Int
ern
at
i
onal
Journ
al of Ele
ctrical
an
d
Co
mput
er
En
gin
eeri
ng
(IJ
E
C
E)
Vo
l.
8
,
No.
6
,
D
ece
m
ber
201
8
, pp.
5153
~
51
61
IS
S
N:
20
88
-
8708
,
DOI: 10
.11
591/
ijece
.
v
8
i
6
.
pp
5153
-
51
61
5153
Journ
al h
om
e
page
:
http:
//
ia
es
core
.c
om/
journa
ls
/i
ndex.
ph
p/IJECE
Analysis
o
f
Mobi
le Ser
vice Provid
ers P
er
f
or
m
ance
U
sing
Naive
Bayes D
ata Mini
ng Tech
niq
ue
M.
A. Burh
anuddin
1
,
R
on
iz
am
Is
mail
2
,
N
urul Iz
z
aimah
3
,
Ali
Ab
dul
-
Jabb
ar M
oham
med
4
,
No
rz
aim
ah
Z
aino
l
5
1,
3,
4
Facul
t
y
of
I
nform
at
ion
and
Com
m
unic
at
ion Te
chno
log
y
,
Uni
ver
siti
Te
kn
ika
l
Malay
s
ia Mel
ak
a,
Ma
lay
si
a
2
,
5
Facul
t
y
of
Sci
enc
e
and
T
ec
hno
log
y
,
Kol
ej
Univ
ersit
i
Islam Mela
ka,
Ma
lay
si
a
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
Ma
r
11
, 201
8
Re
vised
Ju
l
4
,
201
8
Accepte
d
J
ul
14
, 2
01
8
Rec
en
tly
,
th
e
m
obil
e
servic
e
provide
rs
have
bee
n
growin
g
rap
idly
in
Malay
s
ia.
I
n
t
his
pape
r,
we
propose
ana
l
yti
c
al
m
et
hod
t
o
find
best
te
l
ec
om
m
unic
at
i
on
provide
r
b
y
visua
li
z
ing
the
ir
p
erf
orm
anc
e
among
te
l
ec
om
m
unic
at
i
on
servic
e
provi
der
s
in
Malay
si
a,
i.e.
TM
Berh
ad,
Celcom
,
Maxis,
U
-
Mobile
,
etc.
Th
is
paperus
es
dat
a
m
ini
ng
te
chni
qu
e
to
e
val
ua
te
the
per
form
anc
eof
t
el
e
comm
unic
at
i
on
servic
e
p
rov
ide
rs
using
thei
r
customers
fee
dba
ck
from
Twit
te
r
Inc
.
It
d
e
m
onstrat
es
on
how
the
s
y
stem
c
ould
proc
ess
and
the
n
in
te
rpr
et
the
b
ig
data
in
to
a
sim
ple
gra
p
h
or
visual
izati
o
n
form
at
.
In
addi
ti
on
,
buil
d
a
computer
i
ze
d
tool
and
rec
o
m
m
end
dat
a
ana
l
y
tic
m
ode
l
base
d
on
the
co
llected
result.
From
pre
pping
the
dat
a
for
pr
e
-
proc
essing
unti
l
conduc
t
ing
anal
y
sis,
th
is
project
is
foc
using
on
the
proc
ess
of
dat
a
sc
ie
nc
e
it
self
wh
ere
Cro
ss
Industr
y
Stan
dar
d
Proce
ss
for
Data
Mining
(
CRIS
P
-
D
M)
m
et
hodolog
y
wi
ll
be
used
as
a
ref
ere
n
ce.
Th
e
ana
l
y
sis
was
deve
lope
d
b
y
using R
la
nguag
e
and
R
Studio
pac
kag
es.
From
the
result
,
i
t
show
s tha
t
Te
lc
o
4
is
the
b
est
as
it
re
ceive
d
hig
hest
positi
v
e
sc
ore
s
from
the
t
wee
t
da
ta.
In
cont
rast
,
T
el
co
3
should
improve
their
per
form
a
nce
as
h
avi
ng
l
e
ss
positi
ve
fee
dba
ck
from
t
hei
r
customers
v
ia
twe
et
da
ta.
T
his
proje
c
t
bring
insight
s
of
ho
w
the
t
el
e
co
m
m
unic
at
ion
in
dustrie
s
c
an
an
aly
z
e
twe
et
d
ata
from
the
i
r
customers.
Mal
a
y
si
a
te
l
ec
om
m
unic
a
ti
on
indust
r
y
will
get
th
e
bene
f
it
b
y
improving
the
ir
customer
sati
sfa
ct
ion
and
busin
ess
growth.
Beside
s,
i
t
wil
l
give
th
e
awa
ren
ess
to
the
t
el
e
co
m
m
unic
at
ion
user
of
upd
at
ed
re
vie
w
from
othe
r
users
.
Ke
yw
or
d:
B
ig d
at
a
D
at
a m
ining
D
at
a scie
nce
M
ob
il
e se
rv
ic
e
s
Naive Bay
es
al
gorithm
T
el
ecom
m
un
icati
on
se
rv
ic
es
Copyright
©
201
8
Instit
ute of
Ad
v
ance
d
Engi
ne
eri
ng
and
Sc
ie
n
ce
.
Al
l
rights re
serv
ed
.
Corres
pond
in
g
Aut
h
or
:
Ali A
bdul
-
Ja
bbar
Mo
ham
m
e
d
,
Faculty
of In
form
ation
a
nd C
omm
un
ic
at
ion
Tech
no
l
og
y,
Unive
rsiti
Tek
nik
al
Mal
ay
sia
Mel
aka
,
Un
i
ver
sit
i
Te
knikal M
al
ay
sia
Mel
aka
,
Hang T
ua
h
Jay
a, 76
100 D
ur
ia
n
T
unggal
, Me
la
ka,
Mal
ay
sia
.
Em
a
il
:
p0
3161
0009@st
ud
e
nt.
utem
.ed
u.
m
y
1.
INTROD
U
CTION
The
m
ai
n
reg
ul
at
or
an
d
gove
rnor
of
te
le
co
m
m
un
ic
at
ion
s
and
it
s
ru
le
s
in
Ma
la
ysi
a
is
the
Ma
la
ysi
an
Com
m
un
ic
at
io
ns
a
nd
Mult
i
m
edia
Com
m
i
ssion
[
1],
[2
]
.
Re
gula
tory
r
eform
s
and
re
hab
il
it
at
ion
a
r
e
ve
ry
i
m
po
rtant
as
pe
ct
s
in
creati
ng
com
petit
ion
eff
ect
ive
ness
a
m
on
g
the
industry
of
te
le
co
m
m
un
ic
at
ion
s.
Corresp
ondi
ngly
,
the
Ma
la
ysi
an
te
le
com
m
u
nicat
ion
s
in
du
stry
has
been
excep
ti
onal
gr
ow
t
h
i
n
recent
ye
ars
[
3]
.
T
her
e
fore,
this
le
ads
to
pro
duce
a
huge
and
div
e
rse
da
ta
set
s
i.e.,
big
data,
w
hich
is
need
a
naly
ti
cs
an
d
inv
est
igati
on
to
disco
ve
r
hi
dd
e
n
co
rr
el
at
ion
s
,
cust
om
er
pr
efe
re
nces,
m
ark
et
trends,
and
f
ur
t
her
va
luable
inf
or
m
at
ion
th
at
m
ay
help
organ
iz
at
io
n
s
m
ake
bette
r
business
decisi
ons.
Pr
oble
m
arises,
with
t
he
gro
wing
fiel
d
of
bi
g
data,
util
iz
at
i
on
of
str
uctu
red
an
d
unstr
uctu
red
data
le
ads
to
worthy
inform
at
io
n
f
or
te
le
com
m
un
ic
a
ti
on
s
in
dustry
in
Ma
la
ysi
a
to
gro
w
e
xpone
ntial
ly
[4]
.
Co
ns
e
qu
e
ntly
,
issues
on
util
iz
at
ion
of
s
tructu
re
d
an
d
un
st
ru
ct
ur
e
d
da
ta
requires
c
riti
cal
and
analy
ti
cal
m
e
tho
ds
t
o
ove
rco
m
e
the
needs
of
in
dustry
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
87
08
In
t J
Elec
&
C
om
p
En
g,
V
ol.
8
, N
o.
6
,
Dece
m
ber
201
8
:
5153
-
5161
5154
grow
t
h
[5
]
,
[
6]
.
Th
ere
a
re
m
any
chall
en
ges
to
be
face
d
for
fin
ding
ou
t
t
he
best
te
le
com
m
un
ic
at
ion
ser
vice
pro
vid
er
si
nce
nowaday
s
the
re
are
to
o
m
a
ny
ch
oices
of
m
ob
il
e
co
m
mu
nicat
io
n
ser
vi
ces
with
a
dif
fer
e
nt
serv
ic
e
rates a
nd spee
ds
[
7]
.
T
he
c
ontrib
ution o
f
t
his stu
dy
is to
giv
e
a
s
olu
ti
on
for
e
va
luat
ing
t
he per
f
or
m
ance of
te
le
com
m
un
ic
a
ti
on
se
rv
ic
e
pr
ov
i
der
s
in
the
Ma
la
ysi
an
te
le
com
m
un
ic
at
ion
s i
ndus
try
,
thi
s is b
y
:
An
al
yz
in
g
hu
ge
and
div
e
rse
data
giv
i
ng
by
the
te
le
co
m
m
un
ic
at
ion
ser
vice
us
e
rs
us
i
ng
t
heir
twit
te
r
accounts
daily
.
Ra
nk
i
ng
the
pe
rfor
m
ance
of
the
te
le
com
m
u
nicat
ion
ser
vic
e
pro
vid
e
rs
i
n
Ma
la
ysi
a
based
on
the
t
weet
s
data of t
heir
use
rs
.
2.
RESEA
R
CH
METHO
D
Fr
om
pr
e
ppin
g
da
ta
f
or
pre
-
proce
ssin
g
unti
l
cond
uctin
g
a
naly
sis,
th
e
sco
pe
of
t
hi
s
pro
j
ect
is
fo
c
us
in
g
on
th
e
process
of
da
ta
sci
ence
it
sel
f.
T
he
m
et
ho
d
use
d
i
n
this
stud
y,
is
base
d
on
Cr
os
s
Ind
us
try
Stand
a
r
d
Proce
ss
for
Data
Mi
ning
(CRI
SP
-
DM)
[
8]
,
as
th
is
m
od
el
is
we
ll
-
known
i
n
th
e
data
m
ining
process
[9]
–
[
11
]
.
T
he
com
plete
pr
oc
ess
diag
ram
of
CR
ISP
-
DM
is
giv
en
in
t
he
Fig
ur
e
1
a
nd
fo
ll
owe
d
by
th
e
descr
i
ption f
or
each
process
in
cl
ud
e
d
in
the
m
od
el
.
Figure
1
.
Cr
os
s
-
I
ndus
try
-
Stan
dard
-
Process
for Data
-
Mi
ning
(CRISP
-
DM
)
m
od
el
[8]
Fr
om
Figure
1
,
the
business
unde
rstan
ding
proces
s
f
ocu
se
s
on
t
he
pur
pose
s
and
requirem
ents
of
the
pro
j
ect
,
w
hich
com
pr
ise
s
unde
rstan
ding
the
bu
si
ness
obj
ect
ives,
s
uccess
c
rite
ria,
pro
j
ect
plan,
a
nd
deliv
eries
[12]
–
[14]
.
T
he
data
unde
rsta
nd
i
ng
pr
ocess
sta
rts
with
an
init
ia
l
-
data
-
c
ollec
ti
on
a
nd
m
a
nag
e
to
procee
d
wit
h
the
data
descr
i
ption
an
d
data
ex
plo
rati
on.
The
data
prep
ara
ti
on
proces
s
inclu
des
data
cl
eanin
g,
sa
m
pl
in
g,
norm
al
iz
a
ti
on
,
and
featu
re
s
el
ect
ion
.
T
he
m
od
el
ing
proc
ess
inclu
des
s
el
ect
m
od
el
ing
te
chn
i
qu
es
,
buil
ding
,
and
trai
ning
the
m
od
el
,
in
ad
diti
on
to
m
ake
pr
e
dicti
on.
The
evaluati
on
pro
cess
includes
t
he
m
od
el
valid
at
ion
,
rev
ie
w
the
re
su
lt
s,
an
d
suc
cess
crit
eria
evaluati
on.
Finall
y,
the
de
plo
ym
ent
pr
ocess
inclu
de
s
resu
lt
visu
al
iz
at
ion,
and
the
re
por
t
creati
on.
T
her
e
fore,
the
m
et
ho
d
that
su
it
s
ou
r
se
nt
i
m
ent
analy
sis
f
or
te
le
com
m
un
ic
a
ti
on
business
operati
on is
de
fined
in t
he wor
kf
l
ow that
giv
e
n
in
Fig
ure
2.
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
& C
om
p
Eng
IS
S
N:
20
88
-
8708
An
alysis
o
f M
obil
e
Service
Pr
oviders
Perfo
r
mance
U
si
ng
Naive
Ba
y
es
D
ata
Mi
ning
…
(
Ali
Ab
dul
-
J
.
M
.
)
5155
Figure
2
.
Se
ntim
ent A
naly
sis
Flow
The
c
om
pu
te
r
pro
gram
in
this
pro
j
ect
is
w
ritt
en
us
in
g
R
Stu
di
o
a
nd
R
la
ngua
ge
w
hich
is
a
pro
gr
am
m
ing
la
ngua
ge
f
or
st
at
ist
ic
al
co
m
p
uting
a
nd
grap
hics.
Wh
il
e
th
e
data
that
will
be
us
e
d
duri
ng
t
he
te
st,
gathe
red
f
ro
m
the
Twitt
er
A
ppli
cat
ion
Plat
fo
rm
In
te
r
ph
a
se
(
API).
F
or
t
he
us
e
r
tha
t
wan
t
to
acce
ss
the
data
from
Twitt
er
API
nee
d
to
hav
e
t
he
Twitt
er
acco
unt.
H
oweve
r,
th
e
first
ste
p
be
fore
be
ginnin
g
the
c
ode,
R
stud
i
o
needs
a
n
API
key
to
s
ynch
ronize
it
with
t
he
T
witt
er
AP
I
.
Af
te
r
t
he
sy
nchr
on
iz
e
su
cce
ss,
the
da
ta
can
be
ga
t
her
e
d
f
re
el
y
fr
om
the
Twitt
er
API,
bu
t
the
R
stud
i
o
c
an
acce
ss
on
ly
the
data
wit
hin
se
ven
days
be
fore
the r
e
quest
d
at
e.
Fo
r
t
he
bi
g
dat
a
analy
sis,
Naï
ve
Ba
ye
s
te
chni
qu
e
is
de
plo
ye
d
in
this
pro
j
ec
t
to
ob
ta
in
t
he
resu
lt
f
ro
m
big
datat
o
pro
du
ce
the
m
os
t
accu
rate
res
ult.
The
Naïve
Ba
ye
s
cl
assifi
er
i
s
a
su
pervise
d
le
arn
in
g
an
d
one
of
the
sim
ple
pr
ob
a
bili
sti
c
cl
assifi
er
te
ch
niques
in
the
Ma
chine
Lea
r
ning
c
ourse
with
str
ong
(
naive
)
ind
e
pende
nce
assum
ption
s
be
tween
the
fea
tures
[
15
]
–
[
17
]
.
The
Fig
ur
e
3
is
sh
ow
in
g
th
e
pr
oce
sses
flo
wch
a
r
t
of N
aï
ve
Ba
ye
s Tech
nique.
Figure
3
.
Naï
ve
Bay
es Tech
ni
qu
e
Flo
wch
a
rt
The
trai
n
cl
ass
ifie
r
can
be
use
d
f
or
trai
ni
ng
the
data
to
cal
culat
e
Ba
ye
s
-
optim
al
esti
m
ates
and
m
ake
pr
e
dicti
on
s
of
the
m
od
el
pa
r
a
m
et
ers
[18]
–
[
20
]
.
The
pr
oce
ss
flo
wc
har
t
of
the
trai
n
cl
as
sifie
r
that
a
ppli
ed
in
this p
roject is
gi
ven
i
n
Fi
gure
4.
Data
Co
llectin
g
Featu
re
/
attribu
te
Selection
Pre
-
p
rocess
in
g
Tr
ain
in
g
Data
Clas
sif
icatio
n
Data
Visu
alizatio
n
Start
Tra
in
Cla
ss
ifier
Te
st C
la
ss
ifi
er
Get
Sent
iment
Exi
t
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
87
08
In
t J
Elec
&
C
om
p
En
g,
V
ol.
8
, N
o.
6
,
Dece
m
ber
201
8
:
5153
-
5161
5156
Figure
4
.
Trai
n C
la
ssifie
r
The
Fig
ure
5
sh
ows
how
N
aï
ve
Ba
ye
s
work
s
in
t
he
te
st
set
cl
assifi
er
for
sentim
ent
data.
This
is
appr
opriat
el
y
represe
ntati
ve
intende
d
f
or
the
unde
rly
ing
recog
niti
on
pro
blem
,
t
hat
le
ads
to
worth
y
inf
or
m
at
ion
for t
el
ecom
m
un
icati
on
s i
ndus
try
in
Ma
la
ysi
a to
grow e
xpone
nt
ia
lly.
Figure
5
.
Test
Cl
assifi
er to
G
et
Sen
ti
m
ent R
esult
Remove
Punctuations
Ta
ke
the proba
b
i
li
t
y
Is i
t in
t
rai
n
da
ta?
Add pri
ority
pro
babi
lit
y
to
th
ese
proba
bil
i
ties
Cal
culat
e
prob
ab
il
ities
using Na
ïve Bayes
Algorit
hm
Add t
he
pro
babil
ity
for
a
ll
the words in
th
e
d
at
a
Yes
No
Input
T
est
Da
ta
Is t
he
posi
ti
ve
proba
bil
i
t
y
h
igh
er
?
Print
Pos
it
ive
Print
Nega
ti
ve
Dum
p
int
o
pic
kl
e
fi
le
Dum
p
int
o
pic
kl
e
fi
le
Start
Remove
Punctuations
The
la
be
l
is
n
egative
Cal
culat
e
th
e
proba
bil
i
ties
Cal
culat
e
th
e
proba
bil
i
ties
Yes
No
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
& C
om
p
Eng
IS
S
N:
20
88
-
8708
An
alysis
o
f M
obil
e
Service
Pr
oviders
Perfo
r
mance
U
si
ng
Naive
Ba
y
es
D
ata
Mi
ning
…
(
Ali
Ab
dul
-
J
.
M
.
)
5157
The
fig
ur
es
a
bove
sho
w
the
m
et
ho
dolo
gy
of
ho
w
to
get
th
e
res
ults
from
our
a
naly
sis.
Con
s
eq
ue
ntly
,
the
f
ollo
wing
i
s
a
bri
ef
e
xpla
nation
inclu
di
ng
ste
p
by
ste
p
of
how
Naï
ve
Ba
ye
s
te
chn
iq
ue
work.
T
his
can
be
detai
le
d
as:
Step
1: d
et
e
rm
i
ning the
test
se
t i
n
o
ur
dataset
as the
foll
ow
i
ng
i
n
Ta
ble
1.
Table
1
.
T
est
S
et
DOC
TE
X
T
CLASS
1
I
lo
v
ed
the servi
ce
+
2
I
h
ated
the servi
c
e
-
3
A great se
rvice,
go
o
d
service
+
4
Po
o
r
serv
ice,
Po
o
r
co
n
n
ectio
n
-
5
A go
o
d
service,
g
r
eat con
n
ectio
n
+
So
,
a total
of 10
un
i
qu
e
wo
rds eg. I
, love
d,
t
he,
ser
vice, a,
gr
eat
,
h
at
e
d,
good,
con
necti
on, p
oor.
Step
2: con
ver
t
ing
t
he data
int
o
a
fr
e
quency t
able,
wh
ic
h
is
giv
e
n
in
Ta
ble
2
as
foll
ows:
Table
2
.
Fr
e
quency Ta
ble
DOC
1
2
3
4
5
I
1
1
lo
v
ed
1
th
e
1
1
serv
ice
1
1
2
1
1
h
ated
1
a
1
1
g
reat
1
1
poor
1
co
n
n
ectio
n
1
1
good
1
1
Clas
s
+
-
+
-
+
Nex
t,
lo
ok at t
he pr
obabili
ti
es p
e
r ou
tc
om
e
(+
or
-
)
Step
3: Com
pute
the prio
rity
P (
+
)
= t
otal o
f
+ cl
ass
P (
-
)
=
total
of
-
cl
ass
Step
4: Com
pute
the con
diti
onal
pr
ob
a
bili
ty
/ p
ossi
bili
ty
o
f e
ach att
rib
ute
P(
I|+);
p(l
oved|
+); P(
t
he|+); P
(
serv
ic
e
|+
);
P(
a|
+); P(g
reat|
+); P(
go
od
|+
); P(
c
onnecti
on
|+
);
P(
w
k.
|+
)
=
nk
:
num
ber
of
tim
es w
ord k
oc
cur
s
in
t
hese
c
ases (+
)
n: num
ber
of
w
ords
i
n (+)
ca
s
e
-
>
14
vo
ca
bula
ry:
tot
al
u
ni
qu
e
wo
rds whil
e test
ing
,
for
unkn
own word
s
we use
nk =
0
a
nd f
i
nd it
s p
r
obabili
ty
b
ei
ng
bo
t
h posi
ti
ve
a
nd n
e
gative.
3.
DA
T
A
ANAL
YS
IS
In
t
his
stu
dy,
we
are
us
i
ng
a
real
data
e
xtr
act
ed
f
ro
m
Twitt
er
API,
a
web
sit
e
us
es
t
o
acce
ss
c
or
e
Twitt
er
data
.
C
on
s
eq
ue
ntly
,
w
e
save
t
he
data into
.c
sv
f
il
e
f
orm
at
as
giv
en
i
n
Fi
gure 6
. N
e
xt,
dataset
is
lo
ad
ed
in R stu
dio
f
or
furthe
r
a
naly
ses.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
87
08
In
t J
Elec
&
C
om
p
En
g,
V
ol.
8
, N
o.
6
,
Dece
m
ber
201
8
:
5153
-
5161
5158
Figur
e
6
.
Data
in .
cs
v form
at
The
dataset
ob
ta
ined
from
th
e
Twitt
er
AP
I
in
our
pro
j
ect
is
con
sist
of
5
f
il
es
of
data
accor
ding
to
5
diff
e
re
nt m
ob
il
e comm
un
ic
at
i
on servic
es
pro
vid
e
rs,
a
nd the
se d
at
a
file
s
, inc
lud
es
:
1.
Ce
lc
om
Tw
eet
D
at
a
2.
Ma
xis T
weet
Data
3.
Digi T
weet
Da
ta
4.
U
-
M
ob
il
e T
we
et
D
at
a
5.
Tu
netal
k
T
wee
t Data
All data
file
s c
on
ta
in
the
sam
e d
at
a att
rib
ute
s,
these
at
trib
utes are give
n
i
n Fi
gure
7
.
Figure
7
.
Data
at
tribu
te
s
Ba
sed
on
the
obta
ined
dataset
and
data
at
tribu
te
s
,
not
al
l
the
data
hav
e
be
en
ap
plied
in
t
he
analy
sis,
on
ly
te
xt
at
tri
bu
te
will
be
s
el
ect
ed
an
d
will
be
us
e
d
for
m
od
el
li
ng
pur
po
s
es.
T
he
pu
rpose
of
the
s
el
ect
ed
at
tribu
te
s is t
o see
the
weig
ht
age
of the
posit
ive,
neg
at
i
ve
a
nd n
e
utral
w
ord.
Fo
r
the
res
ult
of
s
entim
ent
analy
sis,
al
l
the
tweet
te
xts
ha
ve
been
scan
ne
d,
a
nd
the
sc
or
e
has
bee
n
giv
e
n.
The
sc
ore
is
base
d
on
their
po
sit
ivit
y
an
d
ne
gativit
y
w
ords,
w
hich
are
based
on
the
posit
ive
file
an
d
neg
at
ive
f
il
e.
T
he
Fi
gure
8
is
s
howing t
he
tw
eet
s
an
d
it
s g
i
ve
n
sc
or
e
s.
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
& C
om
p
Eng
IS
S
N:
20
88
-
8708
An
alysis
o
f M
obil
e
Service
Pr
oviders
Perfo
r
mance
U
si
ng
Naive
Ba
y
es
D
ata
Mi
ning
…
(
Ali
Ab
dul
-
J
.
M
.
)
5159
Figure
8
.
Twee
ts t
hat alrea
dy
hav
e
sc
or
e
These
sc
or
es
a
nd
resu
lt
s
can
be
us
e
d
to
im
pro
ve
the
cust
om
er
exp
erie
nc
e
and
bu
si
nes
s
grow
t
h
by
disco
ver
i
ng
un
known
c
orrelat
ion
s
,
hi
dd
e
n
pa
tt
ern
s,
c
us
tom
er
prefe
rence
s,
m
ark
et
tren
ds
,
and
f
ur
the
r
valuab
le
inf
or
m
at
ion
th
at
m
ay
help
organ
iz
at
io
ns
m
ake
bette
r
bu
si
ne
ss
decisi
on
s
.
The
te
c
hn
i
qu
e
that
de
plo
ye
d
i
n
this
pro
j
ect
is
the
Naïve
Ba
ye
s
,
wh
ic
h
a
ble
to
pro
vid
e
st
ron
g
ind
e
pe
ndence
assum
ption
s
betwee
n
t
he
fe
at
ur
es
relat
ed
t
o
the
s
entim
ent
analy
sis.
F
ur
t
her
m
or
e,
it
giv
es
the
rob
us
t
s
olu
ti
on
am
on
g
te
le
c
om
m
un
ic
at
ion
s
erv
ic
e
pro
vid
e
r
s
[
10]
.
4.
FIN
DINGS
A
ND R
ES
ULT
S
Af
te
r
t
he
sc
or
e
had
gi
ven,
the
resu
lt
s g
ra
ph
is
plo
t ba
se
d
on
their
ne
gativit
y
and
posit
ivit
y
po
la
rity
as
sh
ow
n
in
Fi
gur
e 9
belo
w.
Figure
9
.
P
olar
it
y of
the
tweet
s
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
87
08
In
t J
Elec
&
C
om
p
En
g,
V
ol.
8
, N
o.
6
,
Dece
m
ber
201
8
:
5153
-
5161
5160
Af
te
r
the
gra
ph
done
pl
otti
ng
,
al
l
t
hese
res
ults
are
tra
ns
fe
rr
e
d
t
o
R
Sh
i
ny
wh
ic
h
is
us
e
d
to
visu
al
iz
e
the
res
ult
in
m
or
e
pro
per
a
nd
creati
ve
way.
R
Sh
iny
ha
d
be
en
ch
os
e
n
as
it
s
easy
interphase
to
unde
rsta
nd
a
nd
us
e
eve
n
f
or
the
ve
ry
first
-
ti
m
e
us
er.
Ba
se
d
on
the
Fi
gur
e
10
belo
w,
w
e
can
see
that
there
are
dif
fe
ren
t
5
boxes
with
dif
f
eren
t
col
or
a
nd
value.
T
he
val
ue
sta
te
d
in
the
box
is
the
am
ount
of
ra
w
da
ta
gather
e
d
f
rom
the
Twitt
er
API
th
at
we
are
deali
ng
with
f
or
thi
s
proj
ect
.
Ba
se
d
on
pola
rity
scor
es
,
te
le
com
m
un
ic
at
ion
ser
vice
pro
vid
er
s r
a
nk
ed
as
Telc
o 1,
Tel
co 2,
T
el
co 3, Te
lc
o 4
a
nd
Tel
co 5.
Figure
10
.
O
ve
rv
ie
w of
data
Fr
om
Figure
10,
highest
twe
et
fr
e
qu
e
ncy
c
om
e
fr
om
Telco
1,
wh
ic
h
is
5000.
Lo
west
is
Tel
co
4,
wh
ic
h
is
540.
It
m
igh
t
be
T
el
co
1
hav
i
ng
highest
num
ber
of
cust
om
ers
in
Ma
la
ysi
a.
The
ov
e
rall
m
odule
create
d
to
m
ake
a
c
om
par
ison
betwee
n
al
l
the
te
le
com
mu
nicat
io
n
se
rv
i
ce
pro
vid
e
rs
in
Ma
la
ysi
a
ba
sed
on
their
posit
ive
po
la
rity
and
ne
gative
pola
rity
.
The
com
par
iso
n
is
plo
tt
ed
in
a
pie
char
t
and
eac
h
of
th
e
te
le
com
m
un
ic
a
ti
on
ser
vice
prov
i
der
s
’
weig
ht
age
are
sta
te
d
in
a
per
centa
ge
value
as
sh
own
in
a
Fig
ur
e
11
as
fo
ll
ows.
Figure
11
.
Sum
m
arization
ba
sed o
n
P
os
it
iv
it
y po
la
rity
Ba
sed
on
the
resu
lt
showe
d
in
Figure
11,
the
te
le
co
m
m
u
nicat
ion
com
pan
y,
Tel
co
4
is
the
best,
wh
ic
h
getti
ng
92%
posit
ive
twit
te
r
com
m
en
ts
fr
om
their
custom
ers.
Lo
w
est
scor
e
is
Te
lc
o
3,
w
hich
is
on
l
y
62%
sco
re
on
po
sit
ive
c
omm
ents.
By
loo
king
at
this
graph,
te
le
com
s
erv
ic
e
pr
ov
i
de
rs
can
e
valuat
e
their
perform
ance easil
y fr
om
their c
us
tom
ers’
t
weet data.
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
& C
om
p
Eng
IS
S
N:
20
88
-
8708
An
alysis
o
f M
obil
e
Service
Pr
oviders
Perfo
r
mance
U
si
ng
Naive
Ba
y
es
D
ata
Mi
ning
…
(
Ali
Ab
dul
-
J
.
M
.
)
5161
5.
CONCL
US
I
O
N
This
pap
e
r
s
hows
on
how
to
analy
ze
a
nd
vi
su
al
iz
e
tw
eet
data,
wh
e
r
e
inf
or
m
at
ion
eff
ect
ively
delivere
d,
es
pe
ci
al
ly
towar
ds
an
in
div
id
ual
with
no
bac
kg
r
ound
in
a
naly
ti
cs
or
relat
ed
s
ubj
ect
.
With
t
he
rig
ht
visu
al
iz
at
ion
a
nd
gra
ph
ic
s
on
tim
e,
we
ca
n
i
m
pr
ove
e
nd
use
r
unde
rstan
di
ng
an
d
at
t
he
sam
e
t
i
m
e
create
s
a
data
interact
io
n
betwee
n
the
us
ers
an
d
t
he
inf
or
m
at
ion
it
sel
f.
Ba
sed
on
th
e
pro
j
ect
res
ult,
the
se
rv
ic
e
pr
ov
i
de
r
com
pan
ie
s
can
see
the
grap
hs
an
d
th
ei
r
se
r
vice
pe
rfo
rm
a
nce
from
twit
t
ers
.
Th
us
,
it
w
il
l
be
able
to
use
this
pro
j
ect
as
a
ref
ere
nce
to
com
pete
wit
h
the
ot
her
te
le
com
m
un
ic
a
ti
on
ser
vice
pro
vid
er
s.
H
oweve
r
,
i
m
pr
ovem
ent
i
s
def
init
el
y
need
ed
in
e
ver
y
syst
e
m
that
is
dev
el
op
e
d.
T
hi
s
is
to
ensu
re
a
gr
a
du
al
incre
ase
in
us
er
sati
sfacti
on a
nd conti
nue
s im
pr
ov
em
ent of the
syst
em
.
ACKN
OWLE
DGE
MENTS
The
a
uthors
w
ou
l
d
li
ke
to
th
ank
t
he
U
niv
e
r
sit
i
Tekn
ikal
Ma
la
ysi
a
Me
lak
a
,
UTeM
Za
m
al
ah
S
chem
e
for
prov
i
ding
the
facil
it
ie
s
to
c
onduct
this
researc
h
pro
je
ct
and
K
olej
U
niv
er
sit
i
Isla
m
Me
la
ka
for
the
fina
ncial
sup
port in
t
his pr
oj
ec
t.
REFERE
NCE
S
[1]
S.
Ja
y
asingh
an
d
U.
Ez
e,
“
An
empiric
a
l
anal
y
s
is
of
consum
er
beha
vior
al
inten
ti
on
to
ward
m
obil
e
coupons
in
Ma
lay
s
ia
”
,
Int
.
J
.
Bus
.
…
,
vol. 4,
no.
2
,
pp
.
221
–
2
42,
2009
.
[2]
“
Suruhanjay
a
K
om
unika
si da
n
Multi
m
edi
a
Mal
a
y
si
a
(MCM
C)”
,
2017.
[3]
“
Axiat
a
Annua
l Report
”
,
Ax
ia
ta
,
2016.
[4]
N.
Ks
het
ri,
“
The
emergi
ng
role
of
Big
Data
in
ke
y
dev
el
opm
ent
i
ss
ues:
Opportun
it
ie
s,
ch
al
l
enge
s,
and
conc
ern
s
”
,
Bi
g
Data
Soc
.
,
v
ol.
1
,
no
.
2
,
p
.
20
53951714564227,
2014
.
[5]
N.
Patil,
P.
Kira
n,
N.
Kir
an,
N.
K
.
,
“
A
Surve
y
o
n
Graph
Dat
aba
s
e
Mana
g
ement
Te
chn
ique
s
for
Huge
Uns
truc
tur
ed
Data
”
,
In
te
rnati
onal
Journal
of
El
e
ct
rica
l
and
C
omputer
Engi
n
e
ering
(
IJE
CE
)
,
Vol
.
8
No.
2
pp
.
1140
-
1149,
201
8
.
[6]
S.
Borodo,
S.
Sham
suddin,
S.
H
.
“
Bi
g
dat
a
pl
at
form
s
and
te
c
hnique
s
”
,
TEL
K
OMNIKA
Indone
sian
Journal
of
El
e
ct
rica
l
Eng
in
ee
ring a
nd
Computer
Sc
ie
nc
e,
V
ol.
1(1)
,
pp
.
191
-
200,
2016
.
[7]
P.S.H.
Leefl
ang,
P.C.
Verho
ef,
P.
Dahlstr
??
m
,
an
d
T.
Freundt,
“
Chal
l
enge
s
and
soluti
ons
for
m
ark
et
ing
in
a
digita
l
era
”
,
Eur.
Mana
g.
J
.
,
vol
.
32
,
no
.
1,
pp.
1
–
12,
201
4.
[8]
R.
W
irt
h
and
J.
Hipp,
“
CRIS
P
-
DM
:
Towa
rds
a
Standa
rd
Proce
ss
Model
for
Da
ta
Mining”
,
App
l.
Knowl.
Disco
v.
Data
Min.
,
no
.
2
4959,
pp
.
29
–
39
,
2000.
[9]
M.
Spruit,
R.
Vroon,
and
R.
Bat
enbur
g
,
“
Towa
rds
hea
lt
h
care
business
int
el
li
g
ence
in
lo
ng
-
te
rm
ca
re
:
An
expl
ora
ti
ve
c
ase
stud
y
in the
Net
her
la
nds”
,
Comp
ut.
Hum
an
Be
ha
v.
,
vol
.
30
,
pp
.
6
98
–
707,
2014
.
[10]
M.N.M.
Ibra
h
i
m
and
M.Z
.
M.
Yus
off,
“
Twit
t
er
senti
m
en
t
c
l
assific
a
ti
on
usin
g
Naive
B
a
y
es
base
d
on
tra
in
er
per
ce
p
ti
on
”
,
in
2015
IEE
E
Con
f
ere
nce
on
e
-
Lea
rning,
e
-
Manag
e
ment
and
e
-
S
erv
ic
es,
IC3
e
2015
,
2016,
pp.
187
–
189.
[11]
P.
Kalgot
ra
and
R.
Sharda
,
“
Progress
ion
ana
l
ysis
of
signal
s:
Ext
endi
ng
CRI
SP
-
DM
to
stream
ana
l
y
tics
”
,
i
n
Proce
ed
ings
-
2
016
IEEE
In
te
rn
ati
onal
Con
fe
ren
ce
on
B
ig
Data
,
Bi
g
Data
2016
,
2016,
pp
.
2880
–
2885.
[12]
A.
Ta
rhin
i,
H.
Amm
ar,
T.
T.
-
I
.
B.
Resea
r
ch,
a
nd
undef
ine
d
2
015,
“
Anal
y
sis
of
the
cr
it
i
cal
succ
ess
factors
f
or
ent
erp
r
ise
r
esourc
e
p
la
nning
impl
ementa
t
ion
from
stake
ho
lde
rs’ p
e
rspec
ti
v
e:
A
s
y
st
emati
c
rev
ie
w”
,
cc
sene
t.
org
.
[13]
A.
B.
-
I.
J.
of
P.
Mana
gement
and
undef
ine
d
2016,
“
The
impact
of
project
m
ana
gement
(PM
)
and
bene
fit
s
m
ana
gement
(B
M) pra
ctic
es
on
proje
c
t
suc
ce
ss
: Towards de
v
el
o
pi
n
g
a
proj
ec
t
be
nef
it
s gov
ern
an
c
e
”
,
E
lsev
i
er
.
[14]
C.
Serra
,
M.
K.
-
I
.
J.
of
P.
Mana
g
e
m
ent
,
and
und
ef
ine
d
2015,
“
Ben
efi
ts r
e
al
isa
ti
on
m
ana
gement
and
it
s inf
lue
n
ce
on
proje
c
t
suc
ce
ss
a
nd
on
th
e execut
ion
of
busin
ess stra
t
egi
es”
,
El
se
vier
.
[15]
P.
Tsanga
r
at
os
and
I.
Il
i
a,
“
Com
par
ison
of
a
log
isti
c
reg
ression
and
Na
ïve
Ba
y
es
class
ifi
er
in
la
ndsl
i
de
sus
ce
pti
bilit
y
as
sess
m
ent
s:
The
infl
uen
ce
o
f
m
odel
s
complex
ity
and
training
d
a
ta
set
size”
,
Cat
e
na
,
vol
.
145
,
pp
.
164
–
179,
2016
.
[16]
L.
De
y
,
S.
Chak
rab
ort
y
,
A.
Bisw
as,
B.
Bose,
and
S.
Ti
wari
,
“
Senti
m
ent
Anal
y
sis
of
Revi
ew
Data
s
et
s
Us
ing
N
aï
ve
Ba
y
es‘
and
K
-
NN
Cla
ss
ifi
er
”
,
In
t.
J. I
nf
.
Eng. E
l
e
ct
ron. Bus.
,
vol
.
8,
no
.
4
,
pp
.
54
–
62,
2016
.
[17]
S.
Shah,
K.
Kum
ar,
R.
S,
“
Senti
m
ent
al
Ana
l
ysis
of
Twit
te
r
Data
Us
ing
Clas
sifie
r
Algorit
h
m
s”
,
Inte
rnatio
nal
Journal
of
Elec
t
rical
and
Computer
Eng
ine
ering
(
IJE
CE)
,
Vol.
6
No.
1,
357
-
366,
2016.
[18]
A.
McCal
lum
an
d
K.
Nigam,
“
A
Com
par
ison
of
Eve
nt
Models
fo
r
Naive
Ba
y
es
T
ext
Cla
ss
ifica
t
io
n”
,
A
AA
I/ICML
-
98
Work. Learn. T
ex
t
Ca
te
g.
,
pp
.
41
–
48,
1998
.
[19]
S.
Rasc
hka
,
“
Na
ive
B
a
y
es
and
T
ext
Cl
assificat
io
n
I
-
Introductio
n
and
The
or
y
”
,
arXiv
Prepr.
ar
Xi
v1410
.
5329
,
p
.
20,
2014
.
[20]
D.
Li
-
guo,
D.
Peng,
L.
A.
,
“
A
n
ew
nai
ve
Ba
y
e
s
te
xt
class
ifi
c
at
io
n
al
gorit
hm
”
,
T
ELKOMNIKA
Indone
sian
Journal
of
E
le
c
tric
al
En
gine
ering
and
C
omputer
Scienc
e
,
Vol
12
No 2
,
p
p.
947
-
952;
201
4
.
Evaluation Warning : The document was created with Spire.PDF for Python.