Indonesi
an
Journa
l
of El
ect
ri
cal Engineer
ing
an
d
Comp
ut
er
Scie
nce
Vo
l.
12
,
No.
3
,
Decem
ber
201
8
, p
p.
1
239
~
1
246
IS
S
N: 25
02
-
4752, DO
I: 10
.11
591/ijeecs
.v1
2
.i
3
.pp
1
239
-
1
246
1239
Journ
al h
om
e
page
:
http:
//
ia
es
core.c
om/j
ourn
als/i
ndex.
ph
p/ij
eecs
Developi
ng Corp
ora usin
g Word2
vec and
Wiki
pedia f
or Word
Sense Di
sambigu
atio
n
Farz
a Nurif
an, R
i
yana
r
t
o
S
ar
no
,
Cah
yani
ng
t
yas
Sek
ar Wah
yuni
Depa
rtment
o
f
I
nform
at
ic
s,
Insti
tut
Te
knolog
i
Se
puluh
Nopem
ber
,
Jal
an
Ra
ya
I
T
S
, K
e
puti
h,
Suk
olil
o,
Ko
ta
Sur
abaya, Ja
wa
Ti
m
ur
6
01
11,
Ind
onesia
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
May
25
, 201
8
Re
vised
A
ug
2
5
, 2
018
Accepte
d
Se
p
7
, 2
018
W
ord
Sense
Dis
ambiguat
ion
(W
SD
)
is
one
of
the
m
ost
diffi
cult
proble
m
s
in
the
art
if
ic
i
al
intell
ig
ence
fi
el
d
or
well
known
as
AI
-
har
d
or
AI
-
complet
e
.
A
lot
of
problem
s
ca
n
be
solve
d
using
word
sense
disambiguation
appr
oa
ch
such
as
senti
m
ent
ana
l
y
sis,
m
a
chi
n
e
tr
ansla
t
io
n,
sea
rch
eng
in
e
rel
ev
ance,
cohe
ren
ce,
an
ap
hora
resolution,
and
inference.
T
his
rese
ar
ch
is
d
one
to
solv
e
W
SD
proble
m
with
two
sm
al
l
c
orpora
.
The
use
of
W
ord2ve
c
an
d
W
iki
pedia
are
proposed
t
o
deve
lop
the
cor
pora
.
Afte
r
deve
lop
ing
t
he
cor
pora
,
the
sim
il
ar
ity
o
f
the
sen
te
nc
e
with
the
cor
por
a
is
m
ea
sured
using
cosin
e
sim
il
ari
t
y
to
d
et
ermine
th
e
m
ea
ning
of
the
ambiguous
word.
La
stl
y
,
to
improve
ac
cu
racy
,
L
esk
al
gor
it
hm
s
and
W
u
Palmer
sim
il
ar
ity
are
used
t
o
dea
l
with
prob
lem
s
when
the
re
is
no
word
from
a
sente
nce
in
the
cor
pus.
The
r
esult
s
of
th
e
rese
arc
h
show
an
85
.
51%
ac
c
ura
c
y
r
ate
and
t
he
sem
ant
i
c
sim
il
ari
t
y
improve
the
accurac
y
rat
e
b
y
8.
02%
in
det
ermining
the
m
ea
ning
of
ambiguous words.
Ke
yw
or
ds:
Lesk
W
i
kip
e
dia
Wor
d
se
ns
e
dis
a
m
big
uatio
n
Wor
d2vec
Wu p
al
m
er
Copyright
©
201
8
Instit
ut
e
o
f Ad
vanc
ed
Engi
n
ee
r
ing
and
S
cienc
e
.
Al
l
rights re
serv
ed.
Corres
pond
in
g
Aut
h
or
:
Ri
ya
nar
to Sa
r
no,
Dep
a
rtm
ent o
f
Inform
at
ic
s,
In
sti
tut Te
knol
og
i
Sepulu
h N
op
em
ber
,
Jal
an
Ra
ya
I
T
S
, K
e
puti
h,
Suk
olil
o,
Ko
ta
Sur
abaya, Ja
wa
Ti
m
ur
6
01
11, In
donesia.
Em
a
il
: riy
anar
to@if
.it
s.ac.id
1.
INTROD
U
CTION
Natu
ral
la
ngua
ges
c
on
ta
i
n
a
few
w
ords
with
a
diff
e
re
nt
m
eaning
in
a
dif
fer
e
nt
c
on
te
xt
[
1]
.
Hu
m
an
can
ea
sil
y
disti
ng
uis
h
it
becau
se
we
hav
e
t
he
a
bili
t
y
to
see
the
c
onte
xt
of
the
se
ntence
a
nd
det
erm
ine
the
m
eaning
of
the
a
m
big
uo
us
w
ords.
T
o
m
ake
com
pu
te
rs
unde
rstan
d
the
m
eaning
of
an
a
m
big
uous
wo
r
d,
it
req
uires
a
ve
ry
diff
ic
ult
te
c
hn
i
qu
e
.
The
refo
re
,
Wo
r
d
Sense
Disam
big
uation
(
WSD)
is
existe
d
to
dete
rm
ine
the m
eaning
of am
big
uous
w
ords
[
2]
. For e
xam
ple, the
am
big
uous w
ord
is
the wo
rd '
bank'
:
1.
“H
e
sa
t
dow
n beside
the
Se
ine rive
r
bank
”
[
3]
.
2.
“H
e
d
e
posit
ed
the
mo
ney
at
the Chase
ba
nk”
[
3]
.
The
w
ord
ba
nk
in
bo
t
h
sent
ences
has
a
di
ff
e
ren
t
m
eaning
.
I
n
the
first
sentence
,
it
means
a
place
ne
ar
the
river,
wh
il
e i
n t
he
sec
ond
se
nt
ence,
it
m
eans
a fin
a
ncial
inst
it
ution
.
Wor
d
se
ns
e
disam
big
uation
i
s
ve
ry
im
po
rtant
pro
blem
becau
se
it
has
m
any
us
e
s
s
uch
as
m
achine
translat
ion
[
4]
or
se
ntim
ent
analy
sis
[5]
.
In
m
achine
translat
io
n,
tra
ns
la
ti
ng
a
se
nt
ence
co
ntaining
t
he
a
m
big
uous
w
ords
can
not
be
done
directl
y
witho
ut
lo
okin
g
the
co
nt
ext.
Othe
r
wise,
it
can
be
wrong.
he
accu
racy
of m
achine
transl
at
ion
in
tra
ns
la
ti
ng
w
ords
ca
n
be
im
pr
ov
e
d
[
6]
.
O
ne
of
the ex
am
ples
is
by
us
in
g
Wor
d
se
ns
e
dis
a
m
big
uatio
n.
Ma
ny
resea
rc
he
rs
ha
ve
pro
posed
va
rio
us
a
ppr
oac
hes
t
o
s
ol
ve
w
ord
se
ns
e
disam
big
uatio
n
pro
blem
s
,
bu
t
none
of
it
c
an
ha
ndle
inexi
ste
nt
w
ords
in
a
corp
us
.
In
[7]
,
propose
d
the
us
es
of
a
dap
te
d
wei
gh
te
d
gra
ph
t
o
so
lve
the
pro
bl
e
m
.
In
[
8]
,
pro
po
s
ed
the
us
es
of
m
achine
le
arn
i
ng
to
s
olve
t
he
pro
blem
.
An
ot
her
way
t
o
so
lve
word
se
ns
e
dis
a
m
big
uatio
n
is
by
us
in
g
co
rpus.
Co
r
pu
s
is
a
set
of
struc
tur
ed
te
xt
that
ha
s
m
any
us
es.
On
e
of
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vo
l.
12
, N
o.
3
,
Dece
m
ber
2
01
8
:
1
239
–
1246
1240
them
can
be
us
e
d
to
cl
assify
e
m
otion
s
f
ro
m
m
us
ic
[9]
,
e
m
otion
s
f
r
om
a
te
xt
[10]
,
an
d
w
ord
sens
e
disam
big
uatio
n.
In
[
3]
,
pr
opos
e
d
a
w
ord
s
ense
disam
biguati
on
s
olu
ti
on
us
in
g
S
kip
-
G
ram
cor
pora.
I
n
[
3]
,
Goo
gle
Wo
r
d
Sens
e
Disam
bi
gu
at
io
n
C
orp
ora
as
the
c
orp
or
a
a
nd
achie
ve
d
a
re
su
lt
wit
h
accu
racy
42.
12%.
Howe
ver,
the
existe
d
m
et
ho
d
us
i
ng
co
rpus
did
not
ha
ndle
pro
blem
if
there
is
no
w
ord
from
sentence
th
at
are
in the c
orp
us
.
In
this
resea
r
ch,
the
us
e
of
W
iki
ped
ia
and
Word2
ve
c
is
pr
opos
e
d
to
dev
el
op
the
corp
or
a
.
Me
anwhil
e,
Le
sk
al
gorit
hm
a
n
d
Wu
Palm
er
si
m
il
arity
are
us
e
d
to
handle
pro
blem
if
th
ere
is
no
wor
d
from
sentence
t
hat
a
re
in
c
orp
us
.
F
irst,
tw
o
co
r
pora
are
de
velo
pe
d
us
in
g
data
f
ro
m
W
i
kip
e
dia
.
The
data
obta
ined
from
W
ikipe
di
a
then
pr
e
pro
cessed
to
m
ini
m
iz
e
the
wor
ds
va
riat
ions
[
11
]
.
Af
te
r
pre
processi
ng
t
he
data,
corp
or
a are cr
e
at
ed
us
in
g Wo
r
d2vec.
Sec
ond,
the co
r
pora are used
to
dete
r
m
ine the
m
ean
ing
of an
am
big
uo
us
word.
To
co
nduct
this,
th
e
si
m
il
arity
of
a
sentence
to
the
first
an
d
th
e
second
co
rpus
is
cal
culat
ed
us
in
g
cosine
sim
il
ari
ty
.
If
the
re
a
re
any
w
ords
fro
m
the
sentence
that
do
no
t
bel
ong
t
o
co
r
pora,
Lesk
al
gorith
m
[12]
and
Wu
Palm
er
[
13]
are
use
d
to
cal
culat
e
the
sim
il
arity.
The
n,
the
m
eaning
of
a
n
am
big
uous
w
ords
is
determ
ined
ba
s
ed on t
he val
ue
of sim
i
la
rity
t
hat h
a
s
b
een
ca
lc
ulate
d.
2.
RESEA
R
CH MET
HO
D
The
m
ai
n
ob
j
e
ct
ive
of
this
r
e
search
is
to
de
velo
p
the
c
orp
or
a
a
nd
t
o
us
e
it
as
a
too
l
to
so
lve
word
sense
disam
bi
gu
at
io
n
pro
blem
.
The
pr
op
ose
d
m
et
ho
d
is
di
vid
ed
i
nto
th
r
ee
par
ts;
the
first
par
t
is
dev
e
lop
i
ng
the
W
i
kip
e
dia
corp
or
a;
the
se
cond pa
rt is d
et
erm
ining
the
re
su
lt
; t
he
thi
rd pa
rt is p
e
rfo
rm
a
nce m
easur
e.
2
.
1.
Dev
el
op
i
ng
Wiki
pedia
Co
r
po
r
a
Figure
1
s
how
s
the
process
of
de
velo
ping
the
co
rpora.
Th
e
two
dataset
s
from
W
ikipe
di
a,
su
c
h
as
a
word
“
ba
nk
(
fina
ncial
)”
an
d
“ba
nk
(g
e
ogra
phy)”,
are
us
ed
as
a
n
inp
ut
to
be
pr
e
processe
d.
Then,
the
pre
process
ed
data
are u
se
d
to
create
tw
o
corp
or
a
us
i
ng w
ord
2vec.
C
orpu
s
1
an
d
c
orp
us
2
a
re
the
out
pu
t o
f
each
dataset
.
Figure
1. Co
rpor
a
d
e
velo
pm
e
nt
2
.
1.
1
D
atase
t
The
W
i
kipe
di
a
arti
cl
e
has
m
any
f
eat
ur
es
inclu
ding
a
t
able
of
c
on
te
nts,
a
rtic
le
refe
ren
ces
an
d
cat
egory
la
bels.
In
this
paper,
the
use
of
cat
ego
ry
la
be
l
featur
e
f
ro
m
W
i
kip
e
dia
ar
ti
cl
e
is
pr
op
ose
d
to
determ
ine
wh
ic
h
arti
cl
e
will
be
sel
ect
ed
as
a
dataset
to
dev
el
op
the
co
rpus.
F
or
e
xam
pl
e,
we
de
velo
p
corp
ora
for
word
“
bank”,
t
he
first
c
orp
us
is
ba
nk
as
a
fina
ncia
l
insti
tuti
on
a
nd
the
seco
nd
co
rpus
is
ba
nk
as
geog
raphy.
T
he
arti
cl
e
sel
ection
is
base
d
on
the
cat
eg
or
y
la
bels
fr
om
W
i
kip
e
dia.
F
or
ba
nk
as
a
fi
nan
ci
al
insti
tuti
on
,
t
he
W
iki
ped
ia
a
r
t
ic
le
s
that
con
ta
ins
word
“
bank”
with
c
at
egory
la
bels
relat
ed
to
finan
ci
al
insti
tuti
on
a
re
sel
ect
ed.
Th
e
cat
eg
or
ie
s
we
fou
nd
th
at
relat
ed
t
o
fina
ncial
ins
ti
tuti
on
are
Ba
nk
s
,
Ba
nk
i
ng,
Leg
al
Entit
ie
s,
Ital
ia
n
Inve
ntio
ns
,
a
nd
Eco
nom
ic
History
of
Ital
y.
For
ba
nk
as
ge
ogra
phy,
the
cat
egory
la
bels
are
Hyd
ro
l
og
y,
Geo
m
orp
ho
l
og
y,
Li
m
no
log
y,
Fre
s
hw
at
e
r
Ecol
ogy,
Fluv
ia
l
La
ndf
or
m
s,
Ri
par
ia
n
Zo
ne,
Ri
ver
s
,
Water
Stream
s,
an
d
Water
a
nd
t
he
En
vironm
ent.
Af
te
r
c
hoos
in
g
the
a
rtic
le
s,
th
en
th
e
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci
IS
S
N:
25
02
-
4752
Develo
ping C
orpora
u
si
ng W
or
d2vec
an
d
W
ik
ipedia
f
or
W
or
d S
ense
D
is
ambi
guatio
n
(
F
ar
z
a
Nu
rif
an
)
1241
con
te
nt
of
the
arti
cl
e
in
the
p
aragrap
h
is
ob
t
ai
ned
.
Pa
ragra
ph
s
ta
ke
n
f
rom
W
ikipe
dia
a
re
the
n
bro
ke
n
do
w
n
into se
ntence
s
base
d on pe
rio
d punctuati
on. T
he pr
ocess
ca
n be see
n
i
n
Fi
gure
2.
Figure
2. A
rtic
le
selec
ti
on
2
.
1.
2
Prepr
oc
essing
The
sente
nces
fr
om
the
dataset
hav
e
s
o
m
any
var
ia
ti
on
s.
T
his
co
nd
i
ti
on
m
akes
the
process
f
or
creati
ng the c
orp
or
a
w
il
l be
le
ss accu
rate. T
he
prep
ro
ce
ss it
sel
f
ha
s six
ste
ps
t
o do.
a)
Lo
wer
case
This
is
the
sim
ple
way
to
m
ake
the
wor
ds
va
riat
ion
s
t
o
be
le
ss.
For
exam
ple,
if
there
are
tw
o
w
ords
“M
on
ey
” a
n
d
“
m
on
ey
” it
w
il
l be
recog
nized a
s sam
e w
ord.
b)
Rem
ov
e
punct
uation
In
buil
di
ng
th
e
corpora
an
d
te
sti
ng
it
wit
h
our
te
sti
ng
data,
we
only
need
the
wor
ds.
Theref
or
e
,
the
punctuati
ons
ar
e rem
ov
ed
.
c)
To
ken
iz
e
We
to
ke
nize
the
in
put
sente
nce
to
m
ake
it
easi
er
to
be
processe
d
at
the
nex
t
ste
p,
w
hich
is
P
OS
ta
ggi
ng
and Lem
m
a
ti
zin
g.
d)
PO
S
tag
ging
To
m
ake
the
da
ta
m
or
e
acc
urat
e
we
us
e
POS
ta
ggin
g.
Part
of
S
peec
h
(
P
OS
)
Ta
gg
i
ng
is
com
m
on
ly
use
d
to d
et
e
rm
ine w
hethe
r
a
w
ord
i
s a
noun wo
rd,
a v
e
rb, a
n
a
dj
e
ct
ive or a
n
a
dverb.
e)
Le
m
m
atize
This
is
the
pa
r
t
i
m
po
rtant
process
t
o
m
ake
the
data
t
o
ha
ve
le
ss
var
ia
ti
on
s
.
We
will
m
ake
w
ords
li
ke
“banks”
to
be
sam
e
as
the
wo
r
d
“ba
nk”.
We
lemm
at
iz
ed
the
words
bas
ed
on
the
POS
ta
gg
in
g
of
the
words, s
o
t
he
l
e
m
m
a
ti
zed wo
rd
s
w
il
l be
m
or
e accu
rate
.
f)
Rem
ov
e the
st
op word
Stop
w
ords
a
r
e
w
ords
that
do
no
t
c
on
ta
i
n
sign
ific
a
nt
m
e
anin
g
wh
e
n
it
is
us
e
d
to
c
rea
te
a
corp
us
.
F
or
exam
ple,
are
"
the"
an
d
"t
o
be
"
w
ords,
both
do
not
prov
i
de
a
sig
nificant
m
eaning
to
th
e
co
ntext
of
t
he
sentences
. Ta
bl
e 1
s
hows
the
e
xam
ple o
f pre
processi
ng r
es
ult.
Table
1.
Pr
e
processin
g
Re
s
ul
t
Inp
u
t
Ou
tp
u
t
My cu
rr
en
t ban
k
d
ep
o
sit acco
u
n
t interest rat
e has
jus
t be
en
cu
t again
.
cu
rr
en
t ban
k
dep
o
s
it accou
n
t interest
rate
cu
t
Mos
t peo
p
le hav
e a
cu
rr
en
t
accou
n
t and
m
o
st b
an
k
s p
ay
v
irtually
n
o
interes
t on
this
m
o
n
e
y
.
p
eo
p
le curr
en
t acc
o
u
n
t ban
k
pays virtu
ally
interest
m
o
n
e
y
2
.
1.
3
Cre
at
e
Word2
vec
Corpor
a
The
c
orpora
is
dev
el
op
e
d
us
i
ng
w
ord2vec
word
em
beddi
ng
te
c
hn
i
qu
e
[
14
]
-
[
16
]
on
G
oogle
us
i
ng
data
ob
ta
ine
d
f
ro
m
the
con
te
nt
of
W
i
kip
e
di
a
arti
cl
es.
Wo
r
d2vec
is
us
e
d
because
it
has
two
la
ye
rs
of
neural
netw
orks
us
e
d
to
pro
du
ce
w
ord
em
beddin
g
in
a
vecto
r
s
pace.
I
n
vecto
r
s
paces,
w
ord
s
that
s
har
e
c
om
m
on
con
te
xts
will
co
nve
rg
e
in
a
dj
acent places
[
14]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vo
l.
12
, N
o.
3
,
Dece
m
ber
2
01
8
:
1
239
–
1246
1242
Ther
e
are
t
wo
ways
to
c
reate
a
co
rpus,
su
c
h
as
t
he
Co
ntinuo
us
Ba
g
of
Wor
ds
(CB
O
W)
m
od
el
a
nd
the
S
kip
-
G
ram
m
od
el
[14]
.
T
he
CB
O
W
m
od
el
pre
dicts
th
e
w
ords
base
d
on
the
giv
e
n
c
on
te
xt
w
he
reas
Sk
i
p
-
gr
am
pr
e
dicts
words
that
s
urrou
nd
t
he
giv
e
n
word
[
14
]
.
The
pr
e
proces
sed
dataset
s
ar
e
us
e
d
as
a
n
i
nput
t
o
Wor
d2vec.
Sin
ce
this
researc
h
does
no
t
ha
ve
la
rg
e
dataset
s
for
trai
ni
ng,
Sk
ip
-
gr
am
m
o
del
is
us
e
d
bec
ause
it
has
a
bette
r
s
olu
ti
on
in
handling
i
nfreque
nt
w
ords
tha
n
CB
O
W
m
od
e
l.
Sk
ip
-
gram
m
od
el
is
us
ed
with
a
hundre
d
-
dim
ension
al
vecto
r
a
nd w
it
h wi
ndow f
i
ve w
ords
a
nd m
ini
m
u
m
w
ord
a
ppear te
n t
i
m
es.
2
.
2
.
Det
ermi
ne t
h
e Re
sult
Figu
re
3
sho
w
s
the
proce
ss
of
determ
ining
the
res
ult.
We
us
e
te
sti
ng
data
from
Ox
f
ord
E
ng
li
s
h
Dict
ion
a
ry
an
d
Y
our
dicti
on
a
r
y.com
to
be
prep
ro
ce
ssed
.
T
hen,
t
he
se
nte
nce
sim
il
arity
with
t
he
c
orp
ora
is
cal
culat
ed
to
det
erm
ine the
re
su
lt
.
Figure
3.
Deter
m
ine the Res
ult
2
.
2
.
1
Te
s
tin
g Data
a)
Oxfor
d
E
ngli
sh
Dict
ion
a
ry
The
O
xfo
rd
E
ng
li
s
h
Dict
iona
ry
(O
E
D)
is
the
la
rg
est
E
ngli
sh
dicti
on
a
ry
widely
us
ed
by
people
t
o
search
f
or
w
ord
de
finiti
ons
or
searc
h
for
sentence
exam
ples
f
r
om
a
word.
T
he
refo
re,
OE
D
is
use
d
as
te
sti
ng
data.
b)
Yourdict
ior
na
r
y.com
Yourdict
io
nar
y
.co
m
is
a
fr
e
e
on
li
ne
E
ngli
sh
dicti
onary
that
has
m
any
sam
ple
sentences,
fam
ou
s
quotes,
a
nd
a
udio
pr
onunci
at
ion
s
.
I
n
this
dic
ti
on
ary,
e
xam
ples
of
sente
nce
s
are
m
ade
by
i
nter
net
us
er
s,
s
o
the
data will
have
m
any sentence
v
a
riat
ion
s.
T
he
refor
e
, it i
s us
ed
to
test
the
pro
posed
m
et
ho
d.
2
.
2
.
2
Prepr
oc
essing
D
ata
The pre
proces
s
ing
ste
p for tes
ti
ng
data is t
he
sam
e as p
reproc
essing st
ep
fo
r devel
op
i
ng c
orp
or
a.
2
.
2
.
3
Simi
larity
to
Co
r
po
r
a
a)
Cosine
sim
il
ari
ty
Cosine
sim
il
ar
it
y
is
the
cal
culat
ion
betwee
n
tw
o
vecto
rs
with
the
res
ult
of
an
a
ngle
betwe
e
n
them
[17]
.
C
osi
ne
sim
i
la
rity
pro
duces
r
es
ults
with
inter
vals
betwee
n
-
1
an
d
1.
The
form
ula
for
cosine
si
m
il
arity is
;
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci
IS
S
N:
25
02
-
4752
Develo
ping C
orpora
u
si
ng W
or
d2vec
an
d
W
ik
ipedia
f
or
W
or
d S
ense
D
is
ambi
guatio
n
(
F
ar
z
a
Nu
rif
an
)
1243
=
cos
(
)
=
∙
‖
‖
‖
‖
=
∑
=
1
√
∑
2
=
1
√
∑
2
=
1
(1)
wh
e
re
an
d
are
co
m
po
ne
nts
of wo
rd2
vec
ve
ct
or
s
an
d
, res
pecti
vely
.
b)
Sentence
sim
i
l
arit
y
Ever
y
word
i
n
a
sentence
e
xc
ept
the
am
big
uous
wor
d
it
sel
f
are
cal
culat
e
d
us
i
ng
c
os
i
ne
si
m
il
ari
t
y
with
t
he
am
big
uo
us
w
ord
c
onta
ined
in
the
corp
us
[18]
.
T
he
am
big
uo
us
word
in
the
se
ntence
an
d
t
he
w
ord
from
sentence
that
is
no
t
in
corp
us
will
be
gi
ven
0
value
.
T
he
w
ords
f
r
om
sentences
t
hat
hav
e b
ee
n
cal
culat
ed
us
in
g
c
os
ine
si
m
il
arity are th
e
n
a
ver
a
ge
d.
=
1
∗
∑
=
1
(2)
wh
e
re,
= num
ber
of
words f
r
om
sen
te
nce
= co
sine sim
ilarity
o
f
t
he words
fro
m
sen
te
nce
with the
a
m
big
uo
us
wor
d
c)
Determ
ine the
resu
lt
The
m
eaning
of
the
am
big
uous
w
ord
i
n
a
se
ntence
is
dete
r
m
ined
by
t
he
va
lue
of
se
nte
nc
e
sim
il
arity
that
has
bee
n
c
al
culat
ed.
If
t
he
value
of
se
ntence
sim
il
ari
ty
to
co
r
pu
s
one
is
higher
t
ha
n
corp
us
tw
o,
t
he
n
the
m
eaning
of
th
e
am
big
uous
word
prese
nt
in
a
se
ntence
i
s
as
de
fine
d
by
corpu
s
one
and
vice
ver
sa
.
For
exam
ple,
Table
2
sho
ws
the
cal
culat
ion
w
it
h
pr
e
process
ed
in
pu
t
se
ntence
“cu
rr
e
nt
ba
nk
de
po
sit
ac
coun
t
interest
r
at
e c
ut
” w
it
h
the
c
orpu
s
1 ‘
ba
nk’
as
a f
i
nan
ci
al
in
sti
tuti
on
a
nd the
corp
us
2 ‘b
a
nk
’
as
ge
ogra
ph
y.
Table
2.
C
os
in
e
Sim
il
arit
y R
e
su
lt
W
o
rd f
ro
m
sen
ten
ce
Co
sin
e si
m
il
arit
y
with
word
‘
b
an
k
’
in
corp
u
s 1
in
corp
u
s 2
Cu
rr
en
t
0
.83
8
0
(
n
o
t in co
rpu
s)
Ban
k
0
(
a
m
b
ig
u
o
u
s wo
rd)
0
(
a
m
b
ig
u
o
u
s wo
rd)
Dep
o
sit
0
.94
9
0
.98
3
Accou
n
t
0
.95
2
0
(
n
o
t in co
rpu
s)
Interest
0
.92
5
0
(no
t in co
rpu
s)
Rate
0
.89
5
0
.98
6
Cu
t
0
(
n
o
t in co
rpu
s)
0
.99
2
Sen
ten
ce si
m
ila
rit
y
0
.65
1
0
.42
3
2
.
2
.
4
Sem
an
ti
c Simi
larit
y
a)
Lesk
al
gorithm
Lesk
al
gorith
m
is
a
cl
assic
al
al
go
rit
hm
fo
r
w
ord
se
ns
e
di
sam
big
uation.
In
t
his
pap
e
r,
t
he
sim
plifie
d
Lesk
al
gorith
m
is
us
ed
bec
ause
it
has
a
bette
r
perf
or
m
ance
[
12
]
.
Thi
s
al
go
rithm
is
sh
ow
n
in
Fig
ure
4.
It
cal
culat
es
the
ov
e
rlap
ping
w
ords
bet
wee
n
the
input
sent
ence
an
d
the
sentence
from
word
def
i
niti
on
a
nd
exam
ple in d
ic
ti
on
ary.
In this
case,
Wordnet
is use
d
as
the
di
ct
ion
ary.
Figure
4. Sim
plifie
d
Les
k Alg
or
i
thm
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vo
l.
12
, N
o.
3
,
Dece
m
ber
2
01
8
:
1
239
–
1246
1244
Sentence
s
that
do
not
ha
ve
a
s
ing
le
w
ord
c
on
ta
ined
i
n
th
e
c
orp
or
a
are
the
n
us
e
d
a
s
th
e
in
pu
t
into
this
al
gorithm
. Th
e
outp
ut of t
his
al
gorithm
is o
ne
of the
words
in
Wordnet a
nd
will
b
e
us
e
d
i
n
the
n
e
xt ste
p.
b)
Wu Palm
er S
im
il
arity
Wu
Palm
er
si
m
il
arity
[13]
is
on
e
of
m
any
al
gorithm
that
m
easur
es
the
sem
antic
s
i
m
i
l
arit
y
of
t
w
o
words
base
d o
n
the
Wordnet
tree.
The fo
rm
ula f
or cal
culat
in
g
si
m
il
arity u
sin
g Wu Pal
m
er is
=
2
∗
ℎ
(
)
(
ℎ
(
)
+
ℎ
(
)
)
(3)
wh
e
re,
= Least
C
omm
on
Subs
um
er
(pare
nt of t
he t
wo
w
ords
se
ar
ched)
= the
first
w
ord
= the
seco
nd
word
The
wor
d
re
su
l
te
d
f
ro
m
Lesk
al
gorithm
then
m
easur
ed
with
the
real
m
eanin
g
of
am
biguous
w
ord
i
n
Wor
dn
et
us
in
g
W
u
Palm
er
si
m
il
arity
and
th
e
ou
t
pu
t
sc
or
e
wil
l
be
us
e
d
to
determ
ine
the
resu
lt
.
F
or
e
xa
m
ple,
the
ou
t
pu
t
f
rom
Lesk
al
go
ri
thm
is
“sl
op
e”,
then
the
w
or
d
“sl
ope”
m
ea
su
re
d
with
the
wo
r
d
“
bank”
as
a
fina
ncial
instit
ution an
d w
ord
“b
a
nk
”
as
ge
ogra
phy i
n Wor
dn
et
.
c)
Determ
ine the
resu
lt
The
m
eaning
of
the
am
big
uous
wor
d
in
a
sentence
is
de
te
rm
ined
by
the
value
of
Wu
Palm
er
si
m
il
arity
that
has
been
cal
cu
la
te
d.
I
f
the
va
lue
of
Wu
Pal
m
er
si
m
il
arity
to
co
rpus
one
i
s
highe
r
tha
n
c
orp
us
two,
the
n
the
m
eaning
of
th
e
am
big
uous
word
prese
nt
in
a
se
nte
nce
i
s
as
def
in
e
d
by
corp
us
on
e
and
vice
ver
sa
2
.
3
.
Perf
orm
ance
M
e
asure
To
e
valuate t
he
prop
os
ed
m
eth
od,
these
foll
ow
i
ng for
m
ulas are use
d
=
(
1
(
1
+
1
)
+
2
(
2
+
2
)
)
2
(4)
=
(
1
(
1
+
2
)
+
2
(
2
+
1
)
)
2
(5)
1
=
(
)
(
+
)
2
(6)
=
1
+
2
(
)
(7)
wh
e
re,
1
=
T
r
ue pre
dict
ion
of the
f
irst
sense
1
= False
pr
e
dic
ti
on
of the
first
sen
se
2
=
T
r
ue pre
dict
ion
of the
seco
nd se
ns
e
2
= False
pr
e
dic
ti
on
of the se
co
nd se
ns
e
3.
RESU
LT
S
A
ND AN
ALYSIS
In
this
pa
per
,
Pyt
hon
pro
gr
a
m
m
ing
la
ngua
ge
is
im
ple
m
e
nted
t
o
pro
pos
e
the
m
et
ho
d.
To
get
t
he
arti
cl
es
fr
om
W
i
kip
e
dia,
we
us
e
con
te
nt
functi
on
fr
om
W
ikipedia
pyth
on
li
br
ary.
T
he
nltk
pyth
on
li
brary
i
s
us
e
d
to
prep
r
ocess
t
he
data
from
W
i
kip
e
dia
an
d
ge
ns
i
m
pytho
n
li
br
ary
is
use
d
to
create
the
w
ord
2v
ec
corp
or
a.
T
he
a
m
ou
nt
of
the
te
sti
ng
data
we
us
e
d
can
be
se
en
in
Table
3.
Table
4
sho
ws
the
exp
e
rim
ent
resu
lt
without
sem
antic
si
m
il
arit
y.
Since
the
re
is
no
w
ord
fro
m
the
sentenc
e
inside
bo
t
h
corp
or
a,
t
he
s
entence
si
m
il
arity
will
hav
e
0
val
ue.
Ther
e
f
or
e,
the
pr
eci
sio
n,
recal
l,
an
d
F
1s
c
or
e
value
ca
nnot
be
cal
culat
ed.
We
c
a
n
on
ly
calc
ulate
t
he
acc
ur
acy
.
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci
IS
S
N:
25
02
-
4752
Develo
ping C
orpora
u
si
ng W
or
d2vec
an
d
W
ik
ipedia
f
or
W
or
d S
ense
D
is
ambi
guatio
n
(
F
ar
z
a
Nu
rif
an
)
1245
Tab
le
3.
T
est
in
g
Data
A
m
b
ig
u
o
u
s wo
rds
Sen
ses
W
ik
ip
ed
ia Dataset
(sen
ten
ces)
Testin
g
Data
(sen
ten
ces)
Ban
k
Fin
an
cial I
n
stitu
tio
n
&
Geog
raph
y
335
138
Plan
t
Facto
ry &
Bio
lo
g
y
298
80
Hear
t
Feelin
g
&
Organ
369
40
Av
erage
-
334
86
Table
4.
E
xper
i
m
ent
Re
su
lt
s
of
Cosi
ne
Sim
i
la
rity
W
it
ho
ut
Sem
antic
Si
m
ilarity
A
m
b
ig
u
o
u
s wo
rd
Un
k
n
o
wn
sen
ten
ce
s
Accurac
y
(
%)
Ban
k
29
7
3
.72
Plan
t
10
8
1
.25
Hear
t
2
7
7
.50
Av
erage
1
3
,6
7
7
.49
Table
5
is
the
s
econd
res
ults
that
prese
nts
th
e
ex
per
im
ent
resu
lt
with
sem
antic
sim
il
arit
y.
Since
the
re
is
no
0
value
of
the
sem
antic
si
m
il
arity,
we
can
cal
c
ulate
the
pr
eci
si
on,
r
ecal
l,
an
d
F
1s
c
or
e
.
As
ca
n
be
seen
i
n
Table
5,
if
w
e
us
e
sem
antic
si
m
il
arity
wh
en
there
is
no
w
ord
from
the
sente
nce
i
ns
ide
both
c
orp
or
a
,
the accu
racy
re
su
lt
is im
pr
ov
e
d by 8.0
2%
.
Table
5.
E
xper
i
m
ent
Re
su
lt
of
Cosi
ne Si
m
ilarity
with
Sem
antic
Sim
i
la
rit
y
A
m
b
ig
u
o
u
s wo
rd
Precisio
n
(
%)
Recall (%
)
F1
Sco
re
(%
)
Accurac
y
(
%)
Ban
k
8
8
.21
8
9
.33
8
8
.76
8
9
.05
Plan
t
8
5
.00
8
5
.00
8
5
.00
8
5
.00
Hear
t
8
2
.50
8
2
.58
8
2
.54
8
2
.50
Av
erage
8
5
.23
8
5
.63
8
5
.43
8
5
.51
4.
CONCL
US
I
O
N
This
researc
h
pro
po
ses
the
use
of
W
iki
pe
dia
and
Wo
r
d2ve
c
to
dev
el
op
the
corp
or
a
.
Th
e
add
it
ion
al
al
gorithm
su
ch
as
Lesk
al
gorithm
and
Wu
P
alm
er
si
m
il
arity
are
us
ed
to
ha
nd
le
ine
xisten
t
wo
r
ds
in
a
co
rpus
.
The
res
ults
of
our
pr
opos
e
d
m
et
ho
d
to
s
olve
word
se
ns
e
di
sa
m
big
uatio
n
pro
blem
s
sh
ow
an
acc
ur
acy
rate
of
85.51% a
nd
th
e se
m
antic
si
mil
arit
y ca
n
i
m
p
rove th
e accu
ra
cy
r
at
e b
y
8.
02
%. F
or f
urt
he
r
researc
h,
the
proces
s
for
handlin
g
w
ords
from
a
sentence
t
hat
are
not
in
t
he
c
orpora
with
c
os
i
ne
sim
il
ari
ty
is
sti
ll
la
cking
s
o
that
it
can
be devel
op
ed
to
achie
ve b
et
te
r
accu
racy.
ACKN
OWLE
DGE
MENTS
The
a
utho
rs
would
li
ke
t
o
than
k
to
I
ns
ti
tut
Tek
no
l
og
i
Sepulu
h
N
op
e
m
ber
,
Direkt
or
at
Rise
t
da
n
Pen
gabdia
n
Masy
ar
ak
at,
Direkt
orat
Jend
e
ra
l
Pen
gu
atan
Rise
t
dan
Pen
ge
mba
ngan
,
the
Mi
ni
stry
of
Re
search
, Tec
hnology, a
nd
Higher
Educat
i
on of
Ind
on
e
sia
for
fina
ncin
g t
he
resea
rc
h.
REFERE
NCE
S
[1]
A.
R.
Pal
,
D.
Sa
ha,
and
S.
K.
Na
skar,
“
W
ord
sense
disambiguat
io
n
in
Beng
al
i
:
A
knowledge
b
ase
d
appr
oa
ch
usin
g
Benga
li
W
ordNet
,
”
in
2017
S
ec
ond
Int
ernational
Confe
ren
c
e
on
Elec
tric
al
,
Computer
and
Comm
unic
ati
o
n
Technol
ogi
es
(
ICECCT)
,
2017,
pp.
1
–
5.
[2]
N.
Bouhriz
,
F.
Bena
bbou,
E
.
Habib,
and
B
.
Lahm
ar,
“
W
ord
Sense
Disam
bigua
ti
on
Approac
h
f
or
Arabi
c
Te
x
t,”
IJA
CSA)
Int
.
J.
Adv
.
Comput.
Sc
i.
Appl.
,
vol. 7, n
o.
4
,
pp
.
381
–
38
5,
2016
.
[3]
S.
Gupta,
A.
Nam
a
var
i,
and
T.
O.
Sm
it
h,
“
W
ord
Sense
Disam
bigua
ti
o
n
Us
ing
Skip
-
Gram
and
LS
TM
Models,
”
2017
.
[4]
Q.
-
P.
Ngu
y
en
,
A.
-
D.
Vo,
J.
-
C.
Shin,
and
C.
-
Y.
Ock,
“
Eff
ec
t
of
W
ord
Sense
Dis
ambiguat
ion
on
Neura
l
Mac
hin
e
Tra
nsla
ti
on:
A C
ase
Stud
y
in
Kor
ea
n,
”
I
EE
E
A
cce
ss
,
vol.
6,
pp.
38
512
–
38523,
201
8.
[5]
B.
S.
Rinty
a
rna
,
R.
Sarno,
and
C.
Fatichah
,
“
E
nhanc
ing
the
p
e
rform
anc
e
of
se
nti
m
ent
an
aly
sis
ta
sk
on
produc
t
rev
ie
ws
b
y
hand
li
ng
both
lo
cal
a
nd
global
cont
ex
t,
”
Int
.
J. I
nf. De
ci
s.
S
ci.
,
vol. 11, 2018.
[6]
H.
Sujai
ni,
K.
Kus
p
ri
y
ant
o
,
A.
Akhm
ad
Ar
m
a
n,
and
A.
Purw
ari
an
ti
,
“
A
Novel
Part
-
of
-
Spee
c
h
Set
Deve
lopi
n
g
Method
for
Stati
stic
al
Mac
hin
e
T
ran
slation,”
TEL
KOMNIKA
(
Tel
ec
omm
unic
at
ion
Comput.
Elec
tron.
Control.
,
vol
.
12,
no
.
3
,
p
.
581
,
Sep.
2014.
[7]
B.
S.
R
inty
arn
a
and
R.
Sarn
o,
“
Adapte
d
weighted
gra
ph
f
or
W
ord
Sense
Disam
bigua
ti
o
n,
”
in
2016
4t
h
Inte
rnational
Co
nfe
renc
e
on
Info
rm
ati
on
and
Co
mm
unic
ati
on
Te
chnol
ogy
(
ICoICT)
,
2016,
pp.
1
–
5.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vo
l.
12
, N
o.
3
,
Dece
m
ber
2
01
8
:
1
239
–
1246
1246
[8]
N.
Sharm
a,
S.
Kum
ar,
and
Dr.
S.
Nira
nja
n
,
“
Us
ing
Mac
hine
Le
ar
ning
Algorit
hm
s
for
W
ord
Sen
se
Disam
bigua
ti
on
:
A Bri
ef
Surv
e
y
,
”
ISSN
Int
.
J. Co
mput.
Techno
l.
El
e
ct
ron.
Eng.
,
v
ol.
2
,
no
.
1
,
pp
.
2
249
–
6343.
[9]
F.
Hasta
ri
ta
Rachm
an,
R.
Sarno,
and
C
.
Fa
ti
ch
a
h,
“
Mus
ic
Emot
ion
Cl
assific
a
ti
o
n
base
d
on
L
y
r
i
cs
-
Audio
using
Corpus ba
sed
E
m
oti
on,
”
In
t. J. El
e
ct
r.
Comput.
Eng.
,
vol
.
8
,
no
.
3,
pp
.
1720
–
173
0,
2018
.
[10]
F.
H.
Rac
hm
an
,
R.
Sarno,
and
C.
Fati
ch
ah,
“
CBE:
Corpus
-
base
d
of
emotio
n
for
emotion
det
e
ct
ion
in
te
x
t
document,
”
in
2016
3rd
Inte
r
nati
onal
Con
fe
r
enc
e
on
Infor
mation
Technology,
Comput
er,
and
Elec
tric
al
Engi
ne
ering
(
ICITACEE
)
,
2016,
pp.
331
–
335
.
[11]
S.
Vijay
ar
ani
,
M.
R.
Jana
ni,
a
nd
A.
Profess
or,
“
TE
XT
MIN
IN
G:
OP
EN
SO
U
RCE
TOKENIZATION
TOOLS
–
AN
AN
ALYS
I
S
,
”
Adv.
Comput
.
Inte
ll. An I
n
t. J.
,
vol.
3
,
no
.
1
,
pp
.
37
–
47,
2016
.
[12]
P.
Basil
e,
A.
C
aput
o,
a
nd
G.
S
emera
ro,
“
An
Enha
nce
d
L
esk
W
ord
Sense
Dis
ambiguat
ion
Al
gorit
hm
through
a
Distribut
ional
Sem
ant
ic
Model
,
”
Proc.
COL
ING
2014,
25th
Int.
Conf.
Comput
.
Linguist.
Tech
.
Pap.
,
pp
.
1591
–
1600,
2014
.
[13]
P.
Sharm
a,
R.
Tr
ipa
thi,
and
R
.
C.
Tri
pat
h
i,
“
Findi
ng
Sim
il
ar
Pate
n
ts
through
Sem
ant
ic
Quer
y
Exp
a
nsion,”
Proc
edi
a
Comput.
Sc
i.
,
vo
l.
54
,
pp
.
390
–
39
5,
Jan
.
2015
.
[14]
T.
Miko
lov,
K
.
Chen,
G.
Corra
d
o,
and
J.
De
an,
“
Eff
icient
Esti
m
ation
of
W
ord
Rep
rese
ntations
in
Vec
tor
Space,”
i
n
Proce
ed
ings o
f
t
he
Int
ernati
ona
l
Confe
renc
e
on
Learning
R
epre
se
ntat
ions (
ICLR 2
013)
,
2013.
[15]
T.
Mikolov,
I
.
Suts
keve
r,
K.
Che
n,
G.
Corra
do,
a
nd
J.
Dea
n,
“
Dis
tri
bute
d
r
epr
ese
n
ta
ti
ons
of
words
and
phra
ses
and
the
ir
compos
it
io
nal
ity
,
”
Proc
ee
d
ings
of
th
e
26
t
h
Inte
rnationa
l
Confe
renc
e
on
Neural
Informat
ion
Proc
essing
Syste
ms
-
Vol
um
e
2
.
Curra
n
As
sociate
s In
c.,
pp
.
3
111
–
3119,
2013
.
[16]
T.
Miko
lov,
W
.
Yih,
and
G.
Z
weig,
“
L
ingui
sti
c
Regu
la
ri
ti
es
i
n
Conti
nuous
S
pac
e
W
ord
Rep
rese
ntations,
”
in
Proce
ed
ings
of
the
2013
Conf
ere
nce
of
the
North
Ame
rican
Chapte
r
o
f
th
e
Associa
ti
on
f
or
Computati
on
al
Linguisti
cs:
Hu
man Language
Technol
og
ie
s
,
201
3,
pp
.
746
–
751
.
[17]
A.
Huang,
“
Simi
la
ri
t
y
Mea
sures
for
Te
xt
Docu
m
ent
Cluste
ring,”
in
Proceedi
ng
s
of
the
New
Zealand
Computer
Sci
en
ce R
ese
arc
h
Stude
n
t
Conf
ere
nc
e
2008
,
2008
,
pp
.
49
–
56
.
[18]
D.
W
al
i
and
N. M
odhe,
“
W
ord
Sense
Disam
bigua
ti
on
Algorit
h
m
s in
Hindi,”
20
15.
BIOGR
AP
HI
ES OF
A
UTH
ORS
Farz
a
Nurifa
n
i
s
now
fourth
y
ea
r
student
of
Inform
at
ic
s
Depa
rtment
at
Insti
tut
Te
kno
logi
Sepuluh
Nopem
ber
.
His
cur
r
ent
i
nte
rests
are i
n
T
ext
Min
ing
and
I
te
rne
t
of
Thi
ngs.
E
-
m
ai
l: fa
r
za
nur
ifa
n@gm
ai
l
.
com
Ri
y
ana
r
to
Sarn
o
recei
v
ed
M.
Sc
and
Ph.D
i
n
Com
pute
r
Sc
ie
nc
e
from
th
e
Univer
sit
y
of
Brunsw
ic
k
Canada
in
1988
and
1
992.
In
2003
he
was
prom
ote
d
to
a
Full
P
rofe
ss
or.
His
t
ea
ch
ing
and
rese
arc
h
i
nte
rests
in
cl
ud
e
s
Inte
rne
t
of
Thi
ngs,
Proc
ess
Aw
are
Inform
at
ion
S
y
s
te
m
s,
Inte
lligen
t
S
y
s
tem
s a
nd
Business Proce
ss
Mana
g
e
m
ent
.
E
-
m
ai
l: ri
y
an
arto@if.i
ts
.
ac.i
d
Cah
y
ani
ngt
y
a
s
Sekar
W
ah
y
un
i
rec
e
ive
d
h
er
bac
he
lor
degr
ee
from
Inform
at
ion
S
y
stem
of
Univer
sita
s
Bra
wijay
a
in
2018
.
She
ac
ti
v
ely
j
oine
d
in
m
an
y
orga
nizati
ons,
c
om
m
it
te
es,
and
competi
ti
ons
.
She
has
won
seve
ral
competi
t
ion
s
in
business
pla
n,
eng
li
sh
spee
ch,
and
engl
ish
deba
t
e.
Now
,
sh
e
is
joi
n
ing
m
ag
iste
r
stud
y
a
t
D
epa
rtment
o
f
Inf
orm
at
ic
s
in
Insti
tut
T
eknol
ogi
Sepuluh
Nopem
ber
,
Surab
a
y
a
.
Her
cur
r
ent
in
te
r
e
sts
are
in
Proc
ess
Mining
and
Bu
siness
Proce
ss
Mana
gement.
E
-
m
ai
l: c
ah
y
aningt
y
as.
seka
r
.
w
@gm
ai
l.
com
Evaluation Warning : The document was created with Spire.PDF for Python.