TELKOM
NIKA Indonesia
n
Journal of
Electrical En
gineering
Vol.12, No.6, Jun
e
201
4, pp. 4685 ~ 4
6
9
1
DOI: 10.115
9
1
/telkomni
ka.
v
12i6.545
2
4685
Re
cei
v
ed
De
cem
ber 2
9
, 2013; Re
vi
sed
March 8, 201
4; Acce
pted
March 20, 20
14
Resear
ch of the Defect Model Based on Similarity an
d
Association Rule
Wanjiang Ha
n*
1
, Lixin Jiang
2
, Xiao
y
a
n
Zhang
3
, Tianbo Lu
4
, Sun Yi
5
, Li
Y
a
n
6
,
Weijian
Li
7
1,3,
4,5,
6
School Of Soft
w
a
re En
gi
neer
ing, Be
iji
n
g
Univ
ersit
y
of Posts and T
e
le
communic
a
tio
n
Beiji
ng, Chi
na
2
Departme
n
t of Emergenc
y R
e
spo
n
se, Ch
in
a Ea
rthqu
ake
Net
w
orks C
ent
er, Beiji
ng, Chi
n
a
7
Internation
a
l S
c
hoo
l, Beiji
ng
Univers
i
t
y
of Post and T
e
leco
mmunicati
on
Beiji
ng, Ch
in
a
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: han
w
a
nj
ia
ng
@bu
p
t.edu.cn
*
1
, jlx
@seis.ac.cn
2
, xiao
ya
n@
bupt.ed
u
.cn
3
,
lutb@ bupt.edu.cn
4
, suny
isse@ bupt.edu.cn
5
, 542289
70
0
@
qq.com
6
,
2
0
112
12
922
@ b
upt.edu.c
n
7
A
b
st
r
a
ct
In ord
e
r to
det
ect defects
effi
ciently
an
d
improve
the
qu
ali
t
y of pro
ducts,
this p
aper
p
u
ts forw
ar
d
the co
nce
p
t a
b
out d
e
fect cl
as
sificatio
n
mo
de
l a
nd
def
ect
a
ssociati
on
mod
e
l
by a
l
o
t of
defect d
a
ta. T
h
e
techno
lo
gy of
similar
i
ty is a
p
p
lie
d to
defect
classifi
cati
on
mo
de
l, an
d the
ide
a
of K
now
l
edg
e Disc
o
ver
y
i
n
Datab
a
se is a
ppli
ed to d
e
fe
ct associatio
n
mo
del.
D
e
fec
t
classificatio
n
mo
del c
an a
naly
z
e th
e d
e
fect
efficiently a
nd
provi
des the b
a
sis of solvi
ng
prob
le
ms qu
ick
l
y w
h
ile defect
associ
ation
mo
del ca
n be us
e
d
to detect ear
ly and
preve
n
t pr
obl
e
m
, w
h
ich can
make
e
ffective i
m
pr
ove
m
e
n
ts to testing a
nd d
e
vel
o
p
m
e
n
t
.
T
h
is pa
per
su
mme
d
u
p
GUI
defect
mo
de
l
base
d
o
n
a
lar
ge
nu
mb
er of
interface
defec
ts. T
he mod
e
l
i
s
useful to
i
m
pr
o
v
e the
accurac
y
of forecast a
nd b
e
us
ed for
test pla
nni
ng
and
i
m
pl
e
m
e
n
tation thr
o
u
gh t
h
e
practice of sev
e
ral pr
ojects.
Ke
y
w
ords
: def
ect mo
del, ass
o
ciati
on rul
e
, d
e
fect cla
ssificat
i
on, defect ass
o
ciati
on,si
mi
lar
i
ty
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
With the
rap
i
d develo
p
m
ent of inform
ati
on technol
ogy, the app
lication
of co
mpute
r
softwa
r
e i
s
more
and m
o
re
widely.
A variety of information
prod
uct
s
e
m
erg
ed an
d
the
requi
rem
ents of pro
d
u
c
t q
uality are al
so increa
sin
g
.
By
Analyzing and studying defect
s
,
build
ing
defect cl
assifi
cation mo
del
and defect
asso
ciation
model can h
e
lp to find so
lutions for
si
mila
r
defect
s
q
u
ickly and effici
e
n
tly, which
provide
s
a g
ood
exploration to im
prov
e the q
uality of
prod
uct. Thi
s
pape
r d
e
termine the
defe
c
t cla
s
sifi
cati
on mo
del b
a
sed on
simil
a
ri
ty techniqu
es by
a large n
u
mb
er of d
e
fect d
a
ta from te
sti
ng. Me
anwhile, it researches de
fect associatio
n
mod
e
l
according to
the asso
ciati
on rul
e
theo
ry. Thr
oug
h the defe
c
t cl
a
ssifi
cation m
odel a
nd d
e
fect
asso
ciation
model, you
ca
n
cla
ssif
y
and
solve
pro
b
lem
o
n
time. It provide
a
g
ood
recomme
ndat
ion for pro
d
u
c
t developm
e
n
t and desi
g
n.
Durin
g
the testing pro
c
ess, there
r
are a
large
of defe
c
t data, it is
neces
sa
ry to merg
e the
same o
r
si
milar defe
c
t to t
he same typ
e
of
defect fo
r u
n
ify solution
ea
sily. And mo
re defe
c
ts
c
an be found
by
a spec
ial defec
t. So, defec
t
s
can b
e
effecti
v
e manage
d by defect cla
s
sificatio
n
mod
e
l and defe
c
t asso
ciation m
odel [1].
By similar de
fect reco
gniti
on, re
peate
d
defec
t
s
o
r
ve
ry simil
a
r
def
ects can
be
removed
from defe
c
t li
bra
r
y. Similar pro
b
lem
s
recognition
nee
d
to use the
kn
owle
dge
of n
a
tural l
angu
a
g
e
manag
eme
n
t to identify the simil
a
ri
ty of sent
en
ce
s, so
a
s
to achi
eve t
he pu
rp
ose
of
cla
ssifi
cation.
Defect a
s
sociation analy
s
is can in
dicate that an em
erge
nce of de
fect may lead
to
one o
r
the
other
defe
c
ts t
o
app
ear.
D
ef
ect cl
assi
fication mod
e
l an
d defe
c
t asso
ciate m
odel
ca
n
improve the q
uality of prod
ucts a
nd give
a better way
to dis
c
o
very is
sues
[2].
2. Related Similarit
y
Method
For th
e rese
arch o
n
d
e
fe
ct cl
assification mo
del, d
e
f
ect de
scripti
on la
ngua
ge
need
s to
be an
alyze
d
to determi
ne
simila
r de
scri
ptions
of
the defec
t, then c
l
as
s
i
fy defec
t
and form
an
effective defe
c
t library. Th
e re
se
arch a
bout def
e
c
t
simila
rity except re
co
gnize
gene
ral
wo
rds
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 6, June 20
14: 4685 – 4
691
4686
and exp
r
e
s
si
ons, involve
s
lexical
analy
s
is in p
r
of
e
s
sional field. T
h
erefo
r
e, we n
eed to i
m
pro
v
e
algorith
m
on the ba
sis of common
word
s simila
rity.
Identification
of gen
eral
word
s n
eed
to
cal
c
ul
ate p
r
i
m
ary
simila
rity of a
sente
n
ce. F
o
r
sente
n
ce simi
larity comp
uting, the gene
ral step
s are a
s
follows:
(1)
Segmentatio
n manag
eme
n
t to stateme
n
t
(2)
Phra
se man
a
gement to sta
t
ement, cal
c
u
l
ate simila
rity about semant
ic
(3)
Import synta
c
tic rule, analy
s
is
sente
n
ce stru
cture
(4) Cal
c
ulate
sim
ilarity
about
the sem
antic
of statement
(5)
For
a fiel
d
compo
s
ed
by
a compl
e
x senten
ce,
cal
c
ulating
co
mpl
e
x se
nten
ce
s or
sente
n
ce gro
up sem
antic
simila
rity
For the
seg
m
entation m
anag
ement, t
here
are
ma
ny sop
h
isti
ca
te system
s
such
as
segm
entation
syste
m
of i
n
formatio
n re
trieval l
ab.
F
o
r the
ma
na
gement
of th
e ph
ra
se
sh
o
u
ld
focu
s on th
e mana
gem
ent of stru
ct
ural a
nalys
i
s
. Then calculate simil
a
ri
ty of the phrase
sema
nt
ic.
Gene
ral state
m
ents su
ch
as gene
ral Chine
s
e
a
r
e
a
of
inform
ation
man
agem
e
n
t,
there
are m
any m
e
thod
s of co
mpari
ng text simila
rity. The ide
a
s
wo
rth lea
r
nin
g
. Its analysi
s
and
introducing word recognition
in
defect
s professional
area
will ma
ke us to find a
suitabl
e defect
simila
rity co
mpari
s
o
n
me
thod. So that
help to
cla
s
sify the p
r
obl
em.The follo
wing
s a
r
e
so
me
s
i
milarity mothods
.
Boolean
mod
e
l is
used to
cal
c
ulate th
e
simila
rity of statement with
fast
spe
ed a
nd hig
h
efficiency. But its calculated result is only tw
o value
s
, either the
same
or diffe
rent. For
so
me
sema
ntically
simila
r wo
rd
s, Boolean mo
del may give wro
ng jud
g
m
ent [3].
TF*IDF used in
the
field of informatio
n
retrie
val, the
method i
s
a
statistical m
e
thod, only
the num
be
r o
f
wo
rds in th
e senten
ce
that contai
n
s
a lot of
relate
wo
rd
s m
a
y
be repe
ated,
the
effects of thi
s
stati
s
tical
method
can
be refle
c
ted.
TF*IDF met
hod
s only co
nsid
er the
word
s
statistical pro
pertie
s
in the
context, without co
nsi
d
e
r
ing the sem
antic informa
t
ion of the word
its
e
lf. It als
o
has
some limit
ations
[4].
Vector
spa
c
e
model is exp
r
essed by seq
uen
ce
s state
m
ents and
senten
ce simil
a
rity is
cal
c
ulate
d
by calculating t
he simil
a
rity of le
xical bet
wee
n
se
que
n
c
e
s
, whi
c
h i
s
base
d
on th
eir
sema
ntic si
mi
larity model. It is a good re
spo
n
se to the senten
ce
se
mantic info
rm
ation [5-7].
In this
pap
er, defect
s
sim
ilarity co
mput
ing mo
del
ba
sed
on
vecto
r
spa
c
e
mod
e
l was
prop
osed.
3. Relate
d Associa
t
ion Rules
Thro
ugh
a la
rge nu
mbe
r
of
test data, d
e
f
ec
t dist
ributi
on can b
e
su
mmed u
p
.We
applie
d
asso
ciation
rules of d
a
ta
mining, a
nd
studied
the
r
e
la
tive
d
e
f
ec
ts
, w
h
ic
h
c
a
n an
a
l
yz
e e
ffic
i
en
tly
and prevent p
r
odu
ct defe
c
ts,then imp
r
ov
e prod
uct qu
a
lity.
Asso
ciatio
n
rules refle
c
t the
rel
a
tion
sh
ip am
ong
data o
r
is a
study
whet
her th
e
gene
ration
of
a data
ca
n
speculate the
gene
ration
of
anothe
r d
a
ta
. Data a
s
so
ci
ation reflect th
e
relation
shi
p
in Databa
se.
The associ
ation deg
ree
can be exp
r
esse
d throu
gh sup
p
o
r
t and
confid
en
ce. T
hus, findin
g
the rel
a
tion
shi
p
amon
g
dat
a in the tran
saction d
a
taba
se i
s
the mini
ng
purp
o
ses of a
s
soci
ation rul
e
[8-10].
Asso
ciatio
n rules in a tra
n
s
a
c
tion data
b
a
se a
r
e defin
ed as follo
ws [8-12]:
Defini
tion 1-1:
Let
}
,i
…
,
,i
{i
=
I
n
2
1
be a
colle
ction of items, the transactio
n
databa
se
}
t
,
…
,
t
,
{t
=
D
n
2
1
is
a seri
es of transactio
n
s with a
uniqu
e
identifier TID.
Each
transactio
n
correspon
ds to
a sub
s
et of I.
The colle
ctio
n of items call
ed itemset
s
.
Defini
tion 1-2
: Suppose
I
I
1
,
the sup
port
of itemset
1
I
in the data set D is the
percenta
ge of
I from transa
c
tion
s that co
ntain
1
I
, that mean
s,
support
I
‖
t∈D
|
I⊆t
‖
‖
D
‖
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Re
sea
r
ch of the Defe
ct Mo
del Base
d on
Sim
ilarity and
Asso
ciation
Rule (Wa
n
jia
ng Ha
n)
4687
Defini
tion 1-3
: For item
set I and tra
n
sa
ction
dat
aba
se
D, all
meet u
s
er-spe
cified
Minsu
ppo
rt itemset
s
in T, whi
c
h is grea
ter than
or eq
ual to Minsup
port non
empt
y subset of I, is
the frequ
ent i
t
emset
s
or l
a
rge item
set
s
. Freq
uent
ite
m
set
s
that d
o
not in
clude
other el
eme
n
ts
from freq
uent
itemset is cal
l
ed maximum
frequent item
sets o
r
maxi
mum itemset.
Defini
tion 1-4
: In I and D
, the definition of asso
ciati
on rul
e
s, such as
2
1
I
I
can
be presented as
credibility
or confidence.
T
he so-called of
ru
les’ confidence
is the
ratio
o
f
numbe
r
of tra
n
sa
ction
s
th
at incl
ude
1
I
,
2
I
an
d the
num
be
r of transactio
n
s th
at in
clu
d
e
1
I
.That
is,
coniden
ce
I
⇒I
s
upport
I
∪I
s
upport
I
Whe
r
e
I
I
,
I
2
1
,
2
1
I
I
.
Defini
tion 1
-
5
:
Strong
a
s
sociatio
n rule
s is
asso
ciatio
n rul
e
s
with
D o
n
I
whi
c
h meet
minimum
sup
port an
d mini
mum co
nfide
n
ce. Commo
nly associ
atio
n rule
s ge
ne
rally refers to the
stron
g
asso
ci
ation rule
s a
s
defined ab
ove.
4. Defe
ct
Cla
ssifica
tion a
nd Ass
o
ciati
on Model
No
w, we stu
d
y a new de
fect model,
whi
c
h is d
e
ri
ved from the
defect analy
s
is a
n
d
summ
ary. Th
is
defe
c
t mo
del pl
ays a
n
importa
nt
rol
e
in
the di
scovery
of mo
re d
e
fect
s. T
h
e
defect m
odel
pro
p
o
s
e
s
ef
ficient imp
r
ov
ement
to th
e
develop
men
t
and te
sting
pro
c
e
s
s, an
d
improve th
e
prod
uct
qualit
y much
bette
r. Defe
ct mod
e
l is
con
c
e
r
n
ed with
apply
i
ng the
simila
rity
theory a
nd th
e ide
a
of
the
Knowl
edg
e
Discove
r
y in
Datab
a
se. It is the
result
of su
mming
up
defect di
strib
u
tion, usi
ng a
s
soci
ation rul
e
s a
nd a
naly
z
ing
relatio
n
ship, whi
c
h
ca
n be u
s
e
d
a
s
the
theoreti
c
al ba
sis fo
r the de
ve
lopment im
provem
ent [13-14].
4.1. Defe
ct
Classifica
tion Model De
finition
Defe
ct simila
rity calculatio
n
model is ba
sed o
n
vector spa
c
e mod
e
l
sho
w
n in Fig
u
re 1.
Figure 1. Def
e
ct Similarity Cal
c
ulat
ion M
odel Based o
n
the Vector
Space
Defect A
Defect B
Lexi
cal
l
a
bel
、
specialized
v
o
c
abu
l
a
r
y an
alys
is
Sentence
ve
ctor A
Sentence
ve
ctor B
Calculate weighted
sem
a
n
tic si
mil
a
rity
C
a
l
c
ul
at
e voca
bul
a
r
y
si
m
ilarit
y
Defect sim
ilari
ty
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 6, June 20
14: 4685 – 4
691
4688
Firstly, the similarity betwe
en defect A and defec
t B is defined as a
value in [0,1], where
0 rep
r
e
s
ent
s no simila
rity, 1 rep
r
e
s
ent
s compl
e
tely
si
milar, the larger value in
di
cate
s the mo
re
simila
r they are.
A sente
n
ce
that states d
e
fect
is
sue ca
n
be
exp
r
essed as a
vecto
r
>
T
,...,
T
,
T
<
T
n
2
1
,among whi
c
h
i
T
rep
r
e
s
ent
s a
single
word after
separate words
manag
eme
n
t in defe
c
t senten
ce
s, these
wo
rd
s a
r
e mai
n
ly n
oun
s, verb
s,
adje
c
tives
and
nume
r
al
s, the se
nten
ce’
s
sema
ntic inf
o
rmatio
n is
d
e
scrib
ed m
a
i
n
ly by the p
a
rt of spee
ch o
f
these type
s o
f
word.
The m
a
in i
d
e
a
of
wo
rd-ba
s
ed
simil
a
rity
model
is:
Fi
rst
cal
c
ulate
the word
s’
semantic
simila
rity in the state
m
ent
s, and th
en t
h
rou
gh
se
m
a
ntic si
milarity
cal
c
ulate
de
fect statem
e
n
t
simila
rity, so that the rich sem
antic i
n
form
atio
n can be taken
into accoun
t. The simila
rity
betwe
en sent
ence T and senten
ce
'
T
can
be obtaine
d by their simil
a
rity matrix
'
TT
M
, a
s
sh
ow
n
in Equation (1).
(1)
Whe
r
e,
)
,y
s(x
i
i
rep
r
ese
n
ts the semantic
simil
a
rity of word
i
x
and wo
rd
i
y
, each
row of th
e m
a
trix re
prese
n
ts the
sema
ntic
simila
rity of a
word
in
se
nten
ce
T
and
ea
ch
wo
rd i
n
sente
n
ce
'
T
.
Suppo
se
n
2
1
w
,
…
,
w
,
w
re
spe
c
tively d
enote
s
the
wei
ghts o
f
n
2
1
x
,
…
x
,
x
in
sente
n
ce T, t
a
ke
the
maxi
mum valu
e of
ea
ch
ro
w
or each colum
n
in
the matrix,
that
is
see
k
i
ng
the maximum
sema
ntic
si
milarity of a word
in
sente
n
ce T
and e
a
c
h
word in senten
ce
'
T
. Will
matrix
'
TT
M
com
p
ress to one
-di
m
ensi
onal, a
nd then
we
ig
hted average
s to these m
a
ximum, as
sho
w
n in Equ
a
tion (2
). So we can get th
e weig
hted semantic
simil
a
rity
'
TT
X
betwe
en sente
n
ce
T
and senten
ce
'
T
.
(2)
Acco
rdi
ng to
the given
sim
ilarity, we
ca
n det
e
r
min
e
simila
r type
s
of defe
c
ts, th
ereby to
cla
ssif
y
t
he d
e
f
e
ct
s.
4.2. Defe
ct
Associa
t
ion M
odel
Relative a
nal
ysis i
n
form
s t
hat a
happ
en
ing of o
ne
d
e
fect may i
n
cur an
othe
r
or m
any
other
defe
c
ts occu
r. Fo
r
example, the
pro
b
lem
of
‘interfa
c
e di
splayin
g
’
is asso
ciated with
‘interface indi
cation p
r
obl
e
m
s’,’ con
s
i
s
te
ncy pro
b
lem
s
’ and ‘bou
nda
ry probl
ems’.
In order to co
nfirm the defe
c
t asso
ciation
m
odel and q
uantify the defect cla
s
ses, the test
obje
c
t’s versi
on is define
d
as tran
sa
ctio
n set T
and the defect category set is defined as
set I at
first. Then
th
e asso
ciatio
n
defect
ea
ch
defec
t m
o
d
e
l co
rrespon
ds i
s
dete
r
m
i
ned. After t
h
e
transactio
n
d
a
taba
se
D is
gene
rated
an
d the mini
mu
m su
ppo
rt de
gree
an
d mini
mum confide
n
ce
11
1
2
1
21
2
2
2
12
(,
)
(
,
)
(,
)
(,
)
(
,
)
(,
)
(,
)
(
,
)
(,
)
m
m
TT
nn
n
m
sx
y
s
x
y
sx
y
sx
y
s
x
y
sx
y
M
sx
y
s
x
y
sx
y
12
1
1
(
(
m
a
x
(
(
,
),
(
,
),
(
,
))))
ii
i
i
TT
i
n
m
i
n
i
w
s
xy
s
x
y
s
xy
X
w
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Re
sea
r
ch of the Defe
ct Mo
del Base
d on
Sim
ilarity and
Asso
ciation
Rule (Wa
n
jia
ng Ha
n)
4689
threshold val
ue are given
,
by applying
the Ap
rio
r
i
algorith
m
the
freque
nt item set an
d t
he
asso
ciation rule will be obt
ained at la
st.
We fo
cu
s on
the strong a
s
soci
ation rul
e
s
whi
c
h
set
D meet
s th
e rule
s fo
r m
i
nimum
sup
port de
gree and mini
m
u
m confid
en
ce degree of set I.
Asso
ciatio
n Rule Minin
g
is sea
r
chin
g fo
r a pro
c
e
s
s sati
sfying the minimum
suppo
rt
degree a
nd
minimum
con
f
idence deg
ree. The
sup
p
o
rt deg
ree
a
nd confiden
ce deg
ree a
r
e
the
one
s that are
useful a
nd m
eanin
g
ful.
This pa
per
will q
uantify
different ve
rsions
of
ea
ch
test
obje
c
t, and
finally
get the
transactio
n
d
a
taba
se D. Then defin
e
two thresh
olds a
s
follo
ws:Mini
m
um
suppo
rt de
gree
(min_
s
u
p
) an
d Minimum confiden
ce
co
efficient (min
_co
n
f).
Confid
en
ce d
egre
e
mea
s
u
r
es the rule’
s
intensity whi
l
e the supp
ort degree me
asu
r
e
s
the turn
up
freque
ncy
of th
e rul
e
. A la
rg
er
confid
en
ce
deg
ree
an
d
a sm
alle
r
sup
port d
e
g
r
ee
can
be appli
ed to typical ca
se
s.
Thus,
step
s to determi
ne a
s
soci
ation mo
del are a
s
foll
ows:
1)
Determine th
e asso
ciation
rule X in defe
c
t cla
ssifi
cati
on datab
ase
2)
Ensu
re a con
d
ition that X.suppo
rt>=min
_
su
p
3)
Ensu
re a con
d
ition that X.confiden
ce
>=mini_conf
At last obtai
n
the d
e
fect
set’s
stron
g
re
la
tion
set whi
c
h i
s
th
e def
ect a
s
so
ciatio
n mod
e
l
we want.
5. A Cas
e
of
GUI De
fec
t
Model
This pie
c
e ,a
pplying the a
nalysi
s
model
di
scu
ssed a
bove based o
n
a typical GUI’s test
result, con
c
lu
des a G
U
I defect model.
5.1. Definitio
n
of De
fec
t
Classify
ing
This
step mai
n
ly focuses
o
n
settling a
n
d
analyzin
g th
e data source. These dat
a, which
is the defe
c
t set, are the
total defects by usin
g sever
a
l methods
such as
t
e
s
t
ing. By fully
analyzi
ng the
de
scription
of thes
e d
e
fects,
extra
c
ting the
u
s
efu
l
inform
ation,
cla
s
sifying
and
building
up
th
e defe
c
t m
o
d
e
l spe
c
ie
s, calcul
ating th
e
simila
rity of
each two d
e
fects a
c
cordin
g to
the Equatio
n
(1) an
d Equ
a
tion (2). Fi
rst
define th
e mi
nimum
simila
rity
Min
Sim
and
com
b
ine t
w
o
defect
s
whi
c
h simil
a
rity is larg
er th
an
Min
Sim
,then
get a minimum def
ect cla
ssifi
cat
i
on set,
finally, construct a prelimin
ary frame of
model [15].
This pie
c
e m
a
tche
d the d
e
fect simila
rit
i
es a
c
cordi
n
g
to the defect data of G
U
I test
result. This
defec
t set is
}
d
,
…
,
d
,
{d
=
D
n
2
1
,where n
=
5
0
. By applying (1) a
nd (2), the
simila
rity is calcul
ated. we
defined
Min
Sim
as 80% in this model. The
n
the defe
c
ts wh
ose
simila
rity abo
ve 80% is
cl
assified to o
n
e
set, form
ed
a minimum
defect
cla
ssifi
cation
set. After
cla
ssifie
d
the
proble
m
types an
d pro
b
l
e
m distri
butio
n, calculated
the si
mila
rity degree, an G
U
I
defect cl
assifi
cation tabl
e is sho
w
ed a
s
T
able 1.
Table 1. GUI Defec
t
Clas
s
i
fic
a
tion
T
e
r
m
Defect
classif
i
cat
i
on
1 Interface
displa
y
problem
2
Interface indication problem
3 Wrong
characte
r
problem
4
Punctuation form
at problem
5
Inconformit
y
sem
antic expression
6 Unreada
ble
code
s
7 Messy
ver
s
ions
8 Inconformit
y
pict
ures
9
Inapprop
riate lea
r
ning problem
10 Sequencing
prob
lem
11 Boundar
y
pro
b
lem
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 6, June 20
14: 4685 – 4
691
4690
5.2. Defe
ct
Associa
t
ion M
odel
The p
r
ocess t
o
determi
ne t
he defe
c
t associatio
n mod
e
l is a p
r
o
c
e
s
s of data mini
ng. Set
the p
r
od
uct
versio
n
as transactio
n
se
t T, qu
ant
ize
d
defe
c
t as
item set I a
nd con
s
truct
a
transactio
n
databa
se
D. After the thre
shol
d values of mini
mum su
ppo
rt and minim
u
m
confid
en
ce
suppo
rt a
r
e
gi
ven, frequ
ent
item
se
ts a
nd a
s
so
ciatio
n rul
e
will b
e
ge
nerated
by
applying Ap
ri
ori a
r
ithmetic.
We fo
cu
s on
the strong
a
s
soci
ation rul
e
whi
c
h i
s
th
e rule
satisfying
the minimum
sup
port an
d minimum con
f
idence co
efficient D o
n
I.
This pa
per q
uantized
26 v
e
rsi
o
n
s
of
th
e te
st
obje
c
t and obtain
e
d
tran
sa
ction
databa
se
D. Two threshold value
s
a
r
e define
d
as
fo
llows acco
rding to previo
us expe
rien
ces.
a)
Minimum sup
port deg
re
e (min_sup
) = 3
8
%
b)
Minimum con
f
idence co
efficient de
gre
e
(min_conf)
= 6
8
%
37 rule
s are
obtained by
using Apri
ori arithm
etic. In orde
r to facilitate the analysi
s
results, the
d
i
git re
sults
were tu
rne
d
to
be
spe
c
ific
probl
em
s. Th
en ba
se
d o
n
the con
s
trai
nts
su
ch as
a strong asso
ciati
on
rule sh
oul
d
in
clud
e a
m
i
nimum
su
pp
ort de
gree
which
is 38%
a
nd a
minimum
co
n
f
idence de
gre
e
which
is 68
%, decid
e
wh
ether th
e o
u
tput a
s
sociati
on rule i
s
stro
ng
asso
ciation rule, result is showed in Fig
u
re 2 [16].
Figure 2. Re
sult on Asso
ci
ation Rul
e
The ho
rizont
al axis rep
r
e
s
ent
s the qu
antize
d
rule
while the vertical axis re
p
r
esents
confid
en
ce d
egre
e
or
sup
port de
gre
e
. In Figure
2, 2
2
stro
ng a
s
sociatio
n rule
s are obtai
ned
by
cal
c
ulatin
g th
e outp
u
t satisfying mi
n_sup a
nd
min_
conf. F
o
r the
asso
ciatio
n
rule
of ’inte
rface
displ
a
y pro
b
l
e
m’ =>’inte
rface i
ndi
catio
n
pro
b
lem’
, it’
s
confide
n
ce
is 70%, the
suppo
rt deg
re
e is
43%, whch indicates the
proba
bility that bot
h ‘interface indi
ca
tion probl
em’
and ‘interfa
c
e
indication problem’ occurs i
s
70%, while
the prob
abilit
y is 43% that
‘interface indi
cation probl
em
’
may occur
whe
n
‘interf
ace di
spl
a
y probl
em’ o
c
curs. Thu
s
we mainly
focus
on these
as
so
ciat
ion
s
.
The co
nfiden
ce de
gre
e
of the 31st a
s
so
ciation rule in
Figure 2 is 1
00% and the
sup
port
degree d
o
e
s
not sati
sfy the minimum
suppo
rt deg
re
e,
hen
ce it is not a st
rong asso
ciation rule.
But whe
n
p
r
e
c
on
dition o
ccurs, th
e
con
s
eque
nce rule
alway
s
a
ppe
ars. It’s ap
propriate
to fo
cus
more on con
s
eque
nce
rul
e
whe
n
precon
dition
o
c
curs.
De
spite it is
not a st
ron
g
asso
ciation
rule
and
th
e sup
p
o
rt
d
egree do
esn’t rea
c
h
a
defined
mi
ni
mum
sup
port
degree, b
u
t a
high
confide
n
ce
coeffici
ent de
gree
can al
so
reflect an im
portant a
s
sociation pro
b
le
m.
Table 2. G
U
I Defe
ct Associ
ation Model
Defect ty
pes
Defect associatio
ns
Interface displa
y
problem
Interface indicati
on problem, cons
istency
p
r
oblem,
boundar
y p
r
oble
m
Interface indication problem
Interfac
e displa
y
problem, consistenc
y
pro
b
lem, bo
undar
y pro
b
lem, messy
codes
Messy
Ve
rsions
Interface displa
y
problem
Grap
hical inconsistency
Interface i
ndication problem, cons
istency
p
r
oblem
Learnabilit
y
problem
Interface indication pr
oblem, cons
istency
p
r
oblem,
boundary
p
r
oble
m
Boundar
y pro
b
lem
Interface displa
y
problem, Inte
rface indication problem
, Consistenc
y
problem
Consistency
p
r
o
b
lem
Interface displa
y
problem, Inte
rface
indication problem, Boundar
y p
r
obl
em, Gra
phical inconsistency
0%
20%
40%
60%
80%
1
00%
1
20%
1
4
7
10
13
16
19
2
2
25
28
31
34
3
7
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Re
sea
r
ch of the Defe
ct Mo
del Base
d on
Sim
ilarity and
Asso
ciation
Rule (Wa
n
jia
ng Ha
n)
4691
Acco
rdi
ng to
the re
sea
r
ch on fre
que
n
t
se
t and
strong a
s
so
ciation rul
e
ab
ove, defect
types a
nd
de
scription
s
for
these
types a
r
e
co
nclu
ded
into a
co
rrespondi
ng ta
ble
whi
c
h
form
s
a
defect a
s
soci
ation model, as in Table 2
,
which list
the defect a
s
sociatio
n with use
r
’s inte
rfa
c
e.
Eligible d
e
fects an
d its a
s
sociatio
ns will
be li
sted i
n
th
e mod
e
l a
nd
defect
dist
rib
u
tion a
nd
def
ect
asso
ciation
can be kno
w
n
from this.
Asso
ciatio
n a
nalysi
s
indi
ca
tes that a p
r
o
b
lem may lea
d
to one o
r
th
e other
pro
b
l
e
ms to
appe
ar. Such as ‘inte
rfa
c
e di
splay p
r
oble
m
’ asso
ciate
s
with ‘i
nterface indi
cation p
r
obl
e
m
’,
‘con
si
sten
cy probl
em’,’ bo
unda
ry pro
b
le
ms’ etc.
6. Conclusio
n
This p
ape
r b
a
se
d on ve
ctor spa
c
e si
mi
larity
cal
c
ulati
on mod
e
l and
the asso
ciati
on rul
e
s
techni
c fo
r
d
a
ta minin
g
,
by re
sea
r
chi
ng la
rge
am
ount of
defe
c
t data,
pro
poses a
def
ect
cla
ssifying m
e
thod
a
nd a defect
a
s
so
ci
ation
mo
del
. In
this way, not
only
the defect
a
nalyzing
efficien
cy is i
m
prove
d
but
also b
a
sed
on defe
c
t asso
ciation m
o
del, pertin
ent
sugg
estio
n
s for
software test
ing and desi
gning
will be presented. Further
st
udy will contin
uously work on
spe
c
iali
zing
the si
milarity
of terminol
ogi
es t
hat d
e
scri
be defe
c
ts
a
nd keep
on
rese
archin
g ot
her
defect a
s
soci
ations a
s
well
.
Ackn
o
w
l
e
dg
ements
This
work
wa
s su
ppo
rted i
n
part by the
National
Nat
u
ral Sci
e
n
c
e
Found
ation o
f
China
(Grant No. 61
1702
73).
Referen
ces
[1]
Glenford J M
y
ers,T
om Badgett,
T
odd M
T
h
o
m
as, Core
y
Sandler. T
he Art
of Soft
w
a
r
e
T
e
sting. 2004.
[2]
Bend
er. Req
u
ir
ements Base
d
T
e
sting Process Overvie
w
.
20
09.
[3] Yi
Guan.
Quan
tifying se
ma
nti
c
simi
larity of
Chin
ese w
o
rds
from How
N
et
. Internatio
na
l C
onfere
n
ce o
n
Machi
ne Le
arn
i
ng a
nd C
y
b
e
rn
etics. 2002: 2
3
4
-23
9
.
[4]
Z
heng-T
ao Yu,
Lei H
u
.
Si
mil
a
rity Co
mp
utat
ion
of Chi
nese
Question B
a
s
ed o
n
Ch
unk.
Internatio
na
l
Confer
ence
on
Machin
e Le
arnin
g
an
d C
y
ber
netics. 200
6: 1
7–2
2.
[5]
Gan KW, Wong PW.
Ann
o
tat
i
on
infor
m
atio
n
structures
in
Chin
ese
texts
usin
g H
o
w
net
.
In Second
Chin
ese L
a
n
g
u
age Proc
essin
g
W
o
rkshop H
ong Ko
ng. 20
0
0
: 85-92.
[6]
Serge
i
Nir
en
b
u
rg, Co
nstanti
ne D
onm
ashn
e
w
,
De
an J
g
r
ann
es.
T
w
o appr
oach
e
s to
Matchin
g
i
n
Exampl
e-bas
e
d
Machin
e T
r
anslati
on.
Proce
ddi
ngs of the fifth Internat
ion
a
l
Confere
n
ce o
n
T
heoretica
l
and Meth
od
olo
g
ical i
n
Mach
in
e T
r
anslation o
f
Natural La
ng
uag
es.19
93: 4
5
-57.
[7]
W
anjia
ng H
an.
Stud
y
o
n
the d
e
fect Classific
a
tion mod
e
l. un
pub
lish
ed.
[8]
Lon
g H, L
ee H
YA.
New
Visu
ali
z
a
t
i
on T
e
c
h
niq
ue for K
n
o
w
ledge
Discov
e
ry in OLAP
. Procee
din
g
s
of
the F
i
rst Pacific-Asia Co
nfere
n
ce on Kn
o
w
l
edg e D
i
scover
y
a
nd Data Mi
n
i
ng. 19
97.
[9]
Jai
w
e
i
Ha
n, Michel
in
ne Ka
mber. Data
Minin
g
: Co
nce
p
ts and T
e
chniq
ues, Morg
a
n
Kaufma
n
n
Publ
ishers, Se
cond Ed
itio
n. 2006.
[10]
Margar
et H Du
nham. Data Mi
nin
g
: Introduct
o
r
y
a
nd Adv
a
n
c
ed T
opics, Pearson Ed
ucati
o
n. 2007.
[11]
Dion
H Goh, R
ebecc
a
P Ang.
An Introductio
n
to A
ssociati
o
n rule mi
ni
ng: An ap
plic
ation
in cou
n
se
lin
g
and h
e
l
p
-seek
i
ng be
hav
ior of ado
lesce
nts.
Behav
iour R
e
se
arch Metho
d
s.
200
7; 39(2): 25
9-26
6.
[12]
Li C, F
an M. G
ener
ating
asso
ciatio
n rul
e
s ba
sed o
n
threa
d
e
d
frequ
ent patt
e
rn tree.
Co
mp
ut Eng Ap
pl
.
200
4; 4: 188-1
92 (in C
h
in
ese)
.
[13]
Hain
in
g Li, Ying
xi
ong So
ng,
Yingch
un Li. Perfo
rmanc
e Anal
ys
is and
Exp
e
rim
ental
Stud
y
o
n
F
l
a
t
Optical Com
b
Generati
on.
T
E
LKOMNIKA
Indon
esi
an Jo
urna
l
of Electri
c
al Eng
i
ne
eri
n
g.
2013; 1
1
(1):
28-3
9
.
[14]
Guihu
a
Z
hen
g
,
Quanlon
g Guan. F
u
zz
y
C
o
mpre
hens
ive
Evaluati
on S
o
ft
w
a
r
e
of T
e
achi
ng Qual
it
y
Based on Entr
opy
.
T
E
LKOMNIKA Indo
nesi
an Jo
urn
a
l
of Electrical
En
gi
neer
ing.
201
3;
11(3): 131
3-
132
1.
[15]
Z
H
AO Qian, Z
H
ENG Jun,
LI Jing. S
o
ft
w
a
r
e
Re
lia
bi
lit
y M
ode
lin
g
w
i
t
h
T
e
sting-Effort F
unctio
n
a
n
d
Imperfect Deb
ugg
ing.
T
E
LK
OMNIKA Indon
esia
n Journ
a
l
of Electrical En
gin
eeri
ng.
20
1
2
; 10(8): 199
2-
199
8.
[16]
W
anjia
ng
H
an,
T
i
anbo
Lu,
S
un Y
i
.
R
e
sear
ch o
n
th
e Pr
o
b
le
m Mod
e
l
of
GUI bas
ed
o
n
Kn
ow
led
g
e
Discovery i
n
D
a
tabas
e
. Proceed
ings of the
2013 Intern
ati
ona
l Confer
en
ce on Soft
w
a
r
e
Engi
ne
erin
g
and C
o
mput
er Scienc
e. 201
3: 5-9.
Evaluation Warning : The document was created with Spire.PDF for Python.