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.
15
,
No
.
4
,
A
u
g
u
s
t 2
0
2
6
,
p
p
.
3
1
4
4
~
3
1
5
3
I
SS
N:
2
2
5
2
-
8
9
3
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijai.v
15
.i
4
.
p
p
3
1
4
4
-
3
1
5
3
3144
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
i
.
ia
esco
r
e.
co
m
So
cia
l news f
a
c
tu
a
lity v
erification
using
larg
e lang
ua
g
e mo
dels
T
ra
n Duc
Duo
ng
1
,
H
a
i H
o
a
n Do
2
1
F
a
c
u
l
t
y
o
f
I
n
f
o
r
m
a
t
i
o
n
T
e
c
h
n
o
l
o
g
y
,
P
o
st
s
a
n
d
Te
l
e
c
o
mm
u
n
i
c
a
t
i
o
n
s
I
n
st
i
t
u
t
e
o
f
T
e
c
h
n
o
l
o
g
y
,
H
a
n
o
i
,
V
i
e
t
n
a
m
2
Ec
o
n
o
mi
c
R
e
s
e
a
r
c
h
I
n
st
i
t
u
t
e
o
f
P
o
s
t
a
l
Te
c
h
n
o
l
o
g
y
,
P
o
st
s
a
n
d
Te
l
e
c
o
m
mu
n
i
c
a
t
i
o
n
s I
n
st
i
t
u
t
e
o
f
Te
c
h
n
o
l
o
g
y
,
H
a
n
o
i
,
V
i
e
t
n
a
m
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
J
u
l 5
,
2
0
2
5
R
ev
is
ed
Ma
y
1
3
,
2
0
2
6
Acc
ep
ted
J
u
n
1
9
,
2
0
2
6
S
o
c
ial
m
e
d
ia
p
latfo
rm
s
h
a
v
e
g
re
a
tl
y
a
c
c
e
lera
ted
th
e
sp
re
a
d
o
f
n
e
ws
,
b
u
t
th
is
ra
p
id
in
f
o
rm
a
ti
o
n
flo
w
a
lso
a
m
p
li
fies
th
e
risk
o
f
m
isin
fo
rm
a
ti
o
n
.
T
ra
d
it
io
n
a
l
a
u
to
m
a
ti
c
d
e
tec
ti
o
n
m
e
th
o
d
s
th
a
t
re
ly
so
lely
o
n
tex
tu
a
l
fe
a
t
u
re
s
o
ften
stru
g
g
le wit
h
n
u
a
n
c
e
d
,
e
m
e
rg
in
g
c
o
n
ten
t.
T
h
is p
a
p
e
r
p
re
se
n
t
a
n
o
v
e
l
p
ip
e
li
n
e
th
a
t
v
e
rifi
e
s
t
h
e
fa
c
tu
a
li
t
y
o
f
so
c
ial
n
e
ws
b
y
c
lu
ste
rin
g
re
late
d
a
rt
icle
s
in
to
e
v
e
n
ts
a
n
d
u
sin
g
larg
e
lan
g
u
a
g
e
m
o
d
e
ls
(LL
M
s)
t
o
e
x
trac
t
a
n
d
v
e
rify
c
laim
s
a
g
a
in
st
tru
ste
d
n
e
ws
so
u
rc
e
s.
Th
e
a
p
p
ro
a
c
h
g
r
o
u
p
s
so
c
ial
-
m
e
d
ia
p
o
sts
a
n
d
m
a
in
stre
a
m
re
p
o
rts
o
n
t
h
e
sa
m
e
e
v
e
n
t,
e
x
trac
ts
a
to
m
ic
c
laim
s
wit
h
a
m
o
d
e
l
li
k
e
G
P
T
-
4
,
a
n
d
c
h
e
c
k
s
e
a
c
h
c
l
a
im’s
tru
th
f
u
ln
e
ss
b
y
c
o
m
p
a
rin
g
it
t
o
t
h
e
c
lu
ste
r’s
re
li
a
b
le
n
e
ws
.
Th
is
p
ip
e
li
n
e
wa
s
e
v
a
lu
a
ted
o
n
a
n
e
wl
y
c
o
n
stru
c
te
d
Vie
tn
a
m
e
se
n
e
ws
d
a
tas
e
t
o
f
1
,
7
6
5
a
rti
c
les
(in
c
l
u
d
i
n
g
7
2
3
so
c
ial
-
m
e
d
ia
it
e
m
s),
m
a
n
u
a
ll
y
a
n
n
o
tatin
g
c
laim
s
a
s
tru
e
o
r
fa
lse
.
Th
e
LLM
-
b
a
se
d
m
e
th
o
d
a
c
h
iev
e
d
h
ig
h
a
c
c
u
ra
c
y
(≈8
8
.
9
%
F
1
-
sc
o
re
o
n
c
laim
v
e
rifi
c
a
ti
o
n
a
n
d
9
2
.
1
%
F1
-
sc
o
re
o
n
o
v
e
ra
ll
n
e
ws
v
e
ri
fica
ti
o
n
).
Th
e
se
re
su
lt
s
d
e
m
o
n
s
trate
th
a
t
c
a
re
fu
ll
y
p
r
o
m
p
ted
LL
M
s,
c
o
m
b
i
n
e
d
wit
h
e
v
e
n
t
-
le
v
e
l
c
lu
ste
rin
g
o
f
e
v
id
e
n
c
e
,
c
a
n
o
u
t
p
e
rfo
rm
trad
it
io
n
a
l
m
e
th
o
d
s
(e
.
g
.
,
b
id
irec
ti
o
n
a
l
e
n
c
o
d
e
r
re
p
re
se
n
tatio
n
s
fro
m
tran
sfo
rm
e
rs
(
BERT
)
-
b
a
se
d
c
las
sifiers
)
in
v
e
rif
y
in
g
n
e
ws
.
Th
e
p
a
p
e
r
d
isc
u
ss
e
s
a
d
v
a
n
tag
e
s
o
f
c
lu
ste
ri
n
g
o
v
e
r
sim
p
le
re
tri
e
v
a
l,
sc
a
lab
il
it
y
c
o
n
si
d
e
ra
ti
o
n
s
fo
r
L
LM
s,
a
n
d
p
ro
s
p
e
c
ts
fo
r
m
u
lt
i
li
n
g
u
a
l
a
n
d
k
n
o
wle
d
g
e
-
e
n
h
a
n
c
e
d
v
e
rif
ica
ti
o
n
.
K
ey
w
o
r
d
s
:
C
laim
ex
tr
ac
tio
n
C
laim
v
er
if
icatio
n
L
ar
g
e
lan
g
u
ag
e
m
o
d
els
New
s
clu
s
ter
in
g
So
cial
n
ews v
er
if
icatio
n
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
:
T
r
an
Du
c
Du
o
n
g
Facu
lty
o
f
I
n
f
o
r
m
atio
n
T
ec
h
n
o
lo
g
y
,
Po
s
ts
an
d
T
elec
o
m
m
u
n
icatio
n
s
I
n
s
titu
te
o
f
T
ec
h
n
o
lo
g
y
Km
.
1
0
Ng
u
y
en
T
r
ai,
Ha
Do
n
g
,
Ha
No
i,
Viet
Nam
E
m
ail:
d
u
cd
t@
p
tit.ed
u
.
v
n
1.
I
NT
RO
D
UCT
I
O
N
T
h
e
p
r
o
life
r
atio
n
o
f
s
o
cial
-
m
ed
ia
n
ews
h
as
d
em
o
cr
atize
d
in
f
o
r
m
atio
n
b
u
t
also
f
ac
ilit
ated
th
e
r
ap
i
d
s
p
r
ea
d
o
f
f
alse n
ews.
Misin
f
o
r
m
atio
n
o
n
lin
e
ca
n
s
way
p
u
b
lic
o
p
in
io
n
an
d
ev
en
in
cite
r
ea
l
-
wo
r
ld
h
ar
m
,
as p
ast
ev
en
ts
h
av
e
s
h
o
wn
.
T
r
a
d
itio
n
al
f
ac
t
-
ch
ec
k
i
n
g
s
ites
(
e.
g
.
,
Po
liti
Fact,
Sn
o
p
es)
r
ely
o
n
la
b
o
r
-
in
ten
s
iv
e
m
an
u
al
r
ev
iew,
wh
ich
is
to
o
s
lo
w
f
o
r
lar
g
e
-
s
ca
le
s
o
cial
f
ee
d
s
.
C
o
n
s
eq
u
en
tly
,
au
to
m
ated
m
et
h
o
d
s
ar
e
n
ee
d
ed
.
E
ar
l
y
f
ak
e
n
ews
d
etec
to
r
s
u
s
ed
co
n
ten
t
-
b
ased
cu
es
(
lin
g
u
is
tic
s
ty
le,
s
en
tim
en
t,
an
d
s
tan
ce
)
to
class
if
y
n
ews
[
1
]
–
[
4
]
.
Ho
wev
er
,
s
u
ch
m
eth
o
d
s
o
f
te
n
lack
ex
ter
n
al
g
r
o
u
n
d
in
g
a
n
d
ca
n
b
e
ea
s
ily
f
o
o
led
b
y
ad
v
er
s
ar
ial
o
r
co
n
tex
t
-
s
h
if
ted
co
n
ten
t.
I
n
r
esp
o
n
s
e,
k
n
o
wled
g
e
-
b
ased
ap
p
r
o
ac
h
es
h
av
e
b
ee
n
d
e
v
elo
p
e
d
:
th
ey
ex
tr
ac
t
ch
ec
k
-
wo
r
th
y
s
tatem
en
ts
an
d
co
m
p
ar
e
th
em
ag
ain
s
t
ev
id
en
ce
f
r
o
m
tr
u
s
ted
s
o
u
r
ce
s
o
r
k
n
o
wled
g
e
b
ase
s
.
Fo
r
ex
am
p
le,
th
e
f
ac
t
ex
tr
ac
tio
n
an
d
v
e
r
if
icatio
n
(
FEVE
R
)
b
en
ch
m
ar
k
d
e
m
o
n
s
tr
ated
a
p
ip
elin
e
o
f
claim
ex
tr
ac
tio
n
,
ev
id
e
n
ce
r
etr
iev
al,
an
d
class
if
icatio
n
[
5
]
.
T
h
ese
m
eth
o
d
s
ty
p
ically
e
m
p
lo
y
p
r
e
-
tr
ain
e
d
tr
an
s
f
o
r
m
er
s
(
e.
g
.
b
id
ir
ec
tio
n
al
en
co
d
er
r
ep
r
esen
tatio
n
s
f
r
o
m
t
r
an
s
f
o
r
m
er
s
(
B
E
R
T
)
)
o
n
f
ea
tu
r
es o
r
ev
id
e
n
ce
p
ass
ag
es
[
6
]
–
[
1
0
]
.
R
ec
en
tly
,
lar
g
e
lan
g
u
ag
e
m
o
d
els
(
L
L
Ms)
(
GPT
-
3
/4
)
h
av
e
s
h
o
wn
r
em
ar
k
a
b
le
r
ea
s
o
n
in
g
an
d
k
n
o
wled
g
e
r
etr
iev
al
ab
ilit
ies,
s
u
g
g
esti
n
g
n
ew
f
ac
t
-
c
h
ec
k
in
g
s
tr
ateg
ies
[
1
1
]
,
[
1
2
]
.
U
n
lik
e
class
if
icatio
n
-
o
n
ly
m
o
d
els,
L
L
Ms
ca
n
b
e
p
r
o
m
p
ted
to
d
ec
o
m
p
o
s
e
a
n
d
v
e
r
if
y
co
m
p
lex
claim
s
b
y
g
en
er
atin
g
ev
id
en
ce
o
r
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
S
o
cia
l n
ews
fa
ctu
a
lity v
erif
ica
tio
n
u
s
in
g
la
r
g
e
l
a
n
g
u
a
g
e
mo
d
els (
Tr
a
n
Du
c
Du
o
n
g
)
3145
ex
p
lan
atio
n
s
[
1
3
]
,
[
1
4
]
.
T
h
is
wo
r
k
in
v
esti
g
ates
s
u
ch
an
L
L
M
-
p
o
wer
ed
p
ip
elin
e
f
o
r
s
o
cia
l
n
ews
v
er
if
icatio
n
.
Sp
ec
if
ically
,
s
o
cial
-
m
ed
ia
ar
ti
cles
(
wh
ich
m
a
y
c
o
n
tain
r
u
m
o
r
s
o
r
m
is
in
f
o
r
m
atio
n
)
ar
e
co
llected
an
d
clu
s
ter
th
em
to
g
et
h
er
with
r
eliab
le
n
ews
r
ep
o
r
ts
o
n
t
h
e
s
am
e
ev
e
n
ts
.
GPT
-
4
is
th
e
n
u
s
ed
to
ex
tr
ac
t
ato
m
ic
claim
s
f
r
o
m
ea
ch
s
o
cial
p
o
s
t
an
d
v
er
i
f
y
th
ese
claim
s
ag
ain
s
t
th
e
clu
s
ter
’
s
tr
u
s
two
r
th
y
n
ews.
T
h
e
m
ain
co
n
tr
ib
u
tio
n
s
o
f
th
is
p
ap
er
ar
e:
i
)
a
n
o
v
el
c
lu
s
ter
in
g
+
L
L
M
p
ip
elin
e
f
o
r
v
er
if
y
in
g
s
o
cial
-
m
ed
ia
n
ews
b
y
cr
o
s
s
-
ch
ec
k
in
g
with
m
ain
s
tr
ea
m
s
o
u
r
ce
s
;
ii
)
d
em
o
n
s
tr
atio
n
th
at
L
L
Ms
ca
n
e
x
ce
ed
tr
ad
itio
n
al
m
o
d
els
f
o
r
t
h
is
task
:
o
u
r
GPT
-
4
b
ased
ap
p
r
o
ac
h
ac
h
iev
es
≈
8
8
.
9
%
F1
-
s
co
r
e
o
n
claim
v
er
if
icatio
n
an
d
9
2
.
1
%
F1
-
s
co
r
e
o
n
o
v
er
all
n
ews
lab
elin
g
,
s
ig
n
if
ica
n
tly
o
u
tp
er
f
o
r
m
in
g
a
B
E
R
T
-
b
ased
b
ase
lin
e
;
an
d
iii
)
i
n
s
ig
h
ts
in
t
o
r
etr
iev
al
s
tr
ateg
ies:
clu
s
ter
in
g
b
y
e
v
en
t is sh
o
wn
t
o
y
ield
m
o
r
e
r
elev
a
n
t e
v
id
e
n
c
e
th
an
n
aiv
e
“to
p
-
N”
s
ea
r
ch
a
n
d
s
ca
les b
etter
.
T
h
e
r
em
ain
d
e
r
o
f
th
is
p
a
p
er
i
s
o
r
g
an
ized
as
f
o
llo
ws.
Sectio
n
2
r
ev
iews
r
elate
d
liter
atu
r
e.
Sectio
n
3
d
escr
ib
es
th
e
p
r
o
p
o
s
ed
m
eth
o
d
o
lo
g
y
.
Sectio
n
4
p
r
esen
ts
th
e
d
ataset,
ex
p
er
im
e
n
tal
s
etu
p
,
an
d
ev
alu
atio
n
r
esu
lts
.
Fin
ally
,
s
ec
tio
n
5
co
n
c
lu
d
es th
e
p
ap
e
r
.
2.
RE
L
AT
E
D
WO
RK
2
.
1
.
T
ex
t
-
ba
s
ed
f
a
k
e
news det
ec
t
io
n
E
a
r
l
y
wo
r
k
o
n
a
u
to
m
a
t
ed
d
ec
e
p
t
i
o
n
d
et
e
c
t
i
o
n
f
o
c
u
s
ed
o
n
l
i
n
g
u
i
s
t
i
c
c
u
e
s
an
d
c
l
a
s
s
i
f
i
ca
t
i
o
n
.
F
o
r
e
x
a
m
p
l
e
,
W
a
n
g
[
1
]
in
t
r
o
d
u
c
ed
t
h
e
L
I
A
R
d
a
t
a
s
e
t
o
f
1
2
.
8
K
l
a
b
e
l
ed
s
t
a
t
e
m
en
t
s
f
r
o
m
P
o
l
it
i
F
a
c
t
a
n
d
s
h
o
w
ed
t
h
a
t
n
e
u
r
a
l
m
o
d
e
l
s
(
c
o
n
v
o
l
u
ti
o
n
a
l
n
eu
r
a
l
n
e
t
w
o
r
k
(
C
N
N
s
)
,
h
y
b
r
id
ar
c
h
i
te
c
t
u
r
e
s
)
ca
n
u
s
e
s
u
r
f
a
c
e
p
a
t
t
e
r
n
s
f
o
r
d
e
t
e
c
t
io
n
.
S
u
p
er
v
i
s
ed
m
o
d
e
l
s
l
i
k
e
B
E
R
T
h
av
e
s
i
n
ce
a
c
h
i
e
v
e
d
s
t
r
o
n
g
r
e
s
u
l
t
s
o
n
s
u
c
h
b
en
ch
m
ar
k
s
.
H
o
w
e
v
er
,
t
h
e
s
e
a
p
p
r
o
a
ch
e
s
c
l
a
s
s
i
f
y
n
e
w
s
b
a
s
e
d
s
o
l
e
ly
o
n
t
h
e
t
e
x
t,
w
i
th
o
u
t
g
r
o
u
n
d
i
n
g
i
n
r
e
a
l
-
w
o
r
l
d
e
v
i
d
en
c
e.
A
s
R
a
z
a
e
t
a
l.
[
4
]
n
o
t
e
t
h
a
t
B
E
R
T
-
l
i
k
e
m
o
d
e
l
s
g
en
e
r
a
l
ly
o
u
t
p
e
r
f
o
r
m
L
L
M
s
in
c
l
a
s
s
i
f
i
c
a
t
i
o
n
t
a
s
k
s
(
i
.
e
.
,
t
ex
t
-
b
a
s
e
d
c
l
a
s
s
i
f
ic
a
t
i
o
n
)
.
I
n
o
th
e
r
w
o
r
d
s
,
f
in
e
-
tu
n
e
d
tr
a
n
s
f
o
r
m
e
r
s
ex
c
e
l
w
h
en
a
t
r
a
in
i
n
g
s
e
t
i
s
a
v
a
i
l
ab
l
e
,
b
u
t
t
h
ey
a
r
e
l
im
i
te
d
wh
en
c
l
a
i
m
s
r
eq
u
ir
e
e
x
ter
n
a
l
co
n
t
ex
t
.
D
i
s
t
a
n
t
ly
s
u
p
e
r
v
i
s
e
d
d
a
ta
s
e
t
s
l
ik
e
N
E
L
A
-
G
T
o
r
F
a
k
ed
d
i
t
l
ab
e
l
a
r
t
i
c
l
e
s
f
r
o
m
s
o
c
ia
l
s
o
u
r
c
e
s
b
y
s
o
u
r
c
e
cr
e
d
ib
i
l
i
t
y
[
4
]
,
b
u
t
t
h
e
s
e
c
an
e
m
b
e
d
b
i
a
s
(
e.
g
.
,
l
ab
e
l
in
g
e
n
t
i
r
e
o
u
t
le
t
s
a
s
f
ak
e
)
a
n
d
m
i
s
s
n
u
an
c
e
.
2
.
2
.
E
v
idence
-
ba
s
ed
f
a
ct
che
ck
ing
T
o
m
o
r
e
r
o
b
u
s
tly
v
er
if
y
clai
m
s
,
r
esear
ch
er
s
h
av
e
d
ev
elo
p
ed
f
ac
t
-
c
h
ec
k
in
g
p
ip
elin
es
th
at
r
etr
iev
e
ev
id
en
ce
.
T
h
e
FEVE
R
d
ataset
ex
em
p
lifie
s
th
is
:
it
co
n
s
is
ts
o
f
1
8
5
K
claim
s
f
r
o
m
W
ik
ip
ed
i
a
with
g
r
o
u
n
d
-
tr
u
th
ev
id
en
ce
s
en
ten
ce
s
[
5
]
.
So
lu
ti
o
n
s
ty
p
ically
ex
tr
ac
t
a
claim
,
r
etr
iev
e
r
elev
an
t
d
o
cu
m
en
ts
(
v
ia
B
M2
5
o
r
n
e
u
r
al
s
ea
r
ch
)
,
a
n
d
ap
p
ly
class
if
ier
to
ju
d
g
e
“Su
p
p
o
r
ted
/R
ef
u
te
d
/No
tEn
o
u
g
h
I
n
f
o
”.
Po
liti
Fact
an
d
o
th
er
f
ac
t
-
ch
ec
k
e
r
s
u
s
e
s
im
ilar
p
r
o
ce
s
s
es
with
n
e
ws
o
r
k
n
o
wled
g
e
b
ases
.
Fo
r
ex
am
p
le,
Y
u
et
a
l.
[
1
5
]
f
u
s
ed
tex
t
a
n
d
r
etr
iev
ed
f
ac
t
-
ch
ec
k
in
g
s
ig
n
als
in
a
d
u
al
-
p
er
ce
p
tio
n
m
o
d
el,
im
p
r
o
v
i
n
g
o
n
ea
r
lier
B
E
R
T
b
aselin
es.
B
u
t
ev
en
th
ese
en
h
an
ce
d
ar
ch
itectu
r
es r
ely
o
n
m
atch
in
g
s
u
r
f
ac
e
tex
t o
r
s
h
allo
w
in
f
er
en
ce
.
2
.
3
.
L
a
rg
e
la
ng
ua
g
e
m
o
del
-
driv
en
f
a
ct
v
er
if
ica
t
io
n
R
ec
en
tly
,
LLM
s
h
av
e
b
ee
n
ap
p
lied
to
f
ac
t
v
e
r
if
icatio
n
in
n
o
v
el
way
s
.
On
e
s
tr
an
d
u
s
es
p
r
o
m
p
tin
g
o
r
ch
ain
-
of
-
th
o
u
g
h
t
t
o
h
a
v
e
a
n
L
L
M
an
al
y
ze
a
claim
.
Z
h
an
g
an
d
Gao
[
1
2
]
in
tr
o
d
u
ce
d
th
e
h
ier
ar
ch
ical
s
tep
-
by
-
s
tep
(
HiSS
)
p
r
o
m
p
tin
g
:
th
ey
d
ec
o
m
p
o
s
e
c
o
m
p
le
x
c
laim
s
in
to
s
u
b
-
claim
s
a
n
d
iter
ativ
ely
v
er
if
y
ea
c
h
u
s
in
g
GPT
-
4
,
ac
h
iev
in
g
s
tate
-
of
-
th
e
-
ar
t
o
n
m
is
in
f
o
r
m
ati
o
n
b
en
ch
m
ar
k
s
[
1
2
]
.
Min
et
a
l.
[
1
4
]
p
r
o
p
o
s
e
F
a
ctSco
r
e
f
o
r
f
in
e
-
g
r
ain
ed
ev
alu
atio
n
o
f
L
L
M
an
s
wer
s
,
h
ig
h
lig
h
tin
g
th
at
L
L
M
ac
cu
r
ac
y
h
in
g
es
o
n
co
n
tex
t
an
d
p
r
o
m
p
t
d
esig
n
.
Me
tr
o
p
o
litan
s
k
y
an
d
L
ar
s
o
n
[
1
6
]
d
e
v
elo
p
ed
C
laim
if
y
,
an
L
L
M
-
b
ased
ex
tr
ac
to
r
th
at
b
r
ea
k
s
an
o
u
tp
u
t
in
to
p
r
ec
is
e
f
ac
tu
al
s
tatem
en
ts
.
E
x
p
e
r
im
en
t
s
s
h
o
w
th
at
u
s
in
g
h
ig
h
-
q
u
ality
ex
tr
ac
ted
claim
s
is
cr
u
cial,
s
in
ce
f
a
u
lty
claim
p
h
r
asin
g
ca
n
u
n
d
e
r
m
in
e
v
er
if
icat
io
n
.
A
m
ajo
r
tr
en
d
is
to
c
o
m
b
in
e
L
L
Ms
with
th
e
r
etr
iev
al
o
f
ex
ter
n
al
k
n
o
wled
g
e.
Kh
aliq
et
a
l.
[
1
7
]
ex
e
m
p
lify
th
is
with
r
etr
iev
al
-
au
g
m
en
ted
g
e
n
er
atio
n
-
au
g
m
en
ted
r
ea
s
o
n
in
g
(
R
AGAR
)
,
wh
ich
u
s
es
a
m
u
ltimo
d
al
L
L
M
p
lu
s
r
etr
iev
al
to
f
ac
t
-
ch
ec
k
p
o
liti
ca
l
claim
s
in
twee
ts
an
d
im
a
g
es.
T
h
e
y
in
tr
o
d
u
ce
c
h
ain
-
of
-
r
etr
iev
al
-
au
g
m
en
ted
g
e
n
er
atio
n
(
R
AG)
an
d
t
r
ee
-
of
-
R
AG
r
ea
s
o
n
in
g
,
wh
er
eb
y
th
e
m
o
d
el
alter
n
ates
b
etwe
en
r
etr
iev
i
n
g
ev
i
d
en
ce
a
n
d
g
en
e
r
atin
g
n
ew
s
u
b
-
q
u
esti
o
n
s
,
ac
h
iev
in
g
a
weig
h
ted
F1
-
s
co
r
e
o
f
0
.
8
5
o
n
a
p
o
liti
ca
l
claim
s
test
s
et.
Fac
tC
h
ec
k
GPT
[
1
8
]
im
p
lem
en
ts
an
eig
h
t
-
s
tep
p
ip
elin
e
o
n
L
L
M
o
u
tp
u
ts
.
I
n
ex
p
e
r
im
en
ts
o
n
th
e
Factch
ec
k
-
B
en
ch
(
6
7
8
cla
im
s
f
r
o
m
C
h
atGPT
r
esp
o
n
s
es),
FactC
h
ec
k
GPT
f
o
u
n
d
th
at
s
tate
-
of
-
t
h
e
-
ar
t
ch
ec
k
er
s
(
FacT
o
o
l,
FactSco
r
e,
an
d
Per
p
lex
ity
)
s
tr
u
g
g
le
–
th
e
b
est
F1
-
s
co
r
e
o
n
f
alse
claim
s
was
o
n
ly
~0
.
5
3
.
An
o
th
e
r
lin
e
tr
ea
ts
L
L
Ms
as
ag
en
ts
.
L
i
et
a
l.
[
1
3
]
p
r
o
p
o
s
e
FactAg
en
t,
wh
ich
em
u
lates
h
u
m
an
ex
p
er
t
wo
r
k
f
lo
ws:
th
e
m
o
d
el
p
er
f
o
r
m
s
r
esear
c
h
s
u
b
task
s
s
eq
u
en
tially
(
e.
g
.
,
s
ea
r
ch
,
s
u
m
m
a
r
izatio
n
,
c
r
o
s
s
-
ch
ec
k
in
g
)
b
ef
o
r
e
g
iv
in
g
a
v
er
d
ict.
Qu
elle
an
d
B
o
v
et
[
1
1
]
d
em
o
n
s
tr
ate
th
at
GPT
-
4
with
r
etr
iev
ed
co
n
te
x
t
an
d
s
o
u
r
ce
-
citatio
n
ca
n
f
ac
t
-
ch
ec
k
with
p
r
o
m
is
in
g
ac
c
u
r
ac
y
,
th
o
u
g
h
a
cc
u
r
ac
y
v
ar
ies
b
ased
o
n
q
u
er
y
lan
g
u
a
g
e
an
d
claim
v
er
ac
ity
.
I
n
s
u
m
m
ar
y
,
L
L
Ms
ca
n
lev
er
ag
e
b
r
o
ad
k
n
o
wled
g
e
an
d
co
m
p
lex
r
ea
s
o
n
i
n
g
,
b
u
t
th
eir
u
n
ch
ec
k
e
d
o
u
tp
u
ts
ca
n
b
e
in
co
n
s
is
ten
t.
T
h
er
ef
o
r
e,
m
a
n
y
r
ec
en
t
s
y
s
tem
s
co
m
b
in
e
L
L
Ms w
ith
r
etr
ie
v
al:
e.
g
.
,
in
te
g
r
atin
g
s
ea
r
ch
r
esu
lts
in
to
L
L
M
p
r
o
m
p
ts
f
o
r
ev
id
e
n
ce
[
1
1
]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
4
,
Au
g
u
s
t 2
0
2
6
:
3
1
4
4
-
3
1
5
3
3146
A
s
ec
o
n
d
f
a
m
ily
au
g
m
en
ts
L
L
Ms
with
s
tr
u
ctu
r
ed
k
n
o
wled
g
e
o
r
s
y
m
b
o
lic
r
ea
s
o
n
in
g
.
Gr
ap
h
C
h
ec
k
[
1
9
]
is
a
p
r
im
e
ex
am
p
le:
it
ex
tr
ac
ts
a
k
n
o
wled
g
e
g
r
a
p
h
(
KG)
f
r
o
m
th
e
d
o
cu
m
en
t
a
n
d
claim
,
en
co
d
es
it
with
a
g
r
ap
h
n
eu
r
al
n
etwo
r
k
,
an
d
f
e
ed
s
th
e
r
esu
ltin
g
em
b
e
d
d
in
g
s
in
to
a
n
L
L
M
p
r
o
m
p
t.
E
m
p
ir
ically
,
Gr
ap
h
C
h
ec
k
im
p
r
o
v
e
d
o
v
e
r
all
ac
cu
r
ac
y
b
y
~7
.
1
%
o
v
e
r
b
aselin
es
o
n
s
ev
en
g
en
e
r
al
an
d
m
e
d
ical
-
d
o
m
ain
b
en
ch
m
ar
k
s
.
An
o
th
er
h
y
b
r
id
s
y
s
tem
,
Hy
b
r
id
FC
,
is
a
h
y
b
r
id
KG
-
b
ased
f
ac
t
-
ch
ec
k
in
g
m
eth
o
d
th
at
e
n
s
em
b
les
m
u
ltip
le
ap
p
r
o
ac
h
es
(
tex
t
-
b
ased
,
p
at
h
-
b
ased
,
r
u
le
-
b
ased
,
em
b
e
d
d
i
n
g
-
b
ased
)
t
o
d
ec
id
e
th
e
v
er
ac
ity
o
f
ass
er
tio
n
s
(
tr
ip
les)
in
KG
s
.
I
t
r
ep
o
r
ts
i
m
p
r
o
v
e
m
en
ts
o
f
0
.
1
4
to
0
.
2
7
in
ar
ea
u
n
d
e
r
th
e
cu
r
v
e
(
AUC
)
o
v
er
th
e
p
r
io
r
s
tate
-
of
-
th
e
-
ar
t o
n
th
e
FactB
en
ch
d
ataset
[
2
0
]
.
R
ec
en
t
wo
r
k
h
as
ex
ten
d
ed
f
ac
t
-
ch
ec
k
in
g
to
m
u
ltimo
d
al
claim
s
(
tex
t
+
im
ag
es/v
id
eo
s
)
.
T
h
e
Fak
eM
ar
k
s
y
s
tem
[
2
1
]
tack
le
s
f
ac
t
-
ch
ec
k
in
g
o
f
n
ews
v
i
d
e
o
s
.
Fak
eM
ar
k
u
s
es
a
q
u
esti
o
n
-
d
r
iv
e
n
p
ip
elin
e:
it
p
ass
es
th
e
v
id
eo
,
titl
e,
an
d
tr
a
n
s
cr
ip
t
th
r
o
u
g
h
a
v
is
io
n
lan
g
u
ag
e
m
o
d
el
(
VL
M)
an
d
L
L
M
(
GPT
-
4o
-
m
in
i
an
d
I
n
ter
n
VL
-
8
B
)
with
tailo
r
e
d
p
r
o
m
p
ts
.
Fak
eM
ar
k
’
s
tr
ai
n
in
g
-
f
r
ee
p
i
p
elin
e
ac
h
iev
ed
~6
2
.
9
%
ac
cu
r
ac
y
(
F1
≈
5
9
.
4
%)
o
n
b
i
n
ar
y
v
er
ac
it
y
.
An
o
t
h
er
m
u
ltimo
d
al
ap
p
r
o
ac
h
is
L
R
Q
-
Fact
[
2
2
]
,
d
esig
n
ed
f
o
r
m
u
ltimo
d
al
m
is
in
f
o
r
m
atio
n
(
im
a
g
es+tex
t)
.
I
t
g
en
er
ates
r
elev
an
t
f
ac
t
-
c
h
ec
k
q
u
esti
o
n
s
b
y
p
r
o
m
p
tin
g
a
VL
M
+
L
L
M
p
ip
elin
e.
A
th
ir
d
ex
am
p
le
is
m
u
ltimo
d
al
au
to
m
ated
f
ac
t
-
c
h
ec
k
in
g
v
ia
te
x
tu
aliza
tio
n
(
MA
FT
)
[
2
3
]
,
a
g
e
n
er
al
m
u
ltimo
d
al
f
ac
t
-
ch
ec
k
er
f
o
r
tex
t,
im
ag
es,
v
id
eo
,
an
d
a
u
d
io
.
MA
FT
f
ir
s
t
tex
tu
alize
s
all
in
p
u
t
m
ed
ia
(
e.
g
.
,
g
e
n
er
atin
g
ca
p
tio
n
s
/tra
n
s
cr
ip
ts
)
,
th
en
ex
tr
ac
ts
claim
s
,
an
d
co
llects ev
id
en
ce
(
v
ia
web
s
ea
r
ch
o
r
KB
)
,
an
d
f
in
ally
g
en
e
r
ates a
r
ep
o
r
t.
T
h
e
p
r
o
p
o
s
ed
m
eth
o
d
d
i
f
f
er
s
f
r
o
m
th
ese
p
r
io
r
ap
p
r
o
ac
h
es
in
s
ev
er
al
way
s
.
First,
s
o
cial
n
e
ws
ar
ticles
ar
e
p
r
e
-
clu
s
ter
ed
b
y
ev
en
t
b
ef
o
r
e
r
etr
iev
al.
T
h
is
d
r
a
m
atica
lly
n
ar
r
o
ws
th
e
s
ea
r
ch
s
p
ac
e:
i
n
s
tead
o
f
r
etr
iev
in
g
f
r
o
m
an
e
n
tire
n
ews
co
r
p
u
s
,
we
v
er
if
y
ea
ch
p
o
s
t
ag
ain
s
t
a
s
m
all
s
et
o
f
ev
en
t
-
r
elate
d
ar
ticles,
s
im
ilar
in
s
p
ir
it
to
th
e
clu
s
ter
in
g
u
s
ed
in
FA
STT
R
A
C
K
[
2
4
]
.
T
h
is
r
ed
u
ce
s
n
o
is
e
an
d
co
m
p
u
tatio
n
al
c
o
s
t
wh
ile
p
r
eser
v
in
g
r
elev
an
t
ev
i
d
en
ce
.
Seco
n
d
,
a
p
lain
tex
t
L
L
M
p
ip
elin
e
(
GPT
-
4
)
is
u
s
ed
with
o
u
t
an
y
f
in
e
-
tu
n
in
g
o
r
en
d
-
to
-
en
d
tr
ain
in
g
,
r
ely
i
n
g
o
n
its
ze
r
o
/f
ew
-
s
h
o
t
ca
p
ab
ilit
ies
to
ex
tr
ac
t
claim
s
an
d
ch
ec
k
th
e
m
.
I
n
c
o
n
tr
ast,
m
an
y
p
r
io
r
s
y
s
tem
s
eith
er
f
in
e
-
tu
n
e
m
o
d
e
ls
(
e.
g
.
,
R
az
a
et
a
l.
[
4
]
f
in
e
-
tu
n
e
B
E
R
T
)
o
r
tr
ain
class
if
icati
o
n
h
ea
d
s
.
T
h
ir
d
,
t
h
e
ap
p
r
o
ac
h
tar
g
ets
s
o
cial
m
e
d
ia
n
ews
p
o
s
ts
in
Vietn
am
ese,
a
d
o
m
ain
wh
e
r
e
f
ew
f
ac
t
-
ch
ec
k
i
n
g
s
tu
d
ies
ex
is
t.
B
y
o
p
er
atin
g
in
Vietn
am
ese
an
d
alig
n
in
g
s
o
cial
p
o
s
ts
wit
h
m
ain
s
tr
ea
m
n
ews
clu
s
ter
s
,
a
lo
w
-
r
eso
u
r
ce
m
u
ltil
in
g
u
al
s
ettin
g
(
co
m
p
ar
a
b
le
to
th
e
d
o
m
ain
ad
ap
tatio
n
in
ViFact
C
h
ec
k
[
2
5
]
)
is
ad
d
r
ess
ed
r
ath
er
th
an
m
o
s
tly
E
n
g
lis
h
p
o
liti
ca
l
n
ews.
Fin
ally
,
u
n
lik
e
m
o
n
o
lith
ic
cla
s
s
if
ier
s
,
th
e
p
ip
elin
e
o
u
tp
u
ts
a
“su
s
p
icio
u
s
”
lab
el
o
n
ly
if
th
er
e
is
co
n
f
lictin
g
ev
i
d
en
ce
b
etwe
en
th
e
p
o
s
t
an
d
it
s
clu
s
ter
;
th
i
s
cr
o
s
s
-
v
alid
atio
n
s
tep
lev
er
ag
es
b
o
th
th
e
L
L
M’
s
k
n
o
wled
g
e
a
n
d
c
o
n
cr
ete
r
etr
iev
e
d
f
ac
ts
,
an
asp
ec
t
s
im
ilar
to
R
AG
b
u
t
f
r
am
ed
th
r
o
u
g
h
e
v
en
t
clu
s
ter
s
.
Ov
er
all,
b
y
co
m
b
in
in
g
L
L
M
r
ea
s
o
n
in
g
with
cl
u
s
ter
-
co
n
s
tr
ain
ed
r
etr
iev
al,
th
e
m
eth
o
d
ac
h
ie
v
es
h
ig
h
er
ac
c
u
r
ac
y
t
h
an
b
aselin
es
wh
ile
r
em
ain
in
g
e
f
f
icien
t.
2
.
4
.
New
s
clus
t
er
ing
a
nd
re
t
riev
a
l
An
o
th
er
k
ey
i
d
ea
is
clu
s
ter
in
g
r
elate
d
n
ews
b
ef
o
r
e
v
e
r
if
icatio
n
.
I
n
s
tead
o
f
r
etr
iev
in
g
to
p
-
K
d
o
cu
m
e
n
ts
f
o
r
a
claim
,
o
n
e
ca
n
clu
s
ter
t
h
e
e
n
tire
n
ews
co
r
p
u
s
i
n
to
e
v
en
ts
.
Pan
c
h
en
d
r
ar
ajan
et
a
l.
[
2
6
]
in
tr
o
d
u
ce
d
Mu
ltiC
laim
Net,
clu
s
ter
in
g
s
im
ilar
claim
s
ac
r
o
s
s
lan
g
u
ag
es
to
r
e
d
u
ce
r
ed
u
n
d
an
cy
an
d
im
p
r
o
v
e
r
etr
iev
al
ef
f
icien
cy
[
2
6
]
.
FASTT
R
AC
K
[
2
4
]
tak
es
a
two
-
s
ta
g
e
ap
p
r
o
ac
h
:
it
b
u
ild
s
a
n
o
f
f
li
n
e
clu
s
ter
in
g
in
d
ex
o
f
th
e
co
r
p
u
s
,
th
en
f
o
r
ea
ch
q
u
er
y
r
etr
ie
v
es
o
n
ly
r
elev
an
t
cl
u
s
ter
s
an
d
u
s
es
an
L
L
M
f
o
r
f
i
n
e
-
g
r
ain
e
d
ev
id
e
n
ce
s
co
r
in
g
.
T
h
is
d
r
am
atica
lly
s
h
r
i
n
k
s
th
e
s
ea
r
ch
s
p
ac
e;
FASTT
R
AC
K
r
ep
o
r
ted
>3
3
×
s
p
ee
d
-
u
p
s
an
d
m
u
ch
h
i
g
h
er
ac
cu
r
ac
y
t
h
an
n
aïv
e
r
etr
iev
a
l.
R
ec
en
t
clu
s
ter
in
g
-
b
ased
R
AG
f
r
am
ewo
r
k
s
s
u
ch
as
C
R
AVE
[
2
7
]
f
u
r
th
e
r
d
em
o
n
s
tr
ate
th
at
o
r
g
an
izin
g
d
o
cu
m
en
ts
in
to
co
h
e
r
en
t
ev
en
t
o
r
n
a
r
r
ativ
e
cl
u
s
ter
s
im
p
r
o
v
es
e
v
id
en
c
e
co
n
s
is
ten
cy
an
d
c
o
n
tex
tu
al
g
r
o
u
n
d
in
g
co
m
p
ar
ed
to
f
lat
to
p
-
K
r
etr
iev
al.
I
n
th
e
p
ip
elin
e
,
s
o
cial
p
o
s
ts
an
d
tr
u
s
two
r
th
y
n
ews
ar
e
s
im
ilar
ly
g
r
o
u
p
ed
b
y
ev
e
n
t,
s
o
th
at
e
ac
h
claim
is
ch
ec
k
ed
ag
ain
s
t
th
e
m
o
s
t
r
elev
an
t
clu
s
ter
o
f
s
o
u
r
ce
s
(
r
at
h
er
th
a
n
m
illi
o
n
s
o
f
d
o
cu
m
en
ts
)
.
As
o
th
er
wo
r
k
s
n
o
te,
clu
s
ter
in
g
b
y
ev
en
t
“i
m
p
r
o
v
es
claim
r
etr
iev
al
a
n
d
v
alid
atio
n
”
b
y
r
ed
u
cin
g
r
ed
u
n
d
an
c
y
[
2
6
]
.
T
h
is
ap
p
r
o
ac
h
also
ad
d
r
ess
es
p
r
ac
tical
is
s
u
e
s
:
s
elec
tin
g
a
f
ix
ed
to
p
-
N
r
esu
lts
ca
n
m
is
s
co
n
tex
t
o
r
in
clu
d
e
n
o
is
e,
wh
er
ea
s
clu
s
ter
in
g
ad
ap
ts
to
t
h
e
d
ata
d
is
tr
ib
u
tio
n
.
3.
M
E
T
H
O
D
T
o
en
s
u
r
e
t
h
e
n
ews
v
er
i
f
icatio
n
p
r
o
ce
s
s
is
b
o
th
s
y
s
tem
atic
an
d
tr
an
s
p
a
r
en
t,
th
e
s
y
s
tem
d
e
v
elo
p
ed
is
s
tr
u
ctu
r
ed
in
to
f
o
u
r
m
ain
,
in
ter
co
n
n
ec
ted
s
tag
es.
T
h
ese
s
tag
es
ar
e
in
clu
d
e:
i
)
n
ews
co
llectio
n
an
d
ev
en
t
clu
s
ter
in
g
,
ii
)
claim
ex
tr
ac
tio
n
,
iii
)
claim
v
er
if
icatio
n
,
an
d
iv
)
f
in
al
n
ews v
er
if
icatio
n
.
3
.
1
.
New
s
co
llect
io
n a
nd
clus
t
er
ing
A
d
ataset
o
f
Vietn
am
ese
n
ews
ar
ticles
f
r
o
m
late
2
0
2
4
was
g
ath
er
e
d
,
in
clu
d
in
g
s
o
cial
-
m
ed
ia
p
o
s
ts
(
e.
g
.
,
v
ir
al
Face
b
o
o
k
,
Yo
u
T
u
b
e
n
ews)
an
d
m
ain
s
tr
ea
m
o
u
tle
ts
(
e.
g
.
,
V
n
E
x
p
r
ess
,
Vietn
am
n
et)
.
I
n
to
tal,
1
,
7
6
5
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
S
o
cia
l n
ews
fa
ctu
a
lity v
erif
ica
tio
n
u
s
in
g
la
r
g
e
l
a
n
g
u
a
g
e
mo
d
els (
Tr
a
n
Du
c
Du
o
n
g
)
3147
ar
ticles
wer
e
co
llected
,
co
v
er
i
n
g
d
o
m
ai
n
s
lik
e
p
o
liti
cs,
d
is
aster
s
,
s
p
o
r
ts
,
an
d
en
ter
tain
m
e
n
t.
So
cial
p
o
s
ts
ar
e
tr
ea
ted
as
“item
s
to
v
er
if
y
”
,
wh
ile
th
e
tr
u
s
ted
n
ews
s
er
v
es
as
ev
id
en
ce
.
T
h
e
tex
t
is
p
r
e
p
r
o
ce
s
s
ed
(
clea
n
in
g
HT
ML
an
d
n
o
r
m
alizin
g
f
o
n
t
s
)
an
d
th
en
clu
s
ter
ar
ticles
b
y
ev
e
n
t.
Fo
r
clu
s
ter
in
g
,
d
e
n
s
ity
-
b
ased
s
p
atial
clu
s
ter
in
g
o
f
a
p
p
licatio
n
s
w
ith
n
o
is
e
(
DB
SC
AN
)
[
2
8
]
is
u
s
ed
o
n
d
o
cu
m
e
n
t
em
b
e
d
d
in
g
s
(
with
tex
t
em
b
ed
d
in
g
s
ex
tr
ac
ted
u
s
in
g
Ph
o
B
E
R
T
[
2
9
]
)
,
wh
ich
g
r
o
u
p
s
to
g
eth
er
ar
ticles
m
en
tio
n
in
g
th
e
s
am
e
e
v
en
t
o
r
to
p
ic.
T
em
p
o
r
al
m
etad
ata
(
p
u
b
licatio
n
d
ate)
is
also
in
co
r
p
o
r
ated
to
p
r
ev
e
n
t
u
n
r
elate
d
s
to
r
ies
with
s
im
ilar
co
n
ten
t
f
r
o
m
b
ein
g
co
n
f
late
d
.
Af
ter
tu
n
i
n
g
,
DB
SC
A
N
ef
f
ec
tiv
ely
is
o
lates
clu
s
ter
s
o
f
a
r
b
itra
r
y
s
h
a
p
e
an
d
d
en
s
ity
,
ca
p
tu
r
in
g
ev
en
ts
r
an
g
in
g
f
r
o
m
n
atu
r
al
d
is
aster
s
to
p
r
ess
co
n
f
er
en
ce
s
.
E
ac
h
clu
s
t
er
th
u
s
co
n
tain
s
all
s
o
cial
an
d
m
ain
s
tr
ea
m
ar
ticles r
elate
d
to
o
n
e
ev
en
t.
B
y
clu
s
ter
in
g
f
ir
s
t,
ev
id
en
ce
r
etr
iev
al
is
r
estricte
d
to
th
e
clu
s
ter
o
f
in
ter
est.
I
n
co
n
t
r
ast
to
“to
p
-
N”
s
ea
r
ch
(
wh
er
e
ch
o
o
s
in
g
N
is
ar
b
itra
r
y
an
d
m
ay
o
m
it
r
elev
an
t
co
n
tex
t)
,
clu
s
ter
in
g
a
d
a
p
tiv
ely
co
llects
all
p
er
tin
en
t
r
ef
e
r
en
ce
s
.
As
p
r
io
r
s
tu
d
ies
s
h
o
w,
clu
s
ter
in
g
ev
e
n
ts
ca
n
s
ig
n
if
ican
tly
r
e
d
u
ce
t
h
e
s
ea
r
ch
s
p
ac
e
an
d
im
p
r
o
v
e
ev
id
e
n
ce
r
elev
an
ce
[
2
4
]
.
Fo
r
ex
am
p
le,
d
u
r
in
g
v
e
r
if
icatio
n
,
we
o
n
ly
lo
o
k
at
th
e
r
e
liab
le
n
ews
ar
ticles
in
th
e
s
am
e
clu
s
ter
as
a
s
o
ci
al
p
o
s
t,
r
ath
e
r
th
a
n
q
u
er
y
i
n
g
a
web
s
ea
r
ch
e
n
g
in
e.
T
h
is
en
s
u
r
es
all
ch
ec
k
e
d
s
o
u
r
ce
s
ar
e
o
n
-
to
p
ic
an
d
o
f
k
n
o
wn
cr
e
d
ib
ilit
y
,
wh
ich
also
m
itig
ates
th
e
is
s
u
e
o
f
co
n
f
lictin
g
o
r
s
p
u
r
io
u
s
r
etr
iev
al
r
esu
lts
.
3
.
2
.
Cla
i
m
ex
t
r
a
ct
io
n
G
i
v
e
n
a
s
o
c
ia
l
-
m
e
d
i
a
ar
t
i
c
l
e
,
a
t
o
m
i
c
c
l
a
im
s
a
r
e
i
d
e
n
t
if
i
e
d
w
i
t
h
i
n
i
t
.
An
a
t
o
m
ic
c
l
a
i
m
i
s
d
e
f
in
e
d
a
s
a
c
o
n
c
i
s
e
s
t
a
t
em
e
n
t
th
a
t
co
n
v
e
y
s
o
n
e
v
er
i
f
i
ab
l
e
p
i
ec
e
o
f
i
n
f
o
r
m
a
t
io
n
[
1
4
]
.
F
o
r
in
s
t
a
n
c
e,
“T
h
e
p
r
e
s
i
d
en
t
w
i
l
l
v
i
s
i
t
J
a
p
a
n
i
n
S
ep
t
e
m
b
e
r
”
i
s
a
t
o
m
i
c,
wh
e
r
e
a
s
a
s
en
t
e
n
c
e
l
is
t
i
n
g
m
u
l
t
ip
l
e
f
a
c
t
s
i
s
n
o
t
.
B
r
e
a
k
i
n
g
c
o
n
t
en
t
in
t
o
a
t
o
m
i
c
c
la
i
m
s
s
i
m
p
l
if
i
e
s
v
er
i
f
i
c
a
t
io
n
an
d
av
o
id
s
p
ar
t
i
a
l
t
r
u
t
h
s
.
G
P
T
-
4
i
s
u
s
ed
t
o
e
x
t
r
a
c
t
c
l
a
i
m
s
by
p
r
o
m
p
t
in
g
t
h
e
m
o
d
e
l
w
i
t
h
t
h
e
a
r
t
ic
l
e
t
e
x
t,
a
s
k
i
n
g
i
t
to
l
i
s
t
t
h
e
k
e
y
f
a
c
t
u
a
l
s
t
a
t
em
e
n
t
s
.
I
n
ac
c
o
r
d
an
c
e
w
i
t
h
r
e
c
e
n
t
wo
r
k
,
a
c
a
r
ef
u
l
ly
d
e
s
i
g
n
ed
p
r
o
m
p
t
(
e
.
g
.
,
f
e
w
-
s
h
o
t
e
x
a
m
p
l
e
s
)
i
s
ap
p
l
i
ed
t
o
e
n
s
u
r
e
h
i
g
h
-
p
r
e
c
i
s
io
n
e
x
t
r
a
c
t
io
n
.
W
o
r
k
s
s
u
c
h
a
s
C
l
a
i
m
i
f
y
[
1
6
]
an
d
f
a
c
t
-
in
-
f
r
a
g
m
e
n
t
s
[
3
0
]
d
em
o
n
s
t
r
a
t
e
t
h
a
t
d
e
co
m
p
o
s
i
n
g
l
o
n
g
-
f
o
r
m
co
n
te
n
t
i
n
to
a
t
o
m
i
c
,
v
er
i
f
i
ab
l
e
u
n
i
t
s
s
i
g
n
if
i
c
a
n
t
ly
im
p
r
o
v
e
s
d
o
wn
s
t
r
ea
m
f
a
c
t
-
c
h
e
ck
i
n
g
r
e
l
i
a
b
i
l
i
ty
.
C
l
a
i
m
i
f
y
[
1
6
]
an
d
r
e
l
a
t
ed
m
e
t
h
o
d
s
h
a
v
e
s
h
o
w
n
th
a
t
t
h
e
q
u
a
l
i
ty
o
f
ex
t
r
ac
t
e
d
c
l
a
i
m
s
s
t
r
o
n
g
ly
a
f
f
e
c
t
s
d
o
wn
s
t
r
e
am
f
ac
t
-
ch
e
ck
i
n
g
.
C
l
a
i
m
s
th
a
t
ar
e
u
n
v
e
r
if
ia
b
l
e
(
e
.
g
.
,
o
p
in
i
o
n
s
o
r
u
n
g
r
o
u
n
d
e
d
s
p
e
c
u
la
t
i
o
n
)
a
r
e
f
u
r
th
er
f
i
l
te
r
ed
o
u
t
,
a
n
d
o
n
l
y
c
h
e
ck
-
wo
r
th
y
f
a
c
tu
a
l
c
l
a
im
s
a
r
e
p
a
s
s
e
d
t
o
t
h
e
n
ex
t
s
t
ag
e
.
T
h
i
s
ex
t
r
ac
t
i
o
n
s
t
e
p
a
l
i
g
n
s
w
i
t
h
p
r
a
c
t
i
c
e
s
in
L
L
M
-
b
a
s
e
d
ev
a
lu
a
t
i
o
n
:
th
e
F
act
S
c
o
r
e
f
r
a
m
e
wo
r
k
a
l
s
o
b
r
ea
k
s
t
e
x
t
i
n
to
a
t
o
m
i
c
f
a
c
t
s
t
o
a
s
s
e
s
s
f
a
c
tu
a
l
i
ty
[
1
4
]
.
B
y
f
o
cu
s
i
n
g
th
e
L
L
M
’
s
ef
f
o
r
t
o
n
e
a
ch
c
l
a
i
m
in
d
e
p
en
d
e
n
t
l
y
,
w
e
i
m
p
r
o
v
e
c
o
n
s
i
s
t
e
n
c
y
an
d
t
r
ac
e
ab
i
l
i
t
y
o
f
t
h
e
v
e
r
if
i
c
a
t
io
n
.
I
n
ex
p
er
im
e
n
t
s
,
s
y
s
t
e
m
s
u
s
i
n
g
ex
t
r
ac
te
d
c
l
a
i
m
s
a
r
e
s
h
o
wn
t
o
o
u
tp
e
r
f
o
r
m
a
t
t
e
m
p
t
s
to
v
e
r
if
y
e
n
t
ir
e
a
r
t
i
c
l
e
s
a
t
o
n
ce
.
F
i
g
u
r
e
1
i
l
lu
s
t
r
a
t
e
s
t
h
e
p
r
o
m
p
t
d
e
s
i
g
n
f
o
r
c
l
a
i
m
ex
t
r
a
ct
i
o
n
.
Fig
u
r
e
1
.
Pro
m
p
t d
esig
n
f
o
r
th
e
claim
ex
tr
ac
tio
n
task
3
.
3
.
Cla
i
m
v
er
if
ic
a
t
io
n
E
ac
h
ex
t
r
ac
ted
at
o
m
ic
claim
is
th
en
v
er
if
ied
ag
ain
s
t
e
v
id
e
n
ce
.
An
L
L
M
(
GPT
-
4
)
is
p
r
o
m
p
ted
to
ev
alu
ate
th
e
tr
u
th
o
f
a
claim
g
iv
en
r
elev
an
t
c
o
n
tex
t.
Fo
r
c
o
n
tex
t,
th
e
tr
u
s
ted
n
ews
ar
ticles
in
th
e
s
am
e
clu
s
ter
ar
e
u
s
ed
.
Sp
ec
if
ically
,
th
e
m
o
d
el
is
p
r
o
v
id
ed
with
t
h
e
claim
tex
t
an
d
th
e
co
n
ca
ten
ated
co
n
ten
ts
(
o
r
s
u
m
m
ar
ies)
o
f
all
clu
s
ter
n
ews,
an
d
ask
it
to
r
esp
o
n
d
“
t
r
u
e”
o
r
“
f
alse”
with
a
b
r
ief
ju
s
tific
atio
n
.
A
m
u
lti
-
s
tep
r
ea
s
o
n
in
g
s
tr
ateg
y
in
s
p
ir
ed
b
y
h
ier
ar
c
h
ical
p
r
o
m
p
tin
g
is
ad
o
p
ted
:
if
a
claim
is
co
m
p
lex
,
th
e
L
L
M
ca
n
d
ec
o
m
p
o
s
e
it a
n
d
ch
ec
k
s
u
b
-
c
o
m
p
o
n
en
ts
(
s
im
ilar
to
HiSS
[
1
2
]
).
I
n
p
r
ac
tice,
p
r
o
v
id
in
g
GPT
-
4
with
a
f
ew
-
s
h
o
t
e
x
em
p
lar
s
(
claim
+
ev
id
en
ce
+
v
er
d
ict)
i
s
f
o
u
n
d
to
im
p
r
o
v
e
ac
c
u
r
ac
y
.
T
h
e
L
L
M
th
u
s
ac
ts
as
a
s
m
ar
t
co
m
p
ar
at
o
r
:
lev
er
ag
in
g
b
o
th
its
in
ter
n
a
l
wo
r
ld
k
n
o
wled
g
e
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.
15
,
No
.
4
,
Au
g
u
s
t 2
0
2
6
:
3
1
4
4
-
3
1
5
3
3148
an
d
th
e
p
r
o
v
id
ed
clu
s
ter
co
n
tex
t.
T
h
is
d
if
f
er
s
f
r
o
m
p
u
r
e
r
etr
iev
al
-
b
ased
ch
ec
k
s
(
wh
ic
h
m
ig
h
t
lack
d
ee
p
in
f
er
en
ce
)
a
n
d
f
r
o
m
n
aiv
e
c
lass
if
icatio
n
(
wh
ich
lack
s
d
y
n
am
ic
ev
id
en
ce
)
.
I
t
is
also
r
elate
d
to
ag
en
tic
ap
p
r
o
ac
h
es:
FactAg
en
t
[
1
3
]
d
em
o
n
s
tr
ated
th
at
b
r
ea
k
in
g
v
er
if
icatio
n
in
to
s
u
b
task
s
(
s
ea
r
ch
,
ju
d
g
m
en
t)
y
ield
s
b
etter
p
er
f
o
r
m
a
n
ce
.
No
m
o
d
el
is
tr
ain
ed
e
n
d
-
to
-
en
d
;
in
s
tead
,
GPT
-
4
’
s
ze
r
o
-
s
h
o
t/fe
w
-
s
h
o
t
ca
p
ab
ilit
y
is
r
elied
upon
.
Fig
u
r
e
2
illu
s
tr
ates th
e
p
r
o
m
p
t d
esig
n
f
o
r
claim
v
er
if
ic
atio
n
.
T
o
ass
ess
v
er
if
icatio
n
r
eliab
ilit
y
,
th
e
L
L
M
is
r
e
q
u
ir
e
d
to
p
r
o
v
id
e
v
e
r
b
atim
e
v
id
en
ce
ci
tatio
n
s
f
o
r
ea
ch
v
er
d
ict.
A
v
er
d
ict
is
co
n
s
id
er
ed
ev
id
e
n
ce
-
g
r
o
u
n
d
ed
o
n
ly
if
all
q
u
o
ted
s
en
ten
ce
s
ap
p
ea
r
v
er
b
atim
in
th
e
r
ef
er
en
ce
d
a
r
ticles
an
d
lo
g
ical
ly
s
u
p
p
o
r
t
th
e
c
o
n
clu
s
io
n
.
T
h
e
h
allu
cin
atio
n
r
ate
is
m
ea
s
u
r
ed
as
th
e
f
r
ac
tio
n
o
f
v
er
d
icts
th
at
eith
e
r
:
i)
lack
v
alid
citatio
n
s
o
r
ii)
cite
s
en
ten
ce
s
th
at
d
o
n
o
t
s
u
p
p
o
r
t
t
h
e
v
e
r
d
i
ct.
Ad
d
itio
n
ally
,
a
d
is
tin
g
u
is
h
is
m
ad
e
b
etwe
en
v
er
d
icts
g
r
o
u
n
d
e
d
in
clu
s
ter
ev
id
en
ce
an
d
th
o
s
e
in
f
er
r
e
d
p
r
im
ar
ily
f
r
o
m
t
h
e
m
o
d
el’
s
in
ter
n
al
k
n
o
wled
g
e.
T
h
is
d
esig
n
alig
n
s
with
r
ec
en
t
ef
f
o
r
ts
to
e
n
f
o
r
ce
citatio
n
-
g
r
o
u
n
d
ed
r
ea
s
o
n
in
g
to
r
ed
u
ce
h
allu
cin
atio
n
in
L
L
M
o
u
tp
u
ts
[
3
1
]
.
R
esu
lts
s
h
o
w
th
at
en
f
o
r
cin
g
citatio
n
s
u
b
s
tan
tially
r
ed
u
ce
s
u
n
s
u
p
p
o
r
ted
ju
d
g
m
en
ts
,
th
o
u
g
h
a
s
m
all
f
r
a
ctio
n
o
f
h
allu
cin
ated
d
ec
is
io
n
s
p
er
s
is
ts
,
co
n
s
is
ten
t
with
p
r
io
r
f
in
d
in
g
s
o
n
L
L
M
f
ac
t
u
ality
lim
its
.
3
.
4
.
New
s
v
er
if
ica
t
io
n
Fin
ally
,
claim
-
lev
el
ju
d
g
m
en
t
s
ar
e
ag
g
r
eg
ated
to
lab
el
ea
c
h
s
o
cial
n
ews
item
.
A
s
tr
ict
a
p
p
r
o
ac
h
is
tak
en
:
a
s
o
cial
n
ews
ar
ticle
is
m
ar
k
e
d
v
e
r
if
ied
o
n
ly
if
all
its
ch
ec
k
-
wo
r
t
h
y
claim
s
a
r
e
tr
u
e;
if
a
n
y
claim
is
f
alse,
th
e
ar
ticle
is
m
ar
k
ed
as
u
n
v
er
if
ied
.
T
h
is
co
n
ju
n
ctiv
e
r
u
le
m
ak
es
s
en
s
e
b
ec
au
s
e
o
n
e
f
alse
ass
er
t
io
n
is
en
o
u
g
h
to
d
ee
m
th
e
n
ews
u
n
r
eliab
le.
T
h
e
m
o
d
el’
s
ju
s
tific
atio
n
s
also
p
r
o
v
id
e
tr
a
n
s
p
ar
en
c
y
f
o
r
ea
ch
claim
.
I
n
th
e
d
ataset
an
n
o
tatio
n
,
h
u
m
a
n
an
n
o
tato
r
s
lik
ewise
lab
eled
ea
ch
claim
an
d
th
en
d
er
iv
e
d
th
e
ar
ticle
lab
el
b
y
th
e
s
am
e
r
u
le.
Fig
u
r
e
2
.
Pro
m
p
t d
esig
n
f
o
r
claim
v
er
if
icatio
n
task
4.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
4
.
1
.
Da
t
a
s
et
A
cu
s
to
m
d
ataset
is
co
n
s
tr
u
cted
f
o
r
ev
alu
atio
n
.
I
t
co
n
tain
s
1
,
7
6
5
Vietn
am
ese
n
ew
s
ar
ticles:
7
2
3
s
o
cial
-
m
e
d
ia
item
s
to
v
e
r
if
y
,
a
n
d
1
,
0
4
2
“tr
u
s
ted
”
m
ai
n
s
tr
ea
m
n
ews
s
er
v
in
g
as
ev
id
en
ce
.
Ar
ticles
s
p
an
No
v
em
b
er
–
Dec
em
b
er
2
0
2
4
a
n
d
co
v
er
e
v
en
ts
in
p
o
liti
cs,
d
i
s
aster
s
,
s
p
o
r
ts
,
an
d
en
ter
tain
m
en
t.
Fo
r
ea
c
h
s
o
cial
ar
ticle,
ato
m
ic
claim
s
ar
e
m
a
n
u
ally
an
n
o
tated
(
3
,
3
7
3
claim
s
in
to
tal)
a
n
d
v
er
if
ied
th
eir
tr
u
th
b
y
c
o
n
s
u
ltin
g
clu
s
ter
n
ews.
I
n
an
n
o
tatio
n
,
a
claim
was
m
ar
k
ed
tr
u
e
if
s
u
p
p
o
r
ted
b
y
ev
id
en
ce
,
o
r
f
a
ls
e
o
th
er
wis
e.
I
t
i
s
en
s
u
r
ed
th
at
ea
ch
claim
h
ad
,
o
n
av
er
ag
e
,
3
s
u
p
p
o
r
tin
g
ar
ticles
(
≥
3
0
0
wo
r
d
s
ea
ch
)
in
th
e
clu
s
ter
.
T
h
e
r
ich
er
ev
id
en
ce
s
et
c
o
n
tr
asts
with
s
o
m
e
p
u
b
lic
d
atasets
wh
er
e
o
n
ly
o
n
e
o
r
two
s
o
u
r
ce
s
a
r
e
g
iv
en
p
e
r
claim
.
T
h
e
an
n
o
tatio
n
s
y
ield
e
d
2
,
6
3
4
tr
u
e
claim
s
an
d
7
3
0
f
alse
o
n
es;
co
r
r
esp
o
n
d
i
n
g
ly
,
3
5
2
s
o
c
ial
ar
ticles
b
ec
am
e
v
er
if
ied
(
all
claim
s
tr
u
e)
,
an
d
3
7
1
b
e
ca
m
e
u
n
v
er
if
ied
(
≥
1
f
al
s
e
claim
)
.
I
t
is
em
p
h
asized
th
at
ex
is
tin
g
Vietn
am
ese
o
r
m
u
ltil
in
g
u
al
f
a
ct
-
ch
ec
k
d
atasets
lack
th
is
c
o
m
b
in
atio
n
o
f
s
o
cial
n
ews
with
m
u
ltip
le
h
ig
h
-
q
u
ality
r
ef
er
en
ce
s
.
Hen
c
e,
th
e
d
ataset
is
cr
ea
ted
s
p
ec
if
ically
f
o
r
th
is
s
tu
d
y
.
T
ab
le
1
s
u
m
m
ar
izes
its
k
ey
s
t
atis
tics
.
Fo
r
r
ep
r
o
d
u
cib
ilit
y
,
t
h
e
ar
ticles
ar
e
en
s
u
r
ed
to
co
m
e
f
r
o
m
late
2
0
2
4
s
o
th
ey
wer
e
n
o
t in
th
e
L
L
M
’
s
tr
ain
in
g
d
ata
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
S
o
cia
l n
ews
fa
ctu
a
lity v
erif
ica
tio
n
u
s
in
g
la
r
g
e
l
a
n
g
u
a
g
e
mo
d
els (
Tr
a
n
Du
c
Du
o
n
g
)
3149
I
n
ad
d
itio
n
,
two
o
th
er
d
ata
s
ets,
Vi
W
ik
iF
C
an
d
ViFact
C
h
ec
k
,
ar
e
also
ch
o
s
en
as
ev
alu
atio
n
b
en
ch
m
ar
k
s
f
o
r
t
h
e
m
et
h
o
d
.
T
h
e
r
ea
s
o
n
s
f
o
r
ch
o
o
s
in
g
th
ese
d
atasets
ar
e
Vietn
am
ese
f
ac
t
-
c
h
ec
k
in
g
d
atasets
in
co
n
tex
ts
s
im
ilar
to
th
e
ta
s
k
a
n
d
well
-
estab
lis
h
ed
in
th
e
lite
r
atu
r
e.
ViW
ik
iFC
is
an
o
p
en
-
d
o
m
ain
Vietn
am
ese
W
ik
ip
ed
ia
f
ac
t
-
ch
ec
k
in
g
c
o
r
p
u
s
(
o
v
er
2
0
,
0
0
0
claim
s
)
[
3
2
]
,
wh
ile
ViFact
C
h
ec
k
is
a
Vietn
am
ese
m
u
lti
-
d
o
m
ain
n
ews
f
ac
t
-
ch
ec
k
in
g
d
ataset
wi
th
7
,
2
3
2
claim
s
[
2
5
]
.
B
o
th
a
r
e
m
an
u
ally
cu
r
ate
d
an
d
in
Viet
n
am
ese,
m
atch
in
g
th
e
lan
g
u
ag
e
a
n
d
s
ty
le
o
f
th
e
d
ata.
B
y
co
n
tr
ast,
p
o
p
u
la
r
E
n
g
lis
h
b
en
ch
m
a
r
k
s
(
e.
g
.
,
FEVE
R
o
n
E
n
g
lis
h
W
ik
ip
ed
ia)
o
p
e
r
ate
o
n
d
if
f
er
en
t
lan
g
u
ag
es
an
d
d
o
m
ain
s
[
5
]
a
n
d
w
o
u
ld
n
o
t
r
ef
lect
V
ietn
am
ese
-
s
p
ec
if
ic
lin
g
u
is
tic
p
h
en
o
m
e
n
a.
T
ab
le
1
.
Data
s
et
s
tatis
tic
s
Ty
p
e
Q
u
a
n
t
i
t
y
S
o
c
i
a
l
n
e
w
s
7
2
3
N
u
mb
e
r
o
f
e
x
t
r
a
c
t
e
d
c
l
a
i
ms
3
,
3
7
3
N
u
mb
e
r
o
f
c
l
a
i
ms
v
e
r
i
f
i
e
d
a
s
“
t
r
u
e
”
2
,
6
4
3
N
u
mb
e
r
o
f
c
l
a
i
ms
v
e
r
i
f
i
e
d
a
s
“
f
a
l
se
”
7
3
0
N
u
mb
e
r
o
f
n
e
w
s c
l
a
ss
i
f
i
e
d
a
s
t
r
u
e
(
a
l
l
c
l
a
i
ms
a
r
e
t
r
u
e
)
3
5
2
N
u
mb
e
r
o
f
n
e
w
s c
l
a
ss
i
f
i
e
d
a
s f
a
l
se
(
a
t
l
e
a
st
o
n
e
c
l
a
i
m
i
s
f
a
l
s
e
)
3
7
1
N
e
w
s fr
o
m
t
r
u
s
t
e
d
s
o
u
r
c
e
s
1
,
0
4
2
To
t
a
l
n
e
w
s
1
,
7
6
5
4
.
2
.
E
x
perim
ent
a
l
s
et
up
E
x
p
er
im
en
ts
ar
e
s
tr
u
ct
u
r
ed
a
r
o
u
n
d
k
ey
r
esear
ch
q
u
esti
o
n
s
:
i)
R
Q1
:
a
r
e
L
L
Ms
ef
f
ec
tiv
e
at
s
o
cial
n
ews
v
er
if
icatio
n
?
th
e
L
L
M
p
ip
elin
e
is
c
o
m
p
ar
e
d
a
g
ain
s
t
a
s
tr
o
n
g
b
aselin
e:
a
B
E
R
T
-
b
ased
m
o
d
el
(
a
r
eim
p
lem
en
tatio
n
o
f
a
Vietn
am
ese
f
ak
e
-
n
ews
d
etec
to
r
[
1
0
]
)
tr
ain
ed
t
o
class
if
y
ar
ticles
d
ir
ec
tly
.
T
h
is
s
im
u
lates
“tr
ad
itio
n
al”
n
atu
r
al
lan
g
u
ag
e
p
r
o
ce
s
s
in
g
(
NL
P
)
v
er
if
icatio
n
.
Pre
cisi
o
n
,
r
ec
all,
an
d
F1
-
s
co
r
e
ar
e
r
ep
o
r
ted
f
o
r
claim
-
lev
el
a
n
d
ar
ticle
-
lev
el
lab
els
.
ii)
R
Q2
:
d
o
es
claim
e
x
tr
ac
tio
n
h
elp
?
th
e
s
am
e
L
L
M
v
er
if
icati
o
n
v
er
if
icatio
n
is
test
ed
wh
e
n
ap
p
lied
to
f
u
ll
ar
ticles
(
with
o
u
t
claim
ex
tr
ac
t
io
n
)
v
e
r
s
u
s
ato
m
ic
claim
s
.
Pr
io
r
wo
r
k
s
u
g
g
ests
v
er
if
y
in
g
c
o
n
cise
claim
s
y
ield
s
m
o
r
e
r
eliab
le
r
esu
lts
[
1
6
]
.
iii)
R
Q3
:
c
lu
s
ter
in
g
v
s
.
to
p
-
N
r
etr
iev
al:
th
e
clu
s
ter
in
g
-
b
ased
ap
p
r
o
ac
h
is
co
m
p
ar
ed
to
a
s
im
p
ler
s
ch
em
e
:
f
o
r
ea
ch
s
o
cial
ar
ticle,
u
s
e
an
o
n
lin
e
s
ea
r
ch
ap
p
licatio
n
p
r
o
g
r
a
m
m
in
g
in
ter
f
ac
e
(
API
)
to
r
etr
iev
e
th
e
to
p
-
3
p
o
ten
tially
r
elate
d
n
ews
(
ig
n
o
r
in
g
clu
s
ter
in
g
)
,
th
e
n
v
e
r
if
y
claim
s
ag
ain
s
t
th
o
s
e.
T
h
is
ev
alu
ates
th
e
b
en
ef
it o
f
o
f
f
lin
e
clu
s
ter
in
g
o
f
all
co
llected
n
ews.
I
n
all
L
L
M
ex
p
e
r
im
en
ts
,
th
e
GPT
-
4
m
o
d
el
is
u
s
ed
.
C
laim
s
an
d
cl
u
s
ter
n
ews
ar
e
f
ed
v
ia
a
L
an
g
C
h
ai
n
in
ter
f
ac
e.
Fo
r
th
e
claim
-
ex
tr
ac
tio
n
p
ip
elin
e,
a
f
ew
-
s
h
o
t
p
r
o
m
p
t
with
GPT
-
4
is
u
s
ed
f
o
r
ato
m
ic
claim
s
(
s
im
ilar
to
th
e
C
laim
if
y
m
eth
o
d
[
1
6
]
).
4
.
3
.
Resul
t
s
T
o
ass
ess
th
e
q
u
ality
o
f
L
L
M
-
b
ased
claim
ex
tr
ac
tio
n
,
ex
tr
a
cted
claim
s
ar
e
m
an
u
all
y
ev
al
u
ated
o
n
a
r
an
d
o
m
s
u
b
s
et
o
f
2
0
0
s
o
cial
n
ews
p
o
s
ts
,
s
tr
atif
ied
ac
r
o
s
s
t
o
p
ics.
E
ac
h
p
o
s
t
was
an
n
o
tated
in
d
ep
en
d
en
tly
b
y
two
h
u
m
a
n
a
n
n
o
tato
r
s
with
e
x
p
er
ien
ce
in
m
is
in
f
o
r
m
atio
n
a
n
aly
s
is
.
Fo
r
ea
ch
p
o
s
t,
an
n
o
ta
to
r
s
f
ir
s
t
id
en
tifie
d
th
e
s
et
o
f
g
o
ld
at
o
m
ic
f
ac
t
u
al
claim
s
p
r
esen
t
in
th
e
tex
t.
E
x
tr
ac
ted
claim
s
wer
e
th
en
m
a
tch
ed
ag
ai
n
s
t
g
o
ld
claim
s
u
s
in
g
s
em
an
tic
e
q
u
iv
a
len
ce
cr
iter
ia.
Pre
cisi
o
n
,
r
ec
a
ll,
an
d
F1
-
s
co
r
e
a
r
e
r
ep
o
r
te
d
in
T
a
b
le
2
.
T
h
ese
r
esu
lts
in
d
icate
th
at
th
e
L
L
M
ex
tr
ac
ts
m
o
s
t
v
er
if
iab
le
cl
aim
s
wh
ile
m
ain
tain
in
g
h
ig
h
p
r
ec
is
io
n
,
wh
ich
is
cr
u
cial
s
in
ce
d
o
wn
s
tr
ea
m
v
er
i
f
icatio
n
co
s
t g
r
o
ws lin
ea
r
ly
with
th
e
n
u
m
b
er
o
f
ex
tr
ac
te
d
clai
m
s
.
T
ab
le
2
.
C
laim
ex
tr
ac
tio
n
ev
al
u
atio
n
M
e
t
r
i
c
S
c
o
r
e
(
%)
P
r
e
c
i
s
i
o
n
8
9
.
7
R
e
c
a
l
l
8
7
.
5
F1
-
sc
o
r
e
8
8
.
6
All
ex
tr
ac
tio
n
e
r
r
o
r
s
a
r
e
ca
te
g
o
r
ized
in
to
f
o
u
r
m
u
tu
ally
e
x
clu
s
iv
e
ty
p
es
.
Miss
in
g
f
ac
ts
a
cc
o
u
n
t
f
o
r
4
2
%
o
f
e
r
r
o
r
s
,
w
h
er
e
th
e
m
o
d
el
ca
p
tu
r
es
th
e
m
ai
n
claim
b
u
t
o
m
its
s
ec
o
n
d
ar
y
v
er
if
ia
b
le
d
e
tails
(
e.
g
.
,
d
ates
o
r
n
u
m
er
ical
v
alu
es),
p
r
im
ar
ily
af
f
ec
tin
g
r
ec
all.
Ov
er
-
s
p
litt
in
g
o
r
r
ed
u
n
d
a
n
cy
r
ep
r
esen
ts
2
8
%,
wh
er
e
a
s
in
g
le
f
ac
tu
al
s
tatem
en
t
is
f
r
a
g
m
en
ted
in
to
o
v
e
r
lap
p
in
g
claim
s
,
co
m
p
licatin
g
alig
n
m
en
t
wit
h
g
o
ld
a
n
n
o
tatio
n
s
.
I
n
clu
d
in
g
o
p
in
i
o
n
s
o
r
s
u
b
jecti
v
e
s
tatem
en
ts
ac
co
u
n
ts
f
o
r
1
8
%
o
f
er
r
o
r
s
an
d
n
eg
ativ
ely
im
p
ac
ts
p
r
ec
is
io
n
.
T
h
e
r
em
ain
in
g
1
2
%
in
v
o
l
v
e
co
n
te
x
tu
al
in
co
m
p
leten
ess
o
r
en
tity
am
b
ig
u
ity
,
w
h
er
e
ex
tr
ac
te
d
cl
aim
s
lack
s
u
f
f
icien
t
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.
15
,
No
.
4
,
Au
g
u
s
t 2
0
2
6
:
3
1
4
4
-
3
1
5
3
3150
s
p
ec
if
icity
.
T
h
is
d
is
tr
ib
u
tio
n
ex
p
lain
s
th
e
s
lig
h
tly
lo
wer
r
e
ca
ll
(
8
7
.
5
%)
co
m
p
ar
ed
to
p
r
e
cisi
o
n
(
8
9
.
7
%)
a
n
d
alig
n
s
with
ty
p
ical
er
r
o
r
p
att
er
n
s
r
ep
o
r
ted
i
n
r
ec
en
t
L
L
M
-
b
ased
claim
ex
tr
ac
tio
n
s
tu
d
ie
s
.
E
v
en
t
clu
s
ter
in
g
q
u
ality
is
ev
alu
ated
u
s
in
g
p
u
r
ity
an
d
n
o
r
m
alize
d
m
u
tu
al
in
f
o
r
m
atio
n
(
NM
I
)
o
n
a
m
an
u
ally
lab
eled
s
u
b
s
et
(
3
0
0
ar
ticles)
an
d
r
e
p
o
r
t
t
h
e
r
esu
lts
in
T
ab
le
3
.
Hig
h
p
u
r
ity
co
n
f
ir
m
s
th
at
e
v
id
en
ce
clu
s
ter
s
ar
e
s
em
an
tically
co
h
er
en
t,
w
h
ich
d
ir
ec
tly
b
en
ef
its
v
er
if
icatio
n
ac
cu
r
ac
y
.
M
ac
r
o
-
av
er
a
g
ed
p
r
ec
is
io
n
,
r
ec
all,
an
d
F1
-
s
co
r
e
f
o
r
claim
v
e
r
if
icatio
n
an
d
ar
ticle
(
n
ews)
v
er
if
icatio
n
ar
e
r
ep
o
r
te
d
in
T
ab
le
4
.
T
o
p
r
ev
en
t
in
f
o
r
m
atio
n
leak
ag
e
,
an
ev
en
t
-
lev
el
tr
ain
/tes
t
s
p
lit
is
a
p
p
lied
.
All
ar
ticles
ar
e
f
ir
s
t
g
r
o
u
p
ed
in
t
o
ev
en
t
cl
u
s
ter
s
,
an
d
en
tire
clu
s
ter
s
ar
e
ass
ig
n
ed
ex
clu
s
iv
ely
to
eith
er
tr
ain
in
g
o
r
test
s
ets.
T
h
is
en
s
u
r
es
th
at
n
o
e
v
en
t
a
p
p
ea
r
s
in
b
o
th
s
p
lits
.
Fo
r
L
L
M
-
b
ased
ze
r
o
-
s
h
o
t
ev
alu
atio
n
,
n
ea
r
-
d
u
p
licate
co
n
ten
t
b
etwe
en
s
o
cial
p
o
s
ts
an
d
tr
u
s
ted
ar
ticles
i
s
f
u
r
th
er
r
em
o
v
e
d
u
s
in
g
s
tr
in
g
s
im
ilar
ity
f
ilte
r
in
g
,
en
s
u
r
in
g
th
at
ev
id
en
ce
is
n
o
t tr
i
v
ially
co
p
ie
d
.
‒
B
aselin
e
(
B
E
R
T
)
:
th
e
B
E
R
T
m
o
d
el
(
Sem
ViQA
ar
ch
itectu
r
e
ad
ap
ted
to
Vietn
am
ese
[
1
0
]
)
ac
h
iev
ed
an
F1
-
s
co
r
e
o
f
o
n
ly
7
7
.
7
%
o
n
claim
-
lev
el
lab
els
an
d
7
4
.
6
%
o
n
n
ews
-
lev
el
lab
els.
T
h
is
r
elativ
ely
lo
w
p
er
f
o
r
m
an
ce
h
i
g
h
lig
h
ts
th
e
c
h
allen
g
e
o
f
ca
p
tu
r
in
g
ev
id
e
n
ce
with
o
u
t c
o
n
tex
t.
‒
L
L
M
(
n
o
ex
t
r
ac
tio
n
)
:
v
e
r
if
y
i
n
g
f
u
ll
ar
ticles
with
GPT
-
4
(
u
s
in
g
all
clu
s
ter
tex
t)
g
iv
es
a
n
F1
-
s
co
r
e
o
f
6
9
.
8
%
o
n
n
ews
-
lev
el
lab
els
(
n
o
claim
s
ar
e
ex
tr
ac
ted
,
s
o
t
h
er
e
is
n
o
r
esu
lt
o
n
claim
-
le
v
el
lab
els).
T
h
e
n
o
t
g
o
o
d
r
esu
lt m
ay
b
e
s
u
f
f
er
e
d
f
r
o
m
v
er
b
o
s
ity
an
d
co
n
f
u
s
io
n
in
lo
n
g
in
p
u
ts
.
‒
L
L
M
(
with
ex
tr
ac
tio
n
,
to
p
-
N
s
ea
r
ch
)
:
wh
en
claim
s
ar
e
f
ir
s
t
ex
tr
ac
ted
an
d
th
e
to
p
3
r
e
lev
an
t
clu
s
ter
ar
ticles
as
ev
id
en
ce
(
HiSS
-
s
ty
le
p
r
o
m
p
t
[
1
2
]
)
,
F1
-
s
co
r
e
in
cr
ea
s
es
to
8
6
.
7
%
o
n
claim
-
lev
el
lab
els
an
d
8
9
.
6
% o
n
n
ews
-
lev
el
lab
els.
T
h
is
s
h
o
ws
th
at
lim
itin
g
co
n
tex
t
to
h
ig
h
-
q
u
ality
s
o
u
r
ce
s
h
elp
s
.
‒
L
L
M
(
with
ex
tr
ac
tio
n
,
clu
s
ter
i
n
g
)
:
th
e
f
u
ll p
ip
elin
e
(
claim
s
+
L
L
M
+
clu
s
ter
ev
id
en
ce
)
ac
h
i
ev
ed
~8
8
.
9
%
F1
-
s
co
r
e
o
n
claim
s
an
d
9
2
.
1
%
F1
-
s
co
r
e
o
n
n
ews.
T
h
is
w
as
th
e
b
est
r
esu
lt,
f
ar
s
u
r
p
ass
in
g
th
e
B
E
R
T
b
aselin
e
an
d
th
e
n
o
n
-
clu
s
ter
v
ar
ian
ts
.
I
n
p
a
r
ticu
lar
,
r
ec
all
of
f
alse
claim
s
was
h
ig
h
,
m
ea
n
i
n
g
m
o
s
t
f
als
e
s
tatem
en
ts
wer
e
ca
u
g
h
t.
T
h
ese
r
esu
lts
co
n
f
ir
m
r
ec
en
t
f
in
d
in
g
s
th
at
well
-
p
r
o
m
p
ted
L
L
Ms
ca
n
m
atch
o
r
ex
c
ee
d
s
u
p
er
v
is
ed
m
o
d
els o
n
f
ac
t
-
ch
ec
k
in
g
[
1
2
]
.
T
h
e
u
s
ag
e
o
f
GPT
-
4
with
co
n
t
ex
tu
al
clu
s
ter
ev
id
en
ce
y
ield
e
d
s
u
b
s
tan
tial g
ain
s
.
I
t
also
v
alid
ates
th
at
clu
s
ter
in
g
is
s
u
p
er
io
r
to
n
aïv
ely
r
etr
ie
v
in
g
f
ix
ed
to
p
-
N
s
o
u
r
ce
s
:
clu
s
ter
in
g
en
s
u
r
ed
th
at
n
o
r
elev
a
n
t
ev
id
e
n
ce
was
m
i
s
s
ed
an
d
th
at
o
f
f
-
to
p
ic
co
n
te
n
t
was
n
o
t
in
clu
d
e
d
.
I
n
ex
p
e
r
im
en
ts
,
clu
s
ter
in
g
im
p
r
o
v
e
d
p
r
ec
is
io
n
b
y
s
ev
er
al
p
o
in
ts
o
v
er
to
p
-
N
r
etr
iev
al
f
o
r
th
e
s
am
e
n
u
m
b
e
r
o
f
s
o
u
r
ce
s
.
As
ex
p
ec
ted
,
th
e
co
m
p
ar
is
o
n
also
s
h
o
ws
th
e
ad
v
an
tag
e
o
f
claim
ex
tr
ac
tio
n
:
v
er
if
y
in
g
co
n
cise
at
o
m
ic
s
tatem
en
ts
s
ig
n
if
ican
tly
o
u
tp
er
f
o
r
m
ed
ev
alu
atin
g
wh
o
l
e
ar
ticles,
co
r
r
o
b
o
r
atin
g
t
h
e
b
e
n
ef
it o
f
f
o
cu
s
ed
claim
d
ec
o
m
p
o
s
itio
n
[
1
6
]
.
T
ab
le
5
r
ep
o
r
ts
r
esu
lts
o
n
th
e
t
est
s
p
lit
o
f
ea
ch
b
en
ch
m
ar
k
.
A
s
tr
o
n
g
b
aselin
e
m
o
d
el
(
I
n
f
o
X
L
M
-
lar
g
e
o
n
ViW
ik
iFC
[
3
2
]
;
Gem
m
a
o
n
ViFactC
h
ec
k
[
2
5
]
)
is
co
m
p
a
r
ed
to
th
e
p
r
o
p
o
s
ed
m
et
h
o
d
.
E
ac
h
s
co
r
e
is
s
h
o
wn
with
a
9
5
%
co
n
f
id
en
ce
in
ter
v
al,
an
d
we
in
clu
d
e
a
p
-
v
alu
e
f
r
o
m
a
p
air
e
d
test
to
ch
ec
k
s
ig
n
if
ican
ce
.
T
h
ese
r
esu
lts
s
h
o
w
th
at
th
e
m
eth
o
d
co
n
s
is
ten
tly
s
lig
h
tly
o
u
tp
er
f
o
r
m
s
th
e
b
aselin
es
o
n
b
o
th
d
atasets
(
alb
eit
b
y
o
n
ly
1
–
2
p
o
in
ts
)
,
wh
ile
th
e
c
o
n
f
id
e
n
ce
in
ter
v
als q
u
a
n
tify
th
e
r
eliab
ilit
y
o
f
ea
ch
s
co
r
e.
T
ab
le
3
.
C
lu
s
ter
p
u
r
ity
an
d
N
MI
M
e
t
r
i
c
S
c
o
r
e
(
%)
P
u
r
i
t
y
9
1
.
4
N
M
I
8
6
.
7
T
ab
le
4
.
E
x
p
er
im
en
t
r
esu
lts
M
e
t
h
o
d
C
l
a
i
m
v
e
r
i
f
i
c
a
t
i
o
n
N
e
w
s v
e
r
i
f
i
c
a
t
i
o
n
P
r
e
c
i
s
i
o
n
R
e
c
a
l
l
F1
-
sc
o
r
e
P
r
e
c
i
s
i
o
n
R
e
c
a
l
l
F1
-
sc
o
r
e
B
ER
T
-
b
a
se
d
b
a
se
l
i
n
e
(
S
e
mV
i
Q
A
[
1
0
]
)
7
5
.
3
8
0
.
2
7
7
.
7
6
3
.
3
9
0
.
7
7
4
.
6
LLM
(
n
o
e
x
t
r
a
c
t
i
o
n
)
(
H
i
S
S
[
1
2
]
)
6
3
.
6
7
7
.
3
6
9
.
8
LLM
-
b
a
s
e
d
w
i
t
h
t
o
p
N
t
r
u
st
e
d
n
e
w
s
9
3
.
6
8
0
.
8
8
6
.
7
9
5
.
3
8
4
.
5
8
9
.
6
LLM
-
b
a
s
e
d
w
i
t
h
n
e
w
s
c
l
u
s
t
e
r
i
n
g
9
3
.
1
8
5
.
1
8
8
.
9
9
3
.
6
9
0
.
7
9
2
.
1
T
ab
le
5
.
Ma
cr
o
-
F1
(
in
%)
o
n
Vietn
am
ese
f
ac
t
-
ch
ec
k
in
g
b
en
ch
m
ar
k
s
D
a
t
a
s
e
t
B
a
se
l
i
n
e
m
o
d
e
l
B
a
se
l
i
n
e
F
1
P
r
o
p
o
se
d
m
e
t
h
o
d
V
i
W
i
k
i
F
C
I
n
f
o
X
LM
-
l
a
r
g
e
6
7
.
0
6
8
.
2
±
0
.
5
V
i
F
a
c
t
C
h
e
c
k
G
e
mm
a
(
L
LM
)
8
9
.
9
9
0
.
7
±
0
.
4
T
h
e
r
esu
lts
in
T
ab
le
5
in
d
icate
th
at
th
e
m
eth
o
d
ac
h
iev
es
a
s
lig
h
t
im
p
r
o
v
e
m
en
t
o
v
e
r
th
e
b
a
s
elin
es
o
n
b
o
th
Vietn
am
ese
b
en
ch
m
a
r
k
s
.
On
ViW
ik
iF
C
,
I
n
f
o
XL
M
-
lar
g
e
s
co
r
ed
ab
o
u
t
6
7
.
0
%
f
o
r
p
i
p
elin
e
(
co
m
b
i
n
atio
n
o
f
b
o
th
task
s
)
,
an
d
p
r
o
p
o
s
e
d
m
eth
o
d
s
co
r
es
~6
8
.
2
%
(
±
0
.
5
)
.
On
ViFactC
h
ec
k
,
Gem
m
a
s
co
r
ed
8
9
.
9
0
%,
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
S
o
cia
l n
ews
fa
ctu
a
lity v
erif
ica
tio
n
u
s
in
g
la
r
g
e
l
a
n
g
u
a
g
e
mo
d
els (
Tr
a
n
Du
c
Du
o
n
g
)
3151
wh
ile
th
e
p
r
o
p
o
s
ed
m
eth
o
d
r
e
ac
h
es
~9
0
.
7
%
(
±
0
.
4
)
.
T
h
e
d
if
f
er
en
ce
s
(
≈
1
.
2
an
d
0
.
8
p
o
i
n
ts
)
ar
e
s
m
all,
b
u
t
th
e
s
tatis
t
ical
test
s
co
n
f
ir
m
th
ey
ar
e
u
n
lik
ely
to
b
e
d
u
e
to
r
an
d
o
m
ch
an
ce
(
p
<0
.
0
5
)
.
R
ep
o
r
tin
g
co
n
f
id
en
c
e
in
ter
v
als
h
elp
s
to
s
ee
th
at
th
e
in
ter
v
als
b
ar
ely
o
v
er
lap
,
s
u
p
p
o
r
tin
g
t
h
e
claim
o
f
a
r
ea
l
g
ain
.
T
h
ese
s
m
all
g
ain
s
ar
e
ex
p
ec
ted
,
g
i
v
en
th
at
th
e
b
aselin
es
ar
e
alr
ea
d
y
s
tr
o
n
g
.
T
h
e
k
ey
p
o
i
n
t
is
th
at
p
r
o
p
o
s
ed
ap
p
r
o
ac
h
p
r
o
v
id
es
co
n
s
is
ten
tly
h
ig
h
er
p
er
f
o
r
m
a
n
ce
in
th
e
s
am
e
Vietn
am
ese
co
n
tex
t.
T
h
is
s
u
g
g
ests
p
r
o
p
o
s
ed
m
eth
o
d
b
etter
h
an
d
les Vietn
am
ese
-
s
p
ec
if
ic
lin
g
u
is
tic
p
atter
n
s
o
r
d
o
m
ain
k
n
o
wled
g
e.
4
.
4
.
Dis
cus
s
io
n
W
h
ile
th
e
p
ip
elin
e
ac
h
iev
ed
h
ig
h
ac
cu
r
ac
y
,
s
ev
er
al
p
r
ac
tica
l
ch
allen
g
es
r
em
ain
.
First,
am
b
ig
u
ity
in
claim
s
ca
n
s
tu
m
p
th
e
L
L
M:
s
o
m
e
s
tatem
en
ts
r
eq
u
ir
ed
n
u
an
c
ed
wo
r
ld
k
n
o
wled
g
e
o
r
co
n
tex
t
b
ey
o
n
d
wh
at
was
p
r
o
v
id
e
d
.
Fo
r
ex
am
p
le,
a
claim
ab
o
u
t
a
p
o
licy
an
n
o
u
n
ce
m
en
t
with
o
u
t
n
a
m
ed
p
ar
ticip
an
ts
n
ee
d
ed
cr
o
s
s
-
ch
ec
k
in
g
.
I
n
teg
r
atin
g
ex
ter
n
al
k
n
o
wled
g
e
(
e.
g
.
,
W
ik
ip
ed
ia
o
r
a
KG
)
co
u
ld
h
el
p
d
is
am
b
ig
u
ate
s
u
ch
claim
s
.
I
n
d
ee
d
,
s
u
r
v
ey
s
s
h
o
w
th
at
L
L
Ms
b
e
n
ef
it
f
r
o
m
s
tr
u
ctu
r
ed
k
n
o
wled
g
e
s
o
u
r
ce
s
to
im
p
r
o
v
e
f
ac
tu
al
g
r
o
u
n
d
in
g
[
3
3
]
.
Seco
n
d
,
c
o
n
f
l
ictin
g
s
o
u
r
ce
s
with
in
a
clu
s
ter
ca
n
lead
to
in
co
n
s
is
ten
t
v
er
d
icts
.
I
n
a
f
ew
ca
s
es,
two
r
ep
u
tab
le
o
u
tlets
q
u
o
ted
c
o
n
tr
ad
icto
r
y
f
ig
u
r
es
f
o
r
an
e
v
e
n
t.
T
o
m
itig
ate
th
is
,
o
n
e
co
u
ld
weig
h
t
s
o
u
r
ce
s
b
y
r
eliab
ilit
y
(
e.
g
.
,
f
av
o
r
in
g
g
o
v
er
n
m
en
t
o
r
m
ajo
r
p
r
ess
)
.
I
n
t
h
e
ex
p
e
r
im
en
ts
,
th
e
L
L
M
s
o
m
etim
es
ar
b
itra
r
ily
ch
o
s
e
o
n
e
s
o
u
r
ce
,
s
o
h
u
m
a
n
-
in
-
th
e
-
l
o
o
p
weig
h
tin
g
o
r
a
n
en
s
em
b
le
o
f
L
L
M
ju
d
g
m
en
ts
co
u
ld
in
c
r
ea
s
e
r
o
b
u
s
tn
ess
.
T
h
ir
d
,
s
ca
lab
ilit
y
an
d
laten
cy
ar
e
c
o
n
ce
r
n
s
.
R
u
n
n
in
g
GPT
-
4
o
n
m
a
n
y
claim
s
is
co
m
p
u
tatio
n
ally
ex
p
en
s
iv
e,
p
o
s
in
g
ch
allen
g
es
f
o
r
r
ea
l
-
tim
e
d
ep
lo
y
m
en
t.
R
e
ce
n
t
in
d
u
s
tr
y
p
r
ac
tice
s
u
g
g
est
s
m
o
d
el
d
is
till
atio
n
o
r
q
u
a
n
tizatio
n
tech
n
iq
u
es
to
s
p
ee
d
u
p
i
n
f
er
en
ce
.
Alter
n
ativ
ely
,
a
h
y
b
r
id
ap
p
r
o
ac
h
m
ig
h
t
u
s
e
a
f
aster
,
s
m
aller
L
L
M
f
o
r
ea
s
y
ca
s
es
an
d
r
ese
r
v
e
GPT
-
4
f
o
r
am
b
ig
u
o
u
s
cl
aim
s
.
T
h
e
FASTT
R
A
C
K
f
r
am
ewo
r
k
s
h
o
ws
th
at
o
f
f
lin
e
clu
s
ter
in
g
ca
n
d
r
am
ati
ca
lly
r
ed
u
ce
q
u
er
y
lo
ad
,
an
d
t
h
e
r
esu
lts
alig
n
with
th
eir
f
in
d
in
g
th
at
cl
u
s
ter
in
g
s
ca
les to
ten
s
o
f
th
o
u
s
an
d
s
o
f
d
o
cu
m
e
n
ts
wh
ile
r
etain
in
g
h
i
g
h
ac
cu
r
ac
y
[
2
4
]
.
Fin
ally
,
m
u
ltil
in
g
u
alis
m
an
d
k
n
o
wled
g
e
in
teg
r
atio
n
ar
e
p
r
o
m
is
in
g
f
u
t
u
r
e
d
i
r
ec
tio
n
s
.
T
h
e
cu
r
r
e
n
t
wo
r
k
f
o
cu
s
es
o
n
Vietn
am
ese
co
n
ten
t,
b
u
t
th
e
ar
c
h
itectu
r
e
c
an
n
atu
r
ally
ex
te
n
d
:
m
a
n
y
L
L
Ms
ar
e
m
u
ltil
in
g
u
al,
an
d
ev
id
en
ce
clu
s
ter
s
co
u
ld
i
n
clu
d
e
s
o
u
r
ce
s
in
an
y
lan
g
u
a
g
e.
Fo
r
in
s
tan
ce
,
ch
ec
k
in
g
a
Vietn
am
ese
claim
ag
ain
s
t
E
n
g
lis
h
W
ik
ip
ed
ia
v
ia
GPT’
s
tr
an
s
latio
n
ab
ilit
y
.
T
h
i
s
cr
o
s
s
-
lin
g
u
al
in
teg
r
atio
n
co
u
ld
f
u
r
th
er
im
p
r
o
v
e
co
v
er
ag
e.
Su
r
v
e
y
s
n
o
te
t
h
at
co
m
b
in
in
g
L
L
Ms
with
d
o
m
ai
n
o
n
to
l
o
g
ies
an
d
k
n
o
wled
g
e
b
ases
ca
n
en
h
an
ce
r
ea
s
o
n
in
g
[
3
3
]
;
p
lan
to
e
x
p
l
o
r
e
co
n
n
ec
tin
g
th
e
p
ip
elin
e
t
o
m
ed
ical,
f
in
an
cial,
o
r
g
eo
g
r
ap
h
ic
KG
s
wh
en
v
er
if
y
in
g
s
p
ec
ialized
n
ews.
T
h
e
p
r
o
p
o
s
ed
f
ac
tu
ality
v
er
if
ica
tio
n
f
r
a
m
ewo
r
k
ca
n
b
e
i
n
teg
r
a
ted
in
to
n
ewsr
o
o
m
s
an
d
s
o
cial
-
m
ed
ia
m
o
d
e
r
atio
n
s
y
s
tem
s
th
r
o
u
g
h
API
s
th
at
s
u
p
p
o
r
t
ev
en
t
clu
s
ter
in
g
,
claim
ex
tr
ac
tio
n
,
an
d
ev
id
en
ce
r
etr
iev
al.
I
n
p
r
ac
tice,
th
e
m
o
d
el
f
u
n
ctio
n
s
as a
h
u
m
an
-
in
-
th
e
-
l
o
o
p
ass
is
tan
t,
au
to
m
atica
lly
g
en
er
atin
g
v
er
if
icatio
n
s
u
m
m
a
r
ies
an
d
c
o
n
f
id
en
ce
s
co
r
es
th
at
ed
ito
r
s
o
r
f
ac
t
-
c
h
ec
k
er
s
ca
n
r
ev
iew
b
ef
o
r
e
p
u
b
licatio
n
.
T
h
is
h
y
b
r
i
d
d
esig
n
i
n
cr
ea
s
es
ef
f
icien
cy
w
h
ile
m
ain
tain
in
g
ed
ito
r
ial
c
o
n
tr
o
l
a
n
d
ac
c
o
u
n
tab
ilit
y
(
a
d
em
o
v
er
s
io
n
is
at
h
ttp
s
://8
am
.
v
n
).
T
o
en
s
u
r
e
eth
ical
an
d
tr
u
s
two
r
th
y
d
e
p
lo
y
m
e
n
t,
p
o
ten
tial
lan
g
u
ag
e
an
d
to
p
ic
b
iases
ar
e
m
itig
ated
th
r
o
u
g
h
b
alan
ce
d
d
ataset
f
in
e
-
t
u
n
in
g
an
d
p
er
io
d
ic
au
d
its
o
f
m
o
d
el
o
u
t
p
u
ts
.
All
r
etr
iev
ed
ev
id
en
ce
is
tr
ac
ea
b
le
to
p
u
b
lic
s
o
u
r
ce
s
,
an
d
u
s
er
d
ata
is
n
ev
e
r
s
to
r
ed
.
B
y
co
m
b
in
in
g
s
ca
lab
le
au
to
m
atio
n
with
tr
an
s
p
ar
en
cy
an
d
h
u
m
a
n
o
v
e
r
s
ig
h
t,
th
e
s
y
s
tem
s
u
p
p
o
r
ts
r
esp
o
n
s
ib
le
u
s
e
o
f
L
L
M
-
b
ased
f
ac
t
-
ch
ec
k
in
g
in
r
ea
l
-
wo
r
ld
n
ews e
n
v
ir
o
n
m
en
ts
.
5.
CO
NCLU
SI
O
N
A
r
o
b
u
s
t
f
r
a
m
e
w
o
r
k
f
o
r
v
er
if
y
i
n
g
th
e
f
a
c
t
u
a
l
i
ty
o
f
s
o
c
i
a
l
-
m
e
d
ia
n
e
w
s
b
y
l
e
v
er
ag
i
n
g
L
L
M
s
a
n
d
e
v
e
n
t
-
b
a
s
ed
c
l
u
s
t
er
i
n
g
h
a
s
b
ee
n
p
r
e
s
e
n
te
d
.
T
h
e
c
o
n
t
r
i
b
u
t
io
n
s
i
n
c
l
u
d
e
d
em
o
n
s
t
r
a
t
i
n
g
t
h
a
t
G
P
T
-
4
,
w
h
e
n
g
u
id
e
d
b
y
h
ig
h
-
q
u
a
l
i
ty
e
v
id
e
n
c
e
c
lu
s
t
e
r
s
a
n
d
a
to
m
i
c
c
l
a
i
m
s
,
c
a
n
g
r
e
a
t
l
y
o
u
tp
e
r
f
o
r
m
t
r
ad
i
t
i
o
n
a
l
B
E
R
T
-
b
a
s
e
d
c
la
s
s
i
f
i
e
r
s
o
n
a
V
i
e
t
n
a
m
e
s
e
n
e
w
s
v
e
r
i
f
ic
a
ti
o
n
t
a
s
k
.
N
e
w
s
c
l
u
s
t
er
i
n
g
wa
s
s
h
o
w
n
to
b
e
a
p
o
w
er
f
u
l
s
t
r
a
t
eg
y
f
o
r
e
v
id
e
n
c
e
r
e
tr
i
e
v
a
l
,
y
i
e
ld
i
n
g
m
o
r
e
r
e
l
e
v
an
t
co
n
t
ex
t
th
a
n
s
i
m
p
le
t
o
p
-
N
s
e
a
r
ch
.
T
h
e
h
i
g
h
ac
c
u
r
a
cy
ac
h
i
ev
e
d
(
≈
9
0
–
9
2
%
F
1
-
s
c
o
r
e
)
u
n
d
e
r
s
c
o
r
e
s
t
h
e
p
o
te
n
t
i
a
l
o
f
A
I
-
d
r
iv
en
f
ac
t
-
c
h
ec
k
in
g
.
A
t
t
h
e
s
a
m
e
t
i
m
e
,
k
e
y
l
i
m
i
t
a
t
i
o
n
s
a
r
e
i
d
e
n
t
if
i
e
d
:
L
L
M
i
n
f
e
r
en
c
e
c
o
s
t
s
c
h
a
l
l
en
g
e
r
e
a
l
-
t
im
e
u
s
e
,
a
n
d
a
m
b
i
g
u
o
u
s
o
r
u
n
s
u
p
p
o
r
t
ed
c
l
a
i
m
s
c
a
n
s
t
i
l
l
s
l
i
p
t
h
r
o
u
g
h
.
Fu
t
u
r
e
r
e
s
e
ar
ch
w
i
l
l
ad
d
r
e
s
s
t
h
e
s
e
b
y
i
n
co
r
p
o
r
a
t
in
g
e
f
f
i
c
i
en
t
m
o
d
e
l
s
er
v
in
g
(
d
i
s
t
i
l
l
a
t
i
o
n
,
ca
c
h
i
n
g
)
an
d
r
i
ch
er
k
n
o
w
l
e
d
g
e
s
o
u
r
c
e
s
(
o
n
to
l
o
g
i
e
s
,
KG
s
)
f
o
r
d
i
s
a
m
b
i
g
u
a
t
io
n
.
A
l
s
o
p
l
an
n
e
d
i
s
ex
p
an
s
i
o
n
to
m
u
l
t
i
l
i
n
g
u
a
l
v
e
r
if
i
c
a
t
io
n
:
tr
a
i
n
in
g
w
i
t
h
c
r
o
s
s
-
l
i
n
g
u
a
l
c
lu
s
t
e
r
s
to
a
l
lo
w
a
V
i
e
t
n
a
m
e
s
e
c
l
a
im
t
o
b
e
c
h
ec
k
e
d
ag
a
i
n
s
t
E
n
g
l
i
s
h
an
d
C
h
i
n
e
s
e
co
n
t
en
t
.
A
n
o
t
h
e
r
a
v
en
u
e
i
s
l
e
v
er
a
g
in
g
L
L
M
s
’
a
b
i
l
i
ty
t
o
c
i
te
s
o
u
r
ce
s
e
x
p
l
i
c
i
t
ly
–
f
o
r
g
r
ea
t
e
r
tr
a
n
s
p
ar
en
c
y
,
t
h
e
m
o
d
e
l
m
i
g
h
t
p
r
o
v
id
e
f
a
c
t
u
a
l
j
u
s
t
i
f
i
c
a
t
io
n
s
w
i
t
h
l
i
n
k
s
o
r
r
e
f
e
r
en
c
e
s
.
I
n
s
u
m
m
a
r
y
,
t
h
i
s
w
o
r
k
d
em
o
n
s
t
r
a
t
e
s
th
a
t
L
L
M
-
b
a
s
ed
p
ip
e
l
i
n
e
s
,
a
u
g
m
e
n
t
ed
w
i
th
e
v
en
t
c
lu
s
te
r
i
n
g
,
a
r
e
a
p
r
o
m
i
s
i
n
g
d
i
r
ec
t
i
o
n
f
o
r
c
o
m
b
a
t
in
g
m
i
s
i
n
f
o
r
m
a
t
io
n
o
n
s
o
c
i
a
l
p
l
a
t
f
o
r
m
s
.
A
s
L
L
M
t
e
ch
n
o
l
o
g
y
e
v
o
l
v
e
s
an
d
b
e
c
o
m
e
s
m
o
r
e
e
f
f
i
c
i
en
t
,
s
u
ch
s
y
s
t
e
m
s
a
r
e
a
n
t
i
c
i
p
a
t
e
d
t
o
p
l
a
y
a
n
i
n
c
r
e
a
s
in
g
ly
p
r
a
c
t
i
c
a
l
r
o
le
i
n
r
ea
l
-
t
i
m
e
n
e
w
s
v
e
r
if
i
c
a
t
io
n
,
u
lt
i
m
a
t
e
ly
co
n
t
r
ib
u
ti
n
g
t
o
a
m
o
r
e
t
r
u
s
t
w
o
r
th
y
i
n
f
o
r
m
a
t
io
n
e
co
s
y
s
t
e
m
.
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.
15
,
No
.
4
,
Au
g
u
s
t 2
0
2
6
:
3
1
4
4
-
3
1
5
3
3152
F
UNDING
I
NF
O
R
M
A
T
I
O
N
Au
th
o
r
s
s
tate
n
o
f
u
n
d
in
g
in
v
o
lv
ed
.
AUTHO
R
CO
NT
RI
B
UT
I
O
NS ST
A
T
E
M
E
N
T
T
h
is
jo
u
r
n
al
u
s
es
th
e
C
o
n
t
r
ib
u
to
r
R
o
les
T
a
x
o
n
o
m
y
(
C
R
ed
iT)
to
r
ec
o
g
n
ize
in
d
iv
i
d
u
al
au
th
o
r
co
n
tr
ib
u
tio
n
s
,
r
ed
u
ce
au
th
o
r
s
h
ip
d
is
p
u
tes,
an
d
f
ac
ilit
ate
co
llab
o
r
atio
n
.
Na
m
e
o
f
Aut
ho
r
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
T
r
an
Du
c
Du
o
n
g
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
Hai
Ho
an
Do
✓
✓
✓
✓
✓
✓
✓
C
:
C
o
n
c
e
p
t
u
a
l
i
z
a
t
i
o
n
M
:
M
e
t
h
o
d
o
l
o
g
y
So
:
So
f
t
w
a
r
e
Va
:
Va
l
i
d
a
t
i
o
n
Fo
:
Fo
r
mal
a
n
a
l
y
s
i
s
I
:
I
n
v
e
s
t
i
g
a
t
i
o
n
R
:
R
e
so
u
r
c
e
s
D
:
D
a
t
a
C
u
r
a
t
i
o
n
O
:
W
r
i
t
i
n
g
-
O
r
i
g
i
n
a
l
D
r
a
f
t
E
:
W
r
i
t
i
n
g
-
R
e
v
i
e
w
&
E
d
i
t
i
n
g
Vi
:
Vi
su
a
l
i
z
a
t
i
o
n
Su
:
Su
p
e
r
v
i
s
i
o
n
P
:
P
r
o
j
e
c
t
a
d
mi
n
i
st
r
a
t
i
o
n
Fu
:
Fu
n
d
i
n
g
a
c
q
u
i
si
t
i
o
n
CO
NF
L
I
C
T
O
F
I
N
T
E
R
E
S
T
ST
A
T
E
M
E
NT
Au
th
o
r
s
s
tate
n
o
co
n
f
lict o
f
in
t
er
est.
DATA AV
AI
L
AB
I
L
I
T
Y
T
h
e
d
ata
th
at
s
u
p
p
o
r
t
th
e
f
in
d
i
n
g
s
o
f
th
is
s
tu
d
y
ar
e
av
ailab
le
f
r
o
m
th
e
co
r
r
esp
o
n
d
in
g
au
th
o
r
,
[
T
DD]
,
u
p
o
n
r
ea
s
o
n
ab
le
r
eq
u
est.
RE
F
E
R
E
NC
E
S
[
1
]
W
.
Y
.
W
a
n
g
,
“
‘
Li
a
r
,
l
i
a
r
p
a
n
t
s
o
n
f
i
r
e
’
:
a
n
e
w
b
e
n
c
h
mar
k
d
a
t
a
s
e
t
f
o
r
f
a
k
e
n
e
w
s
d
e
t
e
c
t
i
o
n
,
”
i
n
5
5
t
h
A
n
n
u
a
l
M
e
e
t
i
n
g
o
f
t
h
e
Asso
c
i
a
t
i
o
n
f
o
r
C
o
m
p
u
t
a
t
i
o
n
a
l
L
i
n
g
u
i
st
i
c
s
,
V
a
n
c
o
u
v
e
r
,
C
a
n
a
d
a
,
2
0
1
7
,
p
p
.
4
2
2
–
4
2
6
,
d
o
i
:
1
0
.
1
8
6
5
3
/
v
1
/
P
1
7
-
2
0
6
7
.
[
2
]
V
.
P
.
R
o
sas,
B
.
K
l
e
i
n
b
e
r
g
,
A
.
Le
f
e
v
r
e
,
a
n
d
R
.
M
i
h
a
l
c
e
a
,
“
A
u
t
o
ma
t
i
c
d
e
t
e
c
t
i
o
n
o
f
f
a
k
e
n
e
w
s
,
”
i
n
C
O
L
I
N
G
2
0
1
8
-
2
7
t
h
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
o
n
C
o
m
p
u
t
a
t
i
o
n
a
l
L
i
n
g
u
i
st
i
c
s,
S
a
n
t
a
F
e
,
N
e
w
M
e
x
i
c
o
,
2
0
1
8
,
p
p
.
3
3
9
1
–
3
4
0
1
.
[
3
]
I
.
A
u
g
e
n
st
e
i
n
e
t
a
l
.
,
“
M
u
l
t
i
F
C
:
a
r
e
a
l
-
w
o
r
l
d
m
u
l
t
i
-
d
o
m
a
i
n
d
a
t
a
s
e
t
f
o
r
e
v
i
d
e
n
c
e
-
b
a
s
e
d
f
a
c
t
c
h
e
c
k
i
n
g
o
f
c
l
a
i
ms,
”
i
n
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
2
0
1
9
C
o
n
f
e
re
n
c
e
o
n
Em
p
i
ri
c
a
l
Me
t
h
o
d
s
i
n
N
a
t
u
ra
l
L
a
n
g
u
a
g
e
Pr
o
c
e
ssi
n
g
a
n
d
t
h
e
9
t
h
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
i
n
t
C
o
n
f
e
re
n
c
e
o
n
N
a
t
u
r
a
l
L
a
n
g
u
a
g
e
Pro
c
e
ss
i
n
g
(
EM
N
L
P
-
I
J
C
N
L
P)
,
H
o
n
g
K
o
n
g
,
C
h
i
n
a
,
2
0
1
9
,
p
p
.
4
6
8
4
–
4
6
9
6
,
d
o
i
:
1
0
.
1
8
6
5
3
/
v
1
/
D
1
9
-
1
4
7
5
.
[
4
]
S
.
R
a
z
a
,
D
.
P
.
P
a
t
t
e
r
so
n
,
a
n
d
C
.
D
i
n
g
,
“
F
a
k
e
n
e
w
s
d
e
t
e
c
t
i
o
n
:
c
o
m
p
a
r
a
t
i
v
e
e
v
a
l
u
a
t
i
o
n
o
f
B
E
R
T
-
l
i
k
e
m
o
d
e
l
s
a
n
d
l
a
r
g
e
l
a
n
g
u
a
g
e
mo
d
e
l
s
w
i
t
h
g
e
n
e
r
a
t
i
v
e
A
I
-
a
n
n
o
t
a
t
e
d
d
a
t
a
,
”
K
n
o
w
l
e
d
g
e
a
n
d
I
n
f
o
rm
a
t
i
o
n
S
y
s
t
e
m
s
,
v
o
l
.
6
7
,
n
o
.
4
,
p
p
.
3
2
6
7
–
3
2
9
2
,
2
0
2
5
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
0
1
1
5
-
024
-
0
2
3
2
1
-
1.
[
5
]
J.
T
h
o
r
n
e
,
A
.
V
l
a
c
h
o
s
,
C
.
C
h
r
i
s
t
o
d
o
u
l
o
p
o
u
l
o
s
,
a
n
d
A
.
M
i
t
t
a
l
,
“
F
EV
ER
:
a
l
a
r
g
e
-
s
c
a
l
e
d
a
t
a
s
e
t
f
o
r
f
a
c
t
e
x
t
r
a
c
t
i
o
n
a
n
d
v
e
r
i
f
i
c
a
t
i
o
n
,
”
i
n
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
2
0
1
8
C
o
n
f
e
r
e
n
c
e
o
f
t
h
e
N
o
r
t
h
Am
e
ri
c
a
n
C
h
a
p
t
e
r
o
f
t
h
e
Asso
c
i
a
t
i
o
n
f
o
r
C
o
m
p
u
t
a
t
i
o
n
a
l
L
i
n
g
u
i
st
i
c
s:
H
u
m
a
n
L
a
n
g
u
a
g
e
T
e
c
h
n
o
l
o
g
i
e
s,
N
e
w
O
r
l
e
a
n
s
,
L
o
u
i
si
a
n
a
,
2
0
1
8
,
p
p
.
8
0
9
–
8
1
9
,
d
o
i
:
1
0
.
1
8
6
5
3
/
v
1
/
N
1
8
-
1
0
7
4
.
[
6
]
A
.
S
o
l
e
i
m
a
n
i
,
C
.
M
o
n
z
,
a
n
d
M
.
W
o
r
r
i
n
g
,
“
B
E
R
T
f
o
r
e
v
i
d
e
n
c
e
r
e
t
r
i
e
v
a
l
a
n
d
c
l
a
i
m
v
e
r
i
f
i
c
a
t
i
o
n
,
”
i
n
A
d
v
a
n
c
e
s
i
n
I
n
f
o
rm
a
t
i
o
n
Re
t
r
i
e
v
a
l
:
4
2
n
d
E
u
ro
p
e
a
n
C
o
n
f
e
ren
c
e
o
n
I
R
Re
s
e
a
r
c
h
(
EC
I
R
2
0
2
0
)
,
Li
s
b
o
n
,
P
o
r
t
u
g
a
l
,
2
0
2
0
,
p
p
.
3
5
9
–
3
6
6
,
d
o
i
:
1
0
.
1
0
0
7
/
9
7
8
-
3
-
030
-
4
5
4
4
2
-
5
_
4
5
.
[
7
]
C
.
K
r
u
e
n
g
k
r
a
i
,
J.
Y
a
ma
g
i
sh
i
,
a
n
d
X
.
W
a
n
g
,
“
A
mu
l
t
i
-
l
e
v
e
l
a
t
t
e
n
t
i
o
n
m
o
d
e
l
f
o
r
e
v
i
d
e
n
c
e
-
b
a
se
d
f
a
c
t
c
h
e
c
k
i
n
g
,
”
i
n
F
i
n
d
i
n
g
s
o
f
t
h
e
Asso
c
i
a
t
i
o
n
f
o
r
C
o
m
p
u
t
a
t
i
o
n
a
l
L
i
n
g
u
i
st
i
c
s:
AC
L
-
I
J
C
N
L
P
2
0
2
1
,
2
0
2
1
,
p
p
.
2
4
4
7
–
2
4
6
0
,
d
o
i
:
1
0
.
1
8
6
5
3
/
v
1
/
2
0
2
1
.
f
i
n
d
i
n
g
s
-
a
c
l
.
2
1
7
.
[
8
]
F
.
Y
a
n
g
e
t
a
l
.
,
“
X
F
a
k
e
:
e
x
p
l
a
i
n
a
b
l
e
f
a
k
e
n
e
w
s
d
e
t
e
c
t
o
r
w
i
t
h
v
i
s
u
a
l
i
z
a
t
i
o
n
s,”
i
n
T
h
e
W
o
rl
d
Wi
d
e
We
b
C
o
n
f
e
r
e
n
c
e
,
N
e
w
Y
o
r
k
,
U
n
i
t
e
d
S
t
a
t
e
s
,
M
a
y
2
0
1
9
,
p
p
.
3
6
0
0
–
3
6
0
4
,
d
o
i
:
1
0
.
1
1
4
5
/
3
3
0
8
5
5
8
.
3
3
1
4
1
1
9
.
[
9
]
N
.
K
o
t
o
n
y
a
a
n
d
F
.
T
o
n
i
,
“
E
x
p
l
a
i
n
a
b
l
e
a
u
t
o
m
a
t
e
d
f
a
c
t
-
c
h
e
c
k
i
n
g
f
o
r
p
u
b
l
i
c
h
e
a
l
t
h
c
l
a
i
ms
,
”
i
n
2
0
2
0
C
o
n
f
e
r
e
n
c
e
o
n
Em
p
i
r
i
c
a
l
Me
t
h
o
d
s
i
n
N
a
t
u
ra
l
L
a
n
g
u
a
g
e
Pr
o
c
e
s
si
n
g
(
E
MN
L
P)
,
2
0
2
0
,
p
p
.
7
7
4
0
–
7
7
5
4
,
d
o
i
:
1
0
.
1
8
6
5
3
/
v
1
/
2
0
2
0
.
e
mn
l
p
-
mai
n
.
6
2
3
.
[
1
0
]
D
.
X
.
Tr
a
n
e
t
a
l
.
,
“
S
e
mV
i
Q
A
:
a
sem
a
n
t
i
c
q
u
e
st
i
o
n
a
n
sw
e
r
i
n
g
s
y
st
e
m
f
o
r
V
i
e
t
n
a
mes
e
i
n
f
o
r
m
a
t
i
o
n
f
a
c
t
-
c
h
e
c
k
i
n
g
,
”
a
rXi
v
:
2
5
0
3
.
0
0
9
5
5
,
2
0
2
5
.
[
1
1
]
D
.
Q
u
e
l
l
e
a
n
d
A
.
B
o
v
e
t
,
“
T
h
e
p
e
r
i
l
s
a
n
d
p
r
o
m
i
ses
o
f
f
a
c
t
-
c
h
e
c
k
i
n
g
w
i
t
h
l
a
r
g
e
l
a
n
g
u
a
g
e
m
o
d
e
l
s
,
”
Fro
n
t
i
e
rs
i
n
Art
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
,
v
o
l
.
7
,
F
e
b
.
2
0
2
4
,
d
o
i
:
1
0
.
3
3
8
9
/
f
r
a
i
.
2
0
2
4
.
1
3
4
1
6
9
7
.
[
1
2
]
X
.
Z
h
a
n
g
a
n
d
W
.
G
a
o
,
“
T
o
w
a
r
d
s
LLM
-
b
a
s
e
d
f
a
c
t
v
e
r
i
f
i
c
a
t
i
o
n
o
n
n
e
w
s
c
l
a
i
ms
w
i
t
h
a
h
i
e
r
a
r
c
h
i
c
a
l
st
e
p
-
by
-
s
t
e
p
p
r
o
mp
t
i
n
g
met
h
o
d
,
”
i
n
1
3
t
h
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
i
n
t
C
o
n
f
e
r
e
n
c
e
o
n
N
a
t
u
ra
l
L
a
n
g
u
a
g
e
P
ro
c
e
ssi
n
g
a
n
d
t
h
e
3
r
d
C
o
n
f
e
re
n
c
e
o
f
t
h
e
Asi
a
-
P
a
c
i
f
i
c
C
h
a
p
t
e
r
o
f
t
h
e
Ass
o
c
i
a
t
i
o
n
f
o
r
C
o
m
p
u
t
a
t
i
o
n
a
l
L
i
n
g
u
i
st
i
c
s
,
N
u
s
a
D
u
a
,
B
a
l
i
,
2
0
2
3
,
p
p
.
9
9
6
–
1
0
1
1
,
d
o
i
:
1
0
.
1
8
6
5
3
/
v
1
/
2
0
2
3
.
i
j
c
n
l
p
-
mai
n
.
6
4
.
[
1
3
]
X
.
L
i
,
Y
.
Zh
a
n
g
,
a
n
d
E.
C
.
M
a
l
t
h
o
u
s
e
,
“
La
r
g
e
l
a
n
g
u
a
g
e
m
o
d
e
l
a
g
e
n
t
i
c
a
p
p
r
o
a
c
h
t
o
f
a
c
t
c
h
e
c
k
i
n
g
a
n
d
f
a
k
e
n
e
w
s
d
e
t
e
c
t
i
o
n
,
”
i
n
2
7
t
h
E
u
ro
p
e
a
n
C
o
n
f
e
re
n
c
e
o
n
Ar
t
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
,
1
9
–
2
4
O
c
t
o
b
e
r
2
0
2
4
,
I
n
c
l
u
d
i
n
g
1
3
t
h
C
o
n
f
e
r
e
n
c
e
o
n
Pre
s
t
i
g
i
o
u
s
Ap
p
l
i
c
a
t
i
o
n
s
o
f
I
n
t
e
l
l
i
g
e
n
t
S
y
s
t
e
m
s (P
AI
S
2
0
2
4
)
,
S
a
n
t
i
a
g
o
d
e
C
o
m
p
o
s
t
e
l
a
,
S
p
a
i
n
,
2
0
2
4
,
d
o
i
:
1
0
.
3
2
3
3
/
F
A
I
A
2
4
0
7
8
7
.
[
1
4
]
S
.
M
i
n
e
t
a
l
.
,
“
F
A
c
t
S
c
o
r
e
:
f
i
n
e
-
g
r
a
i
n
e
d
a
t
o
m
i
c
e
v
a
l
u
a
t
i
o
n
o
f
f
a
c
t
u
a
l
p
r
e
c
i
s
i
o
n
i
n
l
o
n
g
f
o
r
m
t
e
x
t
g
e
n
e
r
a
t
i
o
n
,
”
i
n
2
0
2
3
C
o
n
f
e
re
n
c
e
o
n
Em
p
i
ri
c
a
l
Me
t
h
o
d
s
i
n
N
a
t
u
r
a
l
L
a
n
g
u
a
g
e
Pro
c
e
ss
i
n
g
,
2
0
2
3
,
p
p
.
1
2
0
7
6
–
1
2
1
0
0
,
d
o
i
:
1
0
.
1
8
6
5
3
/
v
1
/
2
0
2
3
.
e
mn
l
p
-
mai
n
.
7
4
1
.
[
1
5
]
W
.
Y
u
e
t
a
l
.
,
“
R
e
s
e
a
r
c
h
o
n
f
a
k
e
n
e
w
s
d
e
t
e
c
t
i
o
n
b
a
s
e
d
o
n
d
u
a
l
e
v
i
d
e
n
c
e
p
e
r
c
e
p
t
i
o
n
,
”
En
g
i
n
e
e
ri
n
g
A
p
p
l
i
c
a
t
i
o
n
s
o
f
Ar
t
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
,
v
o
l
.
1
3
3
,
J
u
l
.
2
0
2
4
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
e
n
g
a
p
p
a
i
.
2
0
2
4
.
1
0
8
2
7
1
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
S
o
cia
l n
ews
fa
ctu
a
lity v
erif
ica
tio
n
u
s
in
g
la
r
g
e
l
a
n
g
u
a
g
e
mo
d
els (
Tr
a
n
Du
c
Du
o
n
g
)
3153
[
1
6
]
D
.
M
e
t
r
o
p
o
l
i
t
a
n
s
k
y
a
n
d
J.
L
a
r
so
n
,
“
T
o
w
a
r
d
s
e
f
f
e
c
t
i
v
e
e
x
t
r
a
c
t
i
o
n
a
n
d
e
v
a
l
u
a
t
i
o
n
o
f
f
a
c
t
u
a
l
c
l
a
i
ms,
”
i
n
6
3
r
d
An
n
u
a
l
Me
e
t
i
n
g
o
f
t
h
e
Asso
c
i
a
t
i
o
n
f
o
r
C
o
m
p
u
t
a
t
i
o
n
a
l
L
i
n
g
u
i
st
i
c
s
,
2
0
2
5
,
p
p
.
6
9
9
6
–
7
0
4
5
.
[
1
7
]
M
.
A
.
K
h
a
l
i
q
,
P
.
Y
.
-
C
.
C
h
a
n
g
,
M
.
M
a
,
B
.
P
f
l
u
g
f
e
l
d
e
r
,
a
n
d
F
.
M
i
l
e
t
i
ć
,
“
R
A
G
A
R
,
y
o
u
r
f
a
l
s
e
h
o
o
d
r
a
d
a
r
:
R
A
G
-
a
u
g
men
t
e
d
r
e
a
so
n
i
n
g
f
o
r
p
o
l
i
t
i
c
a
l
f
a
c
t
-
c
h
e
c
k
i
n
g
u
si
n
g
m
u
l
t
i
m
o
d
a
l
l
a
r
g
e
l
a
n
g
u
a
g
e
m
o
d
e
l
s,”
i
n
S
e
v
e
n
t
h
Fa
c
t
Ex
t
r
a
c
t
i
o
n
a
n
d
VER
i
f
i
c
a
t
i
o
n
W
o
r
k
sh
o
p
(
FEVE
R)
,
M
i
a
m
i
,
F
l
o
r
i
d
a
,
2
0
2
4
,
p
p
.
2
8
0
–
2
9
6
.
[
1
8
]
Y
.
W
a
n
g
e
t
a
l
.
,
“
F
a
c
t
c
h
e
c
k
-
b
e
n
c
h
:
f
i
n
e
-
g
r
a
i
n
e
d
e
v
a
l
u
a
t
i
o
n
b
e
n
c
h
mar
k
f
o
r
a
u
t
o
ma
t
i
c
f
a
c
t
-
c
h
e
c
k
e
r
s,”
i
n
F
i
n
d
i
n
g
s
o
f
t
h
e
Ass
o
c
i
a
t
i
o
n
f
o
r
C
o
m
p
u
t
a
t
i
o
n
a
l
L
i
n
g
u
i
st
i
c
s:
EM
N
L
P
2
0
2
4
,
M
i
a
mi
,
F
l
o
r
i
d
a
,
2
0
2
4
,
p
p
.
1
4
1
9
9
–
1
4
2
3
0
.
[
1
9
]
Y
.
C
h
e
n
e
t
a
l
.
,
“
G
r
a
p
h
C
h
e
c
k
:
b
r
e
a
k
i
n
g
l
o
n
g
-
t
e
r
m
t
e
x
t
b
a
r
r
i
e
r
s
w
i
t
h
e
x
t
r
a
c
t
e
d
k
n
o
w
l
e
d
g
e
g
r
a
p
h
-
p
o
w
e
r
e
d
f
a
c
t
-
c
h
e
c
k
i
n
g
,
”
i
n
6
3
r
d
An
n
u
a
l
M
e
e
t
i
n
g
o
f
t
h
e
Ass
o
c
i
a
t
i
o
n
f
o
r
C
o
m
p
u
t
a
t
i
o
n
a
l
L
i
n
g
u
i
st
i
c
s
,
2
0
2
5
,
p
p
.
1
4
9
7
6
–
1
4
9
9
5
,
d
o
i
:
1
0
.
1
8
6
5
3
/
v
1
/
2
0
2
5
.
a
c
l
-
l
o
n
g
.
7
2
9
.
[
2
0
]
U
.
Q
u
d
u
s,
M
.
R
ö
d
e
r
,
M
.
S
a
l
e
e
m
,
a
n
d
A
.
-
C
.
N
.
N
g
o
m
o
,
“
H
y
b
r
i
d
F
C
:
a
h
y
b
r
i
d
f
a
c
t
-
c
h
e
c
k
i
n
g
a
p
p
r
o
a
c
h
f
o
r
k
n
o
w
l
e
d
g
e
g
r
a
p
h
s
,
”
i
n
T
h
e
S
e
m
a
n
t
i
c
We
b
(
I
S
WC
2
0
2
2
)
,
C
h
a
m
,
S
w
i
t
z
e
r
l
a
n
d
:
S
p
r
i
n
g
e
r
,
2
0
2
2
,
p
p
.
4
6
2
–
4
8
0
,
d
o
i
:
1
0
.
1
0
0
7
/
9
7
8
-
3
-
0
3
1
-
1
9
4
3
3
-
7
_
2
7
.
[
2
1
]
B
.
Y
a
n
g
,
D
.
X
u
,
K
.
N
i
u
,
W
.
Li
u
,
Z
.
W
a
n
g
,
a
n
d
M
.
K
a
n
k
a
n
h
a
l
l
i
,
“
A
n
e
w
d
a
t
a
s
e
t
a
n
d
b
e
n
c
h
m
a
r
k
f
o
r
g
r
o
u
n
d
i
n
g
m
u
l
t
i
m
o
d
a
l
mi
si
n
f
o
r
m
a
t
i
o
n
,
”
i
n
3
3
rd
A
C
M
I
n
t
e
rn
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
M
u
l
t
i
m
e
d
i
a
,
N
e
w
Y
o
r
k
,
U
S
A
,
O
c
t
.
2
0
2
5
,
p
p
.
1
2
5
7
1
–
1
2
5
7
7
,
d
o
i
:
1
0
.
1
1
4
5
/
3
7
4
6
0
2
7
.
3
7
5
8
1
9
1
.
[
2
2
]
A
.
B
e
i
g
i
e
t
a
l
.
,
“
C
a
n
L
LM
s
i
m
p
r
o
v
e
mu
l
t
i
mo
d
a
l
f
a
c
t
-
c
h
e
c
k
i
n
g
b
y
a
sk
i
n
g
r
e
l
e
v
a
n
t
q
u
e
s
t
i
o
n
s?,
”
i
n
2
0
2
5
I
E
EE
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
Bi
g
D
a
t
a
(
B
i
g
D
a
t
a
)
,
D
e
c
.
2
0
2
5
,
p
p
.
2
7
3
2
–
2
7
4
1
,
d
o
i
:
1
0
.
1
1
0
9
/
B
i
g
D
a
t
a
6
6
9
2
6
.
2
0
2
5
.
1
1
4
0
0
8
4
6
.
[
2
3
]
K
.
K
a
k
i
z
a
k
i
,
Y
.
M
a
t
s
u
n
a
g
a
,
a
n
d
R
.
F
u
r
u
k
a
w
a
,
“
M
A
F
T
:
m
u
l
t
i
m
o
d
a
l
a
u
t
o
m
a
t
e
d
f
a
c
t
-
c
h
e
c
k
i
n
g
v
i
a
t
e
x
t
u
a
l
i
z
a
t
i
o
n
,
”
i
n
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
3
9
t
h
A
n
n
u
a
l
AA
AI
C
o
n
f
e
re
n
c
e
o
n
Art
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
,
2
0
2
5
,
p
p
.
2
9
6
4
6
–
2
9
6
4
8
,
d
o
i
:
1
0
.
1
6
0
9
/
a
a
a
i
.
v
3
9
i
2
8
.
3
5
3
5
4
.
[
2
4
]
S
.
C
h
e
n
,
F
.
K
a
n
g
,
N
.
Y
u
,
a
n
d
R
.
J
i
a
,
“
F
A
S
TTR
A
C
K
:
r
e
l
i
a
b
l
e
f
a
c
t
t
r
a
c
i
n
g
v
i
a
c
l
u
s
t
e
r
i
n
g
a
n
d
LL
M
-
p
o
w
e
r
e
d
e
v
i
d
e
n
c
e
v
a
l
i
d
a
t
i
o
n
,
”
i
n
Fi
n
d
i
n
g
s
o
f
t
h
e
Ass
o
c
i
a
t
i
o
n
f
o
r
C
o
m
p
u
t
a
t
i
o
n
a
l
L
i
n
g
u
i
s
t
i
c
s:
E
MN
L
P
2
0
2
4
,
M
i
a
m
i
,
F
l
o
r
i
d
a
,
2
0
2
4
,
p
p
.
5
8
2
1
–
5
8
3
6
,
d
o
i
:
1
0
.
1
8
6
5
3
/
v
1
/
2
0
2
4
.
f
i
n
d
i
n
g
s
-
e
mn
l
p
.
3
3
4
.
[
2
5
]
T.
T.
H
o
a
,
T.
Q
.
D
u
y
,
K
.
Q
.
Tr
a
n
,
a
n
d
K
.
V
.
N
g
u
y
e
n
,
“
V
i
F
a
c
t
C
h
e
c
k
:
a
n
e
w
b
e
n
c
h
m
a
r
k
d
a
t
a
set
a
n
d
me
t
h
o
d
s
f
o
r
m
u
l
t
i
-
d
o
m
a
i
n
n
e
w
s
f
a
c
t
-
c
h
e
c
k
i
n
g
I
n
V
i
e
t
n
a
m
e
se,
”
i
n
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
3
9
t
h
AAAI
C
o
n
f
e
re
n
c
e
o
n
Ar
t
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
,
2
0
2
5
,
p
p
.
3
0
8
–
3
1
6
,
d
o
i
:
1
0
.
1
6
0
9
/
a
a
a
i
.
v
3
9
i
1
.
3
2
0
0
8
.
[
2
6
]
R
.
P
a
n
c
h
e
n
d
r
a
r
a
j
a
n
,
R
.
M
.
P
é
r
e
z
,
a
n
d
A
.
Z
u
b
i
a
g
a
,
“
M
u
l
t
i
C
l
a
i
m
N
e
t
:
a
m
a
s
s
i
v
e
l
y
m
u
l
t
i
l
i
n
g
u
a
l
d
a
t
a
s
e
t
o
f
f
a
c
t
-
c
h
e
c
k
e
d
c
l
a
i
m
c
l
u
s
t
e
r
s
,
”
i
n
A
s
s
o
c
i
a
t
i
o
n
f
o
r
C
o
m
p
u
t
a
t
i
o
n
a
l
L
i
n
g
u
i
s
t
i
c
s
:
E
M
N
L
P
2
0
2
5
,
2
0
2
5
,
p
p
.
1
1
2
0
3
–
1
1
2
1
5
,
d
o
i
:
1
0
.
1
8
6
5
3
/
v
1
/
2
0
2
5
.
f
i
n
d
i
n
g
s
-
e
m
n
l
p
.
5
9
9
.
[
2
7
]
A
.
U
.
D
e
y
,
M
.
J.
A
w
a
n
,
G
.
C
h
a
n
n
i
n
g
,
C
.
S
.
D
.
W
i
t
t
,
a
n
d
J.
C
o
l
l
o
m
o
sse,
“
F
a
c
t
-
c
h
e
c
k
i
n
g
w
i
t
h
c
o
n
t
e
x
t
u
a
l
n
a
r
r
a
t
i
v
e
s
:
l
e
v
e
r
a
g
i
n
g
r
e
t
r
i
e
v
a
l
-
a
u
g
me
n
t
e
d
L
LM
s
f
o
r
so
c
i
a
l
me
d
i
a
a
n
a
l
y
s
i
s,”
I
E
EE
T
ra
n
s
a
c
t
i
o
n
s
o
n
C
o
m
p
u
t
a
t
i
o
n
a
l
S
o
c
i
a
l
S
y
s
t
e
m
s
,
v
o
l
.
1
3
,
n
o
.
3
,
p
p
.
3
3
6
5
–
3
3
7
6
,
Ju
n
.
2
0
2
6
,
d
o
i
:
1
0
.
1
1
0
9
/
T
C
S
S
.
2
0
2
6
.
3
6
6
9
7
9
9
.
[
2
8
]
R
.
J.
G
.
B
.
C
a
m
p
e
l
l
o
,
D
.
M
o
u
l
a
v
i
,
a
n
d
J
.
S
a
n
d
e
r
,
“
D
e
n
si
t
y
-
b
a
s
e
d
c
l
u
st
e
r
i
n
g
b
a
s
e
d
o
n
h
i
e
r
a
r
c
h
i
c
a
l
d
e
n
s
i
t
y
e
st
i
m
a
t
e
s
,
”
i
n
A
d
v
a
n
c
e
s
i
n
K
n
o
w
l
e
d
g
e
D
i
s
c
o
v
e
r
y
a
n
d
D
a
t
a
Mi
n
i
n
g
(
PA
K
D
D
2
0
1
3
)
,
2
0
1
3
,
p
p
.
1
6
0
–
1
7
2
,
d
o
i
:
1
0
.
1
0
0
7
/
9
7
8
-
3
-
642
-
3
7
4
5
6
-
2
_
1
4
.
[
2
9
]
D
.
Q
.
N
g
u
y
e
n
a
n
d
A
.
T
.
N
g
u
y
e
n
,
“
P
h
o
B
E
R
T:
p
r
e
-
t
r
a
i
n
e
d
l
a
n
g
u
a
g
e
m
o
d
e
l
s
f
o
r
V
i
e
t
n
a
m
e
se
,
”
i
n
Fi
n
d
i
n
g
s
o
f
t
h
e
Ass
o
c
i
a
t
i
o
n
f
o
r
C
o
m
p
u
t
a
t
i
o
n
a
l
L
i
n
g
u
i
s
t
i
c
s:
EM
N
L
P
2
0
2
0
,
2
0
2
0
,
p
p
.
1
0
3
7
–
1
0
4
2
,
d
o
i
:
1
0
.
1
8
6
5
3
/
v
1
/
2
0
2
0
.
f
i
n
d
i
n
g
s
-
e
m
n
l
p
.
9
2
.
[
3
0
]
L.
Zh
e
n
g
e
t
a
l
.
,
“
F
a
c
t
i
n
f
r
a
g
me
n
t
s:
d
e
c
o
n
st
r
u
c
t
i
n
g
c
o
m
p
l
e
x
c
l
a
i
ms
v
i
a
LLM
-
b
a
s
e
d
a
t
o
mi
c
f
a
c
t
e
x
t
r
a
c
t
i
o
n
a
n
d
v
e
r
i
f
i
c
a
t
i
o
n
,
”
Ex
p
e
rt
S
y
st
e
m
s w
i
t
h
Ap
p
l
i
c
a
t
i
o
n
s
,
v
o
l
.
3
0
2
,
M
a
r
.
2
0
2
6
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
e
sw
a
.
2
0
2
5
.
1
3
0
5
7
2
.
[
3
1
]
K
.
W
u
e
t
a
l
.
,
“
A
n
a
u
t
o
m
a
t
e
d
f
r
a
m
e
w
o
r
k
f
o
r
a
sses
si
n
g
h
o
w
w
e
l
l
LL
M
s c
i
t
e
r
e
l
e
v
a
n
t
me
d
i
c
a
l
r
e
f
e
r
e
n
c
e
s
,
”
N
a
t
u
r
e
C
o
m
m
u
n
i
c
a
t
i
o
n
s
,
v
o
l
.
1
6
,
n
o
.
1
,
A
p
r
.
2
0
2
5
,
d
o
i
:
1
0
.
1
0
3
8
/
s
4
1
4
6
7
-
0
2
5
-
5
8
5
5
1
-
6.
[
3
2
]
H
.
T.
Le
,
L.
T.
T
o
,
M
.
T.
N
g
u
y
e
n
,
a
n
d
K
.
V
.
N
g
u
y
e
n
,
“
V
i
W
i
k
i
F
C
:
f
a
c
t
-
c
h
e
c
k
i
n
g
f
o
r
V
i
e
t
n
a
m
e
se
w
i
k
i
p
e
d
i
a
-
b
a
se
d
t
e
x
t
u
a
l
k
n
o
w
l
e
d
g
e
s
o
u
r
c
e
,
”
2
0
2
4
,
a
rX
i
v
:
2
4
0
5
.
0
7
6
1
5
.
[
3
3
]
W
.
Y
a
n
g
,
L.
S
o
m
e
,
M
.
B
a
i
n
,
a
n
d
B
.
K
a
n
g
,
“
A
c
o
m
p
r
e
h
e
n
s
i
v
e
s
u
r
v
e
y
o
n
i
n
t
e
g
r
a
t
i
n
g
l
a
r
g
e
l
a
n
g
u
a
g
e
mo
d
e
l
s
w
i
t
h
k
n
o
w
l
e
d
g
e
-
b
a
se
d
met
h
o
d
s,
”
K
n
o
w
l
e
d
g
e
-
B
a
se
d
S
y
s
t
e
m
s
,
v
o
l
.
3
1
8
,
Ju
n
.
2
0
2
5
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
k
n
o
sy
s
.
2
0
2
5
.
1
1
3
5
0
3
.
B
I
O
G
RAP
H
I
E
S O
F
AUTH
O
RS
Tr
a
n
Duc
Duo
n
g
h
o
l
d
s
a
Do
c
to
r
o
f
C
o
m
p
u
ter
E
n
g
i
n
e
e
rin
g
d
e
g
re
e
fro
m
P
o
sts
a
n
d
Tele
c
o
m
m
u
n
ica
ti
o
n
s In
stit
u
te
o
f
Tec
h
n
o
lo
g
y
,
Vie
t
n
a
m
in
2
0
1
7
.
He
a
lso
re
c
e
iv
e
d
h
is B.
S
c
.
fro
m
Vie
tn
a
m
Na
ti
o
n
a
l
U
n
iv
e
rsity
o
f
Ha
n
o
i
(I
n
fo
rm
a
ti
o
n
Tec
h
n
o
lo
g
y
)
a
n
d
M
.
S
c
.
(In
fo
rm
a
ti
o
n
S
y
ste
m
s)
fro
m
Un
iv
e
rsit
y
o
f
Lee
d
s,
Un
it
e
d
Kin
g
d
o
m
i
n
1
9
9
9
a
n
d
2
0
0
4
,
re
sp
e
c
ti
v
e
ly
.
He
is
c
u
rre
n
t
ly
a
lec
tu
re
r
a
t
F
a
c
u
lt
y
o
f
I
n
fo
rm
a
ti
o
n
Tec
h
n
o
l
o
g
y
,
P
o
sts
a
n
d
Tele
c
o
m
m
u
n
ica
ti
o
n
s
I
n
stit
u
te
o
f
Tec
h
n
o
l
o
g
y
,
Ha
n
o
i,
Vie
tn
a
m
.
Hi
s
re
se
a
rc
h
in
c
lu
d
e
s
m
a
c
h
in
e
lea
rn
in
g
,
d
e
e
p
lea
rn
i
n
g
,
ima
g
e
a
n
d
n
a
t
u
ra
l
lan
g
u
a
g
e
p
r
o
c
e
ss
in
g
,
a
n
d
LLM
s.
He
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
d
u
c
d
t@
p
ti
t.
e
d
u
.
v
n
.
H
a
i
H
o
a
n
Do
re
c
e
iv
e
d
a
Do
c
to
r
o
f
Eco
n
o
m
ics
fro
m
t
h
e
Vie
tn
a
m
Ac
a
d
e
m
y
o
f
S
o
c
ial
S
c
ien
c
e
s,
Ha
n
o
i
,
Vie
tn
a
m
a
n
d
M
a
ste
r
o
f
P
ro
jec
t
M
a
n
a
g
e
m
e
n
t
fro
m
F
o
re
ig
n
Trad
e
Un
iv
e
rsity
,
Ha
n
o
i,
Vie
t
n
a
m
in
2
0
1
8
a
n
d
2
0
1
3
,
re
sp
e
c
ti
v
e
ly
.
S
h
e
is
c
u
rre
n
tl
y
a
lec
tu
re
r
a
t
t
h
e
Eco
n
o
m
ic
Re
se
a
rc
h
In
stit
u
te
o
f
P
o
sta
l
Tec
h
n
o
lo
g
y
,
P
o
sts
a
n
d
Tele
c
o
m
m
u
n
ica
ti
o
n
s
I
n
stit
u
te
o
f
Tec
h
n
o
l
o
g
y
,
Ha
n
o
i,
Vie
tn
a
m
.
He
r
re
se
a
rc
h
in
c
lu
d
e
s
so
c
ial
e
n
trep
re
n
e
u
rsh
i
p
,
s
o
c
ial
c
o
m
m
u
n
ica
ti
o
n
s,
a
n
d
so
c
ial
p
re
ss
.
S
h
e
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
h
o
a
n
d
h
@p
ti
t.
e
d
u
.
v
n
.
Evaluation Warning : The document was created with Spire.PDF for Python.