I
A
E
S
I
n
t
e
r
n
at
io
n
al
Jou
r
n
al
of
A
r
t
if
ic
ia
l
I
n
t
e
ll
ig
e
n
c
e
(
I
J
-
AI
)
V
ol
. 14, No. 5, O
c
to
be
r
2025
, pp.
4271
~
4278
I
S
S
N
:
2252
-
8938
,
D
O
I
:
10.11591/
ij
a
i.
v
14
.i
5
.pp
4271
-
4278
4271
Jou
r
n
al
h
om
e
page
:
ht
tp
:
//
ij
ai
.
ia
e
s
c
or
e
.c
om
A
I
-
d
r
i
ve
n
h
yp
e
r
-
p
e
r
son
al
i
z
at
i
on
an
d
t
r
an
sf
e
r
l
e
ar
n
i
n
g f
or
p
r
e
c
i
si
on
r
e
c
r
u
i
t
m
e
n
t
N
ou
r
A
lq
u
d
ah
, Q
u
s
ai
Q
.
A
b
u
e
i
n
, M
oh
am
m
e
d
Q
.
S
h
at
n
aw
i
D
e
pa
r
t
m
e
nt
of
C
om
put
e
r
I
nf
or
m
a
t
i
on
S
ys
t
e
m
s
,
F
a
c
ul
t
y of
C
om
put
e
r
a
nd I
nf
or
m
a
t
i
on T
e
c
hnol
ogy,
J
or
da
n U
ni
ve
r
s
i
t
y of
S
c
i
e
nc
e
a
nd T
e
c
hnol
ogy, I
r
bi
d,
J
or
da
n
A
r
t
ic
le
I
n
f
o
A
B
S
T
R
A
C
T
A
r
ti
c
le
h
is
to
r
y
:
R
e
c
e
iv
e
d
M
a
y 1, 2024
R
e
vi
s
e
d
J
un 17, 2025
A
c
c
e
pt
e
d
J
ul
10, 2025
The
research
study
demonstrates
how
artificial
intell
igence
(
AI
)
-
p
owered
models
can
transform
the
hiring
process
by
maximizing
the
match
between
candidates
and
jobs,
leading
to
better
hiring
option
s
and
increased
worker
productivity.
Our
research
develops
highly
personalized
AI
-
p
owered
recruitment
application
s.
By
using
hyper
-
personalization
to
tail
or
job
recommendatio
ns
based
on
job
compatibi
lity
and
big
five
personality
traits,
this
study
leverages
AI
to
improve
job
matching.
Unlike
traditional
recruitmen
t
models
that
depend
only
on
complex
skill
matching,
hyper
-
personalization
combines
soft
skills
and
personality
dimensions
to
ac
hieve
a
more
precise
candidate
-
job
alignment.
Transformer
-
based
models,
inc
luding
bidirectional
encoder
represe
ntations
from
transformers
(
BERT
)
,
RoB
ERTa,
and
cross
-
lingual
language
model
(
XLM
)
-
RoBERTa,
have
shown
exception
al
performance
in
natural
language
processin
g
(NLP
)
and
classifi
cation
tasks;
thus,
we
apply
them.
Transfer
learning
helps
us
t
o
fine
-
tune
these
models
to
improve
the
accura
cy
of
personality
classifi
cation.
Compared
to
conventi
onal
models,
experim
ental
data
achiev
es
up
t
o
80%
accuracy
in
binary
classifi
cation
and
72%
in
multi
-
class
classificati
on.
By
demonstrating
job
-
candidate
compatib
ility
,
this
study
emphasizes
the
potential
of
AI
-
driven
models
to
transform
recruitment,
leading
to
better
hiring
decisions
and
workforce
productivity.
Our
outcomes
play
a
crucial
role in advan
cing hyper
-
personalized AI applica
tions in talent.
K
e
y
w
o
r
d
s
:
B
ig
f
iv
e
pe
r
s
ona
li
ty
F
unc
ti
ona
l
a
r
e
a
s
H
ype
r
-
pe
r
s
ona
li
z
a
ti
on
S
of
t
s
ki
ll
s
T
r
a
ns
f
or
m
e
r
m
ode
ls
This is an
open
acce
ss artic
le unde
r the
CC BY
-
SA
license.
C
or
r
e
s
pon
di
n
g A
u
th
or
:
Q
us
a
i
Q
.
A
bue
in
D
e
pa
r
tm
e
nt
of
C
om
put
e
r
a
nd I
nf
or
m
a
ti
on S
ys
te
m
s
,
F
a
c
ul
ty
of
C
om
put
e
r
a
nd I
nf
or
m
a
ti
on T
e
c
hnol
ogy
J
or
da
n U
ni
ve
r
s
it
y of
S
c
ie
nc
e
a
nd
T
e
c
hnol
ogy
P.
O.
B
ox 3030, I
r
bi
d
22110, J
or
da
n
E
m
a
il
:
qa
bue
in
@
ju
s
t.
e
du.j
o
1.
I
N
T
R
O
D
U
C
T
I
O
N
T
he
s
tu
dy
a
im
s
to
e
nha
nc
e
th
e
r
e
c
r
ui
tm
e
nt
pr
oc
e
s
s
by
de
ve
lo
pi
ng
a
r
ti
f
ic
ia
l
in
te
ll
ig
e
nc
e
(
A
I
)
m
ode
ls
ba
s
e
d
on
pe
r
s
ona
l
tr
a
it
s
a
nd
s
of
t
s
ki
ll
s
r
a
th
e
r
th
a
n
s
ol
e
ly
r
e
ly
in
g
on
te
c
hni
c
a
l
qua
li
f
ic
a
ti
ons
a
nd
ha
r
d
s
ki
ll
s
a
s
in
tr
a
di
ti
ona
l
r
e
c
r
ui
t
m
e
nt
.
H
ype
r
-
pe
r
s
ona
li
z
a
ti
on
r
e
f
e
r
s
to
pe
r
s
ona
li
z
in
g
jo
b
r
e
c
om
m
e
nda
ti
ons
ba
s
e
d
on
th
e
c
a
ndi
da
te
'
s
pe
r
s
ona
li
ty
tr
a
it
s
a
nd
s
of
t
s
ki
ll
s
,
s
uc
h
a
s
c
om
m
uni
c
a
ti
on,
le
a
de
r
s
hi
p,
pr
obl
e
m
-
s
ol
vi
ng,
e
m
ot
io
na
l
in
te
ll
ig
e
nc
e
,
a
nd
te
a
m
w
or
k
[
1]
.
T
r
a
di
ti
ona
l
hi
r
in
g
m
e
th
ods
ty
pi
c
a
ll
y
e
m
pha
s
iz
e
hi
r
in
g
c
a
ndi
da
te
s
ba
s
e
d
s
ol
e
ly
on
th
e
ir
te
c
hni
c
a
l
e
xpe
r
ti
s
e
,
but
th
is
doe
s
not
f
in
d
a
p
r
ope
r
f
it
be
twe
e
n
th
e
a
ppl
ic
a
nt
a
nd
th
e
jo
b
or
w
or
k
e
nvi
r
onm
e
nt
.
T
hi
s
f
oc
us
c
a
n
le
a
d
to
e
m
pl
oye
e
a
da
pt
a
ti
on
is
s
ue
s
,
r
e
s
ul
ti
ng
in
jo
b
tu
r
nove
r
o
r
poo
r
pe
r
f
or
m
a
nc
e
.
T
oda
y,
w
it
h
di
gi
ta
l
a
dva
n
c
e
m
e
nt
s
a
nd
AI
,
th
e
r
e
is
a
s
hi
f
t
to
w
a
r
d
a
m
or
e
pe
r
s
on
a
li
z
e
d
hi
r
in
g
e
xpe
r
ie
nc
e
,
w
hi
c
h
in
vol
ve
s
m
a
tc
hi
ng
e
a
c
h
p
e
r
s
on
w
it
h
a
pos
it
i
on
th
a
t
a
li
gns
w
it
h
th
e
ir
pe
r
s
ona
li
ty
a
nd
s
ki
ll
s
.
P
e
r
s
ona
li
z
a
ti
on
de
pe
nds
on
da
ta
c
om
p
a
ni
e
s
c
ol
le
c
t
f
r
om
pe
opl
e
'
s
in
te
r
a
c
ti
ons
a
nd
b
e
ha
vi
or
s
,
he
lp
in
g
th
e
m
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
,
V
ol
. 14, No. 5, O
c
to
be
r
2025
:
4271
-
4278
4272
bui
ld
r
e
la
ti
ons
hi
ps
ba
s
e
d
on
tr
us
t
a
nd
unde
r
s
ta
ndi
ng
[
2]
.
M
or
e
ove
r
,
ne
w
te
c
hnol
ogi
e
s
e
na
bl
e
us
to
a
na
ly
z
e
va
s
t
a
m
ount
s
of
i
nf
or
m
a
ti
on a
nd ga
in
a
m
or
e
pr
e
c
is
e
unde
r
s
t
a
n
di
ng of
i
ndi
vi
dua
ls
.
F
or
e
xa
m
pl
e
,
G
om
e
z
e
t
al
.
[
3]
de
m
ons
tr
a
te
d
th
a
t
m
ul
ti
li
ngua
l
na
tu
r
a
l
la
ngua
ge
pr
oc
e
s
s
in
g
(
N
L
P
)
te
c
hni
que
s
c
a
n
a
c
c
ur
a
te
ly
e
xt
r
a
c
t
e
m
pl
oye
e
opi
ni
ons
f
r
om
da
ta
in
di
f
f
e
r
e
nt
la
ngua
ge
s
,
a
id
in
g
hum
a
n
r
e
s
our
c
e
s
(
H
R
)
m
a
na
ge
m
e
nt
de
c
i
s
io
ns
.
A
ddi
ti
ona
ll
y,
J
os
hi
e
t
al
.
[
4]
s
how
e
d
th
a
t
us
in
g
m
ul
ti
li
ngua
l
N
L
P
m
ode
ls
in
c
r
e
a
s
e
s
th
e
a
c
c
ur
a
c
y
of
e
m
pl
oye
e
e
nga
ge
m
e
nt
a
s
s
e
s
s
m
e
nt
s
a
nd
c
ont
r
ib
ut
e
s
to
de
ve
lo
pi
ng
c
om
pr
e
he
ns
iv
e
H
R
s
tr
a
te
gi
e
s
.
M
a
c
hi
ne
le
a
r
ni
ng
(
M
L
)
a
nd
d
e
e
p
le
a
r
ni
ng
(
D
L
)
ha
ve
e
m
e
r
ge
d
a
s
pow
e
r
f
ul
to
ol
s
t
o t
a
c
kl
e
m
a
ny c
ha
ll
e
nge
s
w
e
f
a
c
e
i
n s
oc
ie
ty
, p
a
r
ti
c
ul
a
r
ly
i
n r
e
c
r
ui
tm
e
nt
[
5]
.
E
m
pl
oye
e
a
nd
or
ga
ni
z
a
ti
ona
l
ha
r
m
ony
is
e
s
s
e
nt
ia
l
f
or
jo
b
s
a
ti
s
f
a
c
ti
on
a
nd
r
e
te
nt
io
n
[
6]
.
A
s
a
r
e
s
ul
t,
e
m
pl
oye
r
s
s
e
e
k
c
a
ndi
da
te
s
w
it
h
th
e
ne
c
e
s
s
a
r
y
a
bi
li
ti
e
s
a
nd
a
pe
r
s
ona
li
ty
th
a
t
c
om
pl
e
m
e
nt
s
th
e
pos
it
io
n
a
nd
th
e
c
om
pa
ny'
s
c
ul
tu
r
e
.
P
e
r
s
ona
li
ty
te
s
t
s
,
bot
h
onl
in
e
a
nd
in
-
pe
r
s
on,
ha
ve
b
e
c
om
e
e
s
s
e
nt
ia
l
to
ol
s
in
th
e
hi
r
in
g
pr
oc
e
s
s
[
7]
.
A
c
c
or
di
ng
to
J
a
in
e
t
al
.
[
8
]
,
hype
r
-
pe
r
s
ona
li
z
a
ti
on
gi
ve
s
c
us
to
m
e
r
s
a
m
or
e
in
di
vi
dua
li
z
e
d
e
xpe
r
ie
nc
e
,
w
hi
c
h
im
pa
c
t
s
th
e
ir
e
nga
ge
m
e
nt
.
T
hi
s
s
tu
dy
lo
oks
in
to
di
gi
ta
l
c
li
e
nt
e
li
ng
f
or
hype
r
-
pe
r
s
ona
li
z
a
ti
on
in
th
e
f
a
s
hi
on
in
dus
tr
y,
w
it
h
c
o
-
c
r
e
a
ti
on
a
s
a
f
a
c
il
it
a
to
r
.
A
s
id
e
f
r
om
s
ta
nda
r
d
s
our
c
e
s
li
ke
c
ove
r
l
e
tt
e
r
s
or
c
ur
r
ic
ul
um
v
it
a
e
s
(
C
V
s
)
, s
oc
ia
l
ne
twor
ki
ng s
it
e
s
l
ik
e
F
a
c
e
book c
a
n pr
ovi
de
i
nf
or
m
a
ti
on a
bout
a
pe
r
s
on'
s
pe
r
s
ona
li
ty
a
nd pr
e
di
c
t
e
m
pl
oym
e
nt
s
uc
c
e
s
s
[
9]
, [
10]
.
O
nl
in
e
pe
r
s
ona
li
ty
f
or
e
c
a
s
ts
a
c
c
ur
a
t
e
ly
pr
e
di
c
t
ge
nui
ne
pe
r
s
ona
li
ty
t
r
a
it
s
[
11]
. T
he
s
e
a
dva
nc
e
s
i
n our
unde
r
s
ta
ndi
ng
of
be
ha
vi
or
a
l
da
ta
a
nd
pe
r
s
ona
li
ty
tr
a
it
s
hi
ghl
ig
h
t
th
e
ne
e
d
f
or
m
o
r
e
s
ophi
s
ti
c
a
te
d
m
e
th
ods
f
or
a
s
s
e
s
s
in
g
c
a
ndi
da
t
e
s
.
R
e
c
e
nt
r
e
s
e
a
r
c
h
ha
s
f
oc
u
s
e
d
on
id
e
nt
if
yi
ng
a
nd
c
a
te
gor
iz
in
g
r
e
le
va
nt
s
ki
ll
s
f
r
om
te
xt
da
ta
t
o i
m
pr
ove
hi
r
in
g m
ode
ls
, a
lt
hough tr
a
di
ti
ona
l
to
ol
s
, s
uc
h a
s
pe
r
s
ona
li
ty
t
e
s
ts
, a
r
e
s
ti
ll
w
id
e
ly
u
s
e
d.
S
a
yf
ul
li
n
a
e
t
a
l
.
[
12]
e
xpl
or
e
d s
e
v
e
r
a
l
s
ki
ll
c
la
s
s
if
ic
a
ti
on m
e
th
od
s
by
a
nnot
a
ti
ng
a
d
a
ta
s
e
t
f
or
ha
r
d
a
nd
s
of
t
ta
l
e
nt
s
.
F
e
w
s
t
udi
e
s
h
a
ve
e
xa
m
i
ne
d
how
hy
pe
r
-
pe
r
s
on
a
li
z
a
ti
o
n,
w
hi
c
h
h
a
s
be
e
n
e
f
f
e
c
ti
v
e
ly
u
s
e
d
i
n
in
dus
tr
i
e
s
li
ke
f
a
s
hi
on
a
nd
m
a
r
ke
ti
n
g,
c
a
n
b
e
u
s
e
d
i
n
r
e
c
r
u
it
m
e
nt
to
e
n
ha
n
c
e
c
a
nd
id
a
t
e
-
r
ol
e
a
li
gnm
e
nt
d
e
s
pi
te
th
e
s
e
d
e
ve
l
opm
e
nt
s
i
n s
ki
ll
c
la
s
s
if
ic
a
ti
on.
T
a
m
bur
r
i
e
t
al
.
[
13]
c
la
s
s
if
ie
d
phr
a
s
e
s
c
ont
a
in
in
g
s
ki
ll
s
by
m
a
nu
a
ll
y
la
be
li
ng
jo
b
de
s
c
r
ip
ti
ons
,
w
h
e
r
e
a
s
th
e
r
e
s
e
a
r
c
h
i
n
[
14]
,
[
15]
a
ppr
oa
c
h
e
d
s
ki
ll
e
xt
r
a
c
t
io
n
a
s
a
m
ul
ti
-
l
a
be
l
c
la
s
s
if
i
c
a
ti
on
pr
obl
e
m
u
s
in
g
bi
d
ir
e
c
ti
ona
l
e
nc
ode
r
r
e
pr
e
s
e
nt
a
ti
ons
f
r
om
tr
a
ns
f
or
m
e
r
s
(
B
E
R
T
)
.
E
s
ta
bl
is
hi
ng
a
r
e
la
ti
on
s
hi
p be
t
w
e
e
n p
e
r
s
o
na
li
t
y c
h
a
r
a
c
te
r
i
s
ti
c
s
,
s
of
t
s
ki
l
ls
,
a
nd
w
or
k f
it
i
s
vi
t
a
l.
T
hi
s
s
t
ud
y
b
ui
ld
s
on
ou
r
p
r
e
vi
ous
r
e
s
e
a
r
c
h
[
16
]
,
w
h
ic
h
e
x
p
lo
r
e
d
th
e
us
e
o
f
s
o
f
t
s
ki
ll
s
a
n
d
b
ig
f
i
ve
pe
r
s
ona
li
ty
tr
a
i
ts
t
o
i
m
p
r
o
ve
c
a
ndi
da
te
-
r
ol
e
m
a
t
c
h
in
g
,
l
e
ve
r
a
gi
ng
th
e
m
to
e
nha
nc
e
th
e
e
f
f
ic
ie
n
c
y
of
in
n
ova
ti
ve
m
od
e
ls
to
m
e
e
t
t
he
r
e
q
ui
r
e
m
e
n
ts
o
f
m
a
t
c
hi
ng
th
e
r
i
gh
t
e
m
pl
oye
e
t
o
t
he
a
pp
r
op
r
ia
t
e
f
unc
ti
ona
l
a
r
e
a
m
o
r
e
a
c
c
ur
a
te
ly
a
nd
e
f
f
e
c
t
iv
e
ly
.
T
he
c
ur
r
e
nt
s
tu
d
y
a
im
s
to
p
r
e
s
e
nt
th
e
r
e
s
u
l
ts
o
f
a
pp
ly
in
g
a
pr
e
-
tr
a
in
e
d
m
o
de
l
bu
il
t
us
i
ng
th
e
s
a
m
e
da
ta
f
r
om
ou
r
p
r
e
v
io
us
s
t
udy
.
A
dd
it
io
na
l
ly
,
th
e
s
tu
dy
f
oc
us
e
s
o
n
pr
ovi
di
ng
p
e
r
s
ona
li
z
e
d
r
e
c
o
m
m
e
nda
ti
ons
to
i
nd
iv
id
ua
ls
w
it
hi
n
th
e
f
u
nc
t
io
na
l
a
r
e
a
.
I
t
e
x
te
n
ds
p
r
e
v
io
us
w
or
k
by
de
e
p
e
ni
ng
o
u
r
un
de
r
s
ta
ndi
ng
o
f
s
u
bt
le
pe
r
s
o
na
l
it
y
t
r
a
it
s
a
nd
t
he
i
r
a
p
pl
ic
a
ti
ons
i
n
r
e
c
r
ui
tm
e
nt
.
W
e
a
ls
o
p
r
e
s
e
nt
ne
w
r
e
s
ul
ts
de
m
o
ns
t
r
a
ti
n
g
h
ow
pr
e
-
t
r
a
in
e
d
m
ode
ls
c
a
n
e
f
f
e
c
t
iv
e
ly
pe
r
s
o
na
l
iz
e
r
e
c
o
m
m
e
nd
a
t
io
ns
a
c
r
os
s
di
f
f
e
r
e
n
t
s
c
e
na
r
i
os
.
I
n
t
hi
s
c
o
nt
e
xt
,
ou
r
s
tu
dy
r
e
l
ie
s
on
m
ode
r
n
t
r
a
ns
f
o
r
m
e
r
-
b
a
s
e
d
m
ode
ls
,
DL
m
o
de
ls
d
e
s
ig
ne
d
t
o
un
de
r
s
ta
n
d
la
n
gua
ge
a
nd
s
u
b
tl
e
m
e
a
ni
ngs
w
it
h
in
t
e
xt
s
[
17
]
.
W
e
e
m
p
lo
ye
d
th
e
s
e
m
od
e
ls
:
B
E
R
T
,
r
obus
tl
y
op
t
im
iz
e
d
B
E
R
T
R
oB
E
R
T
a
,
a
n
d
c
r
os
s
-
l
in
g
ua
l
la
n
gua
ge
m
ode
l
(
X
L
M
)
-
R
oB
E
R
T
a
,
w
h
ic
h
s
up
po
r
t
a
dva
nc
e
d
c
a
p
a
b
il
it
ie
s
f
o
r
p
r
e
c
is
e
pe
r
s
o
na
l
iz
a
ti
o
n
ba
s
e
d
o
n
m
u
lt
ip
le
c
on
te
x
ts
[
1
8]
,
ty
p
ic
a
ll
y b
a
s
e
d
on
pe
r
s
o
na
l
it
y
tr
a
i
ts
.
T
hi
s
s
t
udy
c
on
t
r
ib
ut
e
s
to
w
o
r
k
pl
a
c
e
hy
pe
r
-
pe
r
s
on
a
l
iz
a
t
io
n
b
y
in
c
r
e
a
s
in
g
k
now
le
d
ge
o
f
h
ow
A
I
,
pa
r
ti
c
ul
a
r
ly
t
r
a
ns
f
or
m
e
r
m
o
de
ls
,
c
a
n
i
m
p
r
o
ve
m
o
de
l
pe
r
f
or
m
a
nc
e
in
c
h
oos
i
ng
t
he
be
s
t
c
a
nd
id
a
te
f
o
r
a
pa
r
ti
c
ul
a
r
f
u
nc
t
io
na
l
a
r
e
a
.
I
t
a
im
s
to
m
a
tc
h
i
nd
iv
i
dua
ls
w
it
h
a
pp
r
op
r
ia
t
e
jo
bs
ba
s
e
d
o
n
th
e
i
r
pe
r
s
o
na
l
it
ie
s
a
nd
c
ha
r
a
c
t
e
r
is
t
ic
s
.
A
dd
i
ti
o
na
l
ly
,
t
he
s
tu
d
y
o
f
f
e
r
s
a
th
o
r
oug
h
a
na
ly
s
is
a
n
d
de
ve
l
op
m
e
n
t
o
f
s
tr
a
te
gi
e
s
f
or
T
r
a
ns
f
o
r
m
e
r
m
ode
ls
to
a
c
hi
e
ve
th
e
be
s
t
p
os
s
ib
le
c
a
n
di
da
t
e
-
to
-
jo
b
m
a
tc
h
a
c
r
os
s
va
r
io
us
c
a
r
e
e
r
f
ie
l
ds
.
A
c
c
o
r
d
in
g
ly
,
i
t
r
e
p
r
e
s
e
n
ts
a
s
i
gn
if
ic
a
nt
s
c
ie
n
ti
f
ic
c
o
nt
r
i
bu
ti
o
n
to
ou
r
u
nde
r
s
ta
nd
in
g
o
f
ho
w
A
I
-
pow
e
r
e
d
m
ode
ls
c
a
n
s
pe
e
d
u
p
a
n
d
i
nc
r
e
a
s
e
th
e
a
c
c
u
r
a
c
y
o
f
ta
le
nt
a
c
q
ui
s
i
ti
on
pr
oc
e
s
s
e
s
.
e
n
ha
nc
e
ta
le
n
t
a
c
qui
s
it
io
n
pr
oc
e
s
s
e
s
d
ir
e
c
t
ly
im
p
a
c
ts
hi
r
i
ng
q
ua
l
it
y
a
n
d
m
a
tc
h
in
g
th
e
r
i
gh
t
pe
r
s
on
to
th
e
r
i
gh
t
jo
b.
T
h
e
a
pp
r
o
a
c
h
in
t
r
odu
c
e
s
a
un
iq
ue
m
e
th
o
d
f
o
r
c
on
ne
c
ti
n
g
j
ob
s
e
e
k
e
r
s
w
it
h
s
ui
ta
b
le
j
obs
b
y
le
v
e
r
a
gi
ng
p
e
r
s
ona
li
ty
va
r
ia
b
le
s
a
s
s
oc
ia
te
d
w
i
th
s
of
t
s
k
il
ls
.
T
hi
s
a
pp
r
o
a
c
h
s
upp
or
ts
th
e
hi
r
i
ng
pr
oc
e
s
s
to
in
c
r
e
a
s
e
c
r
e
a
t
iv
it
y
a
n
d
di
ve
r
s
it
y
a
nd
m
a
tc
h
pe
r
s
o
na
l
it
ie
s
t
o
jo
b
r
e
qui
r
e
m
e
n
ts
,
r
e
ga
r
dl
e
s
s
o
f
e
x
pe
r
ie
nc
e
.
I
t
a
im
s
to
r
e
du
c
e
f
a
v
or
it
is
m
,
pr
om
ot
e
e
m
pl
oye
e
r
e
te
nt
io
n,
a
nd
i
m
p
r
o
ve
c
om
pa
ny
e
f
f
ic
ie
nc
y
.
T
he
s
t
udy
de
ve
l
ops
p
r
e
d
ic
ti
v
e
la
ng
ua
g
e
m
od
e
ls
w
i
th
T
r
a
ns
f
o
r
m
e
r
M
od
e
ls
t
o
m
a
ke
i
t
e
a
s
ie
r
to
f
i
nd
id
e
a
l
c
a
ndi
da
te
s
ba
s
e
d
o
n
th
e
ir
s
o
f
t
s
ki
l
ls
f
o
r
i
d
e
a
l
j
obs
[
1
9
]
.
2.
M
E
T
H
O
D
T
hi
s
s
e
c
ti
on
c
ont
a
in
s
a
d
e
ta
il
e
d
de
s
c
r
ip
ti
on
of
our
s
tu
dy'
s
da
ta
s
e
ts
,
m
od
e
ls
,
a
nd
m
ode
li
ng
m
e
th
odol
ogi
e
s
.
O
ur
r
e
s
e
a
r
c
h
a
im
s
to
im
pr
ove
knowle
dge
of
t
he
c
om
pl
e
x
r
e
la
ti
on
s
hi
ps
b
e
twe
e
n
s
of
t
s
ki
ll
s
,
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
AI
-
dr
iv
e
n hy
pe
r
-
pe
r
s
onal
i
z
at
io
n and tr
ans
fe
r
le
ar
ni
ng f
o
r
pr
e
c
i
s
io
n r
e
c
r
ui
tme
nt
(
N
our
A
lq
udah
)
4273
pe
r
s
ona
li
ty
c
ha
r
a
c
te
r
is
ti
c
s
,
a
nd
or
ga
ni
z
a
ti
ona
l
f
unc
ti
ona
l
r
ol
e
s
.
U
s
in
g
tr
a
ns
f
or
m
e
r
m
ode
ls
,
w
e
hope
to
d
r
a
w
c
r
uc
ia
l
in
s
ig
ht
s
a
nd
pr
om
ot
e
in
f
or
m
e
d
de
c
is
io
n
-
m
a
ki
ng
in
w
or
k
s
c
e
na
r
io
s
.
T
hi
s
c
om
pr
e
he
ns
iv
e
m
e
th
od
e
na
bl
e
s
u
s
to
id
e
nt
if
y
pa
tt
e
r
ns
a
nd
c
or
r
e
la
ti
ons
th
a
t
m
ig
ht
be
c
r
it
ic
a
l
in
m
a
tc
hi
ng
in
di
vi
dua
ls
to
th
e
m
os
t
a
ppr
opr
ia
te
e
m
pl
oym
e
nt
oppor
tu
ni
ti
e
s
ba
s
e
d on the
ir
uni
que
t
r
a
it
s
.
2.1.
D
at
a s
e
t
d
e
s
c
r
ab
t
io
n
C
e
nt
r
a
l
to
our
r
e
s
e
a
r
c
h
is
th
e
H
Y
P
s
ki
ll
da
ta
s
e
t
[
16]
,
a
m
e
ti
c
ul
ous
ly
c
ur
a
te
d
c
ol
le
c
ti
on
c
om
pr
is
in
g
9,802
di
s
ti
nc
t
s
e
nt
e
nc
e
s
s
our
c
e
d
f
r
om
a
va
r
ie
ty
of
r
e
pos
it
or
ie
s
.
T
he
s
e
in
c
lu
de
m
a
nu
a
ll
y
s
a
m
pl
e
d
r
e
s
um
e
s
f
r
om
di
ve
r
s
e
ba
c
kgr
ounds
a
nd
H
R
da
ta
ba
s
e
s
,
a
lo
ngs
id
e
jo
b
d
e
s
c
r
ip
ti
ons
e
xt
r
a
c
te
d
f
r
om
pl
a
tf
o
r
m
s
s
uc
h
a
s
G
la
s
s
door
a
nd K
a
ggl
e
'
s
A
m
a
z
on j
obs
da
ta
s
e
t.
C
r
uc
ia
ll
y, t
he
da
t
a
s
e
t
is
unde
r
pi
nne
d by the
e
xt
e
ns
iv
e
H
Y
P
s
ki
ll
L
e
xi
c
on,
e
nc
om
pa
s
s
in
g
2,190
c
a
te
gor
iz
e
d
te
r
m
s
s
pa
nni
ng
do
m
a
in
s
li
ke
I
nt
e
r
na
ti
ona
l
P
e
r
s
ona
li
ty
I
te
m
P
oo
l
(
I
P
I
P
)
, j
ob pos
ti
ngs
, r
e
s
um
e
s
, i
nt
e
r
vi
e
w
s
, a
nd U
R
L
s
.
O
ur
da
ta
s
e
t
und
e
r
w
e
nt
m
e
ti
c
ul
ous
a
nnot
a
ti
on
by
th
r
e
e
P
h.D
.
l
e
ve
l
e
xpe
r
ts
in
ps
yc
hol
ogy
,
ut
il
iz
in
g
th
e
r
ig
or
ous
I
P
I
P
-
N
E
O
-
120
tr
a
it
s
a
nd
tr
a
it
s
s
tu
dy
f
r
a
m
e
w
or
k
.
T
hi
s
pr
oc
e
s
s
in
vol
ve
d
bi
na
r
y
c
la
s
s
if
ic
a
ti
on
to
id
e
nt
if
y
s
e
nt
e
nc
e
s
c
ont
a
in
in
g
s
of
t
s
ki
ll
s
,
a
nd
m
ul
ti
-
c
la
s
s
c
la
s
s
i
f
ic
a
ti
on
to
a
s
s
ig
n
s
e
nt
e
nc
e
s
to
one
of
th
e
f
iv
e
bi
g
f
iv
e
pe
r
s
ona
li
ty
tr
a
it
c
a
te
gor
ie
s
.
M
or
e
ove
r
,
our
s
tu
dy
in
t
r
oduc
e
s
a
nove
l
a
ppr
oa
c
h
by
m
a
ppi
ng
th
e
s
e
pe
r
s
ona
li
ty
t
r
a
it
s
t
o s
pe
c
if
ic
f
unc
ti
ona
l
a
r
e
a
s
w
it
hi
n or
ga
ni
z
a
ti
ons
:
−
O
pe
nne
s
s
w
it
h
pr
oduc
ti
on
−
C
ons
c
ie
nt
io
us
n
e
s
s
w
it
h
f
in
a
nc
e
−
E
xt
r
a
ve
r
s
io
n w
it
h
m
a
r
ke
ti
ng/
s
a
le
s
−
A
gr
e
e
a
bl
e
ne
s
s
w
it
h
HR
−
E
m
ot
io
na
l
s
ta
bi
li
ty
w
it
h
ope
r
a
ti
ons
T
hi
s
in
f
or
m
a
ti
on
s
e
ts
th
e
f
ounda
ti
on
of
our
r
e
s
e
a
r
c
h,
pr
ovi
di
ng
in
s
ig
ht
s
in
to
th
e
da
ta
s
e
t,
a
nnot
a
ti
on,
a
nd
c
la
s
s
if
ic
a
ti
on
ta
s
ks
c
onduc
te
d
in
our
pr
e
vi
ous
s
tu
dy
[
16]
.
T
he
s
e
c
om
pone
nt
s
w
e
r
e
c
r
uc
ia
l
in
s
h
a
pi
ng
th
e
m
ode
l
de
s
ig
n
a
nd
th
e
m
a
ppi
ng
of
s
of
t
s
ki
ll
s
to
pe
r
s
ona
li
ty
tr
a
it
s
.
B
ui
ld
in
g
th
is
gr
oundwor
k
a
ll
ow
e
d
us
to
e
xt
e
nd our
pr
e
vi
ous
f
in
di
ngs
i
nt
o m
or
e
a
dva
nc
e
d a
ppl
ic
a
ti
ons
u
s
in
g t
r
a
ns
f
or
m
e
r
-
ba
s
e
d m
ode
ls
.
2.2.
D
at
a
p
r
e
p
r
oc
e
s
s
in
g
I
n
th
is
pa
r
t,
w
e
w
il
l
de
s
c
r
ib
e
s
om
e
pr
e
-
pr
oc
e
s
s
in
g
pr
oc
e
dur
e
s
th
a
t
w
e
r
e
us
e
d
in
th
i
s
s
tu
dy.
V
a
r
io
us
pr
e
-
pr
oc
e
s
s
in
g
s
te
p
s
ha
v
e
be
e
n
te
s
t
e
d,
a
nd
m
a
ny
a
ppr
oa
c
h
e
s
ha
ve
be
e
n
us
e
d
.
T
a
bl
e
1
di
s
pl
a
ys
in
s
ta
nc
e
s
of
th
e
s
e
pr
oc
e
s
s
e
s
. M
or
e
e
xa
m
pl
e
s
of
w
ha
t
pr
e
-
pr
oc
e
s
s
in
g pr
oc
e
s
s
e
s
w
e
r
e
e
m
pl
oye
d a
r
e
a
s
f
ol
lo
w
s
.
T
a
bl
e
1. E
xa
m
pl
e
s
of
pr
e
-
pr
oc
e
s
s
in
g s
te
ps
T
e
c
hni
que
E
xa
m
pl
e
s
R
e
m
ovi
ng punc
t
ua
t
i
ons
“
!
, +, :, ;
, ?
, @
”
R
e
m
ovi
ng i
de
nt
i
f
i
e
r
s
“
t
he
”
,
“
a”
,
a
nd
“
an
”
R
e
m
ovi
ng s
t
opw
or
ds
“
H
e
”
,
“
T
he
y”
,
“
i
s
”
,
a
nd
“
on
”
E
xpa
ndi
ng a
bbr
e
vi
a
t
i
ons
“
I
’
m
”
,
“
c
a
n’
t
”
,
“
i
nt
o
,
“
I
a
m
”
,
a
nd “
C
a
n not
”
L
e
m
m
a
t
i
z
a
t
i
on
“
be
e
n”
, “
ha
d”
,
“
i
nt
o”
,
“
be
”
,
a
nd
“
ha
s
/
ha
v
e
”
2.2.1.
T
h
e
ove
r
s
a
m
p
li
n
g t
e
c
h
n
iq
u
e
I
n
a
ddr
e
s
s
in
g
our
s
tu
dy'
s
c
ha
ll
e
nge
s
,
w
e
e
nc
ount
e
r
e
d
th
e
is
s
ue
of
a
n
im
ba
la
nc
e
d
da
ta
s
e
t
[
20]
,
w
he
r
e
c
e
r
ta
in
c
la
s
s
e
s
ha
d
s
ig
ni
f
ic
a
nt
ly
f
e
w
e
r
in
s
ta
nc
e
s
c
om
pa
r
e
d
to
ot
he
r
s
.
T
hi
s
im
ba
la
nc
e
c
a
n
s
e
ve
r
e
ly
im
pa
c
t
c
la
s
s
if
ic
a
ti
on
m
ode
l
p
e
r
f
or
m
a
nc
e
,
of
te
n
le
a
di
ng
to
bi
a
s
e
d
p
r
e
di
c
ti
ons
f
a
vor
in
g
th
e
m
a
jo
r
it
y
c
la
s
s
[
21]
.
T
o
m
it
ig
a
te
th
is
is
s
u
e
,
w
e
im
pl
e
m
e
nt
e
d
ove
r
s
a
m
pl
in
g
s
pe
c
if
ic
a
ll
y
f
or
our
m
ul
ti
-
c
la
s
s
c
la
s
s
if
ic
a
ti
on
m
ode
ls
,
w
hi
c
h
yi
e
ld
e
d
im
pr
ove
d
r
e
s
ul
t
s
.
T
hi
s
te
c
hni
qu
e
,
a
w
e
ll
-
e
s
ta
bl
is
he
d
te
xt
a
ugm
e
nt
a
ti
on
m
e
th
od,
in
vol
ve
s
dupl
ic
a
ti
ng
in
s
ta
nc
e
s
in
th
e
m
in
or
it
y
c
la
s
s
e
s
to
e
qua
li
z
e
th
e
ir
r
e
pr
e
s
e
nt
a
ti
on
w
it
h
th
e
m
a
jo
r
it
y
c
la
s
s
e
s
.
W
e
e
m
pl
oye
d
a
r
a
ndom
ove
r
s
a
m
pl
in
g
a
ppr
oa
c
h
to
a
c
hi
e
ve
a
ba
la
nc
e
d
c
la
s
s
di
s
tr
ib
ut
io
n.
T
a
bl
e
2
il
lu
s
tr
a
te
s
th
e
i
ni
ti
a
l
a
nd pos
t
-
r
e
s
a
m
pl
in
g c
ount
s
of
t
r
a
in
in
g da
ta
s
e
nt
e
n
c
e
s
a
c
r
os
s
e
a
c
h c
la
s
s
.
T
a
bl
e
2.
T
he
numbe
r
of
da
ta
in
e
a
c
h c
l
a
s
s
a
f
te
r
t
he
r
e
s
a
m
pl
in
g
t
e
c
hni
que
B
i
g F
i
ve
t
r
a
i
t
s
N
um
be
r
f
r
om
t
he
t
r
a
i
ni
ng da
t
a
(
s
e
nt
e
nc
e
s
)
be
f
or
e
r
e
-
s
a
m
pl
i
ng
N
um
be
r
f
r
om
t
he
t
r
a
i
ni
ng da
t
a
(
s
e
nt
e
nc
e
s
)
a
f
t
e
r
r
e
-
s
a
m
pl
i
ng
O
pe
nne
s
s
1,298
-
C
ons
c
i
e
nt
i
ous
n
e
s
s
648
1,000
E
xt
r
a
ve
r
s
i
on
1,040
-
A
gr
e
e
a
bl
e
ne
s
s
522
1,000
E
m
ot
i
ona
l
s
t
a
bi
l
i
t
y
537
1,000
T
ot
a
l
4,045
5,080
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
,
V
ol
. 14, No. 5, O
c
to
be
r
2025
:
4271
-
4278
4274
2.3. T
r
an
s
f
e
r
l
e
ar
n
in
g m
od
e
l
s
B
ui
ld
in
g
on
th
e
s
uc
c
e
s
s
of
tr
a
ns
f
e
r
le
a
r
ni
ng
f
r
om
p
r
e
-
tr
a
in
e
d
m
ode
ls
in
N
L
P
ta
s
ks
,
our
r
e
s
e
a
r
c
h
in
c
lu
de
s
f
in
e
-
tu
ni
ng
th
e
s
e
m
ode
ls
to
id
e
nt
if
y
s
of
t
s
ki
ll
s
in
a
bi
na
r
y
c
la
s
s
if
ic
a
ti
on
te
s
t
a
nd
pe
r
s
on
a
li
ty
c
ha
r
a
c
te
r
is
ti
c
s
ba
s
e
d
on
th
e
bi
g
f
iv
e
in
a
m
ul
ti
-
c
la
s
s
c
la
s
s
if
ic
a
ti
on
jo
b.
W
e
e
va
lu
a
te
d
th
e
e
f
f
ic
a
c
y
of
m
ul
ti
li
ngua
l
pr
e
-
t
r
a
in
e
d
m
ode
ls
by
te
s
ti
ng
m
a
ny
of
th
e
m
,
in
c
lu
di
ng
B
E
R
T
,
R
oB
E
R
T
a
,
a
nd
X
L
M
-
R
oB
E
R
T
a
.
T
he
pur
pos
e
of
th
e
bi
n
a
r
y
c
la
s
s
if
ic
a
ti
on
e
xe
r
c
is
e
w
a
s
to
d
e
te
c
t
th
e
e
xi
s
te
nc
e
of
s
of
t
s
ki
ll
s
in
pr
ovi
de
d
te
xt
s
.
T
he
m
ul
ti
-
c
la
s
s
c
la
s
s
if
ic
a
ti
on
e
xe
r
c
is
e
a
tt
e
m
pt
e
d
to
c
a
te
gor
iz
e
phr
a
s
e
s
b
a
s
e
d
on
th
e
bi
g
f
iv
e
pe
r
s
ona
li
ty
qua
li
ti
e
s
of
ope
nne
s
s
, c
ons
c
i
e
nt
io
us
ne
s
s
, e
xt
r
a
ve
r
s
io
n, a
gr
e
e
a
bl
e
ne
s
s
,
a
nd e
m
ot
io
na
l
s
ta
bi
li
ty
.
T
he
e
va
lu
a
ti
ons
w
e
r
e
p
e
r
f
or
m
e
d
done
in
th
e
G
oogl
e
C
ol
a
b
e
nvi
r
onm
e
nt
,
w
hi
c
h
ut
il
iz
e
d
it
s
c
om
put
in
g
c
a
pa
bi
li
ti
e
s
f
or
m
ode
l
tr
a
in
in
g
a
nd
f
in
e
-
tu
ni
ng.
W
e
us
e
d
th
e
s
im
pl
e
tr
a
ns
f
or
m
e
r
s
p
a
c
ka
ge
,
w
hi
c
h
s
im
pl
if
ie
s
th
e
de
s
ig
n
of
T
r
a
ns
f
or
m
e
r
-
ba
s
e
d
m
ode
ls
a
nd
e
na
b
le
s
r
a
pi
d
f
in
e
-
tu
ni
ng
of
pr
e
-
tr
a
in
e
d
la
ngua
ge
m
ode
ls
.
T
he
da
ta
s
e
t
w
a
s
di
vi
de
d
in
to
tr
a
in
in
g
a
nd
te
s
ti
ng
s
e
ts
to
c
a
r
e
f
ul
ly
a
na
ly
z
e
th
e
m
ode
l'
s
pe
r
f
or
m
a
nc
e
.
T
he
tr
a
in
in
g
s
e
t
w
a
s
ut
il
iz
e
d
to
f
in
e
-
tu
ne
th
e
m
ode
l
s
,
w
hi
le
t
he
te
s
ti
ng
s
e
t
w
a
s
r
e
ta
in
e
d
f
or
a
s
s
e
s
s
in
g
th
e
m
ode
l'
s
ge
ne
r
a
li
z
a
ti
on
s
ki
ll
s
.
T
he
s
c
ik
it
-
le
a
r
n
P
yt
hon
pa
c
ka
ge
w
a
s
us
e
d
to
pr
oduc
e
e
va
lu
a
ti
on
m
e
a
s
ur
e
s
s
u
c
h
a
s
a
c
c
ur
a
c
y,
r
e
c
a
ll
,
pr
e
c
is
io
n,
a
nd
F
1
-
s
c
or
e
,
e
ns
ur
in
g
a
th
or
o
ugh
a
s
s
e
s
s
m
e
nt
of
th
e
m
ode
ls
'
pe
r
f
or
m
a
nc
e
.
A
c
c
ur
a
c
y
e
v
a
lu
a
te
s
th
e
m
od
e
l'
s
ove
r
a
ll
a
c
c
ur
a
c
y,
w
he
r
e
a
s
p
r
e
c
is
io
n
a
nd
r
e
c
a
ll
of
f
e
r
in
f
or
m
a
ti
on
on
th
e
m
ode
l'
s
pe
r
f
or
m
a
nc
e
in
c
e
r
ta
in
c
la
s
s
e
s
.
T
he
F
1
-
s
c
or
e
,
w
hi
c
h
is
th
e
ha
r
m
oni
c
m
e
a
ns
of
a
c
c
ur
a
c
y
a
nd
r
e
c
a
ll
,
w
a
s
ve
r
y i
m
por
ta
nt
i
n de
te
r
m
in
in
g t
he
opt
im
a
l
ba
la
nc
e
of
pr
e
c
is
io
n a
nd r
e
c
a
ll
.
W
e
a
ls
o
a
dd
r
e
s
s
e
d
th
e
is
s
u
e
of
c
l
a
s
s
i
m
b
a
la
nc
e
,
pa
r
ti
c
ul
a
r
ly
in
th
e
m
ul
ti
-
c
la
s
s
c
la
s
s
if
ic
a
ti
on
jo
b,
by
us
in
g
ove
r
s
a
m
pl
in
g
te
c
hn
iq
ue
s
t
o
gua
r
a
nt
e
e
th
a
t
m
in
or
it
y
c
la
s
s
e
s
w
e
r
e
a
pp
r
op
r
ia
t
e
ly
r
e
pr
e
s
e
n
te
d
i
n
t
he
tr
a
in
in
g
s
e
t.
T
hi
s
m
e
th
od
he
l
ps
to
r
e
duc
e
t
he
pos
s
ib
il
it
y
o
f
th
e
m
o
de
l
be
in
g
bi
a
s
e
d
to
w
a
r
d
th
e
m
a
jo
r
i
ty
c
la
s
s
,
b
oos
ti
ng
t
he
m
o
de
l
'
s
r
ob
us
t
ne
s
s
a
nd
f
a
ir
ne
s
s
in
pr
e
di
c
ti
o
ns
.
I
n
c
o
nc
l
us
i
on
,
ou
r
w
o
r
k
us
e
d
t
r
a
ns
f
e
r
le
a
r
ni
ng
a
n
d
f
in
e
-
t
un
in
g
of
p
r
e
-
tr
a
in
e
d
m
o
de
ls
to
ob
ta
in
c
o
n
s
id
e
r
a
b
le
ou
tc
o
m
e
s
i
n
bi
na
r
y
a
nd
m
u
lt
i
-
c
la
s
s
c
la
s
s
i
f
ic
a
t
io
n
pr
ob
le
m
s
.
W
e
de
m
ons
t
r
a
te
d
t
he
a
bi
l
it
y
of
a
dva
nc
e
d
m
ode
ls
s
uc
h
a
s
B
E
R
T
,
R
o
B
E
R
T
a
,
a
nd
X
L
M
-
R
oB
E
R
T
a
to
de
te
c
t
s
o
f
t
s
k
il
ls
a
nd
pe
r
s
ona
li
ty
t
r
a
it
s
,
c
ont
r
ib
ut
in
g
va
lu
a
bl
e
i
ns
ig
ht
s
to
t
he
f
i
e
ld
o
f
N
L
P
a
nd
H
R
a
na
ly
t
ic
s
.
2.3.1.
B
E
R
T
(
b
as
e
)
G
oogl
e
'
s
B
E
R
T
[
22]
,
a
tr
a
ns
f
or
m
e
r
-
ba
s
e
d
pr
e
-
tr
a
in
e
d
m
ode
l,
w
a
s
us
e
d
f
or
our
ta
s
ks
.
B
E
R
T
is
a
va
il
a
bl
e
in
va
r
io
us
s
i
z
e
s
, s
uc
h
a
s
la
r
ge
a
nd
ba
s
e
,
di
f
f
e
r
in
g
in
hi
dde
n
la
ye
r
s
,
a
tt
e
nt
io
n
he
a
d
s
,
hi
dde
n
s
i
z
e
,
a
nd
pa
r
a
m
e
te
r
s
. F
or
t
hi
s
t
he
s
is
, t
he
(
ba
s
e
)
m
ode
l
w
a
s
f
in
e
-
tu
ne
d f
or
bot
h t
a
s
ks
.
2.3.2.
M
u
lt
il
in
gu
al
B
E
R
T
m
od
e
l
M
ul
ti
li
ngua
l
B
E
R
T
m
ode
l
[
23]
is
pr
e
-
t
r
a
in
e
d
in
th
e
to
p
104
la
ngua
ge
s
w
it
h
th
e
m
os
t
W
ik
ip
e
di
a
e
nt
r
ie
s
.
I
t
is
a
v
a
il
a
bl
e
in
two
ve
r
s
io
ns
:
ol
d
(
not
r
e
c
om
m
e
nde
d
)
a
nd
ne
w
(
r
e
c
om
m
e
nde
d)
.
T
he
n
e
w
ve
r
s
io
n
s
uppor
ts
104
la
ngua
g
e
s
,
a
ddr
e
s
s
in
g
nor
m
a
li
z
a
ti
on
c
onc
e
r
ns
in
va
r
io
us
la
ngua
ge
s
,
m
a
ki
ng
it
hi
ghl
y
r
e
c
om
m
e
nde
d,
e
s
pe
c
ia
ll
y
f
or
non
-
la
ti
n
a
lp
ha
be
ts
s
uc
h
a
s
A
r
a
bi
c
.
I
n
th
is
th
e
s
is
,
th
e
ne
w
ve
r
s
io
n
dow
nl
oa
de
d
f
r
om
t
he
T
e
ns
or
F
lo
w
hub wa
s
ut
il
iz
e
d.
2.3.3.
R
oB
E
R
T
a (
b
as
e
)
F
a
c
e
book'
s
R
oB
E
R
T
a
[
24]
,
a
n
e
nha
nc
e
d
ve
r
s
io
n
of
B
E
R
T
,
w
a
s
pr
opos
e
d
a
s
a
r
obus
tl
y
opt
im
iz
e
d
m
ode
l.
I
t
di
f
f
e
r
s
f
r
om
B
E
R
T
in
th
e
tr
a
in
in
g
pr
oc
e
s
s
a
nd
in
t
e
r
na
l
c
om
pos
it
io
n.
R
oB
E
R
T
a
om
it
s
th
e
n
e
xt
s
e
nt
e
nc
e
pr
e
di
c
ti
on
(
N
S
P
)
obj
e
c
ti
ve
,
f
e
a
tu
r
e
s
a
dyna
m
ic
m
a
s
k
th
a
t
tr
a
de
s
to
ke
ns
be
twe
e
n
tr
a
in
in
g
it
e
r
a
ti
ons
,
a
nd w
a
s
t
r
a
in
e
d on a
m
or
e
e
xt
e
nd
e
d s
e
qu
e
nc
e
of
t
oke
ns
w
it
h a
l
a
r
ge
r
ba
tc
h s
iz
e
c
om
pa
r
e
d t
o
B
E
R
T
.
2.3.4.
X
L
M
-
R
oB
E
R
T
a (
b
as
e
)
T
he
X
L
M
-
R
oB
E
R
T
a
m
ode
l
in
tr
oduc
e
d
by
r
e
s
e
a
r
c
h
in
[
24]
,
[
25]
.
I
t
is
ba
s
e
d
on
F
a
c
e
book'
s
R
oB
E
R
T
a
m
ode
l.
X
L
M
-
R
oB
E
R
T
a
is
a
la
r
ge
m
ul
ti
li
ngua
l
la
ngua
ge
m
ode
l
tr
a
in
e
d
us
in
g
2.5
T
B
of
f
il
te
r
e
d
C
om
m
onC
r
a
w
l
da
ta
.
2.4.
P
r
e
-
t
r
ai
n
e
d
m
od
e
li
n
g t
e
c
h
n
iq
u
e
T
he
pr
e
-
tr
a
in
e
d
m
ode
li
ng
t
e
c
hni
que
f
ol
lo
w
s
a
s
tr
a
ig
ht
f
or
w
a
r
d
a
ppr
oa
c
h.
W
e
f
ir
s
t
pr
e
pa
r
e
th
e
da
t
a
,
th
e
n
f
e
e
d
pr
e
-
pr
oc
e
s
s
e
d
s
e
nt
e
nc
e
s
in
to
a
c
ho
s
e
n
pr
e
-
tr
a
in
e
d
m
ode
l.
F
in
a
ll
y,
th
e
m
ode
l
de
li
ve
r
s
it
s
out
put
.
T
hi
s
pr
oc
e
s
s
is
il
lu
s
tr
a
te
d
in
F
ig
ur
e
1.
T
o
e
ns
ur
e
c
ons
is
te
nt
tr
a
in
in
g,
w
e
e
m
pl
oye
d
s
om
e
s
ta
nda
r
d
s
e
tt
in
gs
.
T
he
tr
a
in
in
g
ba
tc
h
s
i
z
e
w
a
s
s
e
t
to
32,
w
hi
le
th
e
e
va
lu
a
ti
on
ba
tc
h
s
iz
e
w
a
s
16.
A
ddi
ti
ona
ll
y,
f
or
tr
a
ns
f
e
r
le
a
r
ni
ng,
im
pl
e
m
e
nt
in
g
e
a
r
ly
s
to
ppi
ng
w
it
h
a
pa
ti
e
nc
e
of
3
w
a
s
c
r
uc
ia
l.
E
a
c
h
m
od
e
l
ha
lt
e
d
tr
a
in
in
g
upon
r
e
a
c
hi
ng
a
s
p
e
c
if
ic
e
poc
h,
a
s
de
ta
il
e
d
in
T
a
bl
e
3.
T
hi
s
ta
bl
e
a
ls
o
out
li
ne
s
th
e
in
di
vi
dua
l
e
a
r
ly
s
to
ppi
ng
a
nd
le
a
r
ni
ng r
a
te
c
onf
ig
ur
a
ti
ons
f
or
e
a
c
h m
ode
l,
a
lo
ng w
it
h t
he
c
or
r
e
s
ponding a
r
c
hi
te
c
tu
r
e
us
e
d.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
AI
-
dr
iv
e
n hy
pe
r
-
pe
r
s
onal
i
z
at
io
n and tr
ans
fe
r
le
ar
ni
ng f
o
r
pr
e
c
i
s
io
n r
e
c
r
ui
tme
nt
(
N
our
A
lq
udah
)
4275
F
ig
ur
e
1. P
r
e
-
t
r
a
in
e
d m
ode
li
ng t
e
c
hni
que
T
a
bl
e
3.
T
r
a
ns
f
e
r
le
a
r
ni
ng mode
ls
’
a
r
c
hi
te
c
tu
r
e
a
nd I
D
s
C
l
a
s
s
i
f
i
c
a
t
i
on T
a
s
k
s
A
r
c
hi
t
e
c
t
ur
e
M
ode
l
'
s
I
D
N
um
be
r
of
e
poc
hs
L
e
a
r
ni
ng
r
a
t
e
B
i
na
r
y c
l
a
s
s
i
f
i
c
a
t
i
on
B
E
R
T
M
ul
t
i
l
i
ngua
l
15
3.00E
-
05
B
i
na
r
y c
l
a
s
s
i
f
i
c
a
t
i
on
B
E
R
T
B
a
s
e
4
5.00E
-
06
B
i
na
r
y c
l
a
s
s
i
f
i
c
a
t
i
on
R
oB
E
R
T
a
B
a
s
e
8
5.00E
-
06
B
i
na
r
y c
l
a
s
s
i
f
i
c
a
t
i
on
X
L
M
-
R
oB
E
R
T
a
B
a
s
e
2
3.00E
-
05
M
ul
t
i
-
c
l
a
s
s
c
l
a
s
s
i
f
i
c
a
t
i
on
B
E
R
T
M
ul
t
i
l
i
ngua
l
5
2.00E
-
06
M
ul
t
i
-
c
l
a
s
s
c
l
a
s
s
i
f
i
c
a
t
i
on
B
E
R
T
B
a
s
e
8
5.00E
-
06
M
ul
t
i
-
c
l
a
s
s
c
l
a
s
s
i
f
i
c
a
t
i
on
R
oB
E
R
T
a
B
a
s
e
10
2.00E
-
05
M
ul
t
i
-
c
l
a
s
s
c
l
a
s
s
i
f
i
c
a
t
i
on
X
L
M
-
R
oB
E
R
T
a
B
a
s
e
5
3.00E
-
05
3.
R
E
S
U
L
T
S
A
N
D
D
I
S
C
U
S
S
I
O
N
T
hi
s
s
e
c
ti
on
e
xpl
a
in
s
th
e
r
e
s
ul
ts
of
r
e
s
e
a
r
c
h
a
nd
a
t
th
e
s
a
m
e
ti
m
e
gi
ve
s
th
e
c
om
pr
e
he
ns
iv
e
di
s
c
us
s
io
n.
T
a
bl
e
4
a
nd
F
ig
ur
e
2
s
ho
w
how
w
e
pr
of
it
e
d
f
r
om
th
e
c
a
pa
bi
li
ti
e
s
a
nd
pe
r
f
or
m
a
nc
e
s
pe
c
if
ic
a
ti
ons
in
tr
a
ns
f
e
r
l
e
a
r
ni
ng mode
ls
. H
uggi
ngF
a
c
e
tr
a
ns
f
or
m
e
r
s
w
e
r
e
us
e
d t
o c
r
e
a
te
t
he
m
ode
ls
[
25]
.
T
a
bl
e
4.
T
he
a
c
c
ur
a
c
y, pr
e
c
is
io
n, r
e
c
a
ll
,
a
nd
F1
-
s
c
or
e
f
or
bi
na
r
y c
la
s
s
if
ic
a
ti
on w
it
h t
r
a
ns
f
e
r
l
e
a
r
ni
ng mode
ls
M
ode
l
M
ode
l
I
D
F1
-
s
c
or
e
R
e
c
a
l
l
P
r
e
c
i
s
i
on
A
c
c
ur
a
c
y
B
E
R
T
M
ul
t
i
l
i
ngua
l
0.79
0.79
0.79
0.79
B
E
R
T
B
a
s
e
0.79
0.79
0.79
0.79
R
oB
E
R
T
a
B
a
s
e
0.79
0.80
0.79
0.80
X
L
M
-
R
oB
E
R
T
a
B
a
s
e
0.79
0.80
0.80
0.80
F
ig
ur
e
2
.
T
he
a
c
c
ur
a
c
y, pr
e
c
i
s
io
n, r
e
c
a
ll
, a
nd F
1
-
s
c
or
e
f
or
bi
na
r
y c
la
s
s
if
ic
a
ti
on us
in
g t
r
a
n
s
f
e
r
l
e
a
r
ni
ng mode
ls
T
a
bl
e
4 a
nd
F
ig
ur
e
2
d
is
pl
a
y t
he
r
e
s
ul
ts
of
bi
na
r
y c
la
s
s
if
ic
a
ti
on, i
ndi
c
a
ti
ng t
ha
t
th
e
X
L
M
-
R
oB
E
R
T
A
ba
s
e
ve
r
s
io
n
m
od
e
l
a
c
hi
e
ve
d
th
e
hi
ghe
s
t
p
e
r
f
or
m
a
nc
e
.
I
t
de
m
ons
tr
a
te
d
a
n
a
c
c
ur
a
c
y
of
80%
,
r
e
c
a
ll
of
80%
,
F1
-
s
c
or
e
m
e
a
s
ur
e
of
79%
,
a
nd
p
r
e
c
is
io
n
of
80%
.
S
im
il
a
r
ly
,
R
oB
E
R
T
A
a
ls
o
e
xhi
bi
te
d
a
n
a
c
c
ur
a
c
y
of
80%
,
s
ig
ni
f
yi
ng
th
a
t
pr
e
tr
a
in
e
d
m
ode
ls
s
ur
pa
s
s
e
d
ta
s
k
-
tr
a
in
e
d
DL
M
ode
ls
.
M
or
e
ove
r
,
T
a
bl
e
5
a
nd
F
ig
ur
e
3
,
f
oc
us
in
g
on
m
ul
ti
-
c
la
s
s
c
la
s
s
if
ic
a
ti
on,
hi
ghl
ig
ht
s
X
L
M
-
R
oB
E
R
T
A
'
s
s
upe
r
io
r
pe
r
f
or
m
a
nc
e
w
it
h
a
n
a
c
c
ur
a
c
y
of
72%
,
out
pe
r
f
o
r
m
in
g
ta
s
k
-
tr
a
in
e
d
DL
M
ode
ls
in
th
is
s
pe
c
if
ic
ta
s
k.
T
he
r
e
s
ul
ts
s
how
e
d
th
a
t
X
L
M
-
R
oB
E
R
T
A
in
bot
h
ta
s
ks
(
bi
na
r
y
a
nd
m
ul
ti
-
c
la
s
s
c
la
s
s
if
ic
a
ti
on
)
obt
a
in
e
d
th
e
hi
ghe
s
t
a
c
c
ur
a
c
y
a
nd
a
c
hi
e
ve
d
th
e
be
s
t
pe
r
f
or
m
a
nc
e
c
om
pa
r
e
d w
it
h ot
he
r
m
ode
ls
.
T
he
s
e
r
e
s
ul
ts
pr
oduc
e
d
a
n
e
xc
e
ll
e
nt
pe
r
f
or
m
a
nc
e
f
or
X
L
M
-
R
oB
E
R
T
a
in
bot
h
c
la
s
s
if
ic
a
ti
on
ta
s
k
s
.
T
he
s
upe
r
io
r
it
y
is
th
a
t
X
L
M
-
R
[
26]
,
[
27]
is
a
c
ut
ti
ng
-
e
dge
m
ul
ti
li
ngua
l
m
a
s
ke
d
la
ngua
g
e
m
ode
l
tr
a
in
e
d
on
2.5
T
B
of
f
r
e
s
hl
y
c
ons
tr
uc
te
d
c
le
a
n
C
om
m
on
C
r
a
w
l
d
a
ta
in
10
0
la
ngua
ge
s
.
I
t
out
pe
r
f
or
m
s
pr
io
r
m
ul
ti
li
ngua
l
0.
78
0.
79
0.
80
0.
81
F1-
Scor
e
R
ec
al
l
Pr
ec
i
s
i
on
A
cc
ur
ac
y
Per
f
or
m
an
ce
m
et
r
i
c(
%
)
M
odel
B
E
R
T
M
ul
t
i
l
i
ng
ua
l
B
E
R
T
B
as
e
R
ob
er
t
a
B
as
e
X
L
M
-
R
ob
er
t
a
B
as
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
,
V
ol
. 14, No. 5, O
c
to
be
r
2025
:
4271
-
4278
4276
m
ode
ls
s
uc
h a
s
m
B
E
R
T
a
nd
X
L
M
in
c
la
s
s
if
ic
a
ti
on,
s
e
que
n
c
e
l
a
be
li
ng,
a
nd
que
s
ti
on a
ns
w
e
r
in
g,
but
it
r
e
ve
a
l
s
th
e
li
m
it
s
of
m
ul
ti
li
ngua
l
M
L
M
s
,
pa
r
ti
c
ul
a
r
ly
th
e
hi
gh
-
r
e
s
our
c
e
ve
r
s
us
lo
w
-
r
e
s
our
c
e
tr
a
de
-
of
f
a
nd
th
e
r
e
le
va
nc
e
of
c
r
it
ic
a
l
hype
r
pa
r
a
m
e
te
r
s
. M
or
e
ove
r
, t
he
X
L
M
-
R
m
ul
ti
li
ngua
l
m
ode
l
is
t
he
f
ir
s
t
to
s
ur
pa
s
s
t
ypi
c
a
l
m
onol
in
gua
l
ba
s
e
li
ne
s
th
a
t
de
p
e
nd
on
pr
e
-
tr
a
in
e
d
m
ode
ls
.
M
ul
ti
li
ngua
l
B
E
R
T
a
nd
X
L
M
m
ode
ls
h
a
ve
c
ha
ll
e
nge
s
le
a
r
ni
ng
us
a
bl
e
r
e
pr
e
s
e
nt
a
ti
ons
f
or
lo
w
-
r
e
s
our
c
e
la
ngua
ge
s
.
X
L
M
-
R
de
m
ons
tr
a
te
s
th
e
s
upe
r
io
r
it
y
of
m
ul
ti
li
ngua
l
m
ode
ls
ove
r
m
onol
in
gua
l
m
ode
ls
f
or
l
ow
-
r
e
s
our
c
e
l
a
ngua
ge
s
[
28]
.
T
a
bl
e
5.
T
h
e
a
c
c
ur
a
c
y,
pr
e
c
i
s
io
n,
r
e
c
a
ll
,
a
n
d F
1
-
s
c
or
e
f
or
m
ul
ti
c
l
a
s
s
c
la
s
s
if
ic
a
ti
on
w
i
th
t
r
a
n
s
f
e
r
l
e
a
r
ni
ng
m
o
de
ls
M
ode
l
M
ode
l
I
D
F1
-
S
c
or
e
R
e
c
a
l
l
P
r
e
c
i
s
i
on
A
c
c
ur
a
c
y
B
E
R
T
M
ul
t
i
l
i
ngua
l
0.70
0.70
0.70
0.70
B
E
R
T
B
a
s
e
0.67
0.67
0.67
0.67
R
obe
r
t
a
B
a
s
e
0.71
0.71
0.71
0.71
X
L
M
-
R
obe
r
t
a
B
a
s
e
0.72
0.72
0.72
0.72
F
ig
ur
e
3
.
T
he
a
c
c
ur
a
c
y, pr
e
c
i
s
io
n, r
e
c
a
ll
a
nd
F
1
-
s
c
or
e
f
or
m
ul
ti
c
la
s
s
c
l
a
s
s
if
ic
a
ti
on w
it
h t
r
a
ns
f
e
r
l
e
a
r
ni
ng
4.
C
O
N
C
L
U
S
I
O
N
T
he
e
xp
e
r
im
e
nt
a
l
f
in
di
ngs
c
le
a
r
ly
s
uppor
t
th
e
obj
e
c
ti
ve
s
out
li
ne
d
in
th
e
in
tr
oduc
ti
on,
pa
r
ti
c
ul
a
r
ly
in
e
nha
nc
in
g
r
e
c
r
ui
tm
e
nt
pr
e
c
is
io
n
us
in
g
tr
a
ns
f
or
m
e
r
-
ba
s
e
d
m
ode
ls
.
T
hi
s
s
tu
dy
a
ddr
e
s
s
e
s
c
r
it
ic
a
l
c
ha
ll
e
nge
s
in
ta
le
nt
a
c
qui
s
it
io
n
by
e
xpl
or
in
g
a
dva
nc
e
d
te
c
hni
que
s
f
or
id
e
nt
if
yi
ng
opt
im
a
l
c
a
ndi
da
te
s
a
c
r
os
s
va
r
io
us
jo
b
c
a
te
gor
ie
s
.
O
ur
in
ve
s
ti
ga
ti
on
in
to
T
r
a
ns
f
or
m
e
r
m
ode
ls
,
in
c
lu
di
ng
B
E
R
T
,
R
oB
E
R
T
a
,
a
nd
X
L
M
-
R
oB
E
R
T
a
,
yi
e
ld
e
d
pr
om
is
in
g
r
e
s
ul
ts
,
w
it
h
X
L
M
-
R
oB
E
R
T
a
a
c
hi
e
vi
ng
80
a
nd
72%
a
c
c
ur
a
c
y
in
bi
na
r
y
a
nd
m
ul
ti
-
c
la
s
s
c
la
s
s
if
ic
a
ti
on
ta
s
ks
,
r
e
s
pe
c
ti
ve
ly
.
A
ke
y
in
nova
ti
on
li
e
s
in
o
ur
m
e
th
od'
s
a
bi
li
ty
to
a
ut
om
a
ti
c
a
ll
y
e
xt
r
a
c
t,
c
a
te
gor
iz
e
,
a
nd
m
a
p
nove
l
s
ki
ll
s
f
r
om
jo
b
d
e
s
c
r
ip
ti
ons
a
nd
r
e
s
um
e
s
to
c
a
r
e
e
r
c
a
te
gor
ie
s
ba
s
e
d
on
pe
r
s
ona
li
ty
tr
a
it
s
.
T
hi
s
s
tr
e
a
m
li
ne
s
th
e
r
e
c
r
ui
tm
e
nt
pr
oc
e
s
s
a
nd
e
nha
nc
e
s
p
e
r
s
ona
li
z
e
d
m
a
tc
hi
ng
be
twe
e
n
c
a
ndi
da
te
s
a
nd
jo
bs
.
H
ow
e
ve
r
,
th
e
s
tu
dy’
s
r
e
li
a
nc
e
on
s
ta
ti
c
da
ta
s
e
ts
m
a
y
a
f
f
e
c
t
a
da
pt
a
bi
li
ty
a
c
r
os
s
di
f
f
e
r
e
nt
li
ngui
s
ti
c
or
c
ul
tu
r
a
l
c
ont
e
xt
s
. F
ut
ur
e
r
e
s
e
a
r
c
h c
oul
d f
oc
us
on ut
il
iz
in
g r
e
a
l
-
t
im
e
a
ppl
ic
a
nt
da
ta
a
nd r
e
in
f
or
c
e
m
e
nt
l
e
a
r
ni
n
g
to
e
nha
nc
e
pr
e
di
c
ti
on
a
c
c
ur
a
c
y
a
nd
a
da
pt
to
m
a
r
ke
t
f
lu
c
tu
a
ti
ons
.
B
y
de
m
ons
tr
a
ti
ng
th
e
e
f
f
e
c
ti
ve
ne
s
s
of
tr
a
ns
f
e
r
le
a
r
ni
ng
in
N
L
P
f
or
r
e
c
r
ui
tm
e
nt
,
th
is
w
or
k
hi
ghl
ig
ht
s
th
e
pot
e
nt
ia
l
of
A
I
in
m
a
ki
ng
hi
r
in
g
m
o
r
e
pr
e
c
is
e
, f
a
ir
, a
nd e
f
f
ic
ie
nt
.
F
U
N
D
I
N
G
I
N
F
O
R
M
A
T
I
O
N
A
ut
hor
s
s
ta
te
t
he
r
e
i
s
no f
undi
ng i
nvol
ve
d.
A
U
T
H
O
R
C
O
N
T
R
I
B
U
T
I
O
N
S
S
T
A
T
E
M
E
N
T
T
hi
s
jo
ur
na
l
us
e
s
th
e
C
ont
r
ib
ut
or
R
ol
e
s
T
a
xonomy
(
C
R
e
di
T
)
to
r
e
c
ogni
z
e
in
di
vi
dua
l
a
ut
hor
c
ont
r
ib
ut
io
ns
, r
e
duc
e
a
ut
hor
s
hi
p di
s
put
e
s
,
a
nd f
a
c
il
it
a
te
c
ol
la
bo
r
a
ti
on.
0.
64
0.
65
0.
66
0.
67
0.
68
0.
69
0.
70
0.
71
0.
72
0.
73
F1-
Scor
e
R
ec
al
l
P
r
eci
s
i
on
A
cc
ur
ac
y
Per
f
or
m
an
ce
m
et
r
i
c
(
%
)
M
odel
B
E
R
T
M
ul
t
i
l
i
ngual
B
E
R
T
B
as
e
R
ober
t
a
B
as
e
X
L
M
-
R
ober
t
a
B
as
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
AI
-
dr
iv
e
n hy
pe
r
-
pe
r
s
onal
i
z
at
io
n and tr
ans
fe
r
le
ar
ni
ng f
o
r
pr
e
c
i
s
io
n r
e
c
r
ui
tme
nt
(
N
our
A
lq
udah
)
4277
N
am
e
o
f
A
u
t
h
or
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
N
our
A
lq
uda
h
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
Q
us
a
i
Q
. A
bue
in
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
M
o
h
a
m
m
e
d
Q
.
S
h
a
t
n
a
w
i
✓
✓
✓
✓
✓
✓
C
:
C
onc
e
pt
ua
l
i
z
a
t
i
on
M
:
M
e
t
hodol
ogy
So
:
So
f
t
w
a
r
e
Va
:
Va
l
i
da
t
i
on
Fo
:
Fo
r
m
a
l
a
na
l
ys
i
s
I
:
I
nve
s
t
i
ga
t
i
on
R
:
R
e
s
our
c
e
s
D
:
D
a
t
a
C
ur
a
t
i
on
O
:
W
r
i
t
i
ng
-
O
r
i
gi
na
l
D
r
a
f
t
E
:
W
r
i
t
i
ng
-
R
e
vi
e
w
&
E
di
t
i
ng
Vi
:
Vi
s
ua
l
i
z
a
t
i
on
Su
:
Su
pe
r
vi
s
i
on
P
:
P
r
oj
e
c
t
a
dm
i
ni
s
t
r
a
t
i
on
Fu
:
Fu
ndi
ng a
c
qui
s
i
t
i
on
C
O
N
F
L
I
C
T
O
F
I
N
T
E
R
E
S
T
S
T
A
T
E
M
E
N
T
A
ut
hor
s
s
ta
te
no c
onf
li
c
t
of
i
nt
e
r
e
s
t.
D
A
T
A
A
V
A
I
L
A
B
I
L
I
T
Y
T
he
da
ta
th
a
t
s
uppor
t
th
e
f
in
di
ngs
of
th
is
s
tu
dy
w
e
r
e
c
r
e
a
te
d
by
th
e
a
ut
hor
s
a
nd
a
r
e
ope
nl
y
a
va
il
a
bl
e
in
t
he
H
Y
P
s
ki
ll
da
ta
s
e
t
on K
a
ggl
e
a
t
ht
tp
s
:/
/ww
w
.ka
ggl
e
.c
om
/d
a
ta
s
e
ts
/n
our
a
lq
uda
h/
hyp
s
ki
ll
-
da
ta
s
e
t.
R
E
F
E
R
E
N
C
E
S
[
1]
G
.
K
om
a
n,
P
.
B
or
š
oš
,
a
nd
M
.
K
ubi
na
,
“
T
he
pos
s
i
bi
l
i
t
i
e
s
of
us
i
ng
a
r
t
i
f
i
c
i
a
l
i
nt
e
l
l
i
ge
nc
e
a
s
a
ke
y
t
e
c
hnol
ogy
i
n
t
he
c
ur
r
e
nt
e
m
pl
oye
e
r
e
c
r
ui
t
m
e
nt
pr
oc
e
s
s
,”
A
dm
i
ni
s
t
r
at
i
v
e
S
c
i
e
nc
e
s
, vol
. 14, no. 7, 2024, d
oi
:
10.3390/
a
dm
s
c
i
14070157.
[
2]
M
.
T
a
j
pour
,
A
.
S
a
l
a
m
z
a
de
h,
a
nd
E
.
H
os
s
e
i
ni
,
“
J
ob
s
a
t
i
s
f
a
c
t
i
on
i
n
I
T
D
e
pa
r
t
m
e
nt
of
M
e
l
l
a
t
B
a
nk:
doe
s
e
m
pl
oye
r
br
a
nd
m
a
t
t
e
r
?
,”
I
P
SI
B
gD
T
r
ans
ac
t
i
ons
on I
nt
e
r
ne
t
R
e
s
e
a
r
c
h
, vol
. 17, no. 1, pp. 15
–
21, 2021
.
[
3]
L
.
J
.
G
.
-
G
om
e
z
,
S
.
M
.
H
.
-
M
unoz
,
A
.
B
or
j
a
,
J
.
D
.
A
z
of
e
i
f
a
,
J
.
N
ogue
z
,
a
nd
P
.
C
a
r
a
t
oz
z
ol
o,
“
A
na
l
yz
i
ng
na
t
ur
a
l
l
a
ngua
ge
pr
oc
e
s
s
i
ng
t
e
c
hni
que
s
t
o
e
xt
r
a
c
t
m
e
a
ni
ngf
ul
i
nf
or
m
a
t
i
on
on
s
ki
l
l
s
a
c
qui
s
i
t
i
on
f
r
om
t
e
xt
ua
l
c
ont
e
nt
,”
I
E
E
E
A
c
c
e
s
s
,
vol
.
12,
pp. 139742
–
139757, 2024, doi
:
10.1109/
A
C
C
E
S
S
.2024.3465409.
[
4]
R
.
J
os
hi
,
V
.
N
a
i
r
,
M
.
S
i
ngh,
A
.
N
.
-
I
.
J
.
of
A
I
,
a
nd
U
.
2021,
“
L
e
ve
r
a
gi
ng
n
a
t
ur
a
l
l
a
ngua
ge
pr
oc
e
s
s
i
ng
a
nd
pr
e
di
c
t
i
ve
a
na
l
yt
i
c
s
f
or
e
nha
nc
e
d A
I
-
dr
i
ve
n l
e
a
d nur
t
ur
i
ng a
nd e
nga
ge
m
e
nt
,”
I
nt
e
r
nat
i
onal
J
our
nal
of
A
I
A
dv
anc
e
m
e
nt
s
, vol
. 10, no. 1,
pp. 1
-
22,
2021
.
[
5]
R
.
M
e
hm
ood
, F
. A
l
a
m
,
N
.
N
.
A
l
bo
ga
m
i
,
I
.
K
a
t
i
b,
A
.
A
l
be
s
hr
i
,
a
nd
S
.
M
.
A
l
t
ow
a
i
j
r
i
,
“
U
T
i
L
e
a
r
n:
a
pe
r
s
ona
l
i
s
e
d
u
bi
qu
i
t
ous
t
e
a
c
hi
ng
a
nd l
e
a
r
ni
ng
s
ys
t
e
m
f
or
s
m
a
r
t
s
oc
i
e
t
i
e
s
,”
I
E
E
E
A
c
c
e
s
s
, vo
l
. 5,
pp.
2615
–
2635,
20
17, do
i
:
10
.1109
/
A
C
C
E
S
S
.2
017.2
66884
0.
[
6]
M
.
T
i
m
s
,
D
.
D
e
r
ks
,
a
nd
A
.
B
.
B
a
kke
r
,
“
J
ob
c
r
a
f
t
i
ng
a
nd
i
t
s
r
e
l
a
t
i
ons
hi
ps
w
i
t
h
pe
r
s
on
-
j
ob
f
i
t
a
nd
m
e
a
ni
ngf
ul
ne
s
s
:
a
t
hr
e
e
-
w
a
ve
s
t
udy,”
J
our
nal
of
V
oc
at
i
onal
B
e
hav
i
or
, vol
. 92, pp. 44
–
53, 2016, doi
:
10.1016/
j
.j
vb.2015.11.007.
[
7]
B
.
R
.
D
i
ne
e
n,
S
.
R
.
A
s
h,
a
nd
R
.
A
.
N
oe
,
“
A
w
e
b
of
a
ppl
i
c
a
nt
a
t
t
r
a
c
t
i
on:
pe
r
s
on
-
or
ga
ni
z
a
t
i
on
f
i
t
i
n
t
he
c
ont
e
xt
of
w
e
b
-
ba
s
e
d
r
e
c
r
ui
t
m
e
nt
,”
J
our
nal
of
A
ppl
i
e
d P
s
y
c
hol
ogy
, vol
. 87, no. 4, pp. 723
–
734, 2002, doi
:
10.1037/
0021
-
9010.87.4.723.
[
8]
G
.
J
a
i
n,
J
.
P
a
ul
,
a
nd
A
.
S
hr
i
va
s
t
a
va
,
“
H
ype
r
-
pe
r
s
ona
l
i
z
a
t
i
on,
c
o
-
c
r
e
a
t
i
on,
d
i
gi
t
a
l
c
l
i
e
nt
e
l
i
ng
a
nd
t
r
a
ns
f
or
m
a
t
i
on,”
J
our
nal
of
B
us
i
ne
s
s
R
e
s
e
a
r
c
h
, vol
. 124, pp. 12
–
23, 2021, doi
:
10.1016/
j
.j
bus
r
e
s
.2020.11.0
34.
[
9]
G
.
N
.
B
ur
ns
,
N
.
D
.
C
hr
i
s
t
i
a
ns
e
n,
M
.
B
.
M
or
r
i
s
,
D
.
A
.
P
e
r
i
a
r
d,
a
nd
J
.
A
.
C
oa
s
t
e
r
,
“
E
f
f
e
c
t
s
of
a
ppl
i
c
a
nt
pe
r
s
ona
l
i
t
y
on
r
e
s
um
e
e
va
l
ua
t
i
ons
,”
J
ou
r
nal
of
B
us
i
ne
s
s
and P
s
y
c
hol
ogy
, vol
. 29, no. 4, pp. 573
–
591,
2014, doi
:
10.1007/
s
10869
-
014
-
9349
-
6.
[
10]
P
.
L
.
R
ot
h,
P
.
B
obko,
C
.
H
.
V
.
I
dde
ki
nge
,
a
nd
J
.
B
.
T
ha
t
c
h
e
r
,
“
S
oc
i
a
l
m
e
di
a
i
n
e
m
pl
oye
e
-
s
e
l
e
c
t
i
on
-
r
e
l
a
t
e
d
de
c
i
s
i
on
s
:
a
r
e
s
e
a
r
c
h
a
ge
nda
f
or
unc
ha
r
t
e
d t
e
r
r
i
t
or
y,”
J
our
nal
of
M
anage
m
e
nt
, vol
. 42, no. 1, pp. 269
–
298, 2016, doi
:
10.1177/
0149206313503018.
[
11]
I
.
N
i
kol
a
ou,
“
S
oc
i
a
l
ne
t
w
or
ki
ng
w
e
b
s
i
t
e
s
i
n
j
ob
s
e
a
r
c
h
a
nd
e
m
pl
oye
e
r
e
c
r
ui
t
m
e
nt
,”
I
nt
e
r
nat
i
onal
J
our
nal
of
Se
l
e
c
t
i
on
an
d
A
s
s
e
s
s
m
e
nt
, vol
. 22, no. 2, pp. 179
–
189, 2014, doi
:
10.1111/
i
j
s
a
.12067.
[
12]
L
.
S
a
yf
ul
l
i
na
,
E
.
M
a
l
m
i
,
a
nd
J
.
K
a
nna
l
a
,
“
L
e
a
r
ni
ng
r
e
pr
e
s
e
nt
a
t
i
ons
f
or
s
of
t
s
ki
l
l
m
a
t
c
hi
ng,”
A
nal
y
s
i
s
of
I
m
age
s
,
Soc
i
al
N
e
t
w
or
k
s
and T
e
x
t
s
,
C
ha
m
, S
w
i
t
z
e
r
l
a
nd:
S
pr
i
nge
r
,
pp. 141
–
152, 2018, doi
:
10.1007/
978
-
3
-
030
-
11027
-
7_15.
[
13]
D
.
A
.
T
a
m
bur
r
i
,
W
.
J
.
V
.
D
.
H
e
uve
l
,
a
nd
M
.
G
a
r
r
i
ga
,
“
D
a
t
a
O
ps
f
or
s
oc
i
e
t
a
l
i
nt
e
l
l
i
ge
nc
e
:
a
da
t
a
pi
pe
l
i
ne
f
o
r
l
a
bor
m
a
r
ke
t
s
ki
l
l
s
e
xt
r
a
c
t
i
on
a
nd
m
a
t
c
hi
ng,”
i
n
2020
I
E
E
E
21s
t
I
nt
e
r
nat
i
onal
C
onf
e
r
e
nc
e
on
I
nf
or
m
at
i
on
R
e
us
e
and
I
nt
e
gr
at
i
on
f
or
D
at
a
Sc
i
e
nc
e
,
I
R
I
2020
, 2020, pp. 391
–
394, doi
:
10.1109/
I
R
I
49571.2020.00063.
[
14]
A
.
B
hol
a
,
K
.
H
a
l
de
r
,
A
.
P
r
a
s
a
d,
a
nd
M
.
Y
.
K
a
n,
“
R
e
t
r
i
e
vi
ng
s
ki
l
l
s
f
r
om
j
ob
de
s
c
r
i
pt
i
ons
:
a
l
a
ngua
ge
m
ode
l
ba
s
e
d
e
xt
r
e
m
e
m
ul
t
i
-
l
a
be
l
c
l
a
s
s
i
f
i
c
a
t
i
on
f
r
a
m
e
w
or
k,”
i
n
C
O
L
I
N
G
2020
-
28t
h
I
nt
e
r
nat
i
onal
C
onf
e
r
e
nc
e
on
C
om
put
at
i
onal
L
i
ngui
s
t
i
c
s
,
P
r
oc
e
e
di
ng
s
of
t
he
C
onf
e
r
e
nc
e
, 2020, pp. 5832
–
5842, doi
:
10.18653/
v1/
2020.c
ol
i
ng
-
m
a
i
n.513.
[
15]
J
.
D
e
vl
i
n,
M
.
-
W
.
C
ha
ng,
K
.
L
e
e
,
a
nd
K
.
T
out
a
nova
,
“
B
E
R
T
:
p
r
e
-
t
r
a
i
ni
ng
of
de
e
p
bi
di
r
e
c
t
i
ona
l
t
r
a
ns
f
or
m
e
r
s
f
o
r
l
a
ngua
ge
unde
r
s
t
a
ndi
ng,”
C
O
L
I
N
G
2020
-
28t
h
I
nt
e
r
nat
i
onal
C
onf
e
r
e
nc
e
on
C
om
put
at
i
onal
L
i
ngui
s
t
i
c
s
,
P
r
oc
e
e
di
ngs
of
t
he
C
onf
e
r
e
nc
e
,
pp. 5832
–
5842, M
a
y 2019, doi
:
10.18653/
v1/
2020.c
ol
i
ng
-
m
a
i
n.513.
[
16]
Q
. Q
. A
bu
e
i
n, M
.
Q
. S
ha
t
n
a
w
i
, a
nd
N
. A
l
qud
a
h, “
I
m
pr
ovi
ng j
ob m
a
t
c
hi
ng
w
i
t
h
de
e
p l
e
a
r
ni
ng
-
ba
s
e
d hype
r
-
pe
r
s
ona
l
i
z
a
t
i
on,”
I
A
E
S
I
nt
e
r
nat
i
onal
J
our
nal
of
A
r
t
i
f
i
c
i
al
I
nt
e
l
l
i
ge
nc
e
, vol
. 13, no. 2, pp. 1711
–
1722, 2024, doi
:
10.11591/
i
j
a
i
.v13.i
2.pp1711
-
1722.
[
17]
A
.
R
a
ha
l
i
a
nd
M
.
A
.
A
khl
ouf
i
,
“
E
nd
-
to
-
e
nd
t
r
a
ns
f
or
m
e
r
-
ba
s
e
d
m
ode
l
s
i
n
t
e
xt
ua
l
-
ba
s
e
d
N
L
P
,”
AI
,
vol
.
4,
no.
1,
pp.
54
–
110,
2023,
doi
:
10.3390/
a
i
4010004.
[
18]
M
.
P
r
yt
ul
a
,
“
F
i
ne
-
t
uni
ng
B
E
R
T
,
D
i
s
t
i
l
B
E
R
T
,
X
L
M
-
R
oB
E
R
T
a
a
nd
U
kr
-
R
oB
E
R
T
a
m
ode
l
s
f
or
s
e
nt
i
m
e
nt
a
na
l
ys
i
s
of
U
kr
a
i
ni
a
n
l
a
ngua
ge
r
e
vi
e
w
s
,”
A
r
t
i
f
i
c
i
al
I
nt
e
l
l
i
ge
nc
e
, vol
. 29, no.
2
, pp. 85
–
97, 2024, doi
:
10.15407/
j
a
i
2024.02.085.
[
19]
S
.
F
a
r
e
r
i
,
N
.
M
e
l
l
us
o,
F
.
C
hi
a
r
e
l
l
o,
a
nd
G
.
F
a
nt
oni
,
“
S
ki
l
l
N
E
R
:
m
i
ni
ng
a
nd
m
a
ppi
ng
s
of
t
s
ki
l
l
s
f
r
om
a
ny
t
e
xt
,”
E
x
pe
r
t
Sy
s
t
e
m
s
w
i
t
h A
ppl
i
c
at
i
ons
, vol
. 184, 2021, doi
:
10.1016/
j
.e
s
w
a
.2021.115544.
[
20]
X
.
L
i
,
X
.
S
un,
Y
.
M
e
ng,
J
.
L
i
a
ng,
F
.
W
u,
a
nd
J
.
L
i
,
“
D
i
c
e
l
os
s
f
or
da
t
a
-
i
m
ba
l
a
nc
e
d
N
L
P
t
a
s
k
s
,”
P
r
oc
e
e
di
ngs
of
t
he
A
nnual
M
e
e
t
i
ng of
t
he
A
s
s
oc
i
at
i
on f
or
C
om
put
at
i
onal
L
i
ngui
s
t
i
c
s
, pp. 465
–
476, 2020,
doi
:
10.18653/
v1/
2020.a
c
l
-
m
a
i
n.45.
[
21]
G
.
H
a
i
xi
a
ng,
L
.
Y
i
j
i
ng,
L
.
Y
a
na
n,
L
.
X
i
a
o,
a
nd
L
.
J
i
nl
i
ng,
“
B
P
S
O
-
A
da
boos
t
-
K
N
N
e
ns
e
m
bl
e
l
e
a
r
ni
ng
a
l
gor
i
t
hm
f
or
m
ul
t
i
-
c
l
a
s
s
i
m
ba
l
a
nc
e
d
da
t
a
c
l
a
s
s
i
f
i
c
a
t
i
on,”
E
ngi
ne
e
r
i
ng
A
ppl
i
c
at
i
ons
of
A
r
t
i
f
i
c
i
al
I
nt
e
l
l
i
ge
nc
e
,
vol
.
49,
pp.
176
–
193,
2016
,
doi
:
10.1016/
j
.e
nga
ppa
i
.2015.09.011.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
,
V
ol
. 14, No. 5, O
c
to
be
r
2025
:
4271
-
4278
4278
[
22]
M
.
V
.
K
or
ot
e
e
v,
“
B
E
R
T
:
a
r
e
vi
e
w
of
a
ppl
i
c
a
t
i
ons
i
n
na
t
ur
a
l
l
a
ngua
ge
pr
oc
e
s
s
i
ng
a
nd
unde
r
s
t
a
ndi
ng,”
ar
X
i
v
-
C
om
put
e
r
Sc
i
e
nc
e
,
pp. 1
-
18, M
a
r
.
2021
.
[
23]
Y
. L
i
u
e
t
al
.
, “
R
oB
E
R
T
a
:
A
r
obus
t
l
y opt
i
m
i
z
e
d B
E
R
T
pr
e
t
r
a
i
ni
ng a
ppr
oa
c
h,”
a
r
X
i
v
-
C
om
put
e
r
Sc
i
e
nc
e
, pp. 1
-
13
,
J
ul
.
2019
.
[
24]
A
.
C
onne
a
u
e
t
al
.
,
“
U
ns
up
e
r
vi
s
e
d
c
r
os
s
-
l
i
ngua
l
r
e
pr
e
s
e
nt
a
t
i
on
l
e
a
r
ni
ng
a
t
s
c
a
l
e
,”
P
r
oc
e
e
di
ngs
of
t
he
A
nnual
M
e
e
t
i
ng
of
t
he
A
s
s
oc
i
at
i
on f
or
C
om
put
at
i
onal
L
i
ngui
s
t
i
c
s
, pp. 8440
–
8451, 2020, doi
:
10.1865
3/
v1/
2020.a
c
l
-
m
a
i
n.747.
[
25]
S
.
J
.
P
a
n
a
nd
Q
.
Y
a
ng,
“
A
s
ur
ve
y
on
t
r
a
ns
f
e
r
l
e
a
r
ni
ng,”
I
E
E
E
T
r
ans
ac
t
i
ons
on
K
now
l
e
dge
and
D
at
a
E
ngi
ne
e
r
i
ng
,
vol
.
22,
no.
10,
pp. 1345
–
1359, 2010, doi
:
10.1109/
T
K
D
E
.2009.191.
[
26]
M
.
M
a
r
s
,
“
F
r
om
w
or
d
e
m
be
ddi
ngs
t
o
pr
e
-
t
r
a
i
ne
d
l
a
ngua
ge
m
ode
l
s
:
a
s
t
a
t
e
-
of
-
t
he
-
a
r
t
w
a
l
kt
hr
ough,”
A
ppl
i
e
d
Sc
i
e
nc
e
s
,
vol
.
12,
no. 17, 2022, doi
:
10.3390/
a
pp12178805.
[
27]
O
.
M
oha
m
e
d,
A
.
M
.
K
a
s
s
e
m
,
A
.
A
s
hr
a
f
,
S
.
J
a
m
a
l
,
a
nd
E
.
H
.
M
oh
a
m
e
d,
“
A
n
e
ns
e
m
bl
e
t
r
a
n
s
f
or
m
e
r
-
ba
s
e
d
m
ode
l
f
or
A
r
a
bi
c
s
e
nt
i
m
e
nt
a
na
l
ys
i
s
,”
Soc
i
al
N
e
t
w
o
r
k
A
nal
y
s
i
s
and M
i
ni
ng
, vol
. 13, no. 1, 2023, doi
:
10.1007/
s
13278
-
022
-
01009
-
0.
[
28]
S
.
M
ut
uvi
,
“
E
pi
de
m
i
c
e
ve
nt
e
xt
r
a
c
t
i
on
i
n
m
ul
t
i
l
i
ngua
l
a
nd
l
ow
‑
r
e
s
our
c
e
s
e
t
t
i
n
gs
,”
P
h.D
.
di
s
s
e
r
t
at
i
on
,
L
a
bor
a
t
oi
r
e
I
nf
or
m
a
t
i
que
,
I
m
a
ge
, I
nt
e
r
a
c
t
i
on (
L
3i
)
, U
ni
ve
r
s
i
t
é
de
L
a
R
oc
he
l
l
e
,
L
a
R
o
c
he
l
l
e
, F
r
a
nc
e
,
2022.
B
I
O
G
R
A
P
H
I
E
S
O
F
A
U
T
H
O
R
S
Nour
Alqudah
holds
master’s
degree
in
data
science
from
Jordan
University
of
Scienc
e
and
Techn
ology
.
She
currently
works
as
a
digital
analytics
engineer
.
Her
research
interests
encompass
AI,
NLP,
big
data,
deep
learning,
digital
and
dat
a
analytics
.
S
he
can
be
contacted
at email
: nsalq
udah18@
cit.jus
t.edu.j
o.
Dr.
Qusai
Q.
Abuein
an
Associate
Professor
in
the
Department
of
Compute
r
Information
Systems,
Jordan
University
of
Sci
ence
and
Technology
.
He
received
his
Ph
.
D
.
and
master’s
degrees
from
Ibaraki
University
,
Hitachi,
Japan
.
He
teac
hes
web
social
analysis,
web
analytics,
and
data
visualization.
His
research
interests
include
web
social
analysis,
web
analytics
,
data
analysis
,
and
informat
ion
retrieval.
He
can
b
e
contacted
at
email:
qabuein@
just.edu.jo
.
Dr.
Mohammed
Q
.
Shatnawi
is
a
computer
science
academic
and
researcher
with
a
strong
background
in
machine
learning,
data
science,
and
their
applications.
He
earned
his
B.Sc.
in
Computer
Scien
ce
from
Yarmouk
University
and
w
ent
on
to
complete
both
his
M.Sc. and Ph.D. in Computer
Science from T
he George
Washington
University, gradua
ting in
January
2007.
His
work
bridges
theoretical
models
with
real
-
world
a
pplications,
contributing
significantly
to
the
advanceme
nt
of
intelligent
systems
and
data
-
driven
technologies.
He
can
be contacted at email: mshatnawi@just.edu.jo
.
Evaluation Warning : The document was created with Spire.PDF for Python.