Mapping User Engagement to States of Affect via an Unobtrusive
Biofeedback Device
A Dynamic Visualization of Real-time Assessment
Anthony Psaltis, Charalampos Rizopoulos and Constantinos Mourlas
Faculty of Communication and Media Studies, National and Kapodistrian University of Athens,
5 Stadiou str, Athens, Greece
Keywords: Biofeedback, Physiological Measurement, Engagement, Affective Interactions, User Evaluation.
Abstract: The elicitation of affect can be regarded as an influencing factor upon a person’s cognition, emotional state,
mood, attention and motivation. It is also recognizable as an inhibited physiological process expressed by
the human brain as induced or suppressed hormonal and neural stimulation that subsequently instigates a
physical and mental level of attentiveness attractiveness or aversiveness. Physical reactions to emotion
causing events and stimuli in affective computing are classified by direct mapping of facial and postural
expressions to corresponding patterns, by using visual and postural observation methods. Despite the fact
that physiological assessment is generally more reliable and less error-prone, a higher amount of research
has been devoted to visual and postural methods due to the greater complexity and specific knowledge
requirements of the former. Concentrating more on the physiological aspect of assessing affect, we have
developed a biofeedback device, sensing reactions instigated by emotion-causing events and results have
been assessed in real-time using suitable visualization methods. In previous attempts to acquire this type of
measurements, human subjects were physically and psychologically impaired by the electrodes and wiring
attachments used for the acquisition of signals and therefore validity of data was to some extent in question.
In order to achieve an uncompromising assessment environment we designed a system that acquires heart
rate and stress measurements via an ordinarily looking computer mouse. Certain combinations of heart rate
precipitation and tonic level / phasic response of stress levels were investigated as reactions to emotion-
inducing events. Corresponding patterns of physiological measurements to a real time affect allocation
model have reached interesting correlations of events with respective states of engagement to an impressive
degree of coincidence.
1 INTRODUCTION
Various studies of emotion and affect ranging from
the areas of psychology (Eysenck, 1982; Frijda,
1986; Strongman, 2003; Scherer, 2005; Eysenck,
2006), clinical physiology (Lykken & Venables,
1971; Hassett, 1978; Venables & Christie, 1980;
Damasio, 1994/2006; Tsatsou, 2006), affective
computing (Picard, 1997), and HCI (Norman, 2004;
Brave & Nass, 2008; Fairclough, 2009; see also the
vast literature on User Experience [UX] design and
game studies – e.g. Kuniavsky, 2010; Calleja, 2011;
Law & Sun, 2012 etc.) have demonstrated that
physiological stimulations of the human brain
elicited by emotional events are direct, instant,
measurable and quantifiable by using fMRI, PET,
electrodermal response, respiration, heart rate or
combinations of the above. Previous work
addressing this issue (e.g. Ark et al., 1999) required
the user to be impaired by wires and electrodes. This
requirement did not only restrict the user’s posture
and mobility, but also had important psychological
consequences, thus affecting the validity of the
measurements. In the implementation described in
this paper, users were completely untethered, since
they only had to use an ordinary computer mouse in
the usual manner; therefore, they were unaffected by
such limitations.
We assessed the concurrent excitation of two of
these quantities, namely electrodermal response
(skin conductance [SC]) and heart rate (HR), via a
specifically designed interface providing strong
emotional stimulation in an attempt to answer the
following questions:
307
Psaltis A., Rizopoulos C. and Mourlas C..
Mapping User Engagement to States of Affect via an Unobtrusive Biofeedback Device - A Dynamic Visualization of Real-time Assessment.
DOI: 10.5220/0004726303070314
In Proceedings of the International Conference on Physiological Computing Systems (PhyCS-2014), pages 307-314
ISBN: 978-989-758-006-2
Copyright
c
2014 SCITEPRESS (Science and Technology Publications, Lda.)
What is the interpretation of the psychosomatic
state of the subject when both heart rate and stress
level are increasing simultaneously?
Is there a correlation between events that occur
during the interaction with the system and the
aforementioned quantities?
Would the combinations provided by the two
states of each of the measured quantities show a
predictable pattern in response to certain emotional
events?
Could this system produce an adequately reliable
mechanism to map a psychosomatic condition onto
a descriptive model that would ideally be in
accordance with one of the established models of
affect, namely Russell’s (1980, 2003) circumplex
model?
Finally, can this system be used efficiently as a
component of an intelligent interface?
The structure of this paper is as follows: a brief
description regarding the biofeedback principles, our
purpose built electronic device and our system setup
is given. Subsequently, the experimental design
methodologies adopted and the development of the
front-end interface are outlined. Presumptions,
limitations, and predictions are clarified in detail in
the psychosomatic analysis section, followed by a
conclusive interpretation of our experimental data
and suggestions as to how the system may be
improved and used in broader areas of interactive
applications.
2 BIOFEEDBACK ACQUISITION
Derived from basic principles of biofeedback
measurements (Venables & Christie, 1980),
psychological stress is expressed in the human body
by a physiological response of brain induced
vasodilatation of sweat glands of the skin. This
alteration causes measurable changes in electrical
skin conductance that produces quantifiable
indicators of stress levels. Our purpose-built
electronic device used for the acquisition of HR and
stress measurements (SC) has been presented in
previous publications of the authors (Psaltis,
Mourlas, 2011). HR is acquired based on the
principle of Infrared Spectroscopy, effectively using
as sensing elements two deflecting infrared sensors
acquiring HR pulses, for optimised error
cancellation and a more reliable reading. For stress
measurement the skin conductance method has been
adopted as opposed to skin resistance, thus
providing a more reliable indication of the measured
quantity by eliminating inaccuracy problems due to
perspiration on the part of the subject.
Stress level is identified by two silver-silver
chloride contact rings measuring skin conductance,
placed on the sides of a computer mouse and at the
points where the thumb and middle fingers are
resting during the typical use of the mouse by a user.
The two HR sensors are situated in the centre of the
SC rings. Hardware redundancy is embedded in the
system allowing for two channel acquisition of
signals of HR and EC simultaneously, eliminating
artefacts and discontinuities caused by movement of
the fingers unless contact with the mouse is
completely lost. In case of lost contact the system
preserves the trait values until the next valid
measurement is detected. Minute losses of contact
with the sensors or erratic movement of the fingers
onto the mouse could produce an error in HR
reading in the scale of milliseconds between pulse
readings, while the SC measurement is unaffected.
Considering that our system does not take into
account detailed pulse shape and cardiac
arrhythmias, but is instead designed for detecting
heart rate and heart rate variability, this error is
within acceptable margins of tolerance of the
system, as it produces an estimated attenuation of
±0.18 of a pulse per reading. The mouse, including a
signal preconditioning circuit, is connected to a
computer via an electronic interfacing circuit that
filters and conditions the above primary signals and
converts them into a form that can be read by the
computer via the two channel audio input. A
purpose-built software suite comprising the
appropriate components required for a system
configuration console, signal processing and display
is the final component of the system. The use of this
console is essential for the initial settings required,
such as time interval between measurements, audio
card selection and settings, as well as data storage
configuration.
HR detection and correction, as well as SC auto-
calibration algorithms, have been implemented,
providing a relative baseline for each subject
independent of the actual stress levels of individuals.
Since our interest is not exactly how much stress is
the user experiencing but instead the state of stress
in relation to the previous level of stress, we devised
a method for auto calibration, envisaging our
reference point (“baseline”) at the mid distance of
the difference between highest and lowest measured
stress value weighed by the trait values. The baseline
is continuously updated based on the measurements
and the deviation from the mean value of SC (“tonic
level”). This was important primarily because it
PhyCS2014-InternationalConferenceonPhysiologicalComputingSystems
308
eliminated problems like the need for initial
calibration and similarly alleviated continuation
inconsistencies during measurements once the user
released the mouse momentarily. Moreover, as an
additional threshold, the baseline provided a method
of distinguishing additional details in the attributes
of our measurements, effectively indicating the zero
point of transition during instantaneous reactions of
the users (“phasic response”).
3 ASSESSMENT STUDY
3.1 General Description
and Hypotheses
The primary objective of our research was to assess
the conditions in which a certain physiological
response is indicative of a strong emotional
stimulus. In our system, given that we had two
important quantities (HR and SC), a reasonable
assumption was to examine conditions where both
physiological quantities measured showed a
common tendency compared to their previous
measurement and also follow the same directional
pattern, i.e. both either increasing or decreasing at
the same time. From the two measured quantities
(HR and SC), the two main system parameters,
namely the SC and the HR gradient, are derived. The
SC gradient refers to the vector form of SC
measurements in relation to the baseline, whereas
the HR Gradient represents the relationship between
two consecutive HR measurements, weighed by a
factor of heart rate variability (HRV). As such, the
gradient value indicates the tangent of the curve that
depicts the user’s psychosomatic state. A gradient
can be positive (i.e. raising pattern with regard to
previous value), negative (i.e. falling pattern when
the new value is lesser than the previous) or zero
representing a flat pattern essentially indicating no
change of the previous and current quantities.
The algorithm used for the classification of
tendencies on each response acquired by the
biofeedback device is illustrated in figure 1.
Evidently, a high precision and detailed
assessment of an exhaustive range of psychosomatic
states of a human subject was not our aim;
furthermore, this is beyond the capabilities of our
system. In our attempt to design an innovative
system, we developed a tool that would be able to
derive reliable indices of alertness as it is
experienced by the user.
The algorithm used for the visual representation
of the active response to strong emotional stimuli
may be seen in figure 2.
Figure 1: Tendency classification algorithm.
Figure 2: Allocation algorithm.
We classified the users’ behavioural patterns
measured via our experimental platform into four
areas corresponding to four dominant states of
engagement, which in turn we hypothesized as
corresponding to the quadrants of Russell’s
circumplex model of affect. It should be noted that
no single gradient is correlated to either dimension
of the circumplex model (i.e. arousal, valence);
rather, data are mapped to the four quadrants in
relation to the states defined below.
State of Focused Involvement / Engagement
(positive arousal and valence), where the user is
satisfied for fulfilling the task successfully.
State of Contentment (negative arousal, positive
valence), where the user is unable to fulfil a task
but maintains a high level of activation.
State of Perceived Difficulty (positive arousal,
MappingUserEngagementtoStatesofAffectviaanUnobtrusiveBiofeedbackDevice-ADynamicVisualizationof
Real-timeAssessment
309
negative valence), where the subject’s focus on a
task does not change significantly, while
satisfaction diminishes.
State of Non-involvement / Apathy (negative
arousal and valence), where an uninterested and
inattentive person performs a task in negative
valence and arousal levels.
We adopted these patterns of interpretation so
that the stimuli produced by our experimental
scenarios are a more realistic representation of the
users’ psychosomatic state compared to attempting
to infer emotion in detail. We reached this decision
because our system does not have the necessary
accuracy to detect specific emotions with precision. .
During the experiments, processed data were
displayed on-screen in real time, representing user
responses mapped as loci onto a screen area
subdivided into four quadrants. Each of those
quadrants dynamically represented the states
described above as interpreted by the indices
produced by measurements of HR and SC. One
locus point was created every two seconds. The
relationship between the measurements and the
pattern of the projected mapping is shown in figure
3.
Figure 3: Mapping of measurements.
4 DESCRIPTION OF
EXPERIMENTAL PROCEDURE
Facing the interesting challenge of designing a front-
end that would take full advantage of the capabilities
of our system to perform emotional assessment in a
thorough and convincing manner and at the same
time explore to an optimal degree the latest concepts
of emotion-inducing techniques in user interaction,
we came up with a visual environment, in which all
stimulating events and reactions of the user, as well
as processed data from our system, were displayed
in real-time. A recording of data, as well as factors
and parameters used for generic data transformation,
were also produced for tracing and post-processing.
Thus the assessment and verification of the
effectiveness of our system was facilitated by the
ability to reproduce as well as observe user
responses at a later stage.
For the experiment, an ordinary desktop PC
(Intel Dual Core processor, 2 GB memory) was
used. The experimental sessions took place in a
computer lab at the University of Athens, Greece,
under stable environmental conditions (approx. 20
degrees Celsius) for all subjects.
Each session of the experiment included two
parts with a total duration of twenty minutes. The
first part consisted of a puzzle game (“Liquid
Measure”, available at http://www.friv.com/). The
game was intended to elicit positive affect as a result
of successful task completion, and negative affect in
cases in which the subject was unable to progress. It
provided an unknown environment with tasks of
progressively increasing difficulty. The first part had
a duration of 10 minutes. No prior instructions were
given; thus, subjects needed to concentrate and
improvise in order to complete each of the twelve
levels of the game. Not all levels were completed by
the participants, as the time limit was not sufficient
in most cases.
The second part consisted of a video of motor
traffic accidents found on Youtube (available at
https://www.youtube.com/watch?v=26gTlQ1FDW4).
It was intended to elicit instinctive physiological
responses to negatively valenced stimuli (essentially
indicative of distress and fear). Some of the
accidents were predictable, whereas others occurred
suddenly. Two separate categories of emotion-
inducing stimuli were included in this video. The
first category consisted of accidents occurring at a
distance, while the second category consisted of
accidents (or near misses) involving the car on
which the camera was located. These two categories
of stimuli were assumed to be evaluated differently.
More specifically, accidents that occurred further
away were assumed to be evaluated as less
threatening compared to accidents occurring
virtually at a first-person view.
In addition to the physiological measurements
obtained, participants were able to express their
estimate of psychosomatic condition at any time by
selecting one of the four states displayed on the
screen throughout the duration of the test session.
It should be noted that additional information
PhyCS2014-InternationalConferenceonPhysiologicalComputingSystems
310
regarding user preferences on pictorial or textual
assessment or psychological profiles was neither
requested nor taken into account in this study. Apart
from demographics (age and gender), the only type
of personal information requested from the subjects
concerned their driving experience.
The sample (N=17) consisted of university
students ranging from 1
st
-year to masters students
(10 women and 7 men). The age ranged from 18 to
47 years of age with a mean value of 26.2 years. The
participants had no prior knowledge of our
experiment setup, were free to decline participating,
and also retained the option to withdraw or refuse to
complete a part at any point during the experiments.
Two subjects were not included in the final
assessment: one female user was inconsistent as far
as the contact with the computer mouse is concerned
and one male participant was very talkative during
the experiment. Although the consistency of the
results obtained from the latter user was better than
83%, he exhibited respiration arrhythmias that we
judged it could have affected stress measurements
considerably; therefore, this participant was
excluded from analysis.
5 DATA ASSESSMENT
AND EVALUATION
Data analysis was performed for each individual
subject in three steps. First, the visual content (both
game and video) was weighed according to the
perceived emotional impact of each particular event,
providing a table of predicted state of engagement,
event number, and timestamp. The participants’
response to the emotional stimuli was then assessed
and the predicted state of engagement for each
specific event was compared to the state of
engagement measured by the system. The accuracy
rate for each participant was determined by the
degree of similarity between the predicted and
observed states of engagement during the course of
the experiment.
In fact, the correlation of the corresponding
values of emotional intensity was assessed in
comparison with values acquired as responses from
our system. For each subject, data were analysed in
order to formulate a distribution table that provided
a more extensive indication of the convergence or
divergence of emotional patterns derived from our
system, thus deducing its accuracy expressed as a
percentage.
During the development of the interface, we
integrated data post-processing algorithms and
produced results in their final form ready for
analysis. For example, instead of ending up after
each experiment with a vast amount of raw data that
then had to be transformed into a meaningful form,
we integrated display of processed data proactively.
As such, the outcome of post-processed data was
presented to the users in real time. At the same time,
detailed data recordings allowed for further
exploration and tracing as required. Data obtained
from all 15 subjects are presented in Table 1.
Table 1: Detailed data presentation.
Subject
ID
Age
No of Events
Response
(Average)
Female Male
Game Video
F18a 18 42 51 91.38% 91.38%
F18b 18 40 62 98.13% 98.13%
F18c 18 48 69 89.08% 89.08%
F23a 23 45 55 94.23% 94.23%
F23b 23 38 60 87.76% 87.76%
F24 24 44 60 88.91% 88.91%
F25 25 43 57 89.40% 89.40%
F27a 27 43 70 92.22% 92.22%
F27b 27 40 67 94.68% 94.68%
M32a 32 44 53 95.41% 95.41%
M32b 32 44 59 90.26% 90.26%
M35 35 48 55 86.74% 86.74%
M37a 37 46 60 91.33% 91.33%
M37b 37 43 51 90.61% 90.61%
M47 47 41 49 94.71% 94.71%
Median 27 43 59 91.33% 91.38% 90.97%
In an additional test conducted with a subset of the
above participants whereby the users were requested
to look away and divert their attention away from
the computer, we observed that values mapped onto
the four quadrants representing the four states of
engagement changed to the states of Perceived
Difficulty and Non-involvement with an accuracy of
98.1%. As soon as the users redirected their line of
sight back to the computer, their state of engagement
returned to the positively valenced state of Active
Involvement.
MappingUserEngagementtoStatesofAffectviaanUnobtrusiveBiofeedbackDevice-ADynamicVisualizationof
Real-timeAssessment
311
6 FINDINGS OF EXPERIMENTS
6.1 Data Interpretation
Data from the experiments indicated a high degree
of coincidence (91.35%) between optimally assigned
values and those produced by the system with very
few No-Reading errors. Users described their
psychosomatic state by selecting manually one of
the states displayed as follows: ‘Focused
Involvement / Engagement’ 71%, ‘Contentment’, 9%
‘Perceived Difficulty’ 13%, ‘Non-involvement’ 7%
of the time of the experiment. The deviation between
the values obtained through self-report and those
obtained through measurements was approximately
17%, although it is worth noting that users were not
inclined to use the manual selection feature very
often. Positive user involvement during the
experiments is represented in our model by the states
of Focused Involvement and Contentment. The
opposite states were actually mapping either the
state where users were changing level of difficulty
between scenarios or while they spent time waiting
for the next event with diminished or diminishing
involvement. Overall results classified all users in
the above four cases at 44.2% (Focused
Involvement), 31.9% (Contentment), 12.6%
(Perceived Difficulty), and 11.3% (Non Involvement
/ Apathy) of the time of the experiment. Detailed
data regarding allocation per quadrant and gender is
shown in table 2.
Table 2: Mapping distributions per quadrant and gender
(AI = Active Involvement, C = Contentment, PD =
Perceived Difficulty, NI = Non-involvement).
Emotional Mapping Distribution (%)
AI C PD NI
Male 44.6 32.2 11.6 11.1
Female 43.9 31.6 11.0 14.2
Mean 44.2 31.9 11.3 12.6
The difference in responses between male and
female participants was small (<1%), although
event-to-response evaluation has shown coincidence
between 87.76% for female and 98.13% for male
users. Coincidences in the game task were similar
for both male and female participants. However,
differences were observed in the video session. In
our view, this small difference has its origins in the
profiles of male participants being affected more by
aversive driving experiences (age of 32-47 with
driving experience) than female participants (age 18-
27 with little or no driving experience whatsoever).
Differences in both male and female participants
between gaming and video sessions indicated less
accurate event / response matching during the
gaming session data than that during visual
observation, which was more accurate by 4.1%. This
is explained by the fact that, during the gaming
session, users had to dislocate their fingers from the
mouse irregularly, although during the video session
the contact of their fingers with the sensing elements
of the mouse was uninterrupted and the quality of
biofeedback acquisition was therefore nearly ideal.
A caption of a typical real time model
representing the mapping during an experiment is
shown in figure 4.
Figure 4: Real time visualisation of data.
The four numbers in the periphery of the graph (in
figure 4 the numbers 45, 251, 117 and 121
respectively) represent the actual number of
occurrences of the biofeedback data acquired from
the participant and classified by our system. The
number in the centre of the graph is the index of
measurement helping to identify the exact time and
event during post data assessment. At the top of the
graph Tonic Level, Phasic Response as well as
Arousal, Stress Level Baseline and Gradients were
displayed on each cycle of measurements. The latest
value is indicated with a characteristic border that
fades out when the next value is mapped. The
relationship of each measurement to the baseline in
each quadrant can be thought as the imaginary
diagonal axis drawn from the centre of the graph
towards the corners of the rectangle. This additional
information may in future development be scaled so
as to provide a more detailed analysis of the
psychosomatic state of the user into eight
subdivisions rather than the existing four.
PhyCS2014-InternationalConferenceonPhysiologicalComputingSystems
312
7 DISCUSSION
State of engagement was the most important psycho-
physiological condition we attempted to identify and
quantify in this research as it is highly regarded in
Affective Computing research (e.g. Picard, 1997).
Results have verified our views that user
engagement transitions between affective activation
states can be reliably detected by simultaneous
changes in HR and SC gradients. Indication is
immediate as expressed by previous research work
assessing similar aspects by means of fMRI and SC
(Tsatsou, 2006). We may conclude that our system
produces a valid and accurate snapshot of the user’s
state of engagement because, as reported earlier,
whenever the users looked away or otherwise
disengaged their attention from the screen, the
system immediately detected this shift of focus
correctly by labelling their state of attention as either
“Perceived Difficulty” or “Non-involvement” – both
of which are negatively valenced when represented
as coordinates for the dimensions of Russell’s
circumplex model of affect.
Throughout the entire experimental process, a
transition from the state of Active Involvement to
that of Contentment immediately after stressful
stimuli was most frequently observed. This
transition is interpreted as the effect of the slow
decay of the stress level, which produces a negative
SC gradient, effectively indicating a reduction of
arousal.
Taken together, the aforementioned points
indicate that a mapping of the users’ state of
engagement onto Russell’s circumplex model of
affect is accurate, at least with respect to placing the
identified state in the correct quadrant (i.e. positive /
negative valence, positive / negative arousal).
From the above, as well as the overall results of
the experiments, we have indications from our
system that the correlation of common gradients of
present and past values of HR and SC shows high
probability of success in determining various states
of engagement. The time interval for each
measurement was crucial for the accuracy of our
system and optimised accordingly in order to
provide enough time for the subjects’ physiological
response to settle into a detectable timeframe.
Additionally, this timeframe allowed for the
normalization of artefacts that could have been
misleadingly accounted as spontaneous reactions of
the participant.
The time frame was chosen following
optimisation deduced by the assessment of data
produced by added measurements into a larger
buffer of data. Time over 2 seconds has shown a
smoother transition between states but also slowed
down the detection of user response and therefore
rejected.
8 CONCLUSIONS AND FUTURE
WORK
Beginning with the somewhat simplistic but sensible
assumption that coinciding gradients of HR and SC
may express some relation to an emotional state, we
assessed the possibilities of detecting basic cognitive
processes such as engagement, which may be
indicative of underlying emotional states. Results
from our experiments have shown that a correlation
exists; however, we are not clear as to the exact
emotion expressed. It is possible that the excitation
state detected and interpreted by our system as a
state of engagement is in fact another possible
combination of emotional state producing similar
measurements, simply coinciding with the
interpretation of our algorithm. Although our system
at this stage is rudimentary for detailed mapping of
various emotional states such as frustration, sadness,
depression etc, we believe that our system can be
used effectively and efficiently as a component of an
intelligent interface for detecting the user’s degree of
involvement in various applications (e.g. educational
assessment, distance learning etc.); however, future
research may lead to further improvements in the
identification of more detailed psychosomatic states.
An additional avenue we are planning to explore
is the validation of the output of our system with the
aid of standardized validation methodologies and
instruments, such as IADS and IAPS (Bradley &
Lang, 2007), as well as appropriate emotion
recognition software (e.g. facial emotion
recognition). Furthermore, we are planning to
implement a visual pupil size detection component,
which is expected to increase the validity of our
system with respect to focusing intensity.
Our research may be applicable to fields such as
e-learning, educational assessment, virtual
environments, and further areas requiring remote
assessment of user psychosomatic condition.
REFERENCES
Ark, W. S., Dryer, D. C., Lu, D. J. (1999). The Emotion
Mouse. In Proceedings of HCI International (the 8th
International Conference on Human-Computer
MappingUserEngagementtoStatesofAffectviaanUnobtrusiveBiofeedbackDevice-ADynamicVisualizationof
Real-timeAssessment
313
Interaction) on Human-Computer Interaction:
Ergonomics and User Interfaces – Volume I, pp. 818-
823.
Bradley, M. M. & Lang, P. J. (2007). The International
Affective Picture System in the Study of Emotion and
Attention. In J. A. Coan and J. J. B Allen (eds.),
Handbook of Emotion Elicitation and Assessment.
Oxford and New York: Oxford University Press, pp.
29-46.
Brave, S. & Nass, C. (2008) Emotion in Human-Computer
Interaction. In Sears, A. & Jacko, J. A. (eds), The
Human-Computer Interaction Handbook:
Fundamentals, Evolving Technologies, and Emerging
Applications. Lawrence Erlbaum Associates, New
York and London, 2
nd
edition, pp. 77-92.
Calleja, G. (2011) In-game: From Immersion to
Incorporation. MIT Press, Cambridge, MA.
Damasio, A. R. (1994/2006) Descartes’ Error: Emotion,
Reason, and the Human Brain, Vintage, London.
Eysenck, M. (1982) Attention and Arousal. Springer-
Verlag, New York.
Eysenck, M. W. (2006) Fundamentals of Cognition.
Psychology Press, Hove and New York.
Fairclough, S. H. (2009) Fundamentals of Physiological
Computing. International Journal of Human-
Computer Interaction, 21, 133-145.
Frijda, N. H. (1986) The Emotions. Cambridge University
Press, Cambridge.
Hassett, J. (1978) A Primer of Psychophysiology. W. H.
Freeman, San Francisco.
Kuniavsky, M. (2010) Smart Things: Ubiquitous
Computing User Experience Design. Morgan
Kaufmann, Burlington, MA.
Law, E. L. C. & Sun, X. (2012) Evaluating User
Experience of Adaptive Digital Educational Games
With Activity Theory. International Journal of
Human-Computer Studies, 70, 478-497.
Lykken, D. T. & Venables, P. H. (1971) Direct
Measurement of Skin Conductance: A Proposal for
Standardization. Psychophysiology, 8, 656-672.
Norman, D. A. (2004) Emotional Design: Why We Love
(or Hate) Everyday Things. Basic Books, New York.
Picard, R. W., 1997. Affective Computing. MIT Press,
Cambridge, MA.
Psaltis, A. & Mourlas C., 2011. A Real Time Attachment
Free, Psycho physiological Stress and Heart Rate
Measurement System. 1, 1-13.
Russell, J. A. (1980) A Circumplex Model of Affect.
Journal of Personality and Social Psychology,
39,1161–1178.
Russell, J. A. (2003) Core Affect and the Psychological
Construction of Emotion. Psychological Review, 110,
145-172.
Scherer, K. R. (2005). What Are Emotions? And How Can
They Be Measured? Social Science Information, 44,
695-729.
Strongman, K. T. (2003) The Psychology of Emotion:
From Everyday Life to Theory. John Wiley & Sons,
Chichester, UK, 5
th
edition.
Tsatsou, K. (2006) A Study of Electrodermal Response to
Bodily Evoked Potentials. Masters Dissertation,
University of Patras, Greece [In Greek]
Venables P. H. & Christie, M. J. (1980) Electrodermal
activity. In Martin, I. & Venables, P. H. (eds),
Techniques in Psychophysiology, New York, John
Wiley & Sons, pp. 2-67.
PhyCS2014-InternationalConferenceonPhysiologicalComputingSystems
314
