TELKOM
NIKA
, Vol. 11, No. 8, August 2013, pp. 44
9
8
~4
504
e-ISSN: 2087
-278X
4498
Re
cei
v
ed Fe
brua
ry 16, 20
13; Re
vised
Ma
y 15, 20
13
; Accepte
d
May 25, 20
13
Artificial Emotion Engine Benchmark Problem based on
Psychological Test Paradigm
Wang Yi*, Wang Zhi-lian
g
Schoo
l of auto
m
ation a
nd El
e
c
trical Eng
i
ne
e
r
ing,
Un
iversit
y
of Science a
n
d
T
e
chnolo
g
y
Beiji
ng
Beiji
ng, Ch
in
a, 100
08
3
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: bkcool
@16
3
.com
A
b
st
r
a
ct
Most of testing and ev
alu
a
ti
ons of
emotio
n mo
del i
n
the field of
affective computi
n
g
are self-
eval
uatio
n, w
h
ich ai
ms at the
appl
ic
atio
n-sp
ecific back
g
ro
u
nd, w
h
ile t
he r
e
searc
h
on th
e prob
le
m of th
e
Bench
m
ark e
m
otion
a
l
mo
del
i
s
scarce. T
h
is
pap
er fi
rstly pr
opos
ed th
e fea
s
ibil
ity of maki
ng psyc
hol
og
ical
test parad
ig
m
a part of artific
i
al Be
nch
m
ark
eng
ine,
a
nd
w
i
th taking ver
s
atility a
nd eff
e
ctiven
ess as
the
evolutional fact
or to jud
ge the
engine
by testing p
sychologic
a
l par
adigm
s
. In add
ition,
an emotional hidden
Markov mod
e
l is built a
nd tes
t
ed base
d
on the Benc
h
m
ark
theory. T
he detail
ed si
mu
lati
on proc
ess of th
e
exper
iment is g
i
ven. T
he testin
g resulta
n
ts ar
e coinc
i
de w
i
th
the real w
o
rld
’
s situation.
Ke
y
w
ords
:
affective co
mp
uti
ng, artificia
l
e
m
ot
ion, be
nch
m
ark, hidd
en
ma
rkov mo
de
l
Copy
right
©
2013 Un
ive
r
sita
s Ah
mad
Dah
l
an
. All rig
h
t
s r
ese
rved
.
1.
Introduc
tion
The first on
e
who
propo
sed the i
dea
of maki
ng co
mputers with
artificial em
otion
is
American pro
f
essor Min
s
ky In 1985 in
his boo
k “T
h
e
soci
ety of
Mind”. The q
uestio
n
is wh
ether
intelligent ma
chin
es
ca
n h
a
ve so
me e
m
otions,
and
whi
c
h
ca
n b
e
intellige
n
t a
s
hu
man
k
in
d. the
probl
em ha
s
become a
cl
assic
referen
c
ed m
odel b
y
most re
sea
r
ch
ers in thi
s
field. In 199
5,
Ro
salin
d
W.
Picard from
MIT Medi
a L
ab p
ubli
s
he
d
his o
r
igin
al
monog
ra
ph
b
e
com
e
s a
ki
n
d
of
early form
of affective co
mputing te
ch
nology.
A
f
t
e
r
sev
e
r
a
l y
e
a
r
s,
af
f
e
ct
iv
e comp
ut
ing a
n
d
artificial e
m
oti
onal i
s
sue att
r
acte
d the att
enti
on of the
worl
d. Since
1990
s, Japa
n
e
se
have b
e
g
u
n
the re
sea
r
ch
work of Kan
s
ei Engine
erin
g, whic
h a
s
a kind
of technolo
g
ical
sci
ence, com
b
in
es
human e
m
oti
on with en
gin
eerin
g togeth
e
r to de
sign g
ood
s manufa
c
turin
g
.
No
wad
a
ys,
Japan
ha
s fo
rmed th
e mo
st adva
n
ced
system
of K
a
sey
engi
ne
ering.
In
2000, Profe
s
sor
Wan
g
Zh
i-liang fro
m
University
of Scien
c
e an
d Tech
nolo
g
y Beijing in Ch
ina
develop
ed th
e concept of
artificial
psy
c
hology, which
is a
ne
w expl
oration
in the
field of affecti
v
e
comp
uting. M
any a
s
sociati
ons of China
have
st
arte
d
doing
the
rel
e
vant re
se
arch
and
re
ach
ed
higher level,
such as the cel
ebration of the first
sessi
on of
affective com
puting intelligent
intera
ction in
ternation
a
l a
c
ad
emic
conf
eren
ce
s
an
d
the establi
s
hment of arti
ficial psy
c
hol
ogy
and a
r
tificial
emotion Profession
al Co
mmittee of
Chin
a Asso
ci
ation for Arti
ficial Intellige
n
ce
indicates.
Becau
s
e
the
emotio
nal t
h
inki
ng i
s
m
easura
b
le, th
ere
are m
a
n
y
re
sea
r
che
r
s try to
simulate
the
gene
ration
and
cha
nge
s of mood f
r
om differe
nt disciplin
es t
hat psy
c
hol
o
g
y,
cog
n
itive scie
n
ce an
d information scien
c
e incl
ude
d [1]. Meantime, due to the complexity of the
emotion
s
a
s
well a
s
the im
perfe
ctne
ss
o
f
the resea
r
ch
of humanity
emotional va
riation, this
wo
rk
is co
mpli
cat
ed, and th
ere i
s
also
a numbe
r of emotion
a
l theory b
u
ilding
s
coe
x
ist.
Cur
r
e
n
tly
,
there have be
e
n
a lot of mo
dels of the mood.
Of cou
r
se, it is not ea
sy to make these
model
s a
c
hie
v
e an exa
c
t quantitative d
e
scriptio
n
of
human
and
a
n
imal em
otio
ns p
e
rfe
c
tly, but
at least
som
e
of the mod
e
l
s
a
c
hieve
a li
mited mi
micry from the fu
nction
al an
gl
e. For exa
m
p
l
e
,
the O
C
C em
otion mo
del,
this
wa
s p
u
t
forwa
r
d
by O
r
tony, Cl
ore
and
Colli
ns a
t
The
Co
gniti
ve
Structu
r
e of Emotions
in 1988.
It
i
s
an
ori
g
inal
emot
ional
co
gnitive mo
del
and
is the
first on
e of
the pra
c
tical model
s towa
rds the
studie
s
of huma
n
emotion
s
, wh
ich is n
o
t onl
y cater to th
e
requi
rem
ent
of Comp
uter-impleme
nted
developm
ent, but also su
mmari
ze
s an
d co
ncl
ude
s
the
corre
s
p
ondin
g
rel
a
tion
by usi
ng
a reg
u
lar
mod
e
l.
Acco
rdi
ng to
analy
s
is of
variou
s eve
n
ts, a
seri
es of em
o
t
ion is t
r
igg
e
red by oth
e
r
e
n
tity
interacti
on [2]. The
O
C
C mod
e
l div
i
ded the
cou
r
se
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
e-ISSN:
2087
-278X
Artificial Em
otion Engine B
enchm
ark Problem
base
d
on Psych
o
log
i
cal Te
st Para
digm
(Wan
g
Yi)
4499
of emotion
i
n
to thre
e
cat
egori
e
s: th
e
results
of th
e event, the
actio
n
of th
e ag
ent a
n
d
the
perceptio
n of the obje
c
t. The Kismet emo
t
ional m
odel [
3
] that is desi
gned by the
MIT C.Brea
zeal
to apply to a
huma
noid
robot Kism
et is ap
plied to
a hum
anoi
d
rob
o
t Kisme
t. The mod
e
l is
con
s
tituted
b
y
four p
a
rt
s: emotion
a
l
stimuli,
emoti
onal
evaluati
on, emotio
na
l activation
a
n
d
emotional expre
ssi
on.
Europea
n
state
space-b
a
sed e
m
otion mo
del
is an
emotio
nal spa
c
e
whi
c
h
rep
r
e
s
ent
s the basi
c
em
otions’ fa
ctors a
s
seve
ra
l el
e
m
ental vecto
r
s, emotion
a
l state de
scrib
e
d
as discrete p
o
ints in this state spa
c
e [
4
].
Affective model [5] based on the p
r
oba
bility spa
c
e
provide
s
a
new metho
d
for affe
ctive computi
ng ma
ch
in
e au
to
ma
tic
a
lly. T
h
e
o
r
e
tic
a
l s
t
u
d
y
gene
rated th
rough M
a
rkov
Chai
n an
d the HM
M em
otional tra
n
sf
er
simulatio
n
con
s
i
s
ts of f
our
step
s that the transfe
rs of
mood stim
ula
t
e the m
ood spo
n
tane
ou
s transfe
r, tran
sfer of emotio
nal
stimuli a
nd e
m
otional
sp
o
n
taneo
us
me
tastasi
s
.
Th
e
emotion
a
l m
odel a
s
an a
r
tificial em
otion
engin
e
play
s
a key
role i
n
the functio
n
o
f
makin
g
intel
ligent ma
chin
es intelli
gent.
But evaluati
o
n
factor of the
engin
e
is still
not exact en
ough.
Thi
s
e
v
aluation pro
c
e
ss n
eed
s
a set of gen
eric
stand
ard
s
a
s
a found
ation f
o
r the p
u
rpo
s
e of getti
ng th
e judg
e re
sult
s of an
alysi
s
and
synthe
sis.
That’s to
say
the evaluati
on of the pe
rforma
nce of
the Affective gen
eratio
n
engine
nee
ds a
grou
p of studi
es on Ben
c
h
m
ark pro
b
lem
.
2.
Relativ
e
Works
The so-calle
d
Bench
m
ark i
s
a matte
r of Baselin
e no
rm, which is
u
s
ed to e
s
tabli
s
h a
set
of relatively impecca
ble m
odel
s. For
high effici
e
n
t tests a
nd evalu
a
tion sy
stem
unde
r the
sa
me
issue
s
, a co
mmon pl
atform for ma
kin
g
co
mpa
r
iso
n
s a
m
ong
di
fferent ki
nd
s of model
s
and
system
s i
s
b
u
ilt. No
wad
a
ys, Ben
c
hm
ark i
s
sue
ha
s
reached
many
scientific fiel
ds. F
o
r
exam
ple,
the st
ru
ctural
dam
age
ide
n
tification te
chn
o
logy
i
s
a p
r
in
cipal
segment
of
structu
r
al
he
alt
h
monitori
ng (S
HM). With deeply improve
the cap
ability of these methods
the experimental costs
decrea
s
e. In
the Internatio
nal Worksho
p
on Stru
ctural Co
ntrol
co
nferen
ce, the
resea
r
ch gro
u
p
con
s
i
s
ting of
Europ
e
, Asi
a
and
USA
has
bee
n p
r
opo
sed
deali
ng with th
e
SHM an
d th
e
Bench
m
ark
structu
r
e h
a
s
been
set up
by Che
n
at
the sam
e
time in o
r
de
r
to make
dire
ct
comp
ari
s
o
n
s among va
ri
ous te
chn
o
lo
gies. Th
e Benchma
r
k
structure was
raise
d
by Black
Ventura
after
the Ame
r
ican
SHM
re
se
arch g
r
ou
p
wa
s
set u
p
u
nde
r t
he joi
n
t-fundi
ng of
IASC a
nd
ASCE. In 1999 for the first ti
me I
A
C&ASCE research
group
resolved to focus on the
establi
s
hm
en
t of a
well-defined
Ben
c
hma
r
k i
s
su
e of the
Ameri
c
an
gro
up thu
s
ca
n
do
comp
arative studie
s
abo
ut
diverse
st
ru
ctural dam
a
g
e
identification
metho
d
s by
optimizin
g to
ols
of test data
a
nd st
ru
ctural
model
cal
c
ul
ation tool
s. T
here
is
also
Bench
m
ark i
s
sue i
n
the
cu
rrent
testing of ste
a
m gen
erato
r
tube re
sea
r
ch are
a
s. The
part-of
-spee
ch taggi
ng B
enchma
r
k u
s
es
CRF
++ m
o
d
e
l an
d Po
cket CRF
mod
e
l. Ben
c
hma
r
k,
used
as
an eval
uatio
n meth
od i
n
the
comp
uter fiel
d has
a long
-term ap
plicati
on, and i
s
wi
dely use
d
in
the in the ha
rdwa
re
su
ch
as
CPU
evaluat
ion, mem
o
ry
, I/O interfa
c
e
s
a
nd
peripherals software fo
r o
p
e
rating
sy
ste
m
evaluation, m
i
ddle
w
are an
d appli
c
atio
n
softwa
r
e i
n
m
a
chi
ne le
arni
ng field,
whi
c
h enh
an
ced
d
a
ta
pro
c
e
ssi
ng capability of the databa
se [6]. University
of California
Irvine (UCI) d
a
te sets in
clu
d
e
189
gro
u
p
s
whi
c
h
co
ntri
bute a
lot t
o
the
test
a
nd p
o
st
-eval
uation
of m
a
chi
ne l
earni
ng
algorith
m
s. I
n
the combin
atorial o
p
timi
zation i
s
sue,
the de
sign
of broa
dcasti
ng net
works, the
swit
chin
g ci
r
c
uit
d
e
sig
n
,
ship
t
r
an
sp
or
t
a
t
i
on
ro
ute plan
s,
wo
rk assignm
ents,
goo
ds pa
cking
scheme, th
e
sho
r
test
path
issue, the
m
a
ximum
(mini
m
um)
sp
anni
ng tre
e
i
s
sue
,
the be
st ed
ge
unrel
ated set, minimum cut
sets a
s
well
as sale
sman
issue are all belon
g to Benchm
ark issu
es
in broa
d sen
s
e. Espe
ciall
y
, in
the smart cal
c
ul
atio
n field, Levy No. 5 Funct
i
on, Shaffer’s F6
Functio
n
, Six-Hump
Camel
-
Back Fun
c
tio
n
and Ge
n
e
ralize
d
Scha
effer’s Pro
b
lem
are contin
uo
us
Benc
hmark
iss
u
es
[7].
It can be seen that in
many scienti
f
ic re
sea
r
ch
fields, Ben
c
hma
r
k issu
e can b
e
interp
reted
a
s
a
mo
del, al
gorithm,
syst
em te
st
an
d
evaluation
fo
rm. What i
s
more,
emotio
nal
model studie
s
in cu
rre
nt
emoti
onal
comp
uters fi
elds
are
ma
inly for sp
e
c
ific a
ppli
c
ati
on
backg
rou
n
d
s
su
ch as
(Hu
m
an Rob
o
t
Intera
ction, HRI),
(Hum
an Rob
o
t
Coope
ration, HRC) and
Huma
noid
Robot a
nd
so
forth. The
re
aso
n
wh
y we have
n
’t se
en p
r
o
s
a
n
d
co
ns evalua
tio
n
outcom
e
of the universe Benchmark issue is that this
kind of research i
s
still rare. In view of this
situation, i
n
t
h
is
pape
r
we
will u
s
e th
e p
s
ych
o
logi
ca
l t
e
st p
a
ra
digm
as
a p
a
rt of
a
r
tificial e
m
oti
o
n
engin
e
Ben
c
hmark i
s
sue
and u
s
e ve
rsatility and e
ffectivene
ss a
s
an eval
uati
on index. Fin
a
lly,
this set of Be
nchm
ark issu
es
shoul
d be
open a
nd hav
e scalability.
Evaluation Warning : The document was created with Spire.PDF for Python.
e-ISSN: 2
087-278X
TELKOM
NIKA
Vol. 11, No
. 8, August 2013: 4498 –
4504
4500
The pap
er i
s
organi
ze
d a
s
follows: in the thir
d secti
on origi
nally studie
s
the testing of
this Be
nchm
ark i
s
sue
on
the ba
si
s
of i
n
trodu
cin
g
so
me Psy
c
hol
o
g
ical
expe
rim
ental p
a
radig
m
s
emotion
-
ge
ne
rating mo
del. In the sectio
n
four it
s effect
iveness thro
u
gh experi
m
en
ts is verified.
3.
Artificial Emotion Engine
Benchma
r
k
Emotional m
odel, a
s
an
artificial e
m
otion en
gine
Based
on
a Psychologi
cal T
e
st
Paradi
gm, which
plays a
key role i
n
th
e fun
c
tion
of
putting intelli
gen
ce into
m
a
chi
n
e
s
but t
heir
perfo
rman
ce evaluation studie
s
are
im
mature, so
in
this pap
er
we ad
opt the
new eval
uat
ion
method
s of model
s by usin
g psychologi
cal experim
e
n
tal para
d
igm a
s
the evaluati
on stan
dard of
model
s. Ho
wever, as
an a
r
tificial em
otion en
gine,
th
e emotion
a
l
model pl
ays
a key role in
the
function
of
putting intelli
gen
ce into
machi
n
e
s
. T
heir
perfo
rm
ance evalu
a
tion stu
d
ie
s
are
immature [8].
3.1. Ps
y
c
hological Exper
i
mental Para
digms
Re
cently, the psy
c
hol
og
y experime
n
t
al para
d
igm
s
[9], inclu
d
i
ng obj
ect
reversal,
child
ren’
s g
a
m
bling ta
sk,
delay of
gratification, wi
nd
ows task,
s
e
lf-ord
ere
d
Poi
n
ting (SOP o
r
SOTP), towe
r of Ha
noi o
r
tower
of Lo
ndon Str
oop
(Day
-Ni
ght Stroop
) sto
p
-si
gnal ta
sk, h
a
nd
game, Wi
sco
n
sin
card so
rting
te
st (WCST),
flex
ibl
e
item
sel
e
ct
ion ta
sk (FIST), dim
e
n
s
io
nal
change
card sorting (DCCS
), latent vari
able
analy
s
is and so
on. I
n
order to
m
a
ke preparati
ons
for the follow-up tests, we only focu
s mainly on the Childre
n’s G
a
mbling Ta
sk .
The chil
dre
n
’
s
gambli
ng task a
s
a meth
od is first cre
a
ted by Bech
ara to mea
s
u
r
e the hot
executive fun
c
tion
s. But, due to its com
p
lexity
, the method is
simp
lified and imp
r
oved by Kerr
and Zel
a
zot in two a
s
pe
ct
s. In the task, two de
cks of
cards
are
used within
one
set vertical b
a
r
pattern
s a
nd
the othe
r de
p
u
ty polka
dot
s. H
appy a
n
d
sad
faces ca
n be
se
en o
n
the both
wro
n
g
and rig
h
t sid
e
s of them. But what is
different
is th
at the former one de
ck
of cars is al
wa
ys
accomp
anie
d
by happy fa
ce
s which a
r
e occa
sio
nall
y
coupl
ed
with a
sad fa
ce.
By contrast,
the
latter on
e de
ck
of cars h
a
s
two
ha
ppy face
s o
n
the
oppo
site
side
s. Sometime
s, there
will be
a
few
sad
fa
ce
s
(4, 5,
6 fa
ce
s et
c.).
Ha
ppy an
d
sad
face
s re
present getting
candie
s
or lo
sing
can
d
ie
s sepa
rately. Faces numb
e
r i
s
e
qual to the
n
u
mbe
r
of the
can
d
ie
s. App
a
rently, for
e
a
ch
test
can
only
sele
ct o
n
e
ca
rd
ca
n
we
get
only
one
can
d
y by
cho
o
si
ng ve
rtical
-ba
r
p
a
ttern
card
s.
On the
cont
rary, by ch
oo
sing
pol
ka-d
ot ca
rd
s we
get two
ca
ndie
s
, re
sult
in losi
ng m
o
re
averag
ely, ab
out four, five,
or
six
can
d
ie
s a
r
e l
o
st
by a si
ngle fail
ure.
Co
nseque
ntly, the vertical
cards ove
r
we
ight the polka
-dot one
s in the long r
un. So the researche
r
s tell the
children to g
e
t
as ma
ny can
d
ies a
s
they
can at
the
en
d of the gam
e (un
der th
e con
d
ition of 5
0
sele
ction
s
are
unkno
wn by childre
n in adv
ance).
3.2. Emotional Model
In som
e
literature
s
[10] ,
on the b
a
si
s of
the p
r
o
bability sp
ace, Emotional
cha
nge
p
r
oc
es
s as
a r
a
nd
o
m
p
r
oc
es
s
,
the
s
t
imu
li tr
an
s
f
er
pro
c
e
s
s of it
can
be
seen
as an
availa
ble
descri
p
tion of
the hidden M
a
rkov model (HMM). Its form is:
ˆ
ˆ
ˆ
,,
,
,
NM
0
PA
B
Whe
r
e,
N i
s
the
emotion
a
l dime
nsi
o
n
is the
stim
ulus types;
0
P
ˆ
i
s
the initial
state
prob
ability di
stribution ve
ct
or; the
matrix
B is the
state
transitio
n p
r
o
bability matrix
, whi
c
h i
s
u
s
e
d
to describe of the
stat
e transition probability.
ˆ
A
is the
evaluatio
n v
a
lue
,
it
c
a
n be
d
e
sc
r
i
p
t
as
follow:
*
*
*
*
*
2
*
2
*
2
*
2
*
1
*
1
*
1
*
1
ˆ
ˆ
)
1
(
ˆ
ˆ
ˆ
ˆ
1
ˆ
ˆ
1
ˆ
ˆ
1
ˆ
ˆ
)
1
(
ˆ
ˆ
ˆ
ˆ
1
ˆ
ˆ
1
ˆ
ˆ
1
ˆ
ˆ
)
1
(
ˆ
ˆ
}
ˆ
{
ˆ
N
N
N
N
N
N
ij
N
N
N
a
A
(1)
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
e-ISSN:
2087
-278X
Artificial Em
otion Engine B
enchm
ark Problem
base
d
on Psych
o
log
i
cal Te
st Para
digm
(Wan
g
Yi)
4501
Whe
r
e,
ˆ
is an un
determ
i
ned pa
ram
e
ters,
*
*
2
*
1
*
ˆ
ˆ
ˆ
ˆ
N
π
is a li
miting
probability,
N
M
B
ˆ
is ob
se
rvation
prob
ability m
a
trix. Here, when M i
s
e
q
u
a
l to N. Con
s
i
derin
g of
the matrix
N
M
B
ˆ
is invertible. It can be obtai
ne
d as:
a
b
b
b
a
b
b
b
a
N
B
B
B
B
ˆ
ˆ
ˆ
ˆ
2
1
(2)
Whe
r
e a, b can be expressed a
s
follo
w:
1
,
1
1
1
r
r
N
b
r
N
r
a
Whe
r
e, r is a
tunable p
a
ra
meter.
3.3. Benchm
ark Tes
t
ing
Process
The chil
dren'
s gambli
ng ta
sk i
s
su
btra
ct
ed as
a Ben
c
hmark of the emotion mo
d
e
l. The
partici
pant
s a
r
e note
d
by a
prob
ability-b
a
se
d de
script
ion of the finite state ma
chi
ne mod
e
l. Each
state of the
model
rep
r
e
s
ents a
solitai
re; bet
w
een
adja
c
ent
cards e
dge
s ext
r
actio
n
of
ch
oice
sho
u
ld be di
spatch
ed [11].
For exa
m
ple:
The p
a
rtici
p
ants h
ad ju
st
extracte
d a
first de
ck of
cards,
and
pl
ans to
extrac
t the next. A
t
this
tim
e
, He
ha
s two choi
ce
s to contin
ue to b
e
dra
w
n fro
m
the first decks of
cards a
s
well as to be diverted from anot
her de
puty extract.
In this model
,
if the positi
on of
edge i
s
given a
certain
degre
e
of probability values
)
2
,
1
,
(
,
j
i
P
ij
. These pro
bability values ca
n be cal
c
ulate
d
. Pumping thro
ugh
comp
re
hen
si
ve
pre
-
licen
sing incom
e
a
nd d
r
ew
in
com
e
. Extracted
f
r
o
m
a d
e
ck
of card
s, after a
calcul
ation, th
ere
are fou
r
optio
ns. The
r
efo
r
e
,
each state o
f
the set of edges i
s
)
2
,
1
(
,
i
Deck
E
i
.
The output is:
)
2
,
1
(
,
1
2
1
i
P
j
ij
The deg
re
e is:
)
2
,
1
(
,
2
i
Deck
E
i
And the pro
b
ability value of the output fits in Probab
ility
i
P
Initial
.
S
e
lect
t
he de
ck
s init
i
a
l
prob
ability is only use
d
in the begi
nning
of the task.
Emotional fe
edba
ck i
s
th
e de
ci
sion
-m
akin
g fo
r th
e em
otional
impa
ct. Thi
s
affe
ct
revenu
e extracted
Solitaire. Emotional
feedba
ck
in
emotion
a
l stimulus i
n
ten
s
ity cal
c
ulati
o
n
model is:
)
2
,
1
(
)
(
)
(
)
(
r
t
C
t
G
t
t
I
(3)
Whe
r
e,
)
(
t
I
is th
e emotio
nal
stimulus i
n
ten
s
ity;
)
(
t
G
is
th
e ca
p
i
ta
l pr
io
r to th
e
dr
aw
;
1
,
0
r
is the refuse
d coefficie
n
t , which i
s
a
co
nstant re
pre
s
ented pa
rticipants sensitive ;
is the stimul
u
s
inten
s
ity gain .The po
sitive and neg
ative emotion
s
can h
a
ve different valu
es
,if
1
corre
s
po
ndin
g
positive affect , and if
2
Should be h
e
ld to negative.
Evaluation Warning : The document was created with Spire.PDF for Python.
e-ISSN: 2
087-278X
TELKOM
NIKA
Vol. 11, No
. 8, August 2013: 4498 –
4504
4502
Con
c
e
r
nin
g
o
f
the po
sitive and n
egative
emoti
onal
sti
m
ulus, th
e po
sitive tran
sfer hidde
n
Markov mo
d
e
l calculated
value
s
thro
ugh th
e em
otional
stimu
l
us fo
rmul
a, Re
co
rde
d
as:
])
1
,
0
[
)
(
(
)
(
t
p
t
p
Em
Em
. Get po
sitive em
otions t
h
rou
gh
emoti
onal fe
edb
ack
Em
p
. His
t
oric
al
earni
ng
s pe
r decks nee
d to be cal
c
ul
ated by ev
aluat
e the score
s
of each of the
card
s.
)
(
)
(
)
(
)
(
j
j
j
j
j
j
j
j
j
lossMax
losses
gainMax
gains
lossMax
losses
gainMax
gains
score
(4)
Whe
r
e,
j
gains
a
nd
j
losses
respe
c
ts th
e num
be
r of
obtaine
d p
o
si
tive and
neg
ative
earni
ng
s re
spectively;
j
gainsMax
and
j
lossesMax
re
sp
ects
th
e s
i
n
g
l
e ma
ximu
m
a
n
d
minimum income respecti
vely.
4. Experiment Proce
s
s
4.1. Building and Testin
g
of Platform
The te
st pl
atform
con
s
titu
ted of Intel
Pentium d
u
a
l
-co
r
e
T3
200
. The
progra
mming
langu
age
is
VC++6.0
an
d the
rel
e
vant co
mpilatio
n
configu
r
ati
on. On
this
basi
s
, the
e
n
tire
pro
c
e
s
s of the child g
a
m
b
ling ta
sk i
s
simul
a
ted.
This emotional
decisi
on-m
a
kin
g
process is
decrypted a
s
follows [12]:
Step 1: According to the ini
t
ial probability
i
P
Initial
of a deck of cards;
Step 2: Extract of cards;
Step 3: Calcu
l
ate
)
(
t
G
and
)
(
t
C
;
Step 4: Updat
e
)
(
t
I
;
Step 5: Calcu
l
ate the emotional value
)
(
t
p
Em
;
Step 6: Calcu
l
ate
j
score
;
Step 7: Calculate the probability
ij
P
and no
rmalize
d
;
4.2. The Experimental Re
sults
Based
on th
e above em
otion model,
firstly test
the deg
ree
of versatility in orde
r to
validate the
model. T
h
is
study a
s
sum
e
s th
at 2
0
ch
ildren
were
selecte
d
. Th
e
two
sam
p
le
s
are
sho
w
n in Ta
b
l
e 1.
Table 1. Assu
mptions Solit
aire
Dra
w
s
1 2 3
4
5
6
7
8
9
10
11 12 13
14 15 16
17 18 19
20
Card1
1
1
1
1
1
-1
1
1
-1
1
1 1 1
1 1 1
-1
1 1
1
Card2
2 2 2 -4
2
2
-5
2
2
2
2
2
-6
2
2
2
-5
2
2
2
Figure 1. Emotional Chan
ges
Figure 2. Selection
Cha
n
g
e
s
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
e-ISSN:
2087
-278X
Artificial Em
otion Engine B
enchm
ark Problem
base
d
on Psych
o
log
i
cal Te
st Para
digm
(Wan
g
Yi)
4503
Figure 3. Inco
me Cha
nge
s
Figure 4. Select Statistics
In Tabl
e 1,
p
o
sitive n
u
mb
er m
ean
s tha
t
there
is a
h
appy fa
ce
on
the
ba
ck of t
he
ca
rd;
negative num
ber me
an
s a sad o
ne. The
big or sm
a
ll of its magnitu
de rep
r
e
s
e
n
ts the num
ber of
fac
e
s
.
This
as
sumption is c
o
ns
is
tent
with t
he testin
g re
quiremen
t
s of the
chil
dren’
s
gambli
n
g
task.
The
ex
perim
ental
re
sult i
s
sho
w
n
in Fi
gur
e 1
to 4 from th
e
co
mpute
r
si
mulation
of t
h
e
pro
c
e
ss of
ch
ildren’
s ga
mb
ling task.
Figure 1 sh
ows a positi
v
e emotional
value wi
th the discrete
-time variation
curve,
Positive emot
ional value i
s
calculated
b
y
the HMM e
m
otion mod
e
l
1
,
0
Em
p
, as is sho
w
n in
Figure 2. By
analo
g
p
a
rtici
pants in total
reven
ue
ve
rsu
s
time, it
can b
e
see
n
f
r
om th
ese two
cha
r
ts, with the game ta
sks b
e
ing carried out, ch
ild
ren were
sin
k
ing in em
otional feedb
ack,
accumul
a
ting
and
en
han
ci
ng its revenu
e an
d p
o
sitiv
e
em
otion
s
.
Figure 3
sho
w
s the
drawi
ng
pro
c
e
ss in dy
namic time
sh
are, refle
c
ting
details the choice de
cks.
Figure 4
i
s
chart
of a
pro
c
e
s
s on
the
entir
e
draw
STATS, It ca
n be
see
n
t
hat the
partici
pant
s
more
likely to choo
se
a
third
de
cks,
b
e
ca
use of th
e ove
r
all i
n
come
of the
d
e
cks.
This
simulatio
n
pro
c
e
ss i
s
coinci
de
s with
the real situ
ation.
5.
Conclusion
This
articl
e d
e
scrib
e
s som
e
of the p
s
ychologi
cal te
st
paradigm
an
d an
emotion
a
l mod
e
l,
and ba
se
d on
this model
showi
ng its test Benchm
a
r
k
issue. The ex
perim
ental re
sults
sho
w
th
at
the re
alisti
c
simulate
in t
he e
n
tire
ga
me in th
e e
x
perime
n
t of
Child
re
n’s
Gamblin
g T
a
sk is
effective. The
new
evaluati
on metho
d
with exper
im
ent
al paradig
m
a
s
the eval
uati
on sta
nda
rd
of
model
s is ad
opted suitabl
y.
Acknowl
e
dg
emen
ts
Than
ks fo
r the financi
a
l su
pport
s
from f
und of the project
s
as follo
ws:
1) The "
N
atio
nal Natu
ral S
c
ien
c
e fou
n
d
a
tion of Chin
a" (No. 61
170
117).
2) The "
N
atio
nal 973 Program of
China"
(No.2
011
CB5
0540
2).
3) The "
C
oop
eration Proje
c
t of Prov
inci
al Ministry" (No.20
11A09
0
2000
08).
4) The "Maj
or proje
c
ts of n
a
tional sc
ie
nce and technol
ogy" (No. 20
1
0
ZX071
02-00
6).
Referen
ces
[1]
Bing W
,
Xia
o
li
W
.
Perceptio
n Ne
ural
Net
w
orks for Active Noise
Control S
y
stems.
TEL
K
OMNIKA
Indon
esi
an Jou
r
nal of Electric
al Eng
i
ne
eri
n
g
.
2012; 1
0
(7): 1
815-
182
2.
[2]
Orton
y
A, Cl
or
e G L,
Co
lli
ns
A. T
he cogn
itiv
e st
ructure
of
e
m
otions.
UK: C
a
mbrid
g
e
U
n
iv
ersit
y
Press,
198
8.
[3]
Brea
z
e
a
l
C.
Emoti
on an
d socia
b
le
hu
ma
noi
d
ro
bot
s. Internati
o
n
a
l J
ourn
a
l of
Hu
ma
n-Co
mpute
r
Studies
. 20
03;
59(1-
2): 119-1
55.
[4]
W
e
i Z
hehu
a. Artificial ps
yc
hol
og
y theor
y res
earch
of the e
m
otion
a
l rob
o
t. Beiji
ng Un
iver
sit
y
of Scie
n
c
e
and T
e
chno
log
y
. 20
02
[5]
T
eng less
w
i
nt
er. Artificia
l
em
otion
mod
e
l
us
ed
in
pers
ona
l
rob
o
ts. Bei
jin
g
Univ
ersit
y
of
Scienc
e a
n
d
T
e
chnolog
y. 2
00.
Evaluation Warning : The document was created with Spire.PDF for Python.
e-ISSN: 2
087-278X
TELKOM
NIKA
Vol. 11, No
. 8, August 2013: 4498 –
4504
4504
[6]
Che
n
Dez
h
i, S
hao Y
i
n, W
a
n
g
T
ao. Edd
y
c
u
rrent
testing of steam
ge
ner
at
or tubes B
enc
hmark mo
de
l
numeric
al ca
lc
ulati
on. No
nde
struct
ive T
e
sting. 2000; 2
2
(10)
: 435-43
8.
[7]
W
ang L
i
n
g
. Intelli
ge
nt Optimi
zation A
l
g
o
rith
m and Its Ap
pl
icatio
n. Beij
in
g: T
s
inghua
Univ
ersit
y
Press
.
200
3.
[8]
UC Irvine Mac
h
in
e Le
arni
ng
Rep
o
sito
r
y
[OL]. Http://arc
hive
.ics.uci.edu/ml/
[9]
Ps
y
c
h
o
l
ogic
a
l
Exp
e
rim
ent
s [OL]. Http:/
/
w
ww
.
c
npsy
.
net.
[10]
Sutikno T
,
F
a
cta M. Brain Em
otio
n
a
l
Lear
nin
g
Base
d Intel
l
i
gent C
ontro
ller
.
T
E
LKOMNIKA Indon
esi
a
n
Journ
a
l of Elec
trical Eng
i
ne
eri
n
g
. 9(2): 20
1-2
02.
[11]
W
ang L
i
an
g. Benchm
ark
Su
mmar
y
of Performanc
e T
e
st.
Co
mput
er En
gin
eeri
ng a
nd
Appl
icatio
ns
.
200
6; (15): 45-
48.
[12]
Xi
an G, Z
eng
B, Yun Q, et al. Non-S
peci
f
ic
Person C
o
ntinu
ous Sp
ee
ch
Identific
atio
n in Sec
o
n
d
Lan
gu
age us
in
g BPR.
T
E
LKOMNIKA Indon
e
s
ian Jo
urn
a
l of
Electrical En
gi
neer
ing
. 2
012;
10(7): 16
04-
160
9.
Evaluation Warning : The document was created with Spire.PDF for Python.