TELKOM
NIKA
, Vol. 13, No. 4, Dece
mb
er 201
5, pp. 1384
~1
389
ISSN: 1693-6
930,
accredited
A
by DIKTI, De
cree No: 58/DIK
T
I/Kep/2013
DOI
:
10.12928/TELKOMNIKA.v13i4.xxxx
1384
Re
cei
v
ed Ma
rch 2
6
, 2015;
Re
vised Sept
em
ber
26, 20
15; Accepted
Octob
e
r 12, 2
015
Text Mining Research Based on Intelligent Computing
in Information Retrieval System
Yong Li
Cho
ng Qin
g
T
e
chno
log
y
and
Busin
e
ss Institute, Chon
g Qin
g
Cit
y
,
4
000
52,
Chin
a
email: li
yo
n
g
0
4
171
51@
12
6.co
m
A
b
st
r
a
ct
W
i
th the pop
ul
arity and ra
pi
d deve
l
op
ment o
f
the In
ternet, w
eb text informati
on h
a
s rap
i
dly gr
ow
n
as w
e
ll. T
o
ad
d
r
ess the
key
pr
obl
e
m
of
text
mi
nin
g
, text cl
u
s
tering
is
inves
t
igated
in
this
s
t
udy. T
he s
huff
l
ed
frog le
ap
ing
al
gorith
m
as a
n
e
w
type of sw
a
r
m i
n
tel
lig
enc
e
opti
m
i
z
at
io
n a
l
gorith
m
c
an
be
use
d
to i
m
pro
v
e
the perfor
m
a
n
c
e of the K-means a
l
g
o
ri
th
m, but the shuffled frog l
e
a
p
in
g alg
o
rith
m is
influ
ence
d
by i
t
s
mov
i
n
g
step l
e
ngth. On the b
a
sis of this i
n
formatio
n
,
the s
huffled fro
g
le
a
p
in
g al
gorith
m
is improv
ed, a
n
d
the K-
me
ans c
l
usteri
ng
alg
o
ri
thm
bas
ed
on
the i
m
prov
e
d
shuffle
d
frog
l
eap
ing
al
gor
ith
m
is intr
oduc
e
d
.
Experi
m
ent re
sults show
that t
he propose
d
scheme can
enh
anc
e the a
b
ilit
y of searc
h
ing for the opti
m
a
l
initia
l cluster
i
n
g
center an
d can
effectively
avoid i
n
stabi
l
i
ty in t
he clus
tering res
u
lts of the K-me
a
n
s
clusteri
ng a
l
gor
ithm. T
he
pro
p
o
sed sc
he
me
also re
duc
es
the ch
ances
of the al
gorit
h
m
f
a
lli
ng i
n
to the
l
o
ca
l
opti
m
u
m
. T
h
e
perfor
m
anc
e
of the prop
os
ed cluster
i
n
g
sche
m
e is fo
u
nd to be
b
e
tte
r than that of the
clusteri
ng al
gor
ithm b
a
se
d on
the shuffle
d
frog lea
p
in
g al
gori
t
hm.
Ke
y
w
ords
: T
e
xt Mining, Shuf
fled F
r
og Le
ap
i
ng Alg
o
rith
m, Clusteri
ng Acc
u
racy
Copy
right
©
2015 Un
ive
r
sita
s Ah
mad
Dah
l
an
. All rig
h
t
s r
ese
rved
.
1. Introduc
tion
With the
rapi
d gro
w
th of
web i
n
form
ati
on, peo
ple u
r
gently nee
d
a type of technolo
g
y
that organi
zes an
d man
age
s inform
ation to
help them find what they need qui
ckly and
accurately; data mining co
mbing with web mining ha
s eme
r
ged a
s
a re
spo
n
se
to this issue
[1].
Text is the m
a
in form of in
formation
on
the we
b,
so t
e
xt mining ha
s be
com
e
a rese
arch h
o
tspot
in re
cent ye
ars.
Chin
ese
text mining
has
bee
n de
veloped
relat
i
vely late, and it falls be
h
i
nd
English
text
mining
in th
e
asp
e
ct
s of
th
eory
re
se
a
r
ch
and
ap
plication. Chine
s
e
text mining
ba
sed
on the web i
s
therefo
r
e selecte
d
as o
u
r re
sea
r
ch
o
b
ject. Text classificatio
n
and clu
s
teri
ng
are
key technol
o
g
ies i
n
text mining. O
r
ga
nizin
g
an
d cl
assifying text data
sets can
si
gnificant
ly
address the probl
em of informatio
n explosio
n. Text classifi
catio
n
and clu
s
tering can al
so
be
applie
d a
s
th
e tech
nical b
a
si
s in
sp
ecif
ic field
s
, such as informat
ion retrieval,
sea
r
ch e
ngin
e
s,
electroni
c lib
rari
es, and t
e
xt database
s
[2]. With
the advent o
f
the information era, te
xt
cla
ssifi
cation
and cl
uste
rin
g
have be
co
me increa
sin
g
ly popula
r
.
Curre
n
t clu
s
t
e
ring
algo
rith
ms a
r
e
cla
s
sified
into
p
a
rtitioning
cl
usteri
ng, hi
erarchical
clu
s
terin
g
, grid-ba
s
e
d
clu
s
terin
g
, and
density-b
a
sed clu
s
teri
ng
. The K-me
ans al
gorith
m
, a
cla
ssi
cal
cl
ustering
algo
rit
h
m, is a l
o
cal
se
ar
ch
sch
e
m
e
with
som
e
seri
ou
s di
sadvantag
es.
The
K-value nee
d
s
to be determined in adv
ance, and t
he cluste
rin
g
result dep
end
s on the sel
e
ction
of the initial
clu
s
terin
g
ce
nter [3]. In this re
ga
rd, m
any re
sea
r
ch
ers have p
r
o
posed
clu
s
tering
method
s ba
sed on the intelligent optimization al
g
o
rithm [4, 5]. This algo
rithm is gra
d
u
a
lly
develop
ed wi
th the use of certai
n simila
rities bet
wee
n
compl
e
x system
s (e.g. natural or
so
ci
al)
and optimi
z
at
ion pro
b
lem
s
. The algorith
m
obtains th
e next feasibl
e
solution
s th
roug
h ope
rati
on
on a
set of
initial sol
u
tio
n
s in th
e se
arch
spa
c
e
according to
ce
rtain rule
s of p
r
ob
abil
i
ty.
Therefore, th
e sea
r
ching
mech
ani
sm o
f
the al
gorith
m
determin
e
s
its optimiza
t
ion perfo
rma
n
ce
[6]. The shuf
fled frog
lea
p
ing al
go
rith
m is
a type
of ne
w intel
ligent optimi
z
ation alg
o
rith
m.
However,
this algorithm
also ha
s
some disadvantages, such as
its poor l
o
cal
search
ability and
slo
w
conve
r
g
ence
spee
d [
7–10]. To
im
prove th
e
co
nverge
nt pe
rf
orma
nce of t
he b
a
si
c
shuf
fled
frog lea
p
ing
a
l
gorithm, a
n
i
m
prove
d
algo
rithm is i
n
trod
uce
d
in thi
s
study. The clu
s
terin
g
meth
o
d
based on int
e
lligent optim
ization
can
convert a pa
rt
icula
r
clu
s
te
ri
ng pro
b
lem t
o
an optimi
z
a
t
ion
probl
em of t
he obj
ective
function,
whi
c
h find
s
the
optimal value
of the obje
c
tive function
to
obtain the o
p
t
imal clu
s
teri
ng sche
me throu
gh rep
e
a
t
ed iteration.
The key tech
nologi
es in u
s
ing
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 13, No
. 4, Decem
b
e
r
2015 : 138
4 – 1389
1385
intelligent
co
mputing
to
so
lve the o
p
timi
zation
p
r
oble
m
in
clude
co
ding
of the
problem
and
d
e
s
ign
of an appro
p
riate fitness function [11
-
14]. In th
is study, the shuffled leapfrog algo
rithm
is
combi
ned
with the
K-me
an
s al
go
rithm. T
he m
a
in
adva
n
tage
of the
prop
osed
sch
e
me i
s
th
at th
e
k-m
ean
s met
hod
can
be
use
d
to imp
r
ove the con
v
ergen
ce
sp
eed, an
d the
global o
p
timal
solutio
n
can
be obtai
ned
by mean
s of
the sh
uffl
ed
frog le
aping
algorith
m
. In pa
rticula
r
,
the
prop
osed scheme combi
nes the gl
ob
al sea
r
chi
ng
cha
r
a
c
teri
stic of the shuffled frog lea
p
i
ng
algorith
m
an
d
the
simpli
city of the
k-m
e
a
n
s
algo
rithm,
and
acco
rdin
gly, an alg
o
rit
h
m with
glo
b
a
l
sea
r
ching
ca
pability is obt
ained [15, 16]
.
In the next section, the
b
a
si
c shuffled
frog le
aping
algorith
m
is investigate
d
. In Sectio
n
3, a type of improve
d
sh
u
ffled frog lea
p
ing alg
o
ri
th
m is intro
d
u
c
ed. In Sectio
n 4, the clu
s
t
e
ring
algorithm based on improv
ed intelligent computing
i
s
proposed. In
Section 5, an experiment is
c
o
nd
uc
te
d
to
te
s
t
th
e p
e
r
f
or
ma
nc
e o
f
the
pr
o
p
o
s
ed
schem
e. In th
e
last
sectio
n,
con
c
lu
sio
n
s a
r
e
provide
d
[17].
2. Basic Shu
ffled Fro
g
Le
aping Algori
t
hm
The fu
nctio
n
optimizatio
n
probl
em i
s
t
r
a
n
sformed
into
the mi
nimum
value
pro
b
le
m of the
objec
tive function
()
f
x
in the feas
ible domain, where
x
is t
he
solutio
n
v
e
ctor.
First,
F
num
ber of
points is sel
e
cted
a
s
the
initial value
from th
e fe
asibl
e
d
o
mai
n
ra
ndo
mly. The i
-
th frog
is
r
e
pr
es
e
n
t
ed
b
y
12
(,
,
,
)
j
jj
j
n
x
xx
x
, and the obje
c
tive function value o
f
each fro
g
is cal
c
ul
ated
.
Each frog is
orde
re
d de
creasi
ngly acco
rding to
it
s o
b
jective fun
c
tion value, an
d then the e
n
t
ire
frog swa
r
m is divided into
S
number of
sub
g
ro
up
s that contain
m
number of fro
g
s.
In the iteration p
r
o
c
e
s
s, the first
soluti
on e
n
ters i
n
to the
first
su
bgro
up, th
e
se
con
d
solutio
n
ente
r
s into th
e second
sub
g
ro
u
p
, and th
e re
st ca
n b
e
co
ndu
cted in
th
e sa
me m
a
n
ner.
Then, the (
s
+1)
-
th
solutio
n
enters into the first
subg
roup ag
ain, a
nd the (
s
+2)-
th
solutio
n
ent
ers
into the se
co
nd su
bg
rou
p
until all the solution
s
are a
ssi
gne
d co
m
p
letely. In each
subg
ro
up,
the
solutio
n
with
the be
st obje
c
tive functio
n
value
and th
at with the worst o
b
je
ctive function valu
e
are l
abel
ed
a
s
12
(,
,
,
)
bb
b
b
n
x
xx
x
and
12
(,
,
,
)
ww
w
w
n
x
xx
x
, res
p
ec
tively. The
s
o
lution
with t
he bes
t
obje
c
tive function in the
swarm i
s
lab
e
le
d as
12
(,
,
,
)
gg
g
g
n
x
xx
x
. In ea
ch
iteration,
w
x
is
update
d
by formula 1.
1
()
j
bw
D
rx
x
(
1
)
'
ww
j
x
xD
.
ma
x
m
a
x
()
j
DD
D
(
2
)
1
(0
,1
)
rU
,
1,
2
,
,
j
S
,
ma
x
D
represent
s the m
a
ximu
m moving
ste
p
length
of the frog. If
the obj
ective
functio
n
val
ue of
'
w
x
is
bet
ter than
that
of
w
x
,
'
w
x
is repl
ace
d
by
w
x
. If th
e
solutio
n
is
no
t improved,
b
x
is re
pla
c
ed
by
g
x
, and form
ul
as 1
and
2 are executed re
peatedly.
If the solution is
still not i
m
proved, a
new
sol
u
ti
on from the feasible
domai
n is generated to
repla
c
e th
e
origin
al
w
x
. The above
ope
ration
s a
r
e p
e
rform
ed
wit
h
in a
spe
c
ifi
ed num
be
r o
f
times, and th
e locatio
n
of the wo
rst frog
in each
grou
p is up
dated.
Acco
rdi
ngly, the first iteration
of the shuffle
d
frog lea
p
in
g algorith
m
is compl
e
ted.
As ob
serve
d
from formul
a
1, the size of the
moving step
dire
ctly affects the gl
ob
al converg
e
n
c
e o
f
the algorith
m
. When the
size is la
rge,
the
frog ca
n co
n
duct glob
al searchin
g, but it may skip
the optimal so
lution. Whe
n
the size is small,
the frog
ca
n
sea
r
ch finel
y in the lo
cal area, but
it can
easily fall into the local
optim
um.
Therefore, m
o
ving the
ste
p
length
affects the
optim
ization
perfo
rmance of th
e sh
uffled frog
leapin
g
algo
ri
thm to a certa
i
n extent. [18, 19]
3. Impro
v
ed
Shuffled Fr
o
g
Leaping Al
gorithm
For e
a
ch sub
g
rou
p
, the
state of su
rrou
nding
frog
s
can affect the
behavio
r of t
he worst
frogs to some
extent, so repulsi
on agai
n
s
t the wo
rst frog occu
rs. Th
e worst frog
moves with t
he
guide
of inf
o
rmatio
n, an
d the frog
s
in
the su
bgroup en
cou
r
a
ge
o
ne anot
her
to
imp
r
o
v
e
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Text Minin
g
Re
sea
r
ch Ba
sed o
n
Intelligent Com
puti
ng in Inform
ation Retri
e
val System
(
Y
ong
L
i
)
1386
perfo
rman
ce
throug
h com
petition and
coo
peration;
in this way, co-evolutio
n
of group
s can be
reali
z
ed.
The improve
d
shuffled fro
g
leaping alg
o
rith
m en
su
res that each
frog bring
s
a certai
n
amount
of charg
e
, which
is d
e
ci
ded
by the opt
im
ized
obje
c
tive fun
c
tion v
a
lue. Th
en,
the
resultant force imposed o
n
the worst frog from the
other fro
g
s in
the subg
rou
p
is cal
c
ulate
d
to
determi
ne the
moving step
length. In the sub
g
ro
up, the
i
q
of frog
i
is ca
lculate
d
by
1
()
(
)
exp
((
)
(
)
)
ig
i
m
kg
k
fx
fx
qn
fx
f
x
(3)
1,
2
,
,
im
,
()
i
f
x
repre
s
e
n
ts the curren
t objec
tive functio
n
valu
e of frog
i
i
n
the
sub
g
ro
up, an
d
()
g
f
x
rep
r
e
s
ent
s the cu
rre
nt o
p
timal obje
c
ti
ve function v
a
lue in the
swarm. The
comp
one
nt force im
po
sed
on the wo
rst f
r
og in the sub
g
rou
p
is calculated by
()
wj
w
j
w
j
F
qq
x
x
(
4
)
After the
re
su
ltant force
w
F
is
cal
c
ulate
d
, th
e force
impo
sed o
n
e
a
ch frog i
s
n
o
rm
ali
z
ed
to ensure the
feasibility of frog shi
ft, and the proposed
update
strategy is
1
()
j
bw
w
Dr
x
x
a
(
5
)
'
ww
j
x
xD
.
ma
x
m
a
x
()
j
DD
D
(
6
)
2
2
/
ww
w
ar
F
F
,
2
(0
,1
)
rU
, and the propo
se
d sch
e
me in
cre
a
ses the dive
rsity of the
sub
g
ro
up
to
prevent
the
o
c
curren
ce
of
prem
atur
e
co
nverge
nce; a
s
a
result, the
entire g
r
ou
p
can
benefit. The p
r
ocess involv
es in the imp
r
oved sh
uffled
frog leapin
g
algorith
m
is a
s
follows:
Step 1. Initial
i
ze th
e swa
r
m and
pa
ram
e
ters,
su
ch
a
s
the total
nu
mber of fro
g
s,
F
; the
numbe
r of
subgroup
s,
S
;
the ite
r
ation
times withi
n
the
sub
g
rou
p
,
I
t
; and
the
hybrid ite
r
ati
o
n
times
,
max
N
.
Step 2. Calcu
l
ate the obje
c
tive function value
()
j
f
x
of each frog.
Step 3. Ord
e
r the
F
nu
mbe
r
of obj
ective
function val
u
es, an
d divid
e
it into
S
nu
mber
of
sub
g
ro
up
s.
Step 4. Dete
rmine th
e in
dividual
b
x
with the be
st o
b
jective fun
c
t
i
on value
an
d the
individual
w
x
with the worst obje
c
tive function value in
each g
r
ou
p. Within the given iteration
times
,
I
t
, updat
e the worst solution a
c
cording to formul
as 5 an
d 6.
Step 5. Fo
r e
a
ch
subg
rou
p
,
arrang
e the
indi
vidual
s in
desce
nding
o
r
de
r a
c
cordin
g to the
obje
c
tive function value to con
s
titute a n
e
w group.
Step 6. Determine whethe
r the terminati
on co
nditi
on i
s
met; if this condition is
satisfied,
the relate
d in
formation of t
he optimal o
b
j
ective f
unctio
n
value is the
output. Othe
rwi
s
e, retu
rn t
o
step 2.
4. Cluste
ring
Algorithm Based on the Im
prov
ed Sh
uffled Fr
og L
eaping Algor
ithm
4.1. Principle of the
K-M
eans Alg
o
rithm
The inp
u
t is
the data set
12
{,
,
,
}
n
Xx
x
x
, and the nu
mber of
clu
s
tering
s i
s
K
. The
output is
K
nu
mb
er
o
f
c
l
us
te
r
i
ng
s
,
j
C
. Th
e process in
volved in the
K-mea
n
s
al
gorithm i
s
a
s
follows
:
Step1.
K
number of ra
nd
om data is ch
ose
n
as the i
n
itial clu
s
terin
g
cente
r
12
,,
,
k
zz
z
from
X
.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 13, No
. 4, Decem
b
e
r
2015 : 138
4 – 1389
1387
Step2. Cal
c
ul
ate the di
sta
n
ce
between
each data
p
o
int
i
x
and
j
C
,
{1
,
2
,
,
}
j
k
. If
the
con
d
ition of
ij
i
m
x
zx
z
,
1,
2
,
,
mK
,
mj
is met, then
ij
x
C
.
Step3. Calcul
ate the new
center poi
nt
**
*
12
,,
,
k
zz
z
with
*
1
ji
ij
XC
i
zX
n
,
1,
2
,
,
iK
(7)
Step4. If
*
ii
zz
,
1,
2
,
,
iK
, the algorithm
stops, an
d the curre
n
t cente
r
point is sele
cted
as the result
of the cluste
ri
ng parti
tion.
Otherwise, re
turn to step 2.
4.2. Cluste
ring Scheme
Bas
e
d on th
e Impro
v
ed Shuffled Fr
o
g
Leaping Al
gorithm
The shuffled
frog leapi
ng
algorithm u
s
es the fitne
s
s value of e
a
ch in
dividua
l in one
popul
ation to
sea
r
ch the
o
p
timal value.
Therefore,
ch
oosi
ng
suitab
le fitness fun
c
tion affe
cts t
h
e
conve
r
ge
nce rate of this al
gorithm. The
fitness fun
c
tio
n
is
2
1
1/
ji
k
ji
ix
C
f
xz
(8)
The clu
s
teri
n
g
algorithm b
a
se
d on the impr
ove
d
sh
uffled frog leaping alg
o
rit
h
m is as
follows
:
Step 1. Set th
e initial para
m
eters of the algorith
m
.
Step 2. Initialize the entire swarm a
nd
K
numbe
r of initial clu
s
terin
g
cente
r
s.
Step 3. Calcu
l
ate the
dista
n
ce
bet
wee
n
X
and
K
num
ber
of cente
r
s co
rrespon
din
g
to the
frog.
X
is cla
s
sified a
c
cordi
ng to distan
ce.
Step 4.
Cal
c
u
l
ate the fitne
s
s valu
e of
ea
ch f
r
og
a
c
cording to
cl
assi
fication, a
nd
arrang
e
the frog
swa
r
m in d
e
sce
n
d
ing o
r
d
e
r
according
to th
e si
ze
of the
fitness value
to gen
erate
j
D
rand
omly.
Step 5. Divide the frog
swarm into
m
number of
sub
g
roup
s. Calcul
ate the be
st solution
b
x
, the wors
t s
o
lution
w
x
, and the global o
p
timal solutio
n
g
x
.
Step 6. Withi
n
the given i
t
eration time
s,
I
t
, update t
he worst sol
u
tion a
c
cordi
ng to
formula
s
5 an
d 6.
Step 7. Each
sub
g
ro
up i
s
combine
d
to fo
rm a n
e
w fro
g
swa
r
m. Arrange th
e fro
g
swarm
in de
scendin
g
ord
e
r a
c
co
rding to it
s fitness val
ue, i
n
crea
se the
global ite
r
atio
n times by o
ne,
and retu
rn to
step 5. Re
pe
at the above pro
c
e
ss u
n
til the maximum
iteration time
is achi
eved.
5. Experiment Re
sults
The Unive
r
sit
y
of California, Irvine (UCI)
data set is commonly u
s
ed in the al
gorithm
perfo
rman
ce
testing of ma
chin
e learnin
g
, info
rmation
processin
g
, and data mini
ng. The data
of
this data set
are st
rictly labeled, so thi
s
data se
t is typically use
d
as the evalua
ting standa
rd
of
several al
gori
t
hms. Th
e Iri
s
a
nd
Wine
d
a
ta sets
of UCI are
sele
ct
ed to te
st the
perfo
rma
n
ce
of
the p
r
opo
se
d
sche
me. Th
e num
be
r of
sub
g
ro
up
s
m
is 3, the
num
b
e
r
of frog
s i
n
the subg
rou
p
n
is 20, the iteration time
I
t
wi
thin the sub
g
roup is 3
0
, an
d t
he global
search time is
1000. Th
e
clu
s
terin
g
pe
rforma
nce of the propo
se
d sch
eme a
nd the traditional k-mea
n
s based on
the
shuffled fro
g
leaping alg
o
rithm is sh
own in T
abl
e 1. This table depi
cts that the average
clu
s
terin
g
a
c
curacy of
the prop
osed sch
e
me
i
s
hig
her than that of t
he k-me
an
s a
l
gorithm
ba
se
d
on the traditio
nal sh
uffled frog leapi
ng al
gorithm.
Table 1. Clu
s
tering pe
rformance co
mp
arison of the two alg
o
rithm
s
Data set
algorithm
right
wrong
accurac
y
Iris
shuffled frog leap
ing algorithm
138
12
92.00%
proposed schem
e
141
9
94.00%
Wine
shuffled frog leap
ing algorithm
146
32
82.02%
proposed schem
e
157
21
88.20%
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Text Minin
g
Re
sea
r
ch Ba
sed o
n
Intelligent Com
puti
ng in Inform
ation Retri
e
val System
(
Y
ong
L
i
)
1388
Table 2. Fitne
ss fun
c
tion va
lue com
p
a
r
ison of the two algorith
m
s
Data set
algorithm
fitness function
i
nner class dista
n
ce
bet
w
een-class distance
Iris
shuffled frog leap
ing algorithm
0.604±0.072
3.493±0.224
0.891±0.037
proposed schem
e
0.409±0.016
3.101±0.109
0.985±0.019
Wine
shuffled frog leap
ing algorithm
1.088±0.015
4.462±0.217
2.967±0.432
proposed schem
e
1.015±0.015
4.197±0.396
3.651±0.306
Figure 1. Con
v
ergen
ce
co
mpari
s
o
n
of two alg
o
rithm
s
on Iri
s
data
s
et
Figure 2. Con
v
ergen
ce
co
mpari
s
o
n
of two alg
o
rithm
s
on Wi
ne dat
aset
The
com
pari
s
on
of the
fitness fu
nctio
n
value
i
s
sh
o
w
n i
n
T
able
2. The
avera
ge fitne
ss
function val
u
e of the p
r
op
ose
d
sch
e
me
is
smalle
r th
an that of the
tradition
al scheme. Th
e in
ner-
cla
s
s di
stan
ce is al
so
sm
aller,
and
th
e bet
we
e
n
-cl
a
ss di
stan
ce
is larger.
T
he
conve
r
ge
nce
comp
ari
s
o
n
o
f
the two al
g
o
rithm
s
in
th
e Iri
s
a
nd
Wi
ne d
a
ta
set
s
is
sho
w
n
in
Figures 1
an
d 2,
respe
c
tively. The
k-m
ean
s algo
rithm b
a
s
ed
on th
e
sh
uffled frog
le
aping
algo
rith
m is
affected
by
the initial value. This scheme can easil
y
fall in
to
the local optimal
solution. Co
mpared with
the
traditional
scheme, th
e p
r
opo
sed
cl
ust
e
ring
alg
o
rith
m ba
se
d o
n
the im
prove
d
shuffled f
r
og
leapin
g
algo
ri
thm has hig
h
e
r co
nvergen
ce spee
d and
higher
clu
s
te
ring a
c
curacy
.
0
20
40
60
80
100
12
0
140
16
0
180
200
0.4
0.5
0.6
0.7
0.8
0.9
1
g
l
o
b
al
i
t
er
ati
o
n
ti
me
s
f
i
t
n
e
s
s
f
unc
t
i
on va
l
u
e
KFL
A
IK
F
L
A
0
20
40
60
80
10
0
120
14
0
16
0
18
0
20
0
1
1.
1
1.
2
1.
3
1.
4
1.
5
1.
6
1.
7
1.
8
g
l
ob
al
i
t
er
ati
o
n t
i
mes
f
i
t
n
e
ss f
u
nc
t
i
on
va
l
u
e
KFL
A
IK
F
L
A
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 13, No
. 4, Decem
b
e
r
2015 : 138
4 – 1389
1389
6. Conclusio
n
To solve the
key p
r
obl
em
of text mining
, we
co
ndu
ct
ed a
study
on
text mining
b
a
se
d o
n
intelligent co
mputing in a
n
informatio
n
retrieval
system. An im
proved
sh
uffled frog le
api
ng
algorith
m
wa
s p
r
op
osed, t
he detail
ed
p
r
ocess
of
this algo
rithm
wa
s p
r
e
s
ente
d
,
and a
novel
K-
mean
s cl
ust
e
ring
metho
d
ba
sed
on
the im
proved shuffled
frog lea
p
ing
algorith
m
wa
s
introdu
ce
d. T
he exp
e
rim
e
nt re
sult
s
sh
ow th
at
the prop
osed scheme ha
s
hi
gher
cl
uste
ri
ng
accuracy
an
d co
nvergen
ce
spe
ed th
an the
k-me
ans
algo
rith
m ba
sed
on
the shuffled
frog
leapin
g
algo
rithm. This study can
se
rve as
a
useful and me
aningful
refe
ren
c
e for te
xt
cla
ssifi
cation
and cl
uste
rin
g
, particul
a
rly
in informatio
n retrieval a
n
d intelligent computing.
Referen
ces
[1]
Samaa
n
M, Rahman F
.
PT
H-200 Ami
no aci
d
profil
es
in pa
tient
w
i
t
h
intest
inal fa
ilur
e
: prel
iminar
y d
a
t
a
on bi
och
e
mica
l
insig
h
ts: Abstract PT
H-200 F
i
gure.
Gut
. 201
5; 64(1): A1-A5
84.
[2]
Roza
nski MR,
Mitchell T
.
Greetings
of Peac
e. Claritas:
Jo
u
r
nal of D
i
al
og
u
e
an
d Cu
lture
. 201
5;
4(2):
5.
[3]
W
ang F
,
Gen
g
C, Su
L. Pa
rameter i
d
e
n
tifica
tio
n
a
nd
pr
edicti
on
of Jil
e
s–Atherton
mo
del for
DC-
bias
ed transf
o
r
m
er usi
ng im
p
r
oved s
huffle
d
frog
le
api
ng al
gorithm and le
ast
squar
e
su
p
port
vector
machi
ne.
IET
Electric Pow
e
r Appl
icatio
ns.
2
015; 9(9): 6
60-
669.
[4]
W
ang T
,
Z
hao
X, Z
h
o
u
Y.
Commu
nit
y
d
e
tection
in
so
cial
net
w
o
rk u
s
ing
shuffle
d
f
r
og-le
ap
in
g
optimis
ation.
In
ternatio
nal J
o
u
r
nal of Secur
i
ty and Netw
orks.
2015; 1
0
(4): 2
22-2
27.
[5]
Lei
D, Guo
X.
A sh
uffled
fro
g
-le
api
ng
al
gor
ithm for
h
y
bri
d
flo
w
sho
p
sc
hed
uli
n
g
w
i
th
tw
o
age
nts.
Expert Systems w
i
th Applicati
ons
. 201
5; 42(
23): 933
3-9
339
.
[6]
Pasha MF
K, Y
easmi
n D, Re
ntch JW
. Dam
-
lake Op
eratio
n to Op
timize Fish
Habitat.
Enviro
nmenta
l
Processes
. 20
15; 2(4): 63
1-6
45.
[7]
Gómez-Gonzá
l
ez M, Ru
iz-Ro
d
rig
uez F
J
, Ju
rado
F
.
Metah
euristic
an
d pr
oba
bil
i
stic tech
niq
ues f
o
r
optima
l
all
o
cat
i
on a
nd siz
e
of biomass d
i
stri
buted g
e
n
e
r
ation i
n
un
bal
ance
d
radi
al s
y
stems.
IET
Ren
e
w
able P
o
w
e
r Generatio
n
. 2015; 9(
6): 653-6
59.
[8]
Geem Z
W
. Multiobj
ective
opti
m
izatio
n of
w
a
t
e
r distri
buti
on
net
w
o
rks
usi
n
g
fuzz
y
t
heor
y
a
nd h
a
rmo
n
y
search.
Wa
te
r.
201
5; 7(7): 361
3-36
25.
[9]
Youa
Z
,
Cao
a
X, W
a
ng
a Y.
An U
n
e
q
u
a
l
Clusteri
n
g
Strateg
y
for
W
S
Ns Bas
ed
Urb
an Inte
lli
ge
nt
T
r
ansportation S
y
stem.
Jour
n
a
l of Informatio
n
and C
o
mp
utation
a
l Scie
nce
. 2015; 12(
10): 400
1-40
12.
[10]
Sedi
ghiz
a
d
eh
M. Optimal R
e
c
onfig
uratio
n
a
nd C
a
p
a
citor
Alloc
a
tion
in
R
adi
al D
i
strib
u
ti
on S
y
stems
Using th
e H
y
br
id Shuffl
ed F
r
o
g
Lea
pi
ng Al
go
rithm in the F
u
zz
y
F
r
ame
w
o
r
k
.
Journal
of Operatio
n an
d
Au
tom
a
ti
on
i
n
Po
we
r En
g
i
ne
eri
n
g
. 20
15; 3(1)
: 56-70.
[11]
Z
hao J, Lv L. T
w
o-P
has
es L
earn
i
ng Sh
uffle
d
F
r
og Lea
pi
ng
Algorithm.
Inte
rnatio
nal Jo
urn
a
l of Hybri
d
Information T
e
chno
logy
. 2
015
; 8(5): 195-20
6
.
[12]
Li Y,
Hu
L. A
fast e
x
act sim
u
lati
on
m
e
tho
d
for a
class
of
Markov
jum
p
process
e
s.
T
h
e Jo
urna
l
of
Che
m
ic
al Phys
ics.
2015; 1
43(
18): 184
10
5.
[13]
W
ang F
,
Gen
g
C, Su
L. Pa
rameter i
d
e
n
tifi
catio
n
a
nd
pr
edicti
on
of Jil
e
s–Atherton
mo
del for
DC-
bias
ed transf
o
r
m
er usi
ng im
p
r
oved s
huffle
d
frog
le
api
ng al
gorithm and le
ast
squar
e
su
p
port
vector
machi
ne.
IET
Electric Pow
e
r Appl
icatio
ns.
2
015; 9(9): 6
60-
669.
[14]
Elbe
ltagi E,
Heg
a
z
y
T
,
Grierson
D. A modifie
d
sh
uffled frog-l
e
a
p
in
g optim
izat
ion a
l
g
o
rithm:
app
licati
ons to
proj
ect manag
ement.
Structure and Infrastru
c
ture Engi
ne
eri
n
g
. 200
7; 3(1): 53-6
0
.
[15]
T
ang Z
.
Improved K-m
e
a
n
s
Cluster
in
g
Algorit
hm b
a
s
ed o
n
Ge
neti
c
Algor
ithm.
T
E
LK
O
M
NI
KA
Indon
esi
an Jou
r
nal of Electric
al Eng
i
ne
eri
n
g
.
2014; 1
2
(3):19
17-1
923.
[16]
Moha
n R
N
VJ, Ra
o KRS. E
fficient K-Me
a
n
s F
u
zzy
Cl
uste
r R
e
li
ab
i
l
i
t
y
o
n
An
gl
e Ori
e
n
t
ed
Face
Reco
gniti
on.
In
ternatio
nal J
o
u
r
nal of Infor
m
at
ics and C
o
mmunic
a
tion T
e
c
h
nol
og
y, 20
12, 2(1):38-
45.
[17]
Niknam
T
,
F
a
rsani E
A
. A h
y
b
r
id se
lf-ad
aptiv
e p
a
rticle s
w
a
r
m optimiz
atio
n
an
d mo
difie
d
shuffled
frog
lea
p
in
g al
gorit
hm for distrib
u
tion fe
eder
reconfi
gurati
o
n
.
Engin
eeri
n
g
Applic
atio
ns
of Artificia
l
Intelli
genc
e
. 20
10; 23(8): 1
340
-134
9.
[18]
Niknam T
,
Nayer
i
p
our M. An efficient mul
t
i-
obj
ective mo
difie
d
shuffle
d
frog leap
in
g alg
o
rithm for
distrib
u
tion f
e
e
der rec
onfig
ura
t
ion pr
obl
em.
Europ
e
a
n
T
r
an
sactions
on E
l
ectrical P
o
w
e
r.
201
1; 21(
1):
721-
739.
[19]
Eusuff MM, Lanse
y
KE. Op
timizatio
n
of
w
a
te
r distri
buti
o
n net
w
o
rk d
e
s
i
gn
usin
g the
shuffled fro
g
lea
p
in
g al
gorith
m
.
Journal of
W
a
ter Resourc
e
s Plan
nin
g
an
d Mana
ge
ment
. 2003; 12
9(3): 210-
225.
Evaluation Warning : The document was created with Spire.PDF for Python.