MANAGING EMOTIONS IN SMART USER MODELS FOR
RECOMMENDER SYSTEMS
Gustavo González, Beatriz López, Josep Lluís de la Rosa
Institut d’Informàtica i Aplicacions, Universitat de Girona, Campus Montilivi - Building P4, E-17071, Girona, Spain.
Keywords: Emotions, recommender systems, smart user models, user modelling.
Abstract: Our research focuses on the development of methodologies that take into account the human factor in user
models. There is an obvious link between personality traits and user preferences - both being indications of
default tendencies in behavior, that can be automated by systems that recommend items to a user. In this
work, we define an emotional component for Smart User Models and provide a methodology to build and
manage it. The methodology contemplates the acquisition of the emotional component, the use of emotions
in a recommendation process and the updating of the Smart User Model according to the recommendation
feedback. The methodology is illustrated with a case study.
1 INTRODUCTION
Emotional Intelligence has been described as an
important part of human decision-making (Goleman,
1995). It has been proved that, at a neurological
level, emotions play a definitive role in the cognitive
process (Joseph, 2001). Emotions should be taken
into account when building a user model since there
is an obvious link between personality traits and user
preferences - both being indications of default
tendencies in behaviour.
There are several approaches to computational
models of emotions. Each one focuses its research
according to its application in particular domains;
some of them are described in (El-Nasr, et al., 1999).
Event appraisal models proposed by (Roseman, et
al., 1990) identify events with emotions. The OCC
model (Ortony, et al.; 1988) provides a taxonomy,
which labels general emotions based on a valence
for the reaction to events and objects. Another
example is the “Six Basic Emotions” model
proposed by (Ekman, 1982) which is based on
research into universal facial expressions. Other
models are based on physiological simulation of
emotions, where each emotion is defined in terms of
the physiological reaction to it (Picard 1997).
(Kopecek, 2001) has defined a user model based
on the personality and emotions, describing it as
dialogue automata with properties of information
systems. This work is supported by another study
by (Green, et al., 2001) on neuronal biophysics and
computation that tackles the matter of how to obtain
the description of the behaviour of a system and how
it can be modified by generating a series of
interactions between the inputs and outputs, while
the internal state of system is changing.
Our research concerns emotion modeling for
recommender systems. Most recommender systems
assist the user in a selection process based on
interest and preferences of a single person or group
of people (Sanguesa, et al., 2000). For doing so,
recommender systems maintains a user model in
which objective features of the products the user has
been interested in the past as well as subjective
features regarding the evaluation of such process are
stored. Then, together with objective and subjective
characteristics, we try to represent the emotional
state of the user in what we call Smart User Models.
Providing user models with an emotional
component, traditional recommender systems can
improve the interaction with the user. Then, if user
requirements regarding a product or service are
satisfied, he/she will probably use the system again.
Moreover, if the answer is complemented with the
experience of having an affective deal with the
lender of the service, the satisfaction degree and
confidence perceived by the user is greater, and so
probabilities of further use of the system (González,
et al., 2002).
187
González G., López B. and Lluís de la Rosa J. (2004).
MANAGING EMOTIONS IN SMART USER MODELS FOR RECOMMENDER SYSTEMS.
In Proceedings of the Sixth International Conference on Enterprise Information Systems, pages 187-194
DOI: 10.5220/0002610901870194
Copyright
c
 SciTePress

In this paper, we introduce our methodology to
build and manage the emotional part of the Smart
User Model. This paper is organised as follows. In
section 2, we provide several definitions regarding
the different representational levels for the Smart
User Model (SUM). Then, we describe in section 3
the methodology proposed to deal with the
emotional component of the SUM. We continue on
section 4 with a case study, and we end in section 5
with some conclusions and discussion.
2 DEFINING EMOTIONAL
FEATURES IN SMART USER
MODELS
A Smart User Model (SUM) is an adaptive user
model, which captures the evolution of the user
regarding his/her emotions (González, 2003). The
SUM should be then an artificial representation of
the user. For achieving it, we distinguish two
representational levels: the computational and the
domain level. Following notation is given to
represent formally the mental features of a user at
the different levels.
2.1 Computational Level
Let be L the set of attributes, which represent the
features, and behaviours of a user at the
computational level composed by three dimensions:
{
}
O
n
O
i
OOOO
aaaaaA ,...,,...,,,
321
=
{
}
S
m
S
j
SSSS
aaaaaA ,...,,...,,,
321
=
{
}
EE
k
EEEE
aaaaaA
l
,...,,...,,,
321
=
A
O
is the finite set of objective attributes of user.
These can be provided by the user or acquired from
any database. Relate the name, age and socio-
demographic information of the user. A
S
is the finite
set of subjective attributes of user. These are the
personal judgment that the user performs according
to her/his impressions, feelings and opinions or an
arbitrary expression of his/her private preferences.
These features can only be acquired through user
interaction with external environment and the
system. Finally, A
E
relates psychological traits and
personality, such as joy, surprise, sadness, anger,
disgust, etc.
Each attribute can take values in a given domain,
using the following notation:
(
)
O
i
O
i
avaluev =
,
(
)
S
j
S
j
avaluev =
,
(
)
E
k
E
k
avaluev =
With those set of attributes, it is possible to
define a Smart User Model as follow:
Definition 1: A Smart User Model, SUM, is the
collection of attributes-value pairs that characterize
the user.
(
)
(
)( )( )
[
]
()()()()
[]
()()()()
[]
EEE
k
E
k
EEEE
S
m
S
m
S
j
S
j
SSSS
vavavava
vavavavaSUM
ll
,,...,,,...,,,,
,,,...,,,...,,,,
2211
2211
=
O
n
O
n
O
i
O
i
OOOO
vavavava ,,...,,,...,,,,
2211
,
In above definition it can distinguish, the
objective component of the SUM,
O
, the
subjective component
S
U
and the emotional
component,
E
U
, as follows:
U
(
)
(
)
(
)
(
)
[
]
O
n
O
n
O
i
O
i
OOOOO
vavavavaU ,,...,,,...,,,,
2211
=
(
)
(
)
(
)
(
)
[
]
S
m
S
m
S
j
S
j
SSSSS
vavavavaU ,,...,,,...,,,,
2211
=
(
)
(
)
(
)
(
)
[
]
EEE
k
E
k
EEEEE
vavavavaU
ll
,,...,,,...,,,,
2211
=
Then, we get the following alternative definition:
ESO
UUUSUM ,,=
The emotional component represents then, the
different moods that the user manifests. Since too
many attributes make the emotional component of
the SUM inoperable for a recommendation process,
we resume the emotional component of the user in a
single value, that we have called the emotional state.
The emotional state allows to know general range
for emotions indicating whether an emoting
individual is feeling pleasant versus unpleasant,
dominating versus vulnerable, and activated versus
quiescent. Such states can be classified in:
-Markedly Negative: This state includes the
affective states or moods typically of a user with bad
humour. As consequence, the suggestions of the
recommender systems have to be carefully studied.
-More Negative: This range of affective states is
a degree more flexible than the first one. In the
same way includes moods with “high sensibility”,
that should be taken into account at the moment of
the recommendations.
-Neutral: Users in these affective states are
doubtful. They don’t crack under pressure but they
may still become anxious, depressed or very nervous
when things become difficult. The users are more
ICEIS 2004 - HUMAN-COMPUTER INTERACTION
188

propensities to receive a wide range of
recommendations than in the previous cases.
Excitatory attributes
Markedly
Negative
More
Negative
Neutral
More
Positive
Markedly
Positive
-More Positive: In this range of moods the user
has a relative self-control. He/she is open to new,
non-expected recommendations.
Price -0.8 -0.3 -0.2 0.2 0.4
Capacity -0.7 -0.2 -0.1 0.1 0.2
Curiosity 0.4 0.5 0.6 0.7 0.8
Food quality 0.3 0.4 0.5 0.6 0.9
Quality/Price relation -0.6 -0.5 -0.4 0.1 0.3
Efficient service -0.8 -0.6 -0.5 0.2 0.3
-Markedly Positive: At this state, any kind of
excitation from the environment, including
unexpected recommendations, are usually
welcomed.
Such emotional information is useful for the
recommendation process, for which we have
developed a mechanism based on attribute activation
and inhibition. Note that the emotional state is
computed as required (see section 3.2 for the
procedure) but it is not explicitly stored at SUM.
2.2 Domain Level
The domain level is the particular environment in
the real world in which the user is modelled. It is
marked by specific characteristics and organization
according to design goals of the software
applications.
Let be D a set of attributes that define a given
domain.Let be
D
A
D
⊂
the set of characteristics,
properties and organization or operation of an item
(object or service) in a given domain D.
{
}
D
p
D
h
DDD
aaaaA ,...,,...,,
21
=
Let be
D
A
I
⊂
a set of interests of a user in
particular objects or services in a domain D.
{
}
,...,,...,,
21
I
p
I
i
III
aaaaA =
Let be
the socio-demographic features of
the user in the domain D, normally introduced in a
“login” procedure.
U
A
{
}
,...,,...,,
21
U
r
U
k
UUU
aaaaA =
Among all the attributes at the domain level, we
distinguish the attributes that represent emotional
connections between the attributes and the emotional
state. They are called excitatory attributes,
D
E
,
{
}
IDD
AAE U⊆
. In each domain, an
activation table that relates emotional states with
excitatory attributes is defined (see Table 1).
Table 1: A possible activation table
Each excitatory attribute, , has an activation
degree,
i
[-1, 1] for a given emotional
state i. A low value close to -1, means inhibition,
that is, the recommender system can ignore it in the
recommendation process. A high value close to 1,
means activation, that is, the recommender systems
should be take especially care of the attribute when
doing the recommendation.
D
e
∈)(
D
eAD
This paper focuses on the definition and
management of emotional features. For details about
the objective and subjective features, see (González,
2003).
3 EMOTIONAL FEATURES
MANAGEMENT
To deal with emotional features of the user we have
defined a methodology based on three stages:
initialisation, update and advice.
The initialisation stage consists in the
acquisition of emotional features of the user to
compound the Smart User Model. The advice stage
proposes a method to help recommender systems to
provide suggestions according to the emotional state
of the user. The update stage consists in the keep
informed the Smart User Model due to the emotional
changes of the user according to the most recent
interactions. So the values of the activation table are
updated.
3.1 Initialisation
The initialisation of emotional features about the
user can be acquired by means of the Emotional
Intelligence Test (EIT) that has a 98% degree of
confidence (Jarabek; 2001). The EIT provide
information about the user which is appropriately
assigned in the different emotional features of the
user at SUM. Thus, the initialisation stage is
conformed by two steps: 1) The emotional
intelligence test and 2) Information distribution.
MANAGING EMOTIONS IN SMART USER MODELS FOR RECOMMENDER SYSTEMS
189

3.1.1 The emotional intelligence test
Table 2: A possible table of relations between
parameters and valences through the moods
The EIT provides a set of parameters from the user,
which can be classified and labelled. Such
parameters are five: Self-conscience; Self-Control;
Goal-Orientation and Motivation; Self-Expression
and Social-ability; and Empathy. Each parameter is
defined in [0, 1] (see Figure 1).
3.1.2 Information distribution
The parameters provide information about the user
from which we wish to compute the current
emotional attributes of the user,
i
a
, of the SUM. In
our model, we define as many emotional attributes
as moods provided by the psychology studies of
(Scherer, 1988). Moods are affective states as anger,
angry, and hopeless. Each parameter of the EIT has
a set of mood related. To know the corresponding
value of each mood (emotional attribute) of the user
from the EIT parameters we perform the distribution
according to the relationship between mood and
parameters shown in table 2.
E
Let be Par the set of parameters, namely:
Par = {Self-conscience, Self-Control, Goal-
Orientation, Self-Expression, Empathy}.
Each parameter p
i
∈
Par has a value, VAL(p
i
).
Let be Mood the set of the all-possible user
moods, Mood =
{
nk321
. A set
of moods is defined for each parameter,
∀ parameter
p
}
)(
mmmmm ,...,,...,,,
i
∈
Par, ∃ a set of moods
i
. At
this step a value of mood
ij
, VAL(
ij
m
), is
computed for each mood of each parameter, that is:
MoodpMod ⊂
m
∀ p
i
∈
Par ; ∀
ij
m
∈
)(
i
pMod
VAL (
) Í VAL(p
ij
m
i
)
Be aware that VAL (p
i
)
∈
[0, 1], so VAL( )
ij
m
∈
[0, 1] too.
At the end of this step, each mood
ij
m
∈
Mood
has a value. Since we define an emotional attribute,
for each mood, we get that , and so
we obtain the emotional component
of the
SUM.
E
i
a
MoodsA
E
=
E
U
Advice
The goal of this stage is to take advantages of the
information of the emotional state of the user,
E
of the SUM, in the domain level.
U
The advice step consists on providing emotional
information to recommender systems in order to
allow the improvement of the recommendations
made to the user. For each domain a set of attributes,
E
D
, will be activated or inhibited depending on the
emotional user state, and accordingly to the
activation table. In order to obtain the emotional
state of the user from the emotional features, we
propose a two-step methodology: Valence
aggregation and labelling.
ICEIS 2004 - HUMAN-COMPUTER INTERACTION
190

Parameter Value
[0, 1]
Overall Score = 86,8
0.368
Self-conscience = 66
0.16
Self-Control = 85
0.35
Goal-orientation and
motivation = 97
0.47
Self-Expression and Social-
ability = 86
0.36
Empathy = 110
0.60
Label Markedly
Negative
More
Negative
eutral More
Positive
Markedly
Positive
N
Figure 1: A sample of the results of the Emotional Intelligence Test
3.2.1 Valence Aggregation
A valence is the degree of attraction or aversion that
a person feels toward a specific object or event.
Possible values of valences range from (+ + ) to (- -).
According to the psychology studies, each mood
(our emotional features) can be labelled with a
valence. For example, eager is +, angry is -. Table 2
shows the complete relationship between moods (on
the cells) and valences (columns). Then, from the
emotional component of the user, we can: 1)
Compute the individual value of each valence; 2)
Compute the global value for all valences. The result
of the second step provides us with a global mood of
the user.
Sub-step 1: Valence computation
At this step, we spread the information of moods
to each valence. For each valence, valence
i,
a set of
moods,
is defined,
)(
j
valenceMod
x
i
[0, 0.2) (0.2, 0.4) (0.4, 0.6) (0.6, 0.8) (0.8, 1]
∀ valence valence
i
∈
Valence, a set of
moods
i
. Then, we
compute the value of the valence, VAL(
)
for each valence ∈ Valence as follow:
∃
MoodvalenceMod ⊂)(
j
valence
∀
j
valence
∀
ij
m
∈
)(
j
valenceMod
VAL (
) =
j
valence
Nm
j=1
mVAL
nj
ij
∑
=
)(
(1)
Where
= Cardinality of ,
and VAL (
valence
)
Nm
)(
j
valenceMod
j
∈
[0,1]
Sub-step 2 : Global mood of the user
At this step, we compute the final value of all the
valences for the user, GlobalMood
,
as
GlobalMood =
Numva
l
valenceVAL
Numvalj
)(
∑
=
j
j
1
=
(2)
Where Numval = Number of valences. The result
GlobalMood is defined in [0, 1].
3.2.2 Labelling
At this step, the global mod value is fuzzyfied in
order to know the emotional state of the user
according to the labels shown in the table 3, which
correspond to categories according to the relatively
temporary state of feelings in the user.
Table 3: Labels for the emotional state
Several membership functions are possible to
define the fuzzy sets. As a start point of our research
and taking into account the computational efficiency
and the posterior discretization of the results we
have chosen trapezoidal membership functions
(MFs). Figure 2 shows the fuzzy values proposed.
After fuzzyfiying the globalmood value of the
previous step, we get the corresponding emotional
state of the user. Such emotional state is used to
access the activation table and know the activation
degree of the emotional attributes. Such emotional
information will be taken into account in a
recommendation process.
3.2 Update
At this stage, the activation degree of excitatory
attributes, E
D
at the domain level are updated
according to the feedback of the recommender
system.
MANAGING EMOTIONS IN SMART USER MODELS FOR RECOMMENDER SYSTEMS
191

A feedback value is provided for each
recommendation, this value is between the [-1,1]
interval. Then, when the excitatory attribute has a
high activation degree, this is updated according to
the following expression:
)1()()( ϕϕ −+=
D
i
D
i
eADeAD
Feedback ;
when
(3)
0)( ≥
D
i
eAD
Otherwise,
)1()()( ϕϕ −−=
D
i
D
i
eADeAD
Feedback ;
when
(4)
0)( <
D
i
eAD
Where
ϕ
is the factor of system evolution
dynamics to reward or to punish the correspondent
excitatory attribute according to the feedback from
the recommendation process (Jonker and Treur,
1999).
The activation equation 3 and 4, guarantee that:
1. Attributes with a high activation degree, with
a positive feedback, will diminish its distance to the
maximum activation degree (value of 1).
2. Attributes with a high activation degree, with
a negative feedback, will become less important in a
recommendation process for the emotional state i.
3. Attributes with a low activation degree, with
a positive feedback, will diminish its activation
degree in the recommendation process for the
emotional state i.
4. Attributes with a low activation degree, with
a negative feedback, will increase the distance to the
minimum value activation degree (value -1).
4 CASE STUDY
In this section, we illustrate with an example the
methodology proposed. We get the emotional
component of the SUM of the user Juan Valdez
®
,
and then use it to recommend to the user a restaurant
by means of a recommender system. Afterwards, the
feedback of the recommendation is used to update
the activation table.
4.1 Initialisation
Markedly
Negative
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
1
0.5
0
More
Negative
Neutral
More
Positive
Markedly
Positive
Fuzzy value of the degree of Global Mood
)(x
µ
Fi
g
ure 2: Membershi
p
functions for the Global Moo
d
Here we illustrate how the emotional component of
Juan Valdez
®
is acquired.
4.1.1Emotional Intelligence Test
Let’s suppose that the Emotional Intelligence Test of
Juan Valdez
®
results are the ones shown in the
Figure 1: (Self-conscience = 0.16, Self-control =
0.35, Goal orientation and motivation = 0.47, Self-
expression and social-ability = 0.36, Empathy =
0.60).
4.1.2Information Distribution
First of all, we distribute the self-conscience value
to all the corresponding moods. That is,
Mod(self-conscience) = {weak, afraid,
anguished, frightened, helpless, scared, confident,
courageous, cowardly, lively, stimulated, happy}.
So, value(weak) = value (afraid) = value
(anguished) = value (frightened) = value (helpless) =
value (scared) = value (confident) = value
(courageous) = value (cowardly) = value (lively) =
value (stimulated) = value (happy) = 0.16
Analogously, we distribute the rest of the
parameters.
4.2 Advice
Here we illustrate how the emotional component of
the user can be applied in a recommender system.
ICEIS 2004 - HUMAN-COMPUTER INTERACTION
192

4.2.1Valence Aggregation
This step is compound by two sub-steps: Valence
computation and Global Mood of the user.
Sub-step 1: Valence Computation
Let’s start with the computation of the (--)
valence. The moods corresponding to this valence,
mood(--), are the following (see table 2):
Mod(--) = {weak, aggressive, desperate, fed up,
intolerant, vengeful, apathetic, dejected, listless,
angry, depressed, sad, unhappy, lonely, offended,
outraged, repelled}
Then, the individual computation of the (--)
valence is performed according to equation (1) as
follows:
VAL(--) = [VAL(weak) + VAL(aggressive) +
…+ VAL(repelled)] /Nm
VAL(--) = [0.16 + 0.35 + 0.35 + 0.35 + 0.35 +
0.35 + 0.47 + 0.47 + 0.47 + 0.36 + 0.36 + 0.36 +
0.36 + 0.60 + 0.60 + 0.60 + 0.60] /17
VAL(- -) = 0.4211
Analogously we can compute the value of the
rest of the valences:
VAL(-) = 0.3924
VAL(- +) = 0.3804
VAL(+) = 0.4119
VAL(+ +) = 0.4062
Sub-step 2: Global Mood of the user
Finally, we compute the global mood of the user
according to equation (2).
GlobalMood = VAL(--) + VAL(-) + VAL(-+) +
VAL(+) + VAL(++)/Numval
GlobalMood = [0.4211 + 0.3924 + 0.3804 +
0.4119 + 0.4062]/5
GlobalMood = 0.4024
4.2.2 Labelling
The global mood of Juan Valdez
®
has been a
positive value = 0.4024. This value corresponds to
the Neutral label as show the Figure 2.
4.2.3 Advice mechanism
Let’s suppose that the excitatory attributes in the
restaurant domain are: price, capacity, curiosity,
food quality, quality/price relation, efficient service.
=
D
E
{price, capacity, curiosity, food quality,
quality/price relation, efficient service}
Remember that our mechanism consists in to
give activation degree between [-1, 1] to all
excitatory attributes. Table 1 summarizes the
activations and inhibitions for the restaurant domain.
Accordingly, we activate or inhibit the excitatory
attributes in relation with the emotional state of the
user performed. Since Juan Valdez
®
has a neutral
emotional state, the following activations and
inhibitions hold:
• Activate: curiosity (0.6), food quality
(0.5)
• Inhibit: price (-0.2), capacity (-0.1),
quality/price relation (-0.4) and efficient
service (-0.5).
With these activations and inhibitions, the Juan
Valdez
®
Smart User Model advises the
recommender system in order to decide more
suitable items in the restaurant domain according to
the current emotional state of Juan Valdez
®
.
4.3 Update
In our example, the recommender system has
suggested to Juan Valdez
®
a restaurant getting the
following feedback from the recommender system:
0.9. We know the global mood of Juan Valdez
®
,
which is Neutral. From table 1, we know the
activation degree for each of excitatory attribute,
which has contributed to the recommendation.
If we take the first excitatory attribute, curiosity,
which has been activated in the recommendation
process, and taking into account that the feedback
has been positive, we update the activation degree of
the curiosity attribute according to equation 3 (with
ϕ
= 0.5):
(
)
=
curiosityAD
i
0.5*0.6 + (1 – 0.5)*0.9 = 0.75
Analogously, we get the new values for food-
quality: 0.70. Regarding price, it has been inhibited
in the recommendation process, so equation 4
applies:
(
)
=
priceAD
i
0.5*(-0.2) – (1 – 0.5)*0.9 = -0.55
MANAGING EMOTIONS IN SMART USER MODELS FOR RECOMMENDER SYSTEMS
193

Analogously, we get the values for capacity (-
0.5), quality_price_rel(-0.65), and efficient_service
(-0.70).
5 CONCLUSIONS
In this paper, we have introduced a new approach to
model the emotional state of the user in what we
define a Smart User Model. The model is based on
attribute-value pairs, in which attributes
corresponding to the emotional component are
computed on the basis of psychological works. This
methodology can be used to both, acquire user
personality traits and deliver the user emotional
features to recommender systems. The methodology
is based on three steps: initialisation, advice and
update. First, initialisation is based on Emotional
Intelligence. Second, the use of the emotional
component of the model in a recommendation
process is achieved through the activation and
inhibition of domain features in a given application.
Finally, the feedback of the recommendation allows
the updating of the connections between applications
that use the Smart User Model and the emotional
state of the user. In addition, we have illustrated the
methodology with case study.
Our next step goes towards integrate our Smart
User Model with the context in where
recommendations are performed (Van de Velde,
1997), (Bianchi-Berthouze and Lisetti, 2002). The
knowledge of the current situation of a user,
combined with the knowledge of his/her Smart User
Model can provide remarkable results in the field of
recommenders systems.
ACKNOWLEDGMENT
This research has been supported through the
Spanish project MCYT DPI2001-2094-C03-01.
REFERENCES
Bianchi-Berthouze, N., Lisetti, C.L. 2002, ‘Modeling
Multimodal Expression of User’s Affective Subjective
Experience’. User Modeling and User-Adapted
Interaction, vol. 12, No. 1, pp. 49-84.
Ekman, P. 1982, Emotion in the human face. 2nd Edition.
Cambridge University Press, Cambridge, UK.
El-Nasr, M.S., Ioerger, T. R. and Yen, J. 1999, ‘PETEEI:
A PET with Evolving Emotional Intelligence’, In
Proceedings of Autonomous Agents Conference ’99,
WA.
Goleman, Daniel. 1997, Emotional Intelligence: Why it
can matter more that IQ. Bantam Books, New York.
González, G., López, B. & de la Rosa J. Ll. 2002. ‘The
Emotional Factor: An Innovative Approach to User
Modelling for Recommender Systems’, In
Proceedings of AH2002 Workshop on
Recommendation and Personalization in e-Commerce,
Málaga, Spain, pp. 90-99.
González, G. 2003. Towards Smart User Models for Open
Environments. M.Sc. Thesis. Institut d'Informàtica i
Aplicacions. Universitat de Girona. Girona, Spain.
Green, S., Hurst, L., Nangle B., Cunningham, P., Somers,
F., Evans, R. 1997, ‘Software Agents: A review’,
Technical Report TCS-CS-1997-06 Departament of
Ciomputer Science. Trinity College, Dublin.
Jarabek, I. (2001). Emotional Intelligence Test, 2nd
Revision, Plomeus Inc. QueenDom.com™ [On-line]
Available at:
http://www.queendom.com/tests/iq/emotional_iq_r2_a
ccess.html
Jonker, C. M. and Treur, J. 1999, ‘Formal analysis of
models for the dynamics of trust based on
experiences’, In Proceedings of MAAMAW’99.
Lecture Notes in AI, vol 1647, Springer Verlag, pp
221–232.
Joseph, S. 2001, ‘AI’, J@pan Inc., [On-line] Available at:
http://www.japaninc.net/article.php?articleID=514
[Accessed January 15
th
2002].
Kopecek, I. 2001, ‘Personality and Emotions - Finite State
Modelling by Dialogue Automata’, In Proceedings of
UM 2001 Workshop on Attitudes, University of Bari,
pp. 1-6.
Ortony, A., Clore, G., Collins, A. 1988, The Cognitive
Structure of Emotions. Cambridge, Cambridge
University Press.
Picard, R. W. 1997, Affective Computing. M.I.T. Press,
Cambridge
Roseman, I.J., Martin S. S. & Paul E.J. 1990, ‘Appraisals
of Emotion-Eliciting Events: Testing a Theory of
Discrete Emotions’ Journal of Personality and Social
Psychology, The American Psychological Association,
Inc. , vol. 59, No. 5, pp. 899-915.
Sanguesa, R., Cortés U., Faltings B. 2000, Workshop on
Recommender Systems, Autonomous Agents 2000,
Barcelona, Spain.
Scherer, K. R., (Ed) 1988, Apendix F. ‘Labels describing
affective states in five major languages’, Facets of
emotion: Recent research, Hillsdale, NJ: Erlbaum, pp.
241-243, [On-line version revised by the members of
the Geneva Emotion Research Group], Available at:
http://www.unige.ch/fapse/emotion/resmaterial/resmat
erial.html
[Accessed May 6
th
2003],
Van de Velde, W. 1997, ‘Co-habited Mixed Realities’, In
Proceedings of IJCAI’97. Workshop on Social
Interaction and Communityware, Nagoya, Japan.
ICEIS 2004 - HUMAN-COMPUTER INTERACTION
194