TELKOM
NIKA Indonesia
n
Journal of
Electrical En
gineering
Vol.12, No.5, May 2014, pp
. 4071 ~ 40
7
8
DOI: http://dx.doi.org/10.11591/telkomni
ka.v12i5.4770
4071
Re
cei
v
ed O
c
t
ober 2
2
, 201
3; Revi
se
d Decem
b
e
r
27, 2013; Accept
ed Ja
nua
ry 1
8
, 2014
The New Algorithms of Weighted Association Rules
Based on Apriori and FP-Growth Methods
Ting Liu
Coll
eg
e of Information Sci
enc
e and T
e
chno
l
o
g
y
, Z
hengz
ho
u Normal U
n
iv
ersit
y
,
Hen
an Z
h
e
ngz
hou, 45
00
44, Chin
a
E-mail: ed
uli
u
ti
ng@
163.com
A
b
st
r
a
ct
In order to improve the freq
uent
ite
m
sets
gen
erate
d
laye
r-w
ise efficienc
y, the paper u
s
es the
Apriori
pro
pert
y
to reduc
e the
search s
pace.
F
P
-grow
algor
ithm for
mini
ng
freque
nt patte
rn steps
mai
n
ly
is
divid
ed
into tw
o steps: F
P
-tree a
nd F
P
-tre
e to constr
uct
a recursiv
e
mi
nin
g
. Algor
ith
m
F
P
-Grow
t
h is to
avoi
d the hi
gh
cost of candi
date it
e
m
sets
gen
eratio
n, few
e
r, more effi
cient scan
n
in
g
.
T
he paper p
u
ts
forw
ard the
ne
w
algorit
hms
of
w
e
ighte
d
ass
o
ciatio
n ru
les
ba
sed
on A
p
rior
i
and
F
P
-Grow
t
h metho
d
s. In th
e
same support, this
method
is
the
most effe
ctive a
n
d
stabl
e
max
i
mu
m fr
equ
ent
ite
m
set
s
mini
ng
ca
pa
city
and
mi
ni
mu
m
executi
on ti
me.
T
h
rough th
eor
etical a
nal
ys
is and ex
peri
m
en
tal simul
a
tion o
f
the performan
c
e
of the alg
o
rith
m is disc
usse
d
,
it is proved t
hat the alg
o
rith
m is feasi
b
l
e
a
nd effective.
Ke
y
w
ords
:
F
P
-grow
,
apriori,
w
e
ighted
asso
ciatio
n rule
Copy
right
©
2014 In
stitu
t
e o
f
Ad
van
ced
En
g
i
n
eerin
g and
Scien
ce. All
rig
h
t
s reser
ve
d
.
1. Introduc
tion
Data mi
ning
asso
ciation
rule mi
ning
is an
impo
rtant re
se
arch topic in t
he field.
Asso
ciatio
n rules
can ge
nerally be di
vided
into
Boolea
n a
s
so
ciation
rul
e
and
qua
ntitative
asso
ciation
rules. Ag
ra
wal
in 1993
mad
e
Boolea
n
a
s
so
ciation
rule
s is
pro
p
o
s
ed
, after cla
ssi
cs
Apriori
and A
p
rio
r
i TID al
g
o
rithm. Multi
valued attri
b
u
t
es are divide
d into catego
ries of attrib
utes
and attribute, many
alg
o
rith
ms
i
n
solving
multi
va
lued
attributes min
i
ng a
s
so
ciatio
n rules,
are th
e
contin
uou
s
n
u
meri
cal
di
scriminatio
n, ge
t the
corr
e
s
p
o
nding
fuzzy d
e
scriptio
n, a
n
d its processi
ng
method is
si
milar to the Boolea
n asso
ci
ation rule
s mi
ning.
As in
the
weighted
a
s
so
ciation
rules ba
sed
on
simultan
eou
s minin
g
p
o
si
tive and
negative associatio
n rule
s,
will pro
d
u
c
e
some co
ntra
diction
s
and
meanin
g
le
ss
rule
s, therefo
r
e,
in the traditio
nal su
ppo
rt, confid
en
ce framework,
introdu
ced in thi
r
d pa
ramete
rs to remove t
h
e
redu
nda
nt ru
les. Ba
sed
o
n
the
correl
a
t
ion
between
positive
and
negat
ive asso
ciation rul
e
s
mining al
gorit
hm ba
sed
on
intere
st and
right: the po
si
tive and ne
ga
tive associ
ation rul
e
s
mini
ng
algorith
m
ba
sed on weight
ed chi
-
squa
re
test; t
he posi
t
ive and nega
tive asso
ciati
on rul
e
s mini
ng
algorith
m
[1]. Whe
n
the d
a
taba
se i
s
very uneve
n
ly distrib
u
ted,
the above fo
r the mini
ng
of
asso
ciation
rules is
not ef
fective. Beca
use
of
the
l
o
w
fre
que
ncy items who
s
e
sup
port
i
s
often
low, a
nd the
r
efore
ra
rely
be exh
u
med.
Aiming at th
is p
r
obl
em, t
he
sup
port
d
egre
e
m
odel
for
different tran
sa
ction
s
with
different min
i
mum su
ppo
rt
threshold, it is more effici
ent use
r
interest
rule
s.
Weig
hted a
s
so
ciation
rule
s with n
egati
v
e supp
ort a
nd co
nfiden
ce of the cal
c
ulated by
introdu
cin
g
n
egative set, suppo
rt de
gre
e
and
conf
id
ence de
gree
cal
c
ulatio
n is facin
g
a
sev
e
re
test
of
its se
arch spa
c
e e
x
ponential growth.
Be
cau
s
e data mi
nin
g
is the
obje
c
t of ma
ss
d
a
ta,
and a
s
soci
ation rule
s a
r
e i
n
focu
s for th
e proje
c
t of it.
Cla
s
s with A
p
riori
algo
rithm
ba
sed
on
FP-Gro
wth
algo
rithm; a cl
ass
based o
n
a
cl
ass of
algorith
m
s. P
r
eviou
s
re
se
arch at hom
e and ab
ro
a
d
have don
e
a lot of work, their typi
cal
algorith
m
s
are pro
p
o
s
ed i
n
MINWAL (O) alg
o
ri
thm
and MINWAL
(W) alg
o
rith
m, the pro
p
o
s
e
d
algorith
m
Ne
w-Ap
rio
r
i 20
0
3
, in ad
dition
to WA
R al
g
o
rithm
and t
he M
W
QA
R
algorith
m
. Th
e
pape
r puts fo
rwa
r
d the ne
w algo
rithms
of weight
ed a
s
soci
ation rul
e
s ba
sed o
n
Apriori a
nd F
P
-
Growth methods
.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 5, May 2014: 4071 – 40
78
4072
2. The Analy
s
is of Weig
h
t
ed Associa
t
ion Rules Te
chnolog
y
With the
rap
i
d develo
p
m
ent of comp
uter
te
chn
o
lo
gy, the so
ci
ety has
ente
r
ed th
e
informatio
n a
ge, esp
e
ci
all
y
the databa
se an
d data
colle
ction te
chnolo
g
y deve
l
opment, vari
ous
indu
strie
s
sto
r
e a lot of d
a
t
a, from the
origin
al
file d
a
ta to the co
mputer
stora
ge. Beca
use
the
databa
se exp
ansi
on cap
a
city,
large
am
ounts of
dat
a
in datab
ase, much
impo
rtant inform
ation
behin
d
, and i
t
is the information is ext
r
acte
d fr
om t
he datab
ase, will create a
lot of potential
value. In the
face
of the
increa
sin
g
e
x
pansi
on of
sea
of data,
the tra
d
ition
a
l data
analy
s
is
method
s hav
e been u
nabl
e to meet the need
s of
the peopl
e, as is
sho
w
n by Fig
u
re1.
Figure 1. Wei
ghted Asso
ci
ation Rul
e
s T
e
ch
nolo
g
y Diagra
m
Since th
e a
ssociatio
n rule
s in d
a
ta mini
n
g
, fi
rst
by Agrawal
was put
forward, pe
o
p
le o
n
the asso
ciati
on rul
e
s i
n
d
a
ta mining te
chn
o
logy
research h
a
s
be
en cond
ucte
d
,
in theory it on a
lot of very fruitful analysi
s
and
re
sea
r
ch, prac
ti
ce
also ma
de
many effective algorith
m
s for
mining a
s
so
ciation rul
e
s f
r
om the
o
ry to appl
i
c
ation
foundation.
Here, ba
sed
on data mi
ning
asso
ciation rules algo
rith
m
introd
uctio
n
.
Associat
io
n rule
minin
g
algorith
m
cl
assic
algo
rith
m is
Apriori
algo
rit
h
m. The
alg
o
rithm i
s
p
r
o
posed by Ag
rawal et al
i
n
199
4, the
main
work i
s
on
asso
ciation rule freq
uent
item set mini
ng, esp
e
ci
all
y
for Boolea
n asso
ciation
rule
s frequ
e
n
t
itemset
s
mini
ng.
It can b
e
a
ttributed to
the followi
ng
two
kind
o
f
thought: o
ne is the
a
ttribute
discrimi
nation
,
it is weighted, the proble
m
is
tran
sformed into a weighted Bool
ean Asso
ciati
o
n
rule
s. The
other m
e
thod i
s
the attri
but
e of dom
ain
is divide
d int
o
overla
ppin
g
interval, th
en
locate
d n
ear t
he b
oun
dary
element
ma
kes it
po
ssible
to s
i
multaneous
l
y in the two intervals
.
A
s
a
result of these eleme
n
ts si
multaneo
usly
to t
he two zone
s betwee
n
cont
ributio
n
s
, there m
a
y be
too much e
m
pha
sis o
n
the
role of these
element
s.
The eig
h
t parameter
persp
ective tran
sfo
r
mati
on m
o
d
e
l althoug
h it can m
o
re a
c
curately
descri
be th
e
came
ra'
s
mo
vement, but it
s
comp
utat
io
nal compl
e
xity greate
r
d
e
g
r
ee.
Con
s
id
ering
the accu
ra
cy with re
al-time
requi
reme
nts, it is the six-pa
ramete
r af
fine model of
came
ra moti
on
cau
s
e
d
by the interface scene chang
e modelin
g [2]. When the scene rel
a
tive depth chang
e is
not sig
n
ifica
n
t, the six-pa
rameter
affine
model
can
well d
e
scri
be
the rotatio
n
of the ca
mera
,
panni
ng and t
r
an
slation
a
l motion.
At present in the asso
ci
ation rul
e
s
a
l
gorit
hm fo
r
freque
nt itemset
s
discov
ery, has
prod
uced va
rious
effective
method. G
e
nerally
spe
a
k
ing, the
s
e
algorith
m
s
are to follow two
step
s: one, t
o
esta
blish t
he fre
que
nt item se
ts
a ca
ndidate
set; t
w
o, in th
e ca
ndidate
set
o
u
t
actually in
clu
des frequ
ent item se
t of all
sub
s
ets. In
asso
ciation
rule metho
d
, Apriori m
e
tho
d
is
most wi
dely used. Th
e Apri
ori metho
d
is
use
d
as
a lay
e
r by layer se
arch iterative
method, and i
t
is usi
ng the p
r
eviou
s
iterati
on of
the freq
uent item sets a
s
after on
e is iteratio
n
of the can
d
id
ate
item sets, an
d then gets th
e final result prun
e.
Whe
r
ein, ea
ch
tran
sa
ction
is
a set of
T
i
s
e
a
ch
tra
n
sa
ction having an
ide
n
tifier TID.
Let
A
be
a set,
T
contai
ns
A when
the asso
ciation rule
s are sha
ped a
s
A
→
B impli
c
ation,
whi
c
h,
A I,
and
only whe
n
the A
T B, I, A
∩
B=
.
Rule A
→
B
in transactio
n
set
D su
ppo
rt s D
tra
n
sacti
on
contai
ns A
→
B percenta
g
e
,
let menu
s b
e
the minim
u
m su
ppo
rt de
gree, if
s
≥
mi
nsu
p
, call
ed f
o
r
freque
nt itemsets. Rule A
→
B in transaction
set D
confid
en
ce le
vel in C is co
ntained in the
D A
transactio
n
contain
s
both the per
ce
ntag
e of B, namely Support (A
→
B
)
=
P
(
A
∪
B), Confide
n
ce
(A
→
B) = P (A|B), as is shown by Equat
ion (1).
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
The Ne
w Alg
o
rithm
s
of Weighted Asso
ciation
Rule
s
Based o
n
Apriori and
… (Ti
ng Liu)
4073
00
0
0
0
,=
,
,
1
,
1
1
(
,
)
1
(
1
,
)
1
(
,
1
)
(
1
,
1
)
fu
v
f
u
v
fu
v
f
u
v
f
uv
f
u
v
fu
v
f
u
v
(1)
Weig
hted fre
quent itemsets do
wn
ward clo
s
ure
ch
ara
c
ter in tra
d
itio
nal asso
ciatio
n rule
s
in tran
sa
ction
,
a lot of, thus creating
eno
rmou
s
sub
s
et
, using
freq
ue
nt set the
do
wn
ward
closu
r
e
cha
r
a
c
ter, to
some
rule
s p
r
unin
g
, for example, in
the
algorithm Ap
riori, if {AB} a
nd {BC} are
not
freque
nt, then {ABC} an
d {
B
CD} a
r
e n
o
t freque
nt.
Improved
wei
ghted a
sso
ci
ation rule
s al
gorithm
an
d Apriori al
gorit
hm the basi
c
steps
like: first find
all weig
hted
sup
port d
egree is not
sma
ller than a u
s
er-sp
e
cifie
d
minimum
sup
port
degree
weig
h
t
ed freq
uent i
t
emset
s
, and
the freq
uent
i
t
emset
s
satisfying the mini
mum weighte
d
confid
en
ce of
all the rules.
A
set
of {AB} affairs wei
ght
cal
c
ulatio
n
f
o
r:
Fo
r (k=2; L
k-1 = do
b
e
g
in;
k++) Ck =Apri
o
ri-
gen (L
k-1); For ea
ch tra
n
s
a
c
tion
s t
∈∈
D For eac
h
trans
ac
tions
t D {Ct
s
ubset (Ck
,
t); For each
c
a
nd
id
a
t
es
C
∈
Ct
{
c.
n
u
m+
+;
c.
weig
ht
=
Weig
ht
(
c
);
/
/
cal
c
ul
ation
of weig
hted sup
port
} Lk ={
c
∈
Ck
| c.
weig
ht ×
c
.num/|D|
≥
minws
u
p
}
}
r
e
tu
rn
L=
∪
K
Lk
; ITW (A, B
)
=0.6
×
0
.9/ (0.6+
0
.9)
=
0
.36.
Weig
hted su
pport data
b
a
s
e contain
s
the proj
ect a
ffairs
cent
rali
zation value summary : WS (X)
=N X
∈∈
Π
( [i[ w]
X] ) Ti [ik
[
w
]] n
Σ
( [i[ w]
X Ti [ik
[w]]
] )K =
1
K =
1
|X |X
|.
Becau
s
e t
he
data the b
o
d
y
itself is not
too mu
ch
sig
n
ifican
ce, it is only a de
scri
ption of
what h
app
en
ed, and
not for de
ci
sion
-m
akin
g provid
e
reliabl
e ba
si
s; through
dat
a analy
s
is to
find
out the relati
onship betwe
en the given
data, some
meanin
g
and
relevan
c
e, it formed the
so-
calle
d info
rm
ation. Althou
gh given
data
in some
mea
n
ingful thin
gs, but it often
and p
eopl
e n
eed
to c
o
mplete the tas
k
without direc
t
c
o
ntac
t, als
o
c
ann
o
t
se
r
v
e as
d
e
c
is
ion
-
ma
k
i
n
g
ba
s
i
s
;
on
ly th
e
informatio
n for furthe
r pro
c
essing, to un
derta
ke
mo
re
thoroug
h an
alysis, in ord
e
r to obtain
more
useful info
rm
ation, namely
kno
w
led
ge, as is
sho
w
n b
y
Equation (2
) [3].
(2)
Ho
wever, the
amount of d
a
ta to the explosiv
e g
r
o
w
t
h
of now
ma
ke
s it difficult for the
use
r
to like
b
e
fore relying
on expe
rien
ce, lar
ge a
m
o
unts of calcul
ation and
hu
man comma
n
d
to
find out abou
t the artificial
data more comprehe
nsiv
e kno
w
le
dge,
the kno
w
led
ge hidd
en in
the
data. Many
can be
found
and
us
ed, the waste
of
resou
r
ce cau
s
ed
by data.
Many de
ci
si
on
make
rs from the databa
se
mining the
s
e
pattern
s ar
e increa
singly in
tereste
d
in, associ
ation rul
e
s
mining
can
p
r
ovide
effecti
v
e de
cisi
on
suppo
rt, wa
s prom
oted on
certai
n
l
e
vel asso
ciation
rules
mining techn
o
logy develo
p
ment.
The p
r
o
b
lem
of minin
g
asso
ciation
rules can
be
formali
z
e
d
as foll
ows: I
}
, is th
e
collection of all items. I
s
all
affairs D M
set (d
ataba
se
), each tran
sa
ction
T i
s
the
numbe
r
of ite
m
s
colle
ction, T i
n
clu
ded in I, each tran
sa
ct
ion ca
n
u
s
e
T I to identify a uniqu
e ide
n
tifier. A D T.
Asso
ciatio
n
rules are
sha
p
ed a
s
B
=!.
Rule a
B B I, B
I, and
A is a
set of thi
n
g
s
,
T contain
s
A
if
and only if A
B implication.
Hori
zo
ntal
weighted
a
s
so
ciation
rul
e
s
mining
met
hod
and
the
improved
weighted
asso
ciation rules minin
g
method com
p
arison re
sults,
whe
r
ein,
sh
ado
w pa
rt is the fre
que
nt item.
The
algo
rithm
of minim
u
m
sup
port
is re
spectively
p
r
ov
ided
with
15
%, 55%, 33%
, to illust
rate t
he
weig
hting fu
nction in mi
ning a
s
soci
ation rule
s [4]. Hori
zontal
weig
hted a
s
sociatio
n rule
s to
gene
rate freq
uent item pro
c
e
ss i
s
first calcul
ated by cou
n
ting su
p
port deg
ree,
namely a set
in
the data
base
the p
r
op
ortio
n
, and th
en
u
s
e th
e item
weig
ht prunin
g
by it. As yo
u can
see
fro
m
Table
1, ho
ri
zontal
wei
ght
ed a
s
soci
atio
n rul
e
s
mini
ng meth
od b
e
ca
use of th
e presen
ce
of
freque
nt item
sub
s
et
s is
not frequ
ent,
so it c
an n
o
t maintain f
r
equ
ent item,
as i
s
sho
w
n
by
Equation (3).
1
1
[(
)
]
[(
)
]
,
i
k
k
kk
i
x
x
I
X
IX
II
LL
LL
(3)
N
M
y
x
F
y
x
R
RMSE
M
i
N
j
j
i
j
i
11
2
)]
,
(
)
,
(
[
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 5, May 2014: 4071 – 40
78
4074
In thing
s
set
D
wa
s e
s
tabli
s
he
d, with
th
e supp
ort of
s
whe
r
ein
s i
s
D thin
gs co
ntains A!,
B (i.e., A and B two) perce
ntage, it is the prob
ability
P (B A!). Rule a B is
the percentag
e of C, it
is the
con
d
itional p
r
ob
abili
ty P (B A |).
Namely: supp
ort (A B)
= P
(A B! ))
= P (A B D |) in thi
n
g
s
with co
nfiden
ce C, if D co
n
t
ains A affairs wh
ile Pack: x [0, 1], called
the membe
r
ship deg
ree.
A
colle
ct
ion
o
f
it
ems
called
it
em s
e
t
s
(I
TE
set
)
.
Cont
ai
ns
k it
em
set
calle
d k
m
se
t
.
A
set
of frequ
en
cy
of occu
rren
ce
is
c
ontai
ning
the item
set t
r
an
sa
ction
nu
mber,
refe
rre
d to a
s
a
set
of
freque
ncy,
a
cou
n
t o
r
cou
n
ting. If the it
em
set i
s
g
r
e
a
ter th
an
or
equal
to the
freque
ncy
of
M
in_sup an
d D in the affairs
of the total nu
mber
of
pro
d
uct, the item
set satisfie
s the sm
all sup
port
M in_su
p
as i
s
sh
own by Equation (4).
s
s
n
n
t
s
n
s
f
t
f
)
2
,
(
)
,
(
~
)
(
(4)
The
u
s
e
of vertical data mining
frequ
ent
pattern a
bove two
me
thods are b
a
s
ed
on
stand
ard d
a
ta format as tran
sa
ction se
t mining freq
uent pattern
s in the algorithm, also can
be
use
d
in a vertical fo
rmat
of data mining [5]. It is the core
of equivalen
c
e tra
n
sfo
r
m
a
tion
algorith
m
. Co
mpared with
the Apfiofi algorithm u
s
e
s
a hori
z
ontal
format data, this metho
d
h
a
s
the follo
wing
advanta
g
e
s
: 1)
Ca
ndid
a
te ge
neration p
r
o
c
e
s
s
doe
s n
o
t p
r
odu
ce th
e d
a
ta
explosi
on, produ
ced o
n
ly a sma
ll nu
mb
er of set
s
of items.
2
Do n
o
t need to scan the datab
ase
to determine
the (K
+ 1
)
it
em sets
(for
k
> = 1
)
of th
e suppo
rt d
e
gree.
The
use of differen
c
e set
techn
o
logy to
furthe
r
redu
ce th
e
storag
e of
lo
ng
TID coll
ectio
n
ov
erhe
ad
and
the inte
rsectio
n
comp
utation.
Asso
ciatio
n rule mining is one of the importa
nt con
t
ent of data
mining field. But the
traditional
a
s
so
ciation
rule
s mi
ning
is o
ften ba
sed
o
n
such
co
ndit
i
ons:
(1
) T
h
e
datab
ase
ea
ch
item of importance a
r
e th
e same; (2
)
The datab
as
e is uniforml
y
distributed:
but the practical
appli
c
ation of
the databa
se is not a
n
id
eal situat
io
n, but users on
variou
s
proje
c
ts for
attention
degree i
s
also different, a
s
in ind
u
st
ry sale
s, so
me
comm
ercial p
r
ofits, so th
e
mall ope
rato
rs to
pay cl
ose att
ention to
it in
it, for such
a
proje
c
t sho
u
l
d
be
given greater wei
ght. The weig
hts are
introdu
ce
d to the mining of asso
ciation rules.
Matrix of wei
ghted a
s
so
ci
ation rul
e
s
mining al
go
rithm this i
s
b
u
ilt on a d
e
scen
din
g
trimming
weights
ba
sed
on 3
2
al
go
ri
thms
de
scrib
ed in
ou
r
F
-
R
- t
me
thod b
a
sed
on
combi
nation
weig
hting m
a
trix of the
rul
e
s
of P
g
wh
o o
model
in
here
n
t cha
r
a
c
teri
stic,
out
of the
diggin
g
matri
x
of weighte
d
asso
ciation
rules
algo
rith
m [6]. Propo
sed alg
o
rithm
basi
c
tho
ught
is:
first stru
cture
F. F digging process u
s
i
ng P
tree in P tree weight
ing desce
ndi
ng find frequ
ent
itemset
s
p
r
un
ing. Th
en
accordin
g to
the
matrix weight
ed
confid
en
ce a
r
e from th
e fre
quent
ite
m
sets in a
s
so
ci
ation rule
s to find exit, as is sho
w
n by Equation (5
).
1
0
1
(,
)
(
,
)
n
i
i
Vk
t
V
k
t
n
(5)
Whe
n
the d
a
t
a is up
dated
quickly ho
w to improve t
he alg
o
rithm,
nume
r
ical variabl
e
pro
c
e
ssi
ng p
r
oble
m
, proje
c
t set in the
weighte
d
ca
se of asso
ci
ation rule mi
ning metho
d
of
ass
o
c
i
ation rules mining f
i
rs
tly [7]. By
Ag.
W Imi
e
linski a
nd S
w
ami p
r
op
ose
d
swe
e
t. Apriori
algorith
m
by A.orawal an
d Sr
ikant propo
sed "!, subsequ
ently
it was
based
on the Apri
ori
algorith
m
ca
rried
out a se
ries of
imp
r
ov
ements,
com
pare
d
to
the
well-kn
own in
clud
e the
u
s
e
of
hash table t
o
improve the efficie
n
cy
of mini
ng
asso
ciation
rules, u
s
e
s
t
he tran
sa
cti
on
comp
re
ssion
technolo
g
y to the scan
n
ed tran
sa
ctio
n set are co
mpacte
d, used the divisi
on
techn
o
logy o
n
tran
sa
ction
set segme
n
tation, usi
ng
sampli
ng te
chnolo
g
y of mining an
d u
s
i
ng
dynamic item
set co
unting
method.
Asso
ciatio
n rule mining
h
a
s ma
ny extensi
o
n
s
, incl
uding m
u
ltilevel associ
ation rul
e
s
mining, asso
ciation rule
mining, mini
ng asso
ci
ati
on rule
s ba
sed o
n
co
n
s
traint
s, peri
odic
asso
ciation
rules minin
g
, t
he mi
ning
of
weig
hted
associatio
n rule
s, there
a
r
e
some
schola
r
s of
these al
gorith
m
s. In additio
n
, associ
atio
n rule mi
ni
ng
technol
ogy n
o
t only can b
e
dire
ctly use
d
as
a
de
cisi
o
n
supp
ort to
ol, ca
n
also
be a
pplie
d t
o
othe
r
data
minin
g
te
ch
nology,
su
ch
as
asso
ciation
rules
mining
tech
nolo
g
y ca
n be
used in
deci
s
io
n tre
e
indu
ction a
n
d
analysi
s
of time
seri
es d
a
ta, classificatio
n
in data mining
technol
ogy.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
The Ne
w Alg
o
rithm
s
of Weighted Asso
ciation
Rule
s
Based o
n
Apriori and
… (Ti
ng Liu)
4075
3. The Res
e
a
r
ch of Aprior
i and FP-Gro
w
t
h
Me
thod
s
This pa
pe
r a
nalyze
s
the F
P
2 tree is ap
plied to the maximum pat
tern minin
g
, can rea
c
h
a high
er
efficiency in
1 th
at the p
r
oble
m
is
m
a
inly becau
se
the
algorith
m
ge
nerate
s
a
la
rge
numbe
r of ca
ndidate m
a
ximum mod
e
, testing
w
heth
e
r they are for maximum
mode n
eed
s
to
spe
nd a lot of time 1 in order to solve t
he pro
b
lem p
u
t forwa
r
d th
e corre
s
p
ondi
ng improvem
ent
measure, the
s
e
metho
d
s i
n
clu
de
FP2 t
r
ee
orde
ring,
re
du
ce
gen
erate
candi
d
a
te Max
scal
e,
redu
ce
the in
spe
c
tion
scop
e 1 we give t
he imp
r
oved
SFP2Max alg
o
rithm, an
d th
e algo
rithm
with
the improve
d
algorith
m
and
the MAF IA
FP
2Max befo
r
e the pe
rformance co
mp
arison [8].
On the Ne
w.
Apriori al
go
rithm time com
p
lexi
ty is too high an
d the numbe
r of ca
ndidate
itemset
s
i
s
to
o la
rge, i
s
ba
sed
on
the
two p
r
uni
ng
alg
o
rithm
WA
RDM, give
s th
e related
the
o
rem
and its pro
o
f. WARDM alg
o
rithm in mini
ng the dat
ab
ase, only nee
ds to scan th
e databa
se o
n
ce
can g
ene
rate
all the weig
h
t
ed frequ
ent itemset
s
, can
redu
ce th
e d
a
taba
se the
numbe
r of visits,
and im
prove
the efficien
cy
of minin
g
. T
h
rou
gh t
w
o ti
mes
of pruni
ng
can
effect
ively redu
ce t
h
e
numbe
r of ca
ndidate item
sets, as
is
sho
w
n by Equati
on (6
).
2
11
1
(,
)
(
,
)
(
,
)
MN
kt
k
t
xy
M
SE
I
I
I
x
y
I
x
y
MN
(6)
FP -gro
w th algorithm for
mining freq
ue
nt patte
rn ste
p
s mainly is
divided into two ste
p
s:
F
P
2
t
r
e
e a
nd F
P
2
t
r
e
e
to
c
o
ns
tr
uc
t a re
c
u
r
s
ive
mini
ng. Con
s
tru
c
tion of FP
- t
r
ee 1:
algo
rith
m
input: transa
c
tion data
b
a
s
e and the mi
nimum supp
ort thre
shol
d
to D m in_ sup. Outp
ut: th
e
freque
nt patt
e
rn t
r
ee
FP
2tree. M
e
tho
d
: mainly
co
nsi
s
ts
of two
step
s: (1) Scan
s tran
sa
ction
databa
se
D, get the frequent 12
se
t F and thei
r su
ppo
rt co
unt. Acco
rdi
ng to F sup
port
desce
nding
sort, get frequ
ent item table L. (2) To
create the FP -tree to
the root node, labe
led
as null " ag
ai
n to scan the
databa
se, in
the D T ran
s ea
ch tra
n
saction exe
c
ut
es the follo
wi
ng
operation: ".
Extraction
of T ra
n s freq
u
ent item a
nd
pre
s
s L
ord
e
r so
rting. To
sort after f
r
eq
u
ent
item table fo
r [P P], where
p is the first
elemen
t, a
n
d
P is th
e rem
nant of the
el
ements of a
list.
Call the in SERT tree
( [P P
], T ).
STL provide
s
a
seri
es of
cont
aine
rs an
d
template
alg
o
rithm,
combi
n
ing th
ese
co
ntainers
and
algo
rith
ms
can
a
c
hie
v
e a vari
ety of utility func
t
i
on. At the p
o
int of
chai
n
stru
cture
will
link
them to have
the same ite
m
_ nam e n
ode. If the
P is not empty, _ recursive i
n
vocatio
n
of in
SERT tree (P, N). Analys
is
: FP -tree c
o
ns
truc
tion
process of tran
sa
ction d
a
taba
se scan
two
times, finally the databa
se
comp
re
ssi
on
storag
e
into a tree, the tree co
ntains f
r
equ
ent patte
rn
mining all th
e informatio
n
.
FP-tree usually fo
r long
mode and i
n
tensive d
a
taba
se with h
i
gh
comp
re
ssion
ratio, but with
a larg
e num
ber of
sho
r
t p
a
ttern data
b
a
s
e
comp
re
ssi
on pe
rform
a
n
c
e
is su
peri
o
r.
Figure 2. The
Apriori an
d F
P
-Growth Me
thods
In the FP-G
ro
w alg
o
rithm, t
he main
data
st
ru
cture in
cl
ude
s Item, Xiang T
oubia
o
Hea
der
and FP
2 tree
. Item head
er table
and
FP-tree
elem
ent
, a sto
r
ag
e
structu
r
e
by d
e
f
ining a
cla
s
s
to
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 5, May 2014: 4071 – 40
78
4076
encap
sulate t
he related
Item data
mem
bers, in
cludi
n
g
a item
nam
e _, n
ode
ch
ain no
de_
lin
k,
sup
po rt_ co
un t suppo
rt cou
n
t, the parent nod
e ch
ain parent t_ link and CH ILD child no
de
chai
n _ lin
k, o
ne child n
ode
chai
n for
set
contai
ner
poi
nter types. Ite
m head
er ta
bl
e is o
ne by th
e
sup
port
cou
n
t
in descen
d
i
ng order li
ne
ar table, d
a
ta
membe
r
s i
n
clud
e item na
me, calle
d th
e
_
node
chai
n n
ode_ lin
k, as i
s
sh
own by Figure 2.
Therefore
th
e existen
c
e
of Aprio
r
i alg
o
rith
m i
s
with the
same
defect. Aimin
g
at the
defect
s
of
th
e Apri
ori
alg
o
rithm, the
p
r
opo
se
d a
ki
nd of
not
ge
neratin
g
can
d
idate ite
m
set
algorith
m
of FP-Growth, to avoi
d the h
i
gh co
st of candid
a
te item
sets g
ene
rati
on, fewe
r, more
efficient scan
ning. The
r
e is a numbe
r of FP-Growt
h al
gorithm
s ba
sed multi level asso
ciation rule
mining al
gorit
hm, the algo
rithm in different deg
r
ee a
nd improves
the mining ef
ficien
cy [9].
But
most of th
ese alg
o
rithm
s
can
only a
c
h
i
eve a laye
r
within th
e mi
ning, u
nable
to find differe
nt
con
c
e
p
tual le
vel asso
ciatio
ns am
ong th
e data it
em
s;
and some al
gorithm ca
n
realize
the cro
ss
level mining, but it can not reflect the con
c
ept
hi
era
r
chy of intrinsi
c co
nstraints rel
a
tion
s,
resulting in th
e FP heade
r betwe
en com
p
lex, can
effe
ctively simplify the proce
ss of mining.
Weig
hted a
s
so
ciation
rule
s propo
se
d b
y
Apriori al
go
rithm in mini
ng freq
uent it
emset
s
whe
n
there a
r
e actu
ally two
big hypoth
e
sis of the
ol
d t: (1)
Of ea
ch it
em in the
dat
aba
se
with th
e
same
natu
r
e
and fun
c
tion,
i.e. the sa
me.
(2
) Th
e im
p
o
r
tance of
ea
ch item in th
e
databa
se
of the
distrib
u
tion is uniform that is the freque
ncy of oc
currence of the same or simil
a
r [10]. Howev
e
r,
in reality worl
d databa
se o
f
more than two hy
pothe
ses are not establis
h
ed. Wh
en the datab
ase
is in proj
ect
of uneven di
stributio
n of frequ
en
cy
differen
c
e i
s
larger, will cau
s
e the minimum
sup
port
deg
ree is hig
h
wit
h
low have
problem
s in
t
w
o chess
su
rfa
c
e, if
set too
high, the
mini
ng
asso
ciation
rules m
a
y not
involve sho
w
ed a lo
we
r
fre
quen
cy of pro
j
ect, as i
s
sho
w
n by Equ
a
tion
(7).
)
1
/(
]
)
4
/
1
(
)
4
/
2
[(
2
2
)
(
1
1
2
/
1
2
/
1
2
1
j
j
j
X
c
df
cf
v
j
j
(7)
In ord
e
r to
overcome th
e
shortcomin
gs
of Ap
rio
r
i alg
o
rithm,
some
schola
r
s prop
ose
d
th
e
mining of wei
ghted a
s
soci
ation rule
s al
gorithm, for e
a
ch item were weig
hted, e
ffective intere
st to
solve the p
r
oject in the
databa
se
with different
importan
c
e
to discu
ss and analy
s
is of
rep
r
e
s
entativ
e weig
hted a
s
soci
ation rul
e
s minin
g
alg
o
rithm nam
ed
New-Apri
ofi algorith
m
.
Weig
hted su
pport mea
s
u
r
e is defined,
combi
ned
with this algorith
m
, an improv
ed Ne
w-
Apriori al
go
rithm is put forward, in whi
c
h the
item set
X weighted suppo
rt mea
s
u
r
e is defin
ed
as:
swoop
( X
) =max{hi, hE, h
k
}X ',
su
p
( X
), where hi
is
ij wei
ghts
[11]
. Ne
w, Apri
ori
algo
rithm
an
d
OD alg
o
rithm
of Ap. Same idea, reali
z
e
d
by tw
o step
s : first find all weighte
d
su
pport d
egree
is
no less than
a user
spe
c
ified minimu
m weighte
d
sup
port wmi
n
su
p con
s
tra
i
nt all weight
ed
freque
nt itemsets, de
noted
as L, then use a wei
ghte
d
freque
nt itemset
s
satisfyi
ng the minim
u
m
weighted reli
ability constraint of wminco
nf all weighted association
rules.
In the minin
g
of weighte
d
asso
ciatio
n
ru
le
s al
gorit
hm New.Ap
ri
ori al
gorithm
is the
cla
ssi
c
algo
ri
t
h
m;
it
ca
n
ef
f
e
ct
iv
ely
b
e
w
e
ighte
d
asso
ciation
rules minin
g
. Based
on
the
Ne
w.Apfiofi algorithm to p
r
odu
ce a la
rge
numbe
r of ca
ndidate
set of
options
and repeate
d
scan
s
the origi
nal
databa
se
de
fects a
r
e im
proved,
thi
s
pape
r p
r
op
ose
s
a
n
im
proved
algo
rithm
WARDM alg
o
r
ithm. Finally throug
h the si
mulati
on exp
e
rime
nts vali
date the WA
RDM al
go
rith
m.
4.
Ne
w
Alg
o
rithms
of Weigh
t
ed Associa
t
ion
Rules
Bas
e
d
on Aprior
i and FP-Gr
o
w
t
h
Metho
d
s
Mining frequ
ent itemsets
is the
cla
s
sical algo
rithm
Apriori
algo
rit
h
m an
d FP - gro
w
th
algorith
m
. M
o
st oth
e
r alg
o
rithm
s
a
r
e
the two
al
gorithm varia
n
ts. Cla
ssi
c
alg
o
rithm
ha
s t
w
o
default a
s
su
mptions: th
e
databa
se
of
each item
i
n
the same
dat
aba
se
and th
e impo
rtan
ce
of
each item in
the distrib
u
tion is unifo
rm
, the frequen
cy of occurre
n
ce of the
same or
simil
a
r.
Therefore, cl
assical algo
ri
thm is equal
to the
consi
s
tent way of dealin
g with databa
se in t
h
e
databa
se
proj
ect. Ho
weve
r the actu
al sit
uation i
s
not
the ca
se, the
importa
nce
of not the sa
me
and un
even
distrib
u
tion. In orde
r to so
lve the
first proble
m
, re
se
arche
r
s p
u
t forward wei
g
h
t
ed
asso
ciation rules al
gorith
m
.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
The Ne
w Alg
o
rithm
s
of Weighted Asso
ciation
Rule
s
Based o
n
Apriori and
… (Ti
ng Liu)
4077
A priori
algo
ri
thm there
are
two maj
o
r p
r
em
ise
hypoth
e
se
s: 1
)
Data
base items th
e sa
me
importa
nce; 2
)
Databa
se
of
ea
ch item
in
the di
st
ributio
n is unifo
rm, t
he fre
que
ncy
of the
same
o
r
simila
r. Howe
ver, in the real worl
d data
base is
often
not so [12].
Whe
n
the da
tabase is in the
proje
c
t i
s
n
o
t
uniform
in
distribution, freq
uen
cy differe
nce
is large
r
,
will le
ad to
su
pport
deg
re
e
of
rea
s
on
able
setting, if high, discovere
d
asso
ci
ation
rules mi
ght
not involve sho
w
e
d
a lo
wer
freque
ncy of proje
c
t, as is
sho
w
n by Eq
uation (8
).
)
,
0
(
2
))
0
(
(
)]
,
0
(
)
0
(
)
,
0
(
[
2
1
)
,
1
(
2
k
x
A
I
k
x
A
k
x
x
k
(8)
The cu
rre
nt data
mini
ng resea
r
ch
focus. With
wei
ghting
a
s
so
ci
ation
rule
s mining
i
s
calle
d weig
ht
ed asso
ciatio
n rule
s minin
g
. Since
199
8, weighte
d
asso
ciation rule mining h
a
s
received exte
nsive attentio
n and stu
d
y, in the mi
ning of weighte
d
asso
ciation
rules al
go
rithm
resea
r
ch at home and a
b
road, a lot of result
s.
This p
ape
r a
nalyze
s
the
without ge
nerati
ng ca
ndid
a
t
e set dire
ctly to generate freque
nt
pattern
alg
o
rithm for FP
2Gro
w th, its b
a
si
c idea
is to the
e
n
tire
datab
ase comp
re
ssi
o
n
rep
r
e
s
ente
d
as tree
stru
cture of FP
2
t
ree, fr
e
quen
t pattern mi
n
i
ng p
r
o
c
e
ss i
n
to re
cu
rsively
gene
rating
co
ndition
s and
con
d
ition
s
of sub Li
brary
o
f
FP 2tree, P Rio RI pe
rfo
r
man
c
e
relati
ve
to A seri
es al
gorithm
ha
s a
larger imp
r
o
v
e. Be
ca
use
FP 2G
ro
w th
algorith
m
d
a
ta st
ru
cture
u
s
ed
to achi
eve m
o
re
com
p
lex, has certain
difficult
y, and
obje
c
t-o
r
ient
ed p
r
og
ramm
ing lan
gua
ge
C+
+ the
STL
st
anda
rd te
mpl
a
te library, wherei
n the
d
a
ta structu
r
e
and
algo
rith
m for
pro
g
ra
m
provide
s
po
werful
tool
s. The FP
2
G
ro
w th al
g
o
rithm d
e
scri
p
tion an
d a
nalysi
s
, and
then
discu
s
ses its
reali
z
ation m
e
thod.
In order to fu
lly refle
c
t the
wei
ghted
a
s
so
ciation
rule
s ta
ke
s
adva
n
tage
of a
n
efficient
algorith
m
. Th
e wei
ghted
p
r
ocess
ca
n n
o
t only di
stin
guish the im
portant
deg
re
e of the mi
ni
ng
proje
c
t an
d the re
sult
s be
come m
o
re reasona
ble,
a
nd in some
case
s al
so
ca
n gre
a
tly improve
the efficien
cy of the algorith
m
, redu
ce mi
ning we
ighte
d
asso
ciation
rule
s re
qui
red
time. Becau
s
e
in the a
s
so
ci
ation rule
s correlation
alg
o
rithm
to ge
nerate
freq
u
ent
item sets,
stage will cost
algorith
m
’s ru
nning time.
The m
a
in
wo
rks
are
a
s
fo
llows: the
Web d
a
ta mi
ni
ng a
s
so
ciatio
n rule
s a
nd
weig
hted
asso
ciation
rule theo
ry is
studie
d
. For
a variet
y of weig
hted a
s
sociatio
n rul
e
s algorithm fo
r in-
depth
re
sea
r
ch, fo
cu
sing
on
analy
s
is of the
prop
ose
d
Apfiofi
algorith
m
for the
Ne
w.Ap
fiofi
algorith
m
time com
p
lexity is too hig
h
an
d the
numb
e
r of candi
date
itemset
s
exce
ssive
sho
r
tag
e
,
is
b
a
sed on
the
two pru
n
i
ng
al
go
rithm WARDM,
giv
e
s th
e
relate
d theo
rem
an
d its proof, u
s
in
g
synthetic d
a
ta sets TIO,
wIOOK an
d T
4011
0D1
00K experim
ent, and the analysi
s
of
experim
ental
result
s. Experim
ental a
nalysi
s
of environm
ent: CPU Intel PE, 1GB memory,
Wind
ows X
P
operating
system
; al
gorithm
usi
n
g Java l
ang
uage, in E
c
lipse
deb
ug
ging
developm
ent platform.
Figure 3. The
Weighte
d
Asso
ciation
Rul
e
s Based on
Apriori a
nd F
P
-Growth Me
thods
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 5, May 2014: 4071 – 40
78
4078
The pa
pe
r p
u
ts forwa
r
d t
he ne
w al
gorithm
s of
wei
ghted a
s
so
ci
ation rul
e
s
b
a
se
d on
Apriori
a
nd
FP-Growth
method
s. Alg
o
rithm i
n
th
e
ca
ndid
a
te it
em
sets Ck,
Lk.1
p
r
uni
n
g
, to
reduce to
connect frequency (k.1
). Thi
s
reduce
s the com
b
inatorial possi
bilities, whi
c
h
reduce
circulatin
g ju
dgment time
s, thereby re
duci
ng th
e
conne
cting
op
eration,
and
to red
u
ce th
e Ck
option set qu
antity, improve the efficien
cy of t
he alg
o
rithm. If the large
datab
ase
of the time
spe
ndin
g
on
data mining e
fficiency is ve
ry obvious.
The expe
rime
nts sh
ow that
the algorithm
c
an effective
l
y avoid singl
e sup
port
s
mi
nimum
weig
hted su
p
port set too l
o
w num
ber o
f
rules is too
large a
nd too
high can n
o
t effectively th
e
probl
em of
mining a
s
so
ciation rule
s.
The expe
rim
ental re
sult
s sho
w
that, unde
r the sa
me
minimum wei
ghted su
ppo
rt
thre
shold
condition
s,
the
WA
RDM
alg
o
rithm
of wei
ghted frequ
e
n
t
itemset
s
num
ber le
ss than
New.Apfiofi algorit
h
m
; WARDM alg
o
rit
h
m to generate a weight
ed
freque
nt itemsets u
s
in
g time is less th
an Ne
w.Apfio
f
i algorithm: WARDM alg
o
rithm ha
s g
ood
stability and
expan
sibility, in larg
e am
ou
nts of d
a
ta
mi
ning i
s
mo
re
high effici
en
cy. Therefo
r
e t
he
WARDM alg
o
r
ithm ha
s better pe
rform
a
n
c
e than
Ne
w.Apfiofi algorit
hm.
5. Conclusio
n
The a
s
sociati
on rule
s in d
a
ta mining te
chn
o
logy e
s
p
e
cially the weighted a
s
so
ciation
rule
s of the
system, com
p
reh
e
n
s
ive, detailed
a
nal
ysis an
d re
search, an
d h
a
s ma
de
certain
achi
evement
s. The
pap
er
puts fo
rward
the ne
w al
g
o
rithms of
weig
hted a
s
soci
ation rules ba
sed
on Ap
riori
an
d FP-G
ro
wth
method
s. In
orde
r to
i
m
prove the frequ
ent itemsets
gene
rated
lay
e
r-
wise efficie
n
cy, uses the
Apriori
prope
rty to re
d
u
ce
the search
spa
c
e. At th
e sa
me time,
the
algorith
m
kee
p
s th
e A P
R
I
O
RI
algo
rith
m of ex
celle
n
t
pro
pertie
s
, h
a
s
better time
co
mplexity a
nd
spa
c
e
com
p
l
e
xity. The experim
ental re
sults
sh
ow
th
at the perfo
rmance of H a
l
gorithm, FP-gro
w
faster than
Apriori
alg
o
ri
thm by a
n
orde
r
of ma
gnitude,
alth
ough
freq
ue
nt pattern m
i
ning
algorith
m
for some n
on de
nse d
a
taba
se
is very effective.
Referen
ces
[1]
Moham
ad F
a
r
han M
o
h
a
mad
Mohsi
n
, Moh
d
He
lm
y
Ab
d
W
ahab, M
ohd
F
a
iruz Z
a
i
y
ad
i, Cik F
a
zi
l
a
h
Hiba
d
u
lla
h. An
Investig
ation
i
n
to Influ
ence
Factor
of Stud
e
n
t Programm
i
n
g
Grade
Usi
n
g
Associati
o
n
Rule Mi
ni
ng.
AISS
. 2010; 2(2):
19 ~
27.
[2]
Guo Hon
g
ji
an.
A Novel F
r
eq
uent Item-set Minin
g
Metho
d
based
on F
r
e
que
nt-pattern
List.
JDCTA
.
201
1; 5(11): 18
2 ~
188.
[3]
Xia
o
ke
W
u
,
Xi
ongfe
i
Li. Mi
ni
ng M
a
xima
l F
r
equ
ent S
ubtre
es b
a
se
d o
n
F
u
sio
n
C
o
mpr
e
s
s
ion
an
d F
P
-
tree.
JDCTA
.
201
1; 5(7): 191
~
200.
[4]
Li Guod
on
g, Xia Ke
w
e
n. An Improved D
a
ta
Mining T
e
chni
que C
o
mbi
n
e
d
Apriori Alg
o
rit
h
m
w
i
t
h
Ant
Colo
n
y
Alg
o
rith
m and its Appl
i
c
ation.
JDCTA
.
2011; 5(8): 2
4
1
~
249
[5]
T
H
AIR A SAL
IH. Performan
c
e Anal
ys
is of
Chaot
ic C
h
ir
p Sprea
d
Spe
c
trum S
y
stem
in Multip
ath
Enviro
nment.
Journ
a
l of T
heor
etical a
nd Ap
pli
ed Infor
m
atio
n
T
e
chno
logy
. 2
010; 18(
2): 101
-108.
[6]
Usman B
aba
wuro, Zou Be
iji.
Kno
w
l
e
d
ge R
epres
entati
on:
A Genera
l
S
u
rve
y
a
nd T
e
chni
ques for
Soun
d Kno
w
l
e
dge Bas
ed S
y
s
t
ems.
IJI
I
P
. 2011; 2(4): 16 ~
22.
[7]
Mingc
han
g Li
u, He
w
a
ng L
i
u.
Rese
arch o
n
Applic
atio
n of Associati
on R
u
le
Minin
g
in C
h
in
ese Athl
ete
s
’
Nutritio
nal a
nd
Bioch
e
mica
l Inde
xes Mo
nitori
ng.
JDCTA
. 20
12; 6(7): 17
4 ~
180.
[8]
Cha
ngj
ian
g
Li,
Xi
anfe
n
g
Yan
g
.
T
he Researc
h
of
Recomm
e
ndati
on S
y
ste
m
s
in E-Comm
erce
B
a
sed
o
n
Associati
on Ru
le Improve
d
-Ap
r
iori Alg
o
rithms
.
IJACT
. 2012; 4(21): 35
4 ~
36
1.
[9]
YiJie C
h
e
n
. T
he Dev
e
lo
pme
n
t
of the Comm
odit
y
F
l
o
w
An
a
l
y
s
is S
y
stem B
a
sed On Ass
o
ciatio
n Ru
le
Mining.
IJACT
. 2012; 4(1
3
): 4
30 ~
436.
[10]
CH Ca
i, W
C
F
u
Ada, CH
Ch
eng,
W
W
K
w
o
ng. Mini
ng
ass
o
ciati
on ru
les
w
i
t
h
w
e
ig
hted i
t
ems. Proc. of
the Int’l
Data
ba
se Engi
ne
erin
g
and App
lic
atio
ns Symp
osi
u
m
. 1998; 68 ~
77.
[11]
Char
u C Agar
w
a
l, Phil
ip S Y
u
.
A New
F
r
amew
ork for Itemset Gen
e
rati
on
. Procee
di
n
g
s of
the 17th
ACM SIGACT
-SIGMOD-SIGART
Sympos
iu
m on Pri
n
ci
ple
s
of Database
Systems (POD
S
’
98)
Seattle,
W
a
shin
gton. ACM Press Publ
isher. 19
98; 18
~
24.
[12]
Song YQ, Z
hu YQ, Sun Z
H
,
Che
n
G. An algorit
hm a
nd its
updati
ng a
l
gor
ithm
base
d
on
F
P
-T
ree for
minin
g
ma
xim
u
m frequent ite
m
sets.
Journal of
Softw
are
. 2003; 14(9): 1
586
-159
2.
Evaluation Warning : The document was created with Spire.PDF for Python.