TELKOM
NIKA
, Vol.14, No
.4, Dece
mbe
r
2016, pp. 15
75~158
5
ISSN: 1693-6
930,
accredited
A
by DIKTI, De
cree No: 58/DIK
T
I/Kep/2013
DOI
:
10.12928/TELKOMNIKA.v14i4.4234
1575
Re
cei
v
ed Se
ptem
ber 19, 2016; Revi
se
d No
vem
ber
14, 2016; Accepted Novem
ber 29, 20
16
Factors Influencing User’s Adoption of Conv
ersational
Recommender System Based on Product Functional
Requirements
Z.K. Abdur
a
h
man Bai
z
al
*
1
, D
w
i
H. Wi
dy
antoro
2
, Nur Ulfa Maulidev
i
3
Schoo
l of Elect
r
ical En
gin
eeri
ng an
d In
forma
tics, Institut
T
e
knol
ogi Ba
nd
u
ng,
Band
un
g, Indo
nesi
a
, telp/fa
x +
62-22-
250
09
35
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: baiza
l@telk
o
m
univ
e
rsit
y
.
ac.id
1
, dw
i@if.itb.
a
c.id
2
, ulfa@if.itb.ac.id
3
A
b
st
r
a
ct
Conv
ersati
on
a
l
reco
mmend
er
system (
CRS)
hel
ps
custo
m
ers get
prod
uc
ts fitted their
n
eeds
by
repe
ated
inter
a
ction
mech
an
isms. W
h
en c
u
stomers w
ant
to buy
pro
d
u
c
ts havi
ng
ma
ny a
nd
hig
h
t
e
ch
features (
e
.g., cars, s
m
artp
hon
es, n
o
teb
o
o
k, etc.), mo
st users
are
n
o
t
fami
li
ar w
i
th
pro
duct tec
h
nical
features. T
h
e
mor
e
n
a
tura
l w
a
y to e
licit c
u
stomers
’
ne
eds
i
s
by ask
i
ng
w
hat they re
al
ly w
ant to
use w
i
th
the
prod
uct they w
ant (w
e call
as prod
uct functio
nal re
qu
ir
e
m
e
n
t
s). In this pap
er, w
e
analy
z
e
four factors, e.g.,
perce
ived
usef
uln
e
ss, perce
iv
ed e
a
se of
use
,
trust and per
c
e
ive
d
en
joy
m
e
n
t associ
ated t
o
user
’
s
inte
nti
on
to ad
opt t
he
in
teraction
mo
de
l (i
n C
R
S)
bas
ed
on
pro
duct function
al
req
u
i
re
ments.
Res
u
lt
of exp
e
ri
me
nt
usin
g tec
hno
lo
gy acc
epta
n
ce
mod
e
l
(T
AM) in
dicates
that
, for us
ers w
h
o ar
en
’
t
fa
mil
i
a
r w
i
th tec
hni
cal
features, perc
e
ives
useful
ne
ss is
a
mai
n
factor influ
e
n
c
ing
users
’
a
dopti
on. Me
an
w
h
ile, perc
e
iv
ed
enj
oy
ment
play
s a rol
e
o
n
us
er
’
s
i
n
tentio
n t
o
ad
opt
this
int
e
ractio
n
mod
e
l
,
for users w
h
o
are fa
mi
liar w
i
t
h
technic
a
l featur
es of product.
Ke
y
w
ords
:
convers
a
tional
recomm
ende
r system
, tec
hnology
acceptanc
e
model,
online s
h
oping, e-
commerce
Copy
right
©
2016 Un
ive
r
sita
s Ah
mad
Dah
l
an
. All rig
h
t
s r
ese
rved
.
1. Introduc
tion
In most
re
commen
der system, user gi
ves
som
e
re
que
sts,
then
syste
m
gives
recomme
ndat
ions, an
d the
sessio
n end
s. In a real li
fe, a recom
m
endatio
n pro
c
e
ss in a
sh
ort
se
ssi
on is rare, beca
u
se most users rarely able
to
expre
ss hi
s real need
s at the beginni
ng
of
intera
ction, a
nd u
s
e
r
s are
rarely sati
sfie
d with
i
n
itial recom
m
en
dati
ons [1]. This
encourage
s t
he
emergen
ce o
f
numero
u
s
studies o
n
co
nversati
onal
recomme
nde
r syste
m
(CRS) as a
syst
em
that
imitates intera
ction b
e
twee
n
custo
m
er and
prof
ession
al sale
s
supp
ort.
CRS allo
ws m
o
re
than on
e inte
ractio
n
se
ssi
on (multiple
shot
s). T
he
system recom
m
end
s p
r
od
u
c
ts th
at fit user
requi
rem
ents
iteratively through inte
ra
ctions g
ene
rate
d by the system.
Re
comm
end
er syste
m
re
quire
s a user model
for p
e
rsonali
z
in
g recom
m
en
dati
on. This
use
r
mo
del i
s
built by expl
oring
user
preferen
ce
as
well a
s
o
b
serving of user
behavio
r [2]. CRS
can b
e
distin
guished by th
e way of syst
em in
buildi
n
g the user m
odel; navigati
on by asking
and
navigation by
prop
osin
g [1
]. In navigation by as
king
(NBA), the
system p
r
ovi
des a
se
rie
s
of
q
u
e
s
t
io
ns
abo
u
t
us
er
ne
ed
s
[3-
5
],
w
h
ile
in
n
a
v
i
gatio
n by p
r
o
p
o
s
in
g (NBP), th
e
system
sugg
ests
certai
n pro
d
u
cts to users and
obtai
ning user n
eed
s in the form of fe
edba
ck on the
recomme
nde
d prod
uct
s
[6-8].
Interactio
n m
odel
s that ha
ve been d
e
velope
d ask u
s
er
prefe
r
e
n
ce based o
n
tech
nical
feature
s
a
s
p
e
cts. Fo
r hig
h
-tec
h produ
cts that have
m
any features, su
ch a
s
notebo
oks, cars,
smartphones, server
s, PCs, cameras,
etc.,
many users
are not
familiar
with the techni
cal
feature
s
of th
e produ
ct. Th
e users m
o
re
easily
expre
ss th
e fun
c
tio
nal requi
rem
ents of
pro
d
u
c
t
that they loo
k
for, e.g., n
e
e
d
a
sm
artph
o
ne fo
r
selfie
s, HD g
a
min
g
,
but not fo
r vi
deo
re
co
rdin
g.
Several
studi
es h
a
ve trie
d
to develo
p
CRS
s in
whi
c
h th
e qu
esti
ons
are in th
e form
of pro
duct
function
al re
q
u
irem
ents,
su
ch a
s
Fi
ndM
e [9] (u
sing
NBP) a
nd Ad
visor Suite [1
0-13]
(exploiti
n
g
NBA).
Combi
n
ing
NBA and
NBP
is a
way to
d
e
velop a
CRS that able to
imitate the in
teractio
n
betwe
en a prosp
e
ctive bu
yer with a pro
f
essi
onal sa
le
s su
ppo
rt [14, 15]. Howeve
r, these stu
d
i
e
s
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 14, No. 4, Dece
mb
er 201
6 : 1575 – 158
5
1576
are
still focused on generating inte
ractions that refers
to product
techni
ca
l
features. We have
introdu
ce
d a
frame
w
o
r
k that wa
s
able to
gen
erate i
n
tera
ctions
based
on fun
c
tio
n
a
l
requi
rem
ents,
by combi
n
in
g the NBA a
nd NBP, in
o
u
r p
r
eviou
s
rese
arch [16].
This p
ape
r a
i
ms
to analyze th
e influen
ce of
intera
ction
model ba
se
d
on pro
d
u
c
t functio
nal re
q
u
irem
ents o
n
tha
t
frame
w
ork
to spe
c
ific user grou
ps,
a
s
well
as
fa
ct
ors
that affect user ad
opt
ion
o
f
this intera
cti
on
model.
We
utilize T
e
chnol
o
g
y Acceptan
ce Mod
e
l (TAM
) to
add
re
ss it. TAM is
a
model to
expl
ain
or
pre
d
ict th
e
user'
s
acce
ptance a
nd
a
doption
of
a
new informati
on te
chn
o
logi
es [1
7, 18].
At
first, this mo
del wa
s dev
elope
d to de
scribe b
ehav
i
o
r of users i
n
usin
g co
m
puter. The m
a
jor
factors a
r
e p
e
rceived u
s
ef
ulne
ss
and
e
a
se
of us
e. We a
dd two
factors,
pe
rceived enj
oyment
[19] and tru
s
t
[20, 21] to our hypoth
e
se
s mod
e
l, fo
r
analyzi
ng the
effect of these two facto
r
s. In
addition, we
also
see t
he influen
ce
of interacti
on mod
e
l b
a
se
d on p
r
odu
ct functi
onal
requi
rem
ents to some fa
ct
ors,
su
ch
a
s
perceived
ea
se
of use, pe
rceive
d u
s
efu
l
ness, p
e
rcei
ved
enjoyment, a
nd trust.
2. Interaction Model
The inte
ra
ction mo
del
co
vers
provi
s
io
n of qu
estio
n
s, p
r
od
uct
recom
m
en
dati
ons
and
explanation
s
of why
ea
ch
prod
uct
is re
comm
end
ed.
Thi
s
inte
ra
ct
ion aim
s
to e
x
plore
the
user
prefe
r
en
ce
s, based on fu
nction
al req
u
i
reme
nts of
the produ
ct. To gen
erate
this interactio
n,
sistem
re
qui
res the
map
p
i
ng of fun
c
tion
al re
quiremen
t
s - te
ch
nical
feature
s
- p
r
o
duct
s
. Ontol
o
gy
is a re
prese
n
tation of kn
owle
dge that
appropri
a
te
to address this ma
pping.
Interactio
ns are
gene
rated
by
traci
ng
sem
a
ntic relation
ships in th
e
on
tology. In this pap
er,
we ta
ke
sma
r
tph
o
n
e
s
as th
e
domai
n of th
e
CRS
. The
syste
m
utilize
s
an
o
n
tology m
ode
l as a
kno
w
le
dge
ba
se,
wh
ich
con
s
i
s
ts of three mai
n
cla
s
se
s [16]; 1) function
al
req
u
i
r
eme
n
ts, 2)
specifi
c
ation, a
nd 3) p
r
od
uct
.
Figure 1 ill
ust
r
ates the
gen
eral
schem
e
of in
tera
ction
betwe
en u
s
e
r
and
syste
m
[
16]. At
the begi
nnin
g
of the inte
ra
ction, t
he
system provid
es
multiple optio
ns of fu
nctio
n
a
l re
quireme
nts
to use
r
. The
n
,
the use
r
ca
n sel
e
ct
som
e
functi
o
nal requireme
nts i
n
the catego
ry of mandato
r
y,
optional
or n
o
t re
quired. I
f
the
p
r
efere
n
ce
s (u
se
r requireme
nts) are e
noug
h
to recomme
nd
prod
uct
s
, the
system
p
r
ovides a li
st of p
r
odu
ct
s
that are re
com
m
e
nded
a
s
well as explanatio
n
s
of why ea
ch
prod
uct is
recom
m
en
ded
. Meanwhile
,
if the preference is
still too gen
eral, t
h
e
system
will deliver more specif
i
c
questi
ons back to user.
Figure 1. Sch
e
me of Intera
ction Use
r
– System
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Facto
r
s Influe
ncin
g User
’
s
Adoption of
Conversation
al
… (Z.K. Abdurahm
an Baizal)
1577
Whe
n
the
system
recom
m
end
s
pro
d
u
cts,
u
s
er h
a
s
opp
ortu
ni
ty to rea
c
t t
o
the
s
e
prod
uct
s
. If u
s
er
sele
cts o
ne pro
d
u
c
t, then intera
ctio
n
is complete.
Ho
wever, if user
sele
cts n
one
or mo
re th
an
one
pro
d
u
c
t, it indicates t
hat the u
s
e
r
is not
sati
sfie
d with th
e re
comm
end
atio
n.
Hen
c
e, the system will ge
nerate inte
ra
ction
s
ba
ck t
o
help user e
x
presse
s hi
s need
s and m
a
ke
deci
s
io
n.
3. Strateg
y
f
o
r Guiding User
Duri
ng the
int
e
ra
ction, u
s
e
r
prefe
r
en
ce
s
are m
odel
ed
in a u
s
e
r
mo
del. Each int
e
ra
ction
step, su
ch a
s
giving que
stions, p
r
odu
ct
recomme
ndat
ion, or explan
ation, is gen
e
r
ated ba
se
d on
con
d
ition of t
he
curre
n
t user m
odel. T
o
deliver
que
sti
ons, th
e
syst
em will
sele
ct som
e
n
ode
s of
ontology as o
p
tions to ask.
Based on u
s
er’s a
n
sw
e
r
, the system
wil
l
maintain the
node
s in use
r
model a
s
a u
s
er p
r
efe
r
en
ce.
3.1. Case
s in Interac
t
ion Process
es
For ea
ch int
e
ra
ct
ion st
e
p
,
t
he system provide
s
at most
qm
ax
options of functio
nal
requi
rem
ents (as a qu
esti
on), whi
c
h
ca
n be sel
e
ct
e
d
by the user. Based o
n
a survey that wa
s
con
d
u
c
ted to
35 re
sp
ond
e
n
ts, the majo
rity of respon
dents li
ke
d
qma
x
=4
(41.9
4
%) follo
wed
by
qm
ax
=3
(25.
81%), 2 (12.
90%), 5 (9.6
8%), more
th
an 6 (6:45%
) and 6 (3:23
%
). Thus, we
use
qm
ax
=4 i
n
the CRS that
we
will evaluate. We
define five cases that
might arise duri
ng
interaction, whic
h a
r
e a
s
follows”
1.
Initial interacti
on (em
p
ty user model)
In initial int
e
raction, the system
provi
des
que
stions
relate
d
to the fu
nction
al
requi
rem
ents
of produ
cts th
at user
want
s, but
still general. Figu
re 2
gives an exa
m
ple of use
r
interface wh
en the syste
m
delivers some init
ial q
uestio
n
s to
use
r
. Que
s
ti
ons a
bout th
e
budg
et, bran
ds, ope
rating
system
s and
types of sma
r
tphone
s rol
e
as ha
rd con
s
traints, an
d it
gene
rally a
p
p
ear in th
e u
s
er'
s
mi
nd. F
u
nction
al requi
reme
nts
sel
e
cted
as "mu
s
t be fulfilled"
are el
ement
s of the set of mandato
r
y function
al
re
qui
reme
nts, whil
e function
al requireme
nt
s
sele
cted a
s
"
better fulfilled
"
is identical
to
the set of optional fun
c
t
i
onal requi
re
ments. Th
e
function
al re
quire
ment
s selecte
d
as "n
ot requi
red"
are in
clud
ed in the set of not required
function
al re
quire
ment
s.
The n
e
xt qu
estion
(b
esi
d
es th
e q
u
e
s
tion at th
e b
e
g
innin
g
of th
e
intera
ction
)
consi
s
ts of onl
y options of functio
nal re
q
u
irem
ents.
2.
More tha
n
on
e prod
uct sel
e
cted by u
s
er
User
requirement is
still too general, so that
the appropriate
product i
s
too m
any. Thus,
the sy
stem
provides mo
re
spe
c
ific qu
est
i
ons than
the
previ
ous on
e
s
a
nd
potenti
a
lly de
sire
d
by user.
3.
No re
co
mme
nded p
r
od
uct
sele
cted by u
s
er
Whe
n
the
sy
stem p
r
ovide
s
a li
st of re
comm
end
ed
prod
uct
s
, u
s
er may li
ke
some
of
these
re
co
m
m
ende
d p
r
o
d
u
cts,
and
he
hesitate
s
to
d
e
cid
e
o
ne
of
them. Fo
r h
e
l
p
ing
user to
make
de
cisi
o
n
, the syste
m
will
present the differe
nce
s
of fu
n
c
tiona
l requi
rem
ent
s supp
orted
by some of
those p
r
o
d
u
c
ts.This fun
c
ti
onal
requi
re
ments a
r
e o
b
tained by l
ooki
ng up th
e
curre
n
t u
s
e
r
prefe
r
en
ce
(mandato
r
y o
r
option
a
l fun
c
tional
re
qui
rements) i
n
th
e u
s
e
r
mo
del
,
and filled b
y
each
pro
duct. Figu
re
3 sh
ows
an exampl
e
of the sp
e
c
ific fun
c
tion
al
requi
rem
ents are
su
ppo
rted by Le
nov
o A536,
b
u
t are
not su
pporte
d by t
he two
othe
r
sma
r
tpho
ne
s (in the sa
me i
n
terp
retation
for Asu
s
Padf
one Infinity 2 and Le
novo
A8-50
)
.
4.
User p
r
eferen
ce
s are
still too gene
ral
It is possible
that user does not like
any
recommended produ
cts. Thus,
the system
will
ask potential
function
al req
u
irem
ents d
e
s
ire
d
by user, but have no
t been aske
d
.
Question
s
are ge
ne
rate
d by searchi
n
g unexplo
r
ed
use
r
prefere
n
c
e no
de
s in u
s
er m
odel.
5.
No produ
ct that fits user’
s
prefe
r
en
ce
s (there a
r
e con
t
radicto
r
y req
u
irem
ents)
There a
r
e
so
me contra
dict
ory
requi
rem
ents
of user,
so
no p
r
od
uct
s
that m
eet.
Thus, th
e
system
will d
i
splay requi
rements that
cau
s
e
no re
comm
end
ed
prod
uct
s
. Th
en, user
can
revise
his
re
quire
ment
s. A set of co
ntradi
ct
ory requireme
nts
is a
sub
s
et
of the use
r
prefe
r
en
ce
s
within u
s
er m
odel.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 14, No. 4, Dece
mb
er 201
6 : 1575 – 158
5
1578
Figure 2. Example of Initial Interaction
Figure 3. Example of Com
pari
s
on of the
S
pecific Fu
n
c
tional
Req
u
irements of Produ
cts
Selected by
Users
3.2. Product
Recommendation
and Explanation Facilit
y
The syste
m
recom
m
en
ds
prod
uct
s
ba
sed on
user p
r
eferen
ce
s that are mainta
ined in
the use
r
mo
del. This u
s
er mod
e
l re
pre
s
ent
s the
history of user p
r
efe
r
en
ce eli
c
ited d
u
ring
intera
ction. Mandato
r
y functio
nal re
q
u
irem
ent
s a
c
t as hard constraints fo
r re
comm
en
ding
prod
uct
s
.
Wh
ile optio
nal fu
nction
al requi
reme
nts
are
use
d
to
dete
r
mine a
d
egre
e
of
confo
r
mi
ty
of a produ
ct with user p
r
eferen
ce, which is
exp
r
e
s
sed as utilit
y value. More individual
s of
optional fun
c
tional re
quire
ments a
r
e m
e
t by a prod
uct, the high
er utility value of a pro
d
u
c
t.
Re
comm
end
ed pro
d
u
c
ts a
r
e obtain
ed b
y
explor
ing n
ode
s and its relation
ship
s in ontology.
An explan
ation of
why
a
pro
d
u
c
t is recom
m
en
ded
, allows
users to
make d
e
ci
sio
n
easi
e
r. An e
x
planation i
s
gene
rated
by tracin
g b
a
ck u
s
er
pre
f
eren
ce
(in t
he u
s
er
mo
del)
sup
porte
d b
y
each recommen
ded
prod
uct. Fi
g
u
re
4 d
epi
cts an
exam
ple of
a list of
recomme
nde
d
produ
cts, a
s
well as
th
e explanatio
n
s
.
Re
com
m
en
d
ed p
r
od
uct
s
a
r
e
so
rted
ba
sed
on the utility value of each
product.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Facto
r
s Influe
ncin
g User
’
s
Adoption of
Conversation
al
… (Z.K. Abdurahm
an Baizal)
1579
Figure 4. Example of Re
co
mmend
at
ion
and Explanati
on Facility Pa
ge
4. Frame
w
o
r
k
for Ev
aluating User’s P
e
rcep
tions
We ap
ply Techn
o
logy Accepta
n
ce Mo
del (TAM) to
analyze the
influence of
produ
ct
function
al re
quire
ment
s - ba
sed i
n
teraction
mo
d
e
l
to u
s
er's p
e
rception,
an
d sim
u
ltane
o
u
sly
analyze facto
r
s th
at affect
use
r
'
s
ad
opti
on of thi
s
inte
ractio
n mo
del
. In TAM, the
s
e fa
ctors
ref
e
r
to so
me
co
nstruct
s
, which
is al
so
a fo
rm
of u
s
e
r
’s pe
rceptio
n of th
e
syste
m
. With
a hyp
o
the
s
e
s
model, we ca
n analyze the
influence b
e
twee
n these
construct
s
. Th
e con
s
tru
c
ts
we ob
se
rved are
as
follows
:
1.
Perceived Usefulne
ss (P
U):
the degree t
o
whi
c
h u
s
er
feels that the intera
ction m
odel is q
u
ite
useful in reso
lving probl
em
s [17]
2.
Perceived ea
se of use
(EO
U
): the deg
re
e to which user feel
s that u
s
ing the inte
raction mo
del
will be free of
effort [17]
3.
Behavi
o
ral Intention
(BI): th
e degree to which u
s
e
r
inte
nds
to u
s
e th
e intera
ction
model in the
future, espec
ially for produc
t s
e
arc
h
[17]
4.
Perceived En
joym
ent
(PE):
the degre
e
to whi
c
h u
s
er
feels interest
ed, comfo
r
tab
l
e, and
guide
d by a intera
ction mo
del offered [1
9]
5.
Tru
s
t
(TR): the degree to which u
s
e
r
tru
s
ts the re
com
m
endatio
ns g
i
ven by syste
m
, where the
explanation f
a
cility plays a
role [20]
Eac
h
c
o
ns
truc
t c
o
ns
is
ts of
s
o
me cons
truc
t it
em
s, wh
ich
woul
d be
a list of q
u
e
s
tions i
n
que
stionn
aire
, as sh
own in
Table 1. Answers of
qu
est
i
ons 1
-
2
3
are
enco
ded o
n
a 5-p
o
int Like
rt
scale rangi
ng
from I stro
n
g
ly disag
r
e
e
(1), I di
sag
r
e
e
(2
), I Neith
e
r ag
ree
nor disag
r
e
e
(3
), I
agre
e
(4
) to the I stron
g
ly agre
e
(5
). Fo
r fairne
ss, we
rando
mize the que
stion
s
1-23, an
d the
s
e
que
stion
s
are
not group
ed
based on e
a
ch con
s
tru
c
t.
4.1. H
y
potheses Mod
e
l
To a
nalyze th
e influe
nce of
interactio
n m
odel
ba
sed
o
n
p
r
od
uct fu
n
c
tional
requi
rements
(we call
as
func-b
ase
d model
, for short) in in
cre
a
sin
g
the po
sitive percept
ion of the user, as
well a
s
fa
ctors that influ
e
n
c
e u
s
e
r
'
s
inte
ntion to a
dop
t this interacti
on mo
del, we define
the
1
1
followin
g
hypotheses,
1.
H1 : Fun
c
-b
a
s
ed mo
del in
cre
a
ses trust
2.
H2 : Fun
c
-b
a
s
ed mo
del in
cre
a
ses p
e
rceived usefuln
e
ss
3.
H3 : Fun
c
-b
a
s
ed mo
del in
cre
a
ses e
a
se
of use
4.
H4 : Fun
c
-b
a
s
ed mo
del in
cre
a
ses p
e
rceived enjoym
ent
5.
H5 : Perceive
d trust po
sitively affects pe
rceive
d usefu
l
ness
6.
H6 : Perceive
d ease of use
posit
ively affects p
e
rceive
d useful
ne
ss
7.
H7 : Perceive
d ease of use
positively affects p
e
rceive
d enjoyment
8.
H8 : Perceive
d enjoyment
positively affects pe
rceived
usefuln
e
ss
9.
H9 : Tru
s
t po
sitively positively affects pe
rceive
d enjoy
ment
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 14, No. 4, Dece
mb
er 201
6 : 1575 – 158
5
1580
10.
H10 : Perceiv
ed usefulne
ss po
si
tively affects behavioral intention
11.
H11 : Perceiv
ed enjoyme
nt posit
ively affects b
ehavio
ral intention
Table 1. Cons
truc
ts
and Its
Items
Code
Percei
v
e
d
U
sef
ulness
(PU)
PU1
The interaction
model improves the qualit
y
of p
r
od
uct search that I
do
Zanker, et al [20]
PU2
The interaction
model offers ma
n
y
bene
fi
ts over features in the p
r
oduct search
PU3
The interaction
model save m
y
time in product se
arch
PU4
The interaction
model makes it easier for me to fi
nd the best matc
hing product
PU5
The interaction
model in the s
y
stem is able to pro
v
ide better searc
h
results
PU6
The interaction
model is quite hel
pful in the process of finding product
PU7
Overall, the inter
a
ction model is
ver
y
useful in the
product search
Percei
v
e
d ea
se
of us
e (E
OU)
EOU1
S
y
stem
w
i
th the i
n
teraction model
is easy
to use
J. Brooke[22]
EOU2
This interaction model does not r
equire a lot of eff
o
rt
EO
U3
I think the interaction model w
ill also be eas
y
to us
e for othe
rs
EOU4
Each step of interaction is quite clear and e
a
s
y
to
understand
EOU5
It is easy
fo
r me t
o
be sklillful at using this interactio
n model
Trust
(TR
)
TR1
I believe w
i
th the
products recom
m
ended b
y
the s
y
st
em
w
i
th this interaction
model
Zanker, et al [20]
TR2
I believe the expl
anation given b
y
t
he interaction m
odel, about
w
h
y
each product
is recommended
TR3
I believe that the order of
recomm
ended pr
o
ducts list is consistent with m
y
needs
TR4
The e
x
planation
given b
y
s
y
stem l
eads me to believe that the reco
mmended
products are suit
able to m
y
ne
eds
Percei
v
e
d Enjoy
m
ent (PE)
PE1
The interaction
model is quite int
e
resting
Liao, et al [19]
PE2
I feel comfortable
w
i
th th
e interaction in these s
y
ste
m
s
PE3
The process of in
teraction in the sy
st
em is quite pleasant
PE4
Interactions in this sy
stem makes
me feel guided
Beha
v
i
oral In
te
ntio
n (B
I)
BI1
If i have access t
o
this sy
stem so
metime in the fut
u
re, I intend to
u
s
e this sy
stem
for prod
uct search
Liao, et al [19]
BI2
I w
ill use the s
y
st
em
w
i
th this interaction model in the future
BI3
I would recomme
nd this sy
stem to
others
Figure 5
de
pi
cts thi
s
hypot
heses mod
e
l. Zan
k
e
r
[20]
have p
r
oved
that the expl
anation
facility on recommender
sy
stem
s ca
n im
prove the perceived useful
ness, perceived
ease of use
and tru
s
t. In this p
ape
r, we
try to evaluate the in
fluen
ce of func-ba
s
ed mod
e
l to i
n
crea
sing
of the
perceived u
s
efulne
ss, pe
rceived e
a
se of use, tru
s
t and pe
rceive
d enjoyment
(Hypoth
e
ses
H1 -
H4). Som
e
st
udie
s
have
re
ported th
e po
sitive
influen
ce between tru
s
t and p
e
rcei
ved usefulne
ss
[21-24],
so
we al
so
eva
l
uate the eff
e
ct bet
wee
n
these t
w
o
con
s
tru
c
t
s
(H5). Pe
rceived
useful
ne
ss, p
e
rceived e
a
se of us
e an
d
behavio
ral int
ention, bei
ng
the co
re
con
s
tru
c
ts i
n
TA
M,
and their
relat
i
onship is the
most wid
e
ly studied in
TA
M studie
s
, sin
c
e intro
d
u
c
ed
by Davis [17].
Figure 5. Hyp
o
theses Mo
d
e
l
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Facto
r
s Influe
ncin
g User
’
s
Adoption of
Conversation
al
… (Z.K. Abdurahm
an Baizal)
1581
Thus we ma
ke
hyp
o
the
s
es H6
an
d H10,
to
eval
uate the rela
tionshi
p between the
s
e th
ree
con
s
tru
c
t
s
. Perceived
enjo
y
ment ha
s a
cru
c
ial
in
flue
nce
on
cu
sto
m
er i
n
tention
and
beh
avio
r, in
the
area of e-comme
rce [23],
[25-26].
The
r
efor
e a
l
so i
n
volve p
e
rceived
enj
oyment in
o
u
r
hypothe
se
s model (hypot
heses
H7, H8
, H9 and H11
)
.
To tes
t
H1 -
H4,
we involv
e 100 us
ers
wh
o are grouped in catego
ries familiar (48%)
and
unfamilia
r (52
%
) with the tech
nical feature
s
of
the produ
ct. The user’
s
ag
e ran
ged bet
ween
18-
55 years. Th
ey were
sma
r
tphone u
s
e
r
s
and familia
r
with we
b-b
a
sed appli
c
atio
n. Each user
tries
to use two ki
nds of intera
ction mod
e
l; 1) func
tio
nal
requi
rem
ents-ba
sed inte
ra
ction mod
e
l we
have d
e
velop
ed, 2
)
flat m
o
del
(for comp
arison
).
Flat
model
is an
i
n
tera
ction
mo
del b
a
sed
on
the
techni
cal feat
ure
s
of the produ
ct is com
m
only
found in e-comme
rce web site
s. For fairn
e
ss, bot
h
system
s h
a
ve the
sam
e
p
r
odu
ct d
a
ta a
nd id
entical
u
s
er inte
rfaces, while
the
differen
c
e
only
i
n
copi
ng strate
gies of ea
ch
intera
ction ca
se. Afte
r usin
g two differe
nt kind
s of interactio
n mo
del,
use
r
s a
r
e a
s
ked to answe
r the questio
n
s in question
n
a
ire. To evalu
a
te H5-H1
1
, we just focus
on
the u
s
er’
s
an
swers of fu
nc-ba
s
ed
mo
de
l and
apply li
near regressi
on a
nalysi
s
b
y
the follo
win
g
formula:
PE
b
EOU
b
TR
b
a
PU
8
6
5
(1)
EOU
b
TR
b
a
PE
7
9
(2)
PE
b
PU
b
a
BI
11
10
(3)
a
=con
stant,
= ran
dom
error,
b
i
=coefficie
n
t
(index
i
co
rresp
ond
s
to hi
pothe
sis
i
th
i
n
hypothe
se
s
model
)
4.2. Validit
y
and Reliabilit
y
Testing
In a first step,
we do the va
lidity testing on 23 con
s
tru
c
t items pre
s
e
n
ted to re
spo
ndent
s.
By using prin
cipal comp
on
ents extra
c
tio
n
(varimax
rotation method with Kais
er Normaliz
ation),
23 co
nst
r
u
c
t items are re
duced to 11
items, as p
r
e
s
ente
d
in Ta
ble 2. Each
resulting fa
cto
r
corre
s
p
ond
s
to each
con
s
tru
c
t. Ho
we
ver, TR2 is
also hi
ghly correlated
with factor PU,
as
que
stion T
R
2
also
can b
e
i
n
terp
reted to
sup
port P
U
.
Subiyakto, et
al [27] and
Z
anker [2
0] no
ted
that the facto
r
loadin
g
s la
rger tha
n
0.40
ar
e quite u
n
derstand
able
and acce
pta
b
le to form the
unde
rlying scale in the factor analysi
s
. As see
n
in
T
able 2, all factor loa
d
ing
s
are high
er than
0.40. The
re
sult of reli
ab
ility testing is re
pr
esente
d
by value
of Cro
nba
ch
Alpha of e
a
ch
con
s
tru
c
t. O’
Rou
r
ke [28]
noted th
at Cronba
ch
Alph
a value
of 0.
50 o
r
gre
a
ter is
suffi
cient
for
resea
r
ch, whi
l
e 0.70
i
s
re
commen
ded
a
nd 0.8
0
i
s
de
sira
ble. A
s
it
can
be
see
n
f
r
om
Tabl
e 2,
all
Cro
nba
ch Alp
ha value
s
are
greate
r
t
han
0.60, even mostly above 0
.
70 and 0.80.
Table 2. Extracted
Con
s
truct Items
Component
Factor 1 (PE
)
Factor 2(P
U
)
Factor 3(BI
)
Factor 4 (
T
R
)
Factor 5 (E
OU
)
PU1
0.319
0.718
0.202
0.171
0.186
PU3
0.257
0.637
0.488
0.033
0.231
PU7
0.486
0.551
0.290
0.133
0.220
Crobanch Alpha
of PU=0.832
EO
U2
0.195
0.236
0.243
0.063
0.754
EO
U5
0.216
0.357
0.323
0.351
0.441
Crobanch Alpha
of EOU=0.6
0
4
TR1
0.342
0.025
0.162
0.802
0.122
TR2
0.093
0.574
0.013
0.497
0.417
TR3
0.251
0.337
0.261
0.578
0.156
Crobanch Alpha
of TR=0.769
PE1
0.684
0.318
0.317
0.112
0.216
PE4
0.597
0.282
0.431
0.243
0.090
Crobanch Alpha
of PE=0.780
BI2
0.180
0.282
0.688
0.345
0.196
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 14, No. 4, Dece
mb
er 201
6 : 1575 – 158
5
1582
5. Results
5.1. Influenc
e of Func
tio
n
al based Interac
t
ion on User’s Per
c
e
p
tion
As se
en in Ta
ble 3, for u
s
ers wh
o are fa
milia
r with th
e
techni
cal fea
t
ures, the
average
of
the rate
s for
func-ba
s
ed
model i
s
g
r
e
a
ter than
th
e
flat model, e
s
pe
cially for f
a
ctors
percei
v
ed
useful
ne
ss, p
e
rceived ea
se of use, trust and
perceiv
ed enjoyme
nt. To evaluate the significa
nce
of the me
an
differen
c
e,
we
used T
-
t
e
st, an
d
al
so involves L
e
vene'
s te
st for e
quality
of
varian
ce
s. T-test re
sult
s i
n
Tabl
e 3
show th
a
t, func-ba
s
ed
mo
del influe
nce
s
si
gnificantl
y
to
increa
se u
s
e
r
's pe
rception
of these fou
r
factor
s (p <0.
001). It su
gg
ests th
at H1,
H2, H3
and
H4
are a
c
cepte
d
. From T
able
3 we
ca
n noti
c
e that
the user’s rate average
for
th
e
fu
nc-ba
s
ed mo
del
is sig
n
ifica
n
tly higher than
the flat model, for these fou
r
factors.
Table 3. Mea
n
of Rates a
n
d
T-Te
sting
Result
s
Fact
or Model
Familiar User
Not
Famili
ar
Us
er
Mean
t
df
p-
v
a
l
u
e
(2-ta
iled
)
Mean
t
df
p-
v
a
l
u
e
(2-ta
iled
)
PU
Func-Based Mod
e
l
4.63
7.24 66.50
0.00
4.47
9.19 82.62
0.00
Flat Model
3.69
3.42
EO
U
Func-Based Mod
e
l
4.30
6.38 85.36
0.00
4.04
6.89 102
0.00
Flat Model
3.34
3.19
TR
Func-Based Mod
e
l
4.21
4.39 94
0.00
4.14
5.26 102
0.00
Flat Model
3.68
3.40
PE
Func-Based Mod
e
l
4.27
4.34 94
0.00
4.17
6.21 102
0.00
Flat Model
3.59
3.25
T-test p
r
oved
that the ave
r
age
of user’
s
rate
s i
s
si
g
n
ificantly different fo
r the
s
e fou
r
factors,
such that
H1, H2, H3
a
nd H4
are acce
pted
(with si
gnifica
nce l
e
vel
= 0.
001). It can
be
con
c
lu
ded th
at func-ba
s
e
d
model
can
improve u
s
e
r
’
s
pe
rception
s
(pe
r
ceive
d
u
s
efuln
e
ss (P
U),
perceived
ea
se
of u
s
e
(E
OU), tru
s
t (T
R) and
p
e
rcei
ved enj
oyme
nt (PE)), for b
o
th u
s
e
r
s wh
o a
r
e
familiar and unfamiliar
with te
chni
cal feat
ures of product.
5.2.
Fac
t
ors
Influencing
User
's
In
ten
t
ion
to
Ad
opt the
Produ
ct Func
tional
Requir
e
ments-
Bas
e
d Inter
a
ction Model
We a
pply a l
i
near
reg
r
e
ssion analy
s
is
to evaluate the effect
s of among fa
ct
ors
and
analyze facto
r
s influ
e
n
c
ing
adoptio
n of this fun
c
-ba
s
e
d
model. T
h
e
analysi
s
i
s
repre
s
e
n
ted b
y
hypothe
se
s H5 - H1
1.
Fi
rst, we analy
z
e
th
e
p
a
th
of hypothe
se
s te
sting
ba
sed o
n
the
u
s
er’s
perceptio
n, regardle
s
s of t
he u
s
e
r
cate
gorie
s.
T
o
ev
aluate th
e g
o
odne
s
of fit o
f
our
hypothe
se
s
model to dat
a, we use
so
me model
-fit indices re
sulted by LISREL path analysis. Liao, et al [19]
note that the
bound
s of
model fit indi
ce
s that i
ndi
cate the goo
d
fit to the data are
GFI >
0.8,
RMSEA <
0.08. Meanwhile, Al-Maghrabi
, et
al., [23] note that chi square/df
≤
5
repre
s
e
n
ts
go
od
model
-fit. The re
sults
re
veal that the chi
-
squa
re
/degre
e
s of
freedo
m (CMIN/DF
) is
1.58,
goodness of
fit index (GFI
) is
0.93, and root m
ean
square
er
ror of approximate (RMSEA)
is
0.076. The re
sults
confirm t
hat the hypotheses mo
del
fit the data reaso
nably well.
Re
sults
of h
y
potheses te
sting a
r
e d
e
p
icted
i
n
the
Figure 6. E
a
ch
edg
e in
Figure 6
pre
s
ent
s re
gression
coefficient (
b
) of ea
ch ind
epen
de
nt variable (see equ
ations
1 - 3). Value
of
b
rep
r
e
s
e
n
ts
degree of infl
uen
ce of a f
a
ctor t
o
othe
r facto
r
. Tru
s
t and pe
rceived ea
se
of u
s
e
affect pe
rceiv
ed e
n
joyment
with
almo
st
the same
de
gree
(H7
an
d
H9
). It me
a
n
s, expl
anati
on
facility (trust)
and ea
se in the func-ba
s
e
d
model
can
make the u
s
e
r
s feel gui
ded
and intere
ste
d
.
Mean
while, trust, pe
rceive
d ea
se of
use
and
perceiv
e
d
enjoym
ent
are
also facto
r
s th
at influen
ce
perceived u
s
efulne
ss (H5,
H6 and H8).
Perceive
d eas
e of u
s
e is
the stron
g
e
r
influen
cing fa
ctor
of perceived
enjoyment
re
lative to trust
(H7,
H9
). Behavioral int
ention (i
ntent
ion of u
s
ers
to
adopt thi
s
in
teractio
n m
o
del) i
s
i
n
flue
nce
d
by p
e
rceived
usefu
l
nes
s (H10) and perceive
d
enjoyment (H11). In this case, the pe
rceived
enjoym
ent has a m
u
ch
stro
nge
r influence than
perceived
u
s
efullne
s
. We
notice th
at p
e
rceived
ea
se of u
s
e
has a g
r
eat
role
in improving
the
perceived
enj
oyment, whil
e the p
e
rceiv
ed enj
oyment
ha
s a
relativ
e
ly stro
ng i
n
fluen
ce o
n
u
s
er
intention of a
dopting the i
n
tera
ction m
odel. Value
of
R
2
indicates the p
o
rtio
n of varian
ce
s in
depe
ndent v
a
riabl
es that
can
be
expla
i
ned by i
nde
pend
ent vari
able
s
in
mod
e
ls
(expla
nat
ory
powers). Fig
u
re 6 sho
w
s
that
the explanatory po
we
rs of de
pend
ent factors are 42 perce
nt fo
r
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Facto
r
s Influe
ncin
g User
’
s
Adoption of
Conversation
al
… (Z.K. Abdurahm
an Baizal)
1583
perceived u
s
efulne
ss, 34
perc
ent for perceived e
n
joyment an
d 23 pe
rcen
t for behavi
o
ra
l
intention.
Figure 6. Paths of Accepte
d
Hypothe
se
s of Whole Users
(a)
(b)
* : significant w
i
th p < 0.05; ** : significant
w
i
th p <
0.01; *** : signifi
cant w
i
th p < 0.
0
0
1
Figure 7 (a). Paths of Accepted
Hypotheses of Users who are Fa
m
iliar with Tech
nical Features
(b). Paths of Accepted
Hy
potheses of User
s who are not Familiar wi
th Technical Features
Paths of
accepted hypotheses
of users
who are familiar
wi
th techni
cal
features are
depi
cted
by pictures 7(a).
Model
-fit indice
s of thi
s
hypothe
se
s model a
r
e,
CMIN/
D
F=0
.
93,
RMSEA=0.00, and
GFI=0.961. The
resu
lt shows that the hypotheses
model fit
to the
data.
As
see
n
in
Figu
re 7(a),
percei
v
ed ea
se
of
use
great
ly a
ffects the
pe
rceived
enjoy
ment (H7
)
. T
h
is
mean
s that t
he ea
se
presented
by fu
n
c
-b
ased m
o
d
e
l re
ally ma
kes u
s
e
r
s feel
intere
sted
a
nd
guide
d (perceived enj
oyment). Me
an
while, p
e
rcei
ved enjoym
e
nt is influ
e
n
c
ed by p
e
rcei
ved
enjoyment
(H8), an
d p
e
rce
i
ved u
s
efulne
ss influen
ce
the b
ehavio
ral
intention
(H1
0
). Th
e result
s
of hypotheses testin
g show that users who
are familiar
wi
th the technical features, intend to
adopt thi
s
int
e
ra
ction
mod
e
l du
e to
(in
d
i
r
ectly) it
s ea
se an
d
ability to gui
de th
e
u
s
er (enjoym
e
nt).
Ho
wever, tru
s
t and pe
rcei
ved useful
ne
ss a
r
e n
o
t fact
ors that affect user'
s
int
ention of ado
pting
this interactio
n model. Inte
ractio
n ba
se
d
on the
fun
c
ti
onal requi
re
ments
of a p
r
odu
ct is not
so
attractive for use
r
s in this
categ
o
ry, because they
are
already famil
i
ar with the techni
cal featu
r
es
of the produ
ct. Howeve
r, t
hese u
s
e
r
s fe
el guid
ed by
all of interacti
ons,
su
ch th
a
t
they intend
to
use thi
s
interaction mo
del
in the future (behavio
ral int
ention).
Hypothe
se
s
paths fo
r cat
egory of u
s
e
r
s wh
o ar
e no
t familiar with
the techni
cal
feature
s
are
sho
w
n in
Figure 7(b
)
. T
he model
-fit indices
of the
hyphote
s
e
s
model fit the data re
asona
bly
well, as
confi
r
med by the
CMIN/DF=0.82,
RMSEA=0.00 and GFI
=
0.941.
The hyphotesis
test
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 14, No. 4, Dece
mb
er 201
6 : 1575 – 158
5
1584
results sh
ow
that pe
rceive
d ea
se
of u
s
e an
d tr
u
s
t
a
ffect the p
e
rceived u
s
eful
n
e
ss
(a
cce
pt
H5
and H6) a
nd
the perceive
d
usefuln
e
ss i
n
fluen
ce be
h
a
vioral intenti
on (a
ccept H9). He
nce, users
feel that the intera
ction m
o
del is u
s
eful
due to t
he e
a
s
e a
nd tru
s
t. Both of these
factors indi
re
ctly
affect u
s
e
r
s’
intention to
a
dopt the
inte
raction
mod
e
l
(in
CRS) in
the future. F
u
rthe
rmo
r
e, t
h
e
explanation f
a
cility based
on functio
nal
requi
rem
ent
s is eno
ugh to
help u
s
ers in
prod
uct
sea
r
ch.
In addition, trust an
d pe
rceived ea
se
of use i
n
fluen
ce the pe
rceived enj
oyment
(acce
p
t H7
a
n
d
H9). Th
us, th
e explanation
facility and ease of u
s
e in
this intera
ctio
n make the u
s
ers feel gui
d
e
d
(pe
r
ceived en
joyment).
7. Conclusio
n
In experime
n
t
s, we involv
e two catego
ries
of
use
r
s as respon
de
nts; 1) u
s
e
r
s
who a
r
e
familiar with t
he product techni
ca
l features of product
, 2) user
s who are not familiar with the
prod
uct te
ch
nical fe
atures of pro
d
u
c
t. The resu
lt sh
ows that inte
ractio
n mo
del
(in
CRS
) ba
se
d
on produ
ct functio
nal re
q
u
irem
ents i
s
able to in
cre
a
se the p
e
rceived ea
se o
f
use, percei
v
ed
enjoyment, trust and pe
rceived useful
n
e
ss for both
use
r
’s
categ
o
r
ies. Ove
r
all (rega
rdl
e
ss of the
categ
o
ry of use
r), the u
s
ers fe
el interest
ed a
nd g
u
ided
(pe
r
cei
v
ed enjoyme
nt) be
cau
s
e
of
perceived
ea
se
of u
s
e
and
also the
tru
s
t
that co
me
s f
r
om i
n
tera
ctio
ns
and
expla
nation
s
. On
th
e
other h
and, p
e
rceived ea
se of use, tru
s
t and
perceiv
ed enjoyme
nt affect
percei
v
ed useful
ne
ss,
and in
dire
ctl
y
affect user’s intentio
n t
o
ado
pt
the
function
al re
quire
ment
s-b
a
se
d inte
ra
ction
model.
Ho
we
ver, we
can n
o
tice th
at pe
rceived
enjoy
ment is stron
ger i
n
fluen
cin
g
facto
r
of
user’
s
intention to a
dopting the in
teractio
n mo
d
e
l relative to perceived u
s
efulne
ss.
If we look at
the use
r
’
s
p
e
rception
of eac
h user
ca
tegory, there
are
some
di
fference
results bet
ween these two categori
es. For users who are familiar wi
th the technical features o
f
the produ
ct, trust i
s
n
o
t a
factor
affectin
g user
ado
ption to the
mo
del. Since th
ey are
alread
y
familiar with the tech
nical feature
s
, expl
anation
fa
cility based on t
he functio
nal requi
rem
ents is
not attra
c
tive. The inte
re
sti
ng thing i
s
, u
s
ers
fe
el com
f
ortable
and
guide
d (p
erceived enj
oyment)
due to ea
se
of use offe
re
d by pro
d
u
c
t functional
re
quire
ment
s-b
a
se
d interacti
on mod
e
l. We
notice
that p
e
r
ceive
d
e
n
joy
m
ent is a fa
ct
or infl
u
e
n
c
ing
user’
s
a
dopti
on o
n
the i
n
tera
ction
mod
e
l
instea
d of perceived u
s
eful
ness.
For
users
who are not fam
iliar
wi
th the t
e
chnical feat
ures of
the product, trust
and ease
of use pl
ays
a role fo
r th
e
adoptio
n of
p
r
odu
ct fu
nctio
nal
requi
rem
ents-ba
sed
in
teractio
n m
o
d
e
l,
indire
ctly. All
of interactio
n pro
c
e
s
ses (with 5
case
s a
ddre
s
sed by intera
ction mo
del in CRS
) a
s
well as explanation facility, make
users feel that the in
teraction model is useful and very helpf
ul,
so the u
s
e
r
s will adopt t
h
is inte
ra
ction model
(CRS) in the future for p
r
o
d
u
ct se
arch.
The
intere
sting thi
ng i
s
, pe
rceived enj
oyment
is n
o
t a
fa
cto
r
affectin
g u
s
er’s ad
option
of the mo
del
of
intera
ction. Although the u
s
ers feel m
o
re co
mfo
r
tabl
e and gui
ded
by functiona
l requi
reme
nts-
based interaction model th
an the flat model (H4
)
, bu
t
it is not a fa
ctor that affe
cts the u
s
e
r
s to
adopt a fun
c
tional-ba
s
ed in
teractio
n mod
e
l.
Referen
ces
[1]
Bridg
e
D, Goker MH, McGinty L, Sm
y
t
h B.
Case-Bas
ed
Recomme
nd
er
S
y
stems.
T
he Know
ledg
e
Engi
neer
in
g R
e
view
. 200
6; 20(3): 315-
32
0.
[2]
Z
anker M, Jessenitsch
nig
M. Case-Studi
es
on E
x
p
l
oiti
ng E
x
pl
icit C
u
stomer Re
qu
irements i
n
Recommender
S
y
stems.
User
Model
ing a
nd
User-Ad
apted I
n
teractio
n
. 200
9; 19(1-2): 13
3
-
166.
[3]
Do
yle M, Cu
nn
ing
ham P.
A Dyna
mic Ap
proa
ch to Reduc
in
g
Dial
og in On-
L
ine D
e
cisi
on Guid
es
. T
he
5th Europ
e
a
n
W
o
rkshop o
n
Case-B
ased R
easo
n
in
g. 200
0: 49-60.
[4]
Schmitt S, D
opic
haj
P, Do
mínguez-M
arín
P.
Entropy-B
ased
Vs. Si
milarity-Influ
enc
e
d
: Attribute
Selecti
on M
e
th
ods for
Di
alo
g
s
T
e
sted
on
Different E
l
ectron
i
c
Co
mmerc
e
D
o
mai
n
s
. T
he 6th Eur
ope
a
n
W
o
rkshop o
n
Case-B
ased R
easo
n
in
g. 200
2: 380-3
9
4
[5]
Gu M, Aamodt A,
Explanati
on-Bo
osted Q
uestio
n
Sel
e
cti
on in
Conv
ers
a
tion
al CB
R
. W
o
rkshop o
n
Exp
l
a
natio
n in
CBR. Madri
d
. 200
4: 105-
114.
[6]
McGint
y
L, Re
ill
y J. On
T
he Evolutio
n Of
Cr
itiqu
i
n
g
Rec
o
mmen
ders. In: Recomme
n
der S
y
stems
Han
dbo
ok. Spr
i
ng
er US. 201
1
:
419-45
3.
[7]
Smy
t
h B, McGint
y
L, Reilly
J, McCarthy
K.
Co
mp
oun
d Cri
t
iques for C
o
n
v
ersatio
nal R
e
commen
d
e
r
System
s
. EE/W
I
C/ACM Internatio
nal C
onfer
ence
o
n
W
eb Intelli
ge
nce. 20
04: 145-
15
1.
[8]
Reilly
J, Smy
t
h B, McGint
y
L, Mc
Carth
y
K.
Critiqu
i
ng
With
Confi
d
e
n
c
e.
Case-B
ase
d
Re
aso
n
in
g
Rese
arch an
d
Devel
o
p
m
ent
. 200
5; 362
0(03)
: 436-45
0.
Evaluation Warning : The document was created with Spire.PDF for Python.