g
with
s
h
o
r
t
tex
ts
f
r
o
m
v
ar
io
u
s
d
o
m
ain
s
.
T
h
ey
e
x
p
lo
r
e
th
e
p
e
r
f
o
r
m
an
c
e
o
f
th
e
B
E
R
T
o
p
ic
alg
o
r
ith
m
,
a
s
tate
-
of
-
t
h
e
-
ar
t
ap
p
r
o
ac
h
,
o
n
s
h
o
r
t
m
u
lti
-
d
o
m
ain
tex
ts
an
d
f
in
d
th
at
it
o
u
tp
er
f
o
r
m
s
laten
t
Dir
ich
let
allo
ca
tio
n
in
ter
m
s
o
f
to
p
ic
co
h
er
en
ce
a
n
d
d
i
v
er
s
ity
.
Ad
d
itio
n
ally
,
th
ey
a
n
aly
ze
th
e
p
er
f
o
r
m
an
ce
o
f
th
e
HDBS
C
AN
clu
s
ter
in
g
a
lg
o
r
ith
m
u
s
ed
b
y
B
E
R
T
o
p
ic
an
d
id
en
tify
a
n
is
s
u
e
wh
er
e
it
ca
teg
o
r
izes
a
s
ig
n
if
ican
t
p
o
r
tio
n
o
f
th
e
d
o
cu
m
en
ts
as
o
u
tlier
s
,
wh
ich
h
in
d
e
r
s
f
u
r
th
er
an
aly
s
is
.
T
o
ad
d
r
ess
th
is
p
r
o
b
lem
,
th
ey
r
ep
lace
HDBS
C
A
N
with
k
-
m
ea
n
s
clu
s
ter
in
g
.
L
iu
et
a
l.
[
5
2
]
co
n
d
u
cted
a
s
tu
d
y
f
o
cu
s
in
g
o
n
th
e
u
tili
za
tio
n
o
f
s
u
m
m
ar
y
g
e
n
er
atio
n
an
d
to
p
ic
m
o
d
elin
g
to
d
is
ce
r
n
f
ac
to
r
s
in
f
lu
en
cin
g
v
ac
cin
e
attitu
d
es
ac
r
o
s
s
r
eg
io
n
s
.
T
h
e
y
g
ath
er
e
d
twee
ts
co
n
ce
r
n
in
g
Sin
o
v
ac
,
Ast
r
aZ
en
ec
a,
a
n
d
Pf
izer
v
ac
cin
es,
ap
p
ly
in
g
B
E
R
T
o
p
ic
clu
s
ter
in
g
t
o
ca
teg
o
r
ize
twee
ts
in
to
to
p
ics.
Ad
d
itio
n
ally
,
th
e
y
u
tili
ze
d
co
n
tr
asti
v
e
lear
n
in
g
(
C
L
)
to
p
r
o
d
u
ce
s
u
m
m
ar
ies
f
o
r
ea
ch
to
p
i
c.
T
h
e
s
tu
d
y
f
o
u
n
d
th
at
B
E
R
T
o
p
ic
o
u
tp
e
r
f
o
r
m
s
laten
t
Dir
ich
let
allo
ca
tio
n
an
d
id
e
n
tifie
d
th
r
ee
c
o
n
s
is
ten
t
f
ac
to
r
s
a
f
f
ec
tin
g
v
ac
cin
e
attitu
d
es:
v
ac
cin
e
-
r
ela
ted
f
ac
to
r
s
,
h
ea
lth
s
y
s
tem
-
r
ela
ted
f
ac
to
r
s
,
an
d
in
d
iv
i
d
u
al
s
o
cial
attr
ib
u
tes.
T
h
e
r
esear
ch
co
n
clu
d
es
th
at
d
ee
p
lear
n
in
g
m
eth
o
d
s
e
f
f
ec
tiv
ely
r
ev
ea
l
th
ese
f
ac
t
o
r
s
,
ass
is
tin
g
p
o
licy
m
a
k
er
s
an
d
m
ed
ical
in
s
titu
tio
n
s
in
f
o
r
m
u
l
atin
g
m
o
r
e
ef
f
icien
t
v
ac
ci
n
atio
n
s
tr
ateg
ies.
Z
a
n
k
ad
i
et
al
.
[
5
3
]
d
is
cu
s
s
ed
th
e
cr
u
cial
r
o
le
o
f
lear
n
er
s
’
in
ter
ests
in
th
e
r
ea
lm
o
f
e
d
u
ca
tio
n
,
s
p
ec
if
ically
f
o
cu
s
in
g
o
n
m
ass
iv
e
o
p
en
o
n
lin
e
co
u
r
s
es
(
MO
OC
s
)
.
I
t
h
ig
h
lig
h
ts
th
e
s
ig
n
if
ican
t
p
o
ten
tial
o
f
n
u
r
tu
r
in
g
an
d
alig
n
in
g
le
ar
n
er
s
’
in
ter
ests
to
en
h
an
ce
th
eir
o
v
e
r
all
MO
OC
lear
n
in
g
ex
p
er
ien
c
es.
T
h
e
s
tu
d
y
d
elv
es
in
to
th
e
co
n
ce
p
t
th
at
lear
n
er
s
o
f
ten
r
ev
ea
l
th
eir
g
e
n
u
in
e
in
ter
ests
an
d
p
r
e
f
er
en
ce
s
o
n
s
o
cial
m
ed
ia
p
latf
o
r
m
s
th
r
o
u
g
h
u
s
e
r
-
g
en
er
ate
d
co
n
te
n
t,
wh
ich
co
n
tain
s
v
alu
a
b
le
h
id
d
en
in
s
ig
h
ts
.
T
h
e
p
r
im
a
r
y
r
es
ea
r
ch
o
b
jectiv
e
is
to
id
e
n
tify
an
d
ex
tr
ac
t
th
ese
to
p
ical
in
ter
ests
f
r
o
m
lear
n
er
s
’
s
o
cial
m
ed
ia
p
o
s
ts
,
th
er
eb
y
en
r
ich
in
g
th
eir
co
u
r
s
e
p
r
ef
er
en
ce
s
with
in
th
e
MO
OC
co
n
tex
t.
T
o
ac
h
iev
e
th
is
,
th
e
s
tu
d
y
em
p
lo
y
s
NL
P
t
ec
h
n
iq
u
es
an
d
lev
er
a
g
es
v
ar
i
o
u
s
to
p
ic
m
o
d
elin
g
m
eth
o
d
s
,
in
clu
d
in
g
laten
t
Dir
ich
let
allo
ca
tio
n
,
laten
t
s
em
an
tic
an
aly
s
is
,
an
d
B
E
R
T
o
p
i
c.
R
em
ar
k
ab
ly
,
th
e
ex
p
er
im
en
tal
f
in
d
i
n
g
s
u
n
d
e
r
s
co
r
e
th
at
B
E
R
T
o
p
ic
o
u
tp
er
f
o
r
m
ed
th
e
o
th
er
m
o
d
els
wh
en
a
p
p
lied
to
th
e
d
ataset
co
llected
f
o
r
t
h
is
s
tu
d
y
.
4.
M
E
T
H
O
D
T
h
is
s
tu
d
y
f
o
cu
s
es
o
n
ev
alu
a
tin
g
th
e
p
er
f
o
r
m
an
ce
o
f
tr
an
s
f
o
r
m
er
-
b
ased
lan
g
u
ag
e
m
o
d
el
s
f
o
r
tex
t
s
u
m
m
ar
izatio
n
u
s
in
g
b
e
n
ch
m
ar
k
d
atasets
.
A
s
tr
u
ctu
r
ed
ap
p
r
o
ac
h
is
ad
o
p
ted
,
lev
e
r
ag
in
g
th
e
T
wen
t
y
New
s
g
r
o
u
p
s
d
ataset
an
d
an
ar
ticles
d
ataset
to
en
s
u
r
e
d
iv
er
s
i
ty
an
d
r
ep
r
esen
tativ
en
ess
in
t
h
e
ev
alu
atio
n
.
T
h
e
d
atasets
wer
e
ca
r
ef
u
lly
p
r
e
p
ar
ed
an
d
a
n
aly
ze
d
t
o
f
ac
ilit
ate
m
ea
n
in
g
f
u
l c
o
m
p
a
r
is
o
n
s
an
d
in
s
ig
h
ts
.
4
.
1
.
Select
io
n o
f
da
t
a
s
et
T
wo
d
is
tin
ct
d
ataset
s
wer
e
u
t
ilized
in
th
is
s
tu
d
y
to
ev
alu
ate
th
e
p
er
f
o
r
m
an
ce
o
f
tr
an
s
f
o
r
m
er
-
b
ased
lan
g
u
ag
e
m
o
d
els
f
o
r
tex
t
s
u
m
m
ar
izatio
n
.
T
h
e
f
ir
s
t
is
th
e
T
wen
ty
New
s
g
r
o
u
p
s
d
ataset,
a
wid
el
y
u
s
ed
co
llectio
n
o
f
1
8
,
8
4
6
d
o
c
u
m
e
n
ts
ca
teg
o
r
ized
in
to
2
0
d
if
f
er
en
t
n
ewsg
r
o
u
p
s
.
T
h
is
d
ataset,
f
etch
ed
u
s
in
g
th
e
f
etch
_
2
0
n
ewsg
r
o
u
p
s
f
u
n
ctio
n
f
r
o
m
th
e
s
cik
it
-
lear
n
lib
r
a
r
y
,
p
r
o
v
id
es a
r
ich
an
d
d
iv
er
s
e
tex
t
u
al
b
ase
f
o
r
n
atu
r
al
lan
g
u
ag
e
p
r
o
ce
s
s
in
g
task
s
.
T
h
e
s
ec
o
n
d
d
ataset,
r
ef
e
r
r
ed
to
as
th
e
Ar
ticles
d
ataset,
co
m
p
r
is
es
tex
tu
al
d
at
a
ex
tr
ac
ted
f
r
o
m
t
h
e
f
ile
a
r
ticles.x
ls
x
,
s
o
u
r
ce
d
f
r
o
m
KDn
u
g
g
ets.
T
h
is
d
ataset
is
n
o
tab
le
f
o
r
its
s
tr
u
ctu
r
ed
a
n
d
co
m
p
r
eh
e
n
s
iv
e
co
n
ten
t,
o
f
f
er
in
g
a
co
m
p
lem
en
tar
y
p
er
s
p
e
ctiv
e
to
th
e
u
n
s
tr
u
ctu
r
ed
n
at
u
r
e
o
f
th
e
T
wen
ty
New
s
g
r
o
u
p
s
d
ataset.
T
o
g
et
h
er
,
th
ese
d
atasets
p
r
o
v
id
e
a
b
alan
ce
d
f
r
am
ewo
r
k
f
o
r
ev
alu
atin
g
tex
t
s
u
m
m
ar
izatio
n
m
o
d
els ac
r
o
s
s
v
ar
ied
tex
tu
al
c
o
n
tex
ts
.
4
.
1
.
1
.
T
went
y
newsg
ro
up
s
d
a
t
a
s
et
T
h
e
T
wen
ty
New
s
g
r
o
u
p
s
d
at
aset,
is
a
co
m
m
o
n
ly
u
s
ed
te
x
t
d
ataset
f
o
r
v
ar
i
o
u
s
n
atu
r
al
lan
g
u
a
g
e
p
r
o
ce
s
s
in
g
task
s
.
T
h
e
2
0
Ne
wsg
r
o
u
p
s
d
ataset
co
n
tain
s
a
t
o
tal
o
f
1
8
,
8
4
6
n
ewsg
r
o
u
p
d
o
cu
m
en
ts
,
wh
ich
ar
e
o
f
ten
r
ef
e
r
r
ed
to
as
“
ar
ticles
“
o
r
g
an
ized
as
a
co
llectio
n
o
f
2
0
d
if
f
er
e
n
t
n
ewsg
r
o
u
p
s
ca
teg
o
r
ies.
T
h
is
d
ataset
i
s
f
etch
ed
f
r
o
m
th
e
s
ci
-
k
it
-
lear
n
lib
r
ar
y
u
s
in
g
th
e
f
etch
_
2
0
n
e
wsg
r
o
u
p
s
f
u
n
ctio
n
an
d
co
n
s
is
ts
o
f
a
to
tal
o
f
2
0
n
ewsg
r
o
u
p
ca
te
g
o
r
ies,
ea
ch
with
m
u
ltip
le
ar
ticles.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
E
xp
lo
r
in
g
to
p
ic
mo
d
ellin
g
:
a
co
mp
a
r
a
tive
a
n
a
lysi
s
o
f tra
d
itio
n
a
l a
n
d
tr
a
n
s
fo
r
mer
-
b
a
s
ed
…
(
A
ye
s
h
a
R
ia
z
)
1941
4
.
1
.
2
.
Art
icles
da
t
a
s
et
I
n
th
is
s
ec
tio
n
,
we
d
escr
ib
e
th
e
p
r
o
ce
s
s
o
f
lo
a
d
in
g
‘
a
r
ticles.xlsx
,
’
wh
ich
was
o
b
tai
n
ed
f
r
o
m
KDn
u
g
g
ets
an
d
s
er
v
es
as
th
e
p
r
im
ar
y
s
o
u
r
ce
o
f
in
f
o
r
m
atio
n
f
o
r
o
u
r
r
esear
ch
.
T
h
is
d
ataset
is
ch
ar
ac
ter
ized
b
y
its
s
ize
an
d
s
tr
u
ctu
r
e,
wh
ich
a
r
e
k
ey
f
ac
to
r
s
in
u
n
d
er
s
tan
d
in
g
th
e
s
co
p
e
o
f
th
e
d
ata.
T
h
e
d
ataset
ex
h
ib
its
th
e
f
o
llo
win
g
ch
a
r
ac
ter
is
tics
.
I
n
T
a
b
le
1
,
ea
ch
r
o
w
co
r
r
esp
o
n
d
s
to
a
u
n
iq
u
e
r
esear
ch
ar
ticle,
an
d
th
e
co
lu
m
n
s
p
r
o
v
id
e
in
f
o
r
m
atio
n
ab
o
u
t
th
e
n
u
m
er
ical
id
en
tifie
r
,
titl
e,
jo
u
r
n
al,
p
u
b
licatio
n
y
ea
r
,
an
d
ab
s
tr
ac
t
o
f
ea
ch
ar
ti
cle.
T
h
is
s
tr
u
ctu
r
ed
r
ep
r
esen
tatio
n
f
ac
ilit
ates
ea
s
y
r
etr
iev
al
an
d
an
aly
s
is
o
f
th
e
ar
ticles
f
o
r
s
u
m
m
ar
izatio
n
task
s
.
B
y
o
r
g
a
n
izin
g
th
e
d
ata
in
a
tab
u
la
r
f
o
r
m
at,
th
e
d
ataset
en
s
u
r
es
clar
ity
an
d
c
o
n
s
is
ten
cy
,
en
ab
lin
g
ef
f
icien
t
p
r
ep
r
o
ce
s
s
in
g
an
d
f
ea
tu
r
e
ex
tr
ac
tio
n
.
T
ab
le
1
.
C
h
ar
ac
ter
is
tics
o
f
ar
ti
cles.x
ls
x
U
n
n
a
me
d
:
0
Ti
t
l
e
Jo
u
r
n
a
l
Y
e
a
r
A
b
st
r
a
c
t
1
[
Ti
t
l
e
1
]
[
Jo
u
r
n
a
l
1
]
2
0
x
x
[
A
b
st
r
a
c
t
1
]
2
[
Ti
t
l
e
2
]
[
Jo
u
r
n
a
l
2
]
2
0
x
x
[
A
b
st
r
a
c
t
2
]
3
[
Ti
t
l
e
3
]
[
Jo
u
r
n
a
l
3
]
2
0
x
x
[
A
b
st
r
a
c
t
3
]
….
….
….
.
….
.
….
.
6
0
0
0
[
Ti
t
l
e
6
0
0
0
]
[
Jo
u
r
n
a
l
6
0
0
0
]
2
0
x
x
[
A
b
st
r
a
c
t
6
0
0
0
]
4
.
2
.
T
ex
t
prepro
ce
s
s
i
ng
Af
ter
th
e
s
elec
tio
n
o
f
th
e
d
at
aset,
th
e
f
o
llo
win
g
tex
t
p
r
ep
r
o
ce
s
s
in
g
s
tep
s
ar
e
p
er
f
o
r
m
ed
to
r
em
o
v
e
n
o
is
e,
s
to
p
wo
r
d
s
,
a
n
d
r
ed
u
n
d
an
cies
f
r
o
m
th
e
d
ataset.
T
h
ese
p
r
ep
r
o
ce
s
s
in
g
s
tep
s
en
s
u
r
e
th
at
th
e
tex
t
is
clea
n
an
d
s
tan
d
ar
d
ized
,
im
p
r
o
v
i
n
g
th
e
q
u
ality
o
f
in
p
u
t
f
o
r
th
e
s
u
m
m
ar
izatio
n
m
o
d
els.
B
y
eli
m
in
atin
g
ir
r
elev
a
n
t
elem
en
ts
,
th
e
p
r
o
ce
s
s
en
h
a
n
c
es
th
e
f
o
c
u
s
o
n
m
ea
n
i
n
g
f
u
l
c
o
n
ten
t,
c
o
n
tr
ib
u
tin
g
t
o
m
o
r
e
a
cc
u
r
ate
an
d
co
n
cise
s
u
m
m
ar
ies.
4
.
2
.
1
.
Da
t
a
prepro
ce
s
s
ing
f
o
r
L
DA
I
n
th
e
p
r
ep
r
o
ce
s
s
in
g
wo
r
k
f
l
o
w
f
o
r
L
DA,
we
f
o
llo
w
a
s
er
ies
o
f
s
tep
s
to
p
r
e
p
ar
e
th
e
te
x
t
d
a
ta
f
o
r
to
p
ic
m
o
d
elin
g
.
First,
we
lev
er
ag
e
th
e
s
p
aCy
lib
r
ar
y
with
th
e
‘
e
n
_
co
r
e_
w
eb
_
s
m
’
m
o
d
el,
d
is
a
b
lin
g
th
e
p
ar
s
er
an
d
n
am
ed
en
tity
r
ec
o
g
n
izer
f
o
r
e
f
f
icien
cy
.
Ad
d
itio
n
ally
,
we
u
ti
lize
n
atu
r
al
lan
g
u
a
g
e
t
o
o
lk
it
(
NL
T
K
)
s
to
p
w
o
r
d
s
lis
t.
T
h
e
p
r
im
ar
y
p
r
ep
r
o
ce
s
s
in
g
s
tep
s
ar
e
as f
o
llo
ws:
a.
T
ex
t
clea
n
in
g
:
W
e
b
eg
in
b
y
r
em
o
v
in
g
s
to
p
wo
r
d
s
an
d
s
h
o
r
t
to
k
en
s
,
wh
ile
also
elim
in
ati
n
g
letter
ac
ce
n
ts
f
r
o
m
th
e
tex
t d
ata.
b.
Bi
-
g
r
am
an
d
tr
i
-
g
r
am
i
m
p
lem
en
tatio
n
:
Nex
t,
we
b
u
ild
b
i
-
g
r
am
s
an
d
t
r
i
-
g
r
am
s
to
ca
p
tu
r
e
m
ea
n
in
g
f
u
l
wo
r
d
co
m
b
in
atio
n
s
in
t
h
e
tex
t.
c.
L
em
m
atiza
tio
n
:
W
e
th
en
ap
p
ly
lem
m
atiza
tio
n
to
th
e
to
k
e
n
s
an
d
f
ilter
f
o
r
s
p
ec
if
ic
p
ar
t
-
of
-
s
p
ee
ch
tag
s
(
NOUN
,
ADJ,
VE
R
B
,
ADV)
.
d.
Fin
al
clea
n
u
p
:
Af
ter
lem
m
atiz
atio
n
,
we
r
em
o
v
e
s
to
p
wo
r
d
s
an
d
s
h
o
r
t
to
k
en
s
o
n
ce
m
o
r
e,
e
n
s
u
r
in
g
th
at
th
e
p
r
o
ce
s
s
ed
tex
t d
ata
is
clea
n
an
d
r
ea
d
y
f
o
r
to
p
ic
m
o
d
elin
g
.
4
.
2
.
2
.
P
re
pro
ce
s
s
ing
f
o
r
B
E
RT
o
pic
T
h
e
f
o
llo
win
g
p
r
e
p
r
o
ce
s
s
in
g
s
tep
s
ar
e
ex
ec
u
ted
:
a.
T
ex
t
to
k
en
izatio
n
:
T
h
e
tex
t
d
a
ta
is
to
k
en
ized
,
co
n
v
er
tin
g
it
in
to
in
d
iv
id
u
al
wo
r
d
s
u
n
its
to
m
ak
e
in
p
u
t
d
ata
s
u
itab
le
f
o
r
B
E
R
T
o
p
ic.
L
ib
r
a
r
ies lik
e
NL
T
K
an
d
s
p
aCy
ar
e
co
m
m
o
n
l
y
em
p
lo
y
ed
f
o
r
th
is
p
u
r
p
o
s
e.
b.
Sto
p
wo
r
d
s
r
em
o
v
al
:
C
o
m
m
o
n
s
to
p
wo
r
d
s
,
s
u
ch
as
‘
an
d
,
’
‘
th
e,
’
an
d
‘
in
,
’
ar
e
elim
in
ated
f
r
o
m
th
e
tex
t
wh
ich
r
ed
u
ce
s
n
o
is
e
an
d
f
o
cu
s
es o
n
m
o
r
e
m
ea
n
in
g
f
u
l w
o
r
d
s
.
c.
UM
AP
d
im
en
s
io
n
r
e
d
u
ctio
n
:
u
n
if
o
r
m
m
a
n
if
o
ld
ap
p
r
o
x
i
m
atio
n
an
d
p
r
o
jectio
n
(
UM
AP)
is
ap
p
lied
to
t
h
e
tex
t
d
ata.
UM
AP
is
a
d
im
e
n
s
io
n
ality
r
e
d
u
ctio
n
tech
n
i
q
u
e
t
h
at
h
elp
s
i
n
r
e
d
u
cin
g
th
e
co
m
p
lex
ity
o
f
t
h
e
d
ataset
wh
ile
p
r
eser
v
in
g
its
in
tr
in
s
ic
s
tr
u
ctu
r
e.
I
t
is
co
n
f
i
g
u
r
ed
with
s
p
ec
if
ic
p
ar
a
m
eter
s
s
u
ch
as
th
e
n
u
m
b
er
o
f
n
eig
h
b
o
r
s
,
c
o
m
p
o
n
en
ts
,
m
in
im
u
m
d
is
tan
ce
,
an
d
m
etr
ic.
T
h
is
d
im
en
s
io
n
r
e
d
u
ctio
n
aid
s
in
o
p
tim
izin
g
th
e
t
o
p
ic
m
o
d
elin
g
p
r
o
ce
s
s
.
4
.
2
.
3
.
P
re
pro
ce
s
s
ing
f
o
r
g
ene
ra
t
ed
s
um
m
a
ries
f
ro
m
G
P
T
3
T
h
e
p
r
ep
r
o
ce
s
s
in
g
o
f
s
u
m
m
ar
i
es
g
en
er
ated
f
r
o
m
GPT
p
lay
s
a
cr
u
cial
r
o
le
in
r
e
f
in
in
g
a
n
d
o
r
g
an
izin
g
th
e
tex
t d
ata
f
o
r
s
u
b
s
eq
u
e
n
t a
n
aly
s
is
.
T
h
e
f
o
llo
win
g
p
r
ep
r
o
ce
s
s
in
g
s
tep
s
ar
e
p
er
f
o
r
m
e
d
:
a.
T
o
k
en
izatio
n
:
in
itially
,
t
h
e
t
ex
t
is
to
k
e
n
ized
,
b
r
ea
k
in
g
it
d
o
wn
in
to
in
d
iv
id
u
al
wo
r
d
s
.
E
ac
h
wo
r
d
is
co
n
v
er
ted
to
lo
wer
ca
s
e
to
e
n
s
u
r
e
co
n
s
is
ten
cy
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
2
,
Ap
r
il
20
25
:
1
9
3
3
-
1
9
4
8
1942
b.
Pu
n
ctu
atio
n
an
d
n
o
n
-
al
p
h
an
u
m
er
ic
r
em
o
v
al:
p
u
n
ctu
atio
n
a
n
d
n
o
n
-
alp
h
a
n
u
m
er
ic
ch
a
r
ac
ter
s
ar
e
r
em
o
v
ed
f
r
o
m
th
e
tex
t.
T
h
is
s
tep
aid
s
in
elim
in
atin
g
n
o
is
e
an
d
s
im
p
lif
y
in
g
th
e
c
o
n
ten
t.
c.
Sto
p
wo
r
d
s
r
em
o
v
al:
co
m
m
o
n
s
to
p
wo
r
d
s
in
th
e
E
n
g
lis
h
lan
g
u
a
g
e
ar
e
r
em
o
v
ed
f
r
o
m
th
e
tex
t
w
h
ich
en
h
an
ce
s
th
e
f
o
cu
s
o
n
ess
en
tial c
o
n
ten
t.
d.
L
em
m
atiza
tio
n
:
lem
m
atiza
tio
n
is
ap
p
lied
to
th
e
wo
r
d
s
in
th
e
tex
t
to
en
s
u
r
e
u
n
if
o
r
m
ity
an
d
im
p
r
o
v
e
th
e
q
u
ality
o
f
t
h
e
tex
t.
e.
R
ec
o
n
s
tr
u
ctio
n
:
af
ter
to
k
e
n
izatio
n
,
r
em
o
v
al
o
f
p
u
n
ctu
atio
n
a
n
d
s
to
p
wo
r
d
s
,
an
d
lem
m
atiza
tio
n
,
th
e
wo
r
d
s
ar
e
r
ec
o
n
s
tr
u
cted
in
t
o
a
s
en
ten
ce
f
o
r
m
at.
C
o
m
m
as
ar
e
u
s
ed
to
s
ep
ar
ate
th
e
lem
m
atize
d
wo
r
d
s
,
p
r
o
v
id
i
n
g
a
clea
n
,
s
tr
u
ctu
r
ed
f
o
r
m
at
f
o
r
ea
ch
s
u
m
m
ar
y
.
4
.
2
.
4
.
P
re
pro
ce
s
s
ing
s
t
eps
o
f
XL
N
et
T
h
e
tex
t
p
r
e
p
r
o
ce
s
s
in
g
w
o
r
k
f
lo
w
in
v
o
l
v
es
a
s
er
ies
o
f
s
tep
s
p
er
f
o
r
m
e
d
b
e
f
o
r
e
an
d
af
te
r
g
en
er
atin
g
s
u
m
m
ar
ies to
p
r
ep
a
r
e
th
e
d
ata
f
o
r
an
al
y
s
is
an
d
en
s
u
r
e
th
e
q
u
ality
o
f
th
e
g
e
n
er
ated
o
u
tp
u
ts
.
a.
B
ef
o
r
e
g
en
er
atin
g
s
u
m
m
ar
ies
−
T
ex
t
to
k
e
n
izatio
n
:
Af
ter
lo
ad
in
g
th
e
d
ata
th
e
tex
t
d
ata
is
to
k
en
ized
in
to
wo
r
d
s
u
s
in
g
th
e
XL
Net
m
o
d
el
’
s
to
k
e
n
izer
.
−
T
ex
t
p
r
o
ce
s
s
in
g
with
th
e
m
o
d
el
:
T
h
e
co
d
e
g
en
er
ates
s
u
m
m
ar
ies
f
o
r
ea
ch
d
o
cu
m
en
t
u
s
in
g
th
e
XL
Net
m
o
d
el.
I
t to
k
en
izes th
e
in
p
u
t t
ex
t,
ad
ju
s
ts
th
e
len
g
th
,
an
d
g
e
n
er
ates th
e
s
u
m
m
ar
y
.
b.
Af
ter
g
en
er
atin
g
s
u
m
m
ar
ies
−
I
n
itializin
g
p
r
ep
r
o
ce
s
s
ed
s
u
m
m
ar
ies
lis
t:
T
h
e
“
p
r
ep
r
o
ce
s
s
ed
s
u
m
m
ar
ies
”
is
in
itialized
t
o
s
to
r
e
th
e
p
r
ep
r
o
ce
s
s
ed
v
er
s
io
n
s
o
f
th
e
g
en
er
ated
s
u
m
m
ar
ies.
−
Sto
p
wo
r
d
s
R
em
o
v
al:
T
h
e
NL
T
K
s
to
p
wo
r
d
s
ar
e
r
e
m
o
v
e
d
f
r
o
m
th
e
g
e
n
er
ated
s
u
m
m
ar
ies
to
elim
in
ate
co
m
m
o
n
wo
r
d
s
th
at
d
o
n
o
t a
d
d
s
ig
n
if
ican
t m
ea
n
i
n
g
.
−
Re
-
to
k
en
izatio
n
:
T
h
e
p
r
e
p
r
o
ce
s
s
ed
s
u
m
m
ar
ies
ar
e
to
k
en
ized
ag
ain
in
to
wo
r
d
s
u
s
i
n
g
NL
T
K
’
s
_
f
u
n
ctio
n
.
−
Pu
n
ctu
atio
n
r
em
o
v
al:
Pu
n
ct
u
atio
n
an
d
n
o
n
-
alp
h
a
n
u
m
er
ic
ch
ar
ac
ter
s
ar
e
r
em
o
v
ed
f
r
o
m
th
e
p
r
ep
r
o
ce
s
s
ed
s
u
m
m
ar
ies.
−
W
o
r
d
lem
m
atiza
tio
n
:
W
o
r
d
s
in
th
e
p
r
ep
r
o
ce
s
s
ed
s
u
m
m
ar
ie
s
ar
e
lem
m
atize
d
to
r
ed
u
ce
t
h
em
to
th
eir
b
ase
f
o
r
m
.
−
J
o
in
in
g
W
o
r
d
s
:
T
h
e
p
r
e
p
r
o
ce
s
s
ed
wo
r
d
s
ar
e
jo
i
n
ed
b
ac
k
i
n
to
a
s
en
ten
ce
with
co
m
m
as
to
cr
ea
te
a
s
in
g
le
s
tr
in
g
f
o
r
ea
c
h
p
r
ep
r
o
ce
s
s
ed
s
u
m
m
ar
y
.
T
h
e
p
r
e
p
r
o
ce
s
s
in
g
s
tep
s
b
ef
o
r
e
g
en
e
r
atin
g
s
u
m
m
ar
ies
ar
e
r
elate
d
to
g
en
er
atin
g
th
e
s
u
m
m
ar
ies
th
em
s
elv
es,
wh
ile
th
e
p
r
ep
r
o
ce
s
s
in
g
s
tep
s
af
ter
g
en
er
atin
g
s
u
m
m
ar
ies
f
o
cu
s
o
n
clea
n
in
g
an
d
r
ef
i
n
in
g
th
e
g
en
er
ate
d
s
u
m
m
ar
ies f
o
r
f
u
r
th
er
an
aly
s
is
o
r
p
r
esen
tatio
n
4
.
3
.
E
v
a
lua
t
io
n
m
et
rics
W
e
s
tu
d
y
th
e
two
m
o
d
els
q
u
a
n
titativ
ely
.
W
e
m
ea
s
u
r
e
th
e
q
u
ality
o
f
t
h
e
to
p
ics
b
ased
o
n
t
wo
m
etr
ics
u
s
ed
in
t
h
e
s
tu
d
y
[
5
4
]
to
p
ic
co
h
er
en
ce
an
d
to
p
ic
d
iv
e
r
s
ity
.
B
o
th
th
ese
m
et
r
ics
ar
e
also
u
s
ed
in
t
h
e
cu
r
r
en
t
s
tu
d
y
.
T
h
e
to
p
ic
co
h
e
r
en
ce
is
th
e
m
ea
s
u
r
e
th
at
s
h
o
ws
h
o
w
f
r
eq
u
en
tly
w
o
r
d
s
co
–
o
cc
u
r
i
n
a
d
ataset
[
5
5
]
.
T
h
e
h
ig
h
er
th
e
to
p
ic
co
h
er
en
ce
th
e
m
o
r
e
in
ter
p
r
etab
le
is
th
e
to
p
ic
m
o
d
el.
Acc
o
r
d
in
g
to
[
5
5
]
,
th
e
co
h
er
e
n
ce
o
f
a
to
p
ic
ca
n
b
e
ca
lcu
lated
as:
=
∑
.
=
2
∑
|
(
,
)
|
+
1
(
)
−
1
=
1
W
h
er
ea
s
,
,
……
is
th
e
to
p
wo
r
d
s
u
n
d
er
th
e
to
p
ic
k
s
o
r
te
d
b
y
p
r
o
b
ab
ilit
y
in
d
escen
d
in
g
o
r
d
er
.
|
(
,
)
|
is
th
e
n
u
m
b
er
o
f
d
o
c
u
m
en
ts
in
a
d
ataset
co
n
tain
in
g
b
o
th
wo
r
d
s
an
d
.
I
n
o
th
er
wo
r
d
s
,
th
e
h
ig
h
th
e
c
o
h
er
e
n
ce
s
co
r
e
th
e
h
ig
h
er
will
b
e
th
e
m
u
tu
al
in
f
o
r
m
atio
n
b
etwe
en
wo
r
d
s
.
All
th
e
m
o
d
els
wh
ich
s
h
o
w
h
i
g
h
c
o
h
er
en
ce
s
co
r
e
g
e
n
er
ate
h
i
g
h
ly
i
n
ter
p
r
et
ab
le
to
p
ics.
W
h
ile
to
p
ic
d
iv
er
s
ity
is
th
e
m
ea
s
u
r
e
th
at
s
h
o
ws
th
at
h
o
w
u
n
iq
u
e
wo
r
d
s
ar
e
in
s
id
e
th
e
to
p
ics,
th
e
v
alu
e
o
f
r
ed
u
n
d
an
t
to
p
ics
will
b
e
n
ea
r
to
0
wh
er
ea
s
th
e
v
alu
e
n
ea
r
t
o
1
i
n
d
icate
u
n
iq
u
e
to
p
ics.
T
h
e
m
ain
id
ea
b
eh
in
d
th
e
d
iv
e
r
s
ity
is
th
e
p
er
ce
n
tag
e
o
f
u
n
iq
u
e
wo
r
d
s
in
all
th
e
to
p
ics,
th
e
v
alu
e
clo
s
e
to
0
i
n
d
icate
s
m
o
r
e
r
e
d
u
n
d
an
t
to
p
ics
wh
er
ea
s
th
e
v
alu
e
cl
o
s
e
to
1
in
d
icate
v
ar
ied
to
p
ics.
T
h
e
b
asic
id
ea
b
e
h
in
d
to
p
ic
co
h
e
r
en
ce
is
th
at
th
e
wo
r
d
s
h
ap
p
e
n
in
s
am
e
d
o
c
u
m
en
t
will
h
av
e
h
ig
h
co
h
er
en
ce
s
co
r
e.
T
h
e
to
p
ic
q
u
ality
is
th
e
r
esu
lt
o
f
m
u
ltip
licatio
n
o
f
to
p
ic
c
o
h
er
en
ce
a
n
d
to
p
ic
d
iv
er
s
ity
.
Evaluation Warning : The document was created with Spire.PDF for Python.