Adapting OCC theory for affect perception in

Adapting OCC theory for affect perception in
educational software
George Katsionis, Maria Virvou
Department of Informatics
University of Piraeus
Piraeus 18534, Greece
[email protected]; [email protected]
Educational Software
 Due to the potential benefits of computer assisted learning
there is a growing interest for educational software from
numerous institutions.
 Educational applications have to incorporate as many
reasoning abilities as possible to be educationally
beneficial.
 In order to do that, such software has to be structured in
such a way as to be highly adaptive and individualised to
the needs of each student.
The role of affect in learning
 How people feel plays an important role on their cognitive
processes as well.
 Researchers seeking to create intelligent applications begin
to realise the importance of emotions in attention,
planning, learning, memory, and decision-making.
 There has been an increasing interest in building
emotionally intelligent interactive systems that are
equipped to express emotions or even understand
emotions.
Emotionally intelligent systems
 Implement an emotion model which is responsible for
emotion synthesis (Roseman, Antoniou & Jose 1996;
Sloman 1999; Ortony, Clore & Collins (1988).
 Most of the research about implementing an emotion
theory concerns the use of emotions for making more
lifelike, believable and pedagogical agents for interactive
systems.
 There are some studies that are focusing on detecting
emotions of the users, but are mainly referring to restricted
environments and specific conditions.
OCC Cognitive Theory of Emotions
 The OCC (Ortony, Clore, & Collins, 1988) model has
established itself as the standard model for emotion
synthesis.
 A large number of studies employed the OCC model to
generate emotions for their embodied characters, and lately
there are some studies, like our case, that are using it to
model user emotional states.
OCC Cognitive Theory of Emotions
This model specifies 22 emotion categories based on:
 Reactions of the subject to an event which is relevant to
his/hers goals or someone else goals depending on the
desirability of the event.
 Reactions of the subject to an action of an accountable
agent (including itself) depending on the praiseworthiness
of the action.
 Reactions of the subject to attitudes of attractive or
unattractive objects.
OCC Cognitive Theory of Emotions
 Goal-based emotions: Depend on the desirability of an
unconfirmed, confirmed or disconfirmed event, whether it was
anticipated, and whether it happened to the agent itself or
someone else (joy, distress, hope, fear, satisfaction,
disappointment, relief, fears-confirmed, happy-for, resentment,
gloating, pity).
 Standard-based emotions: Depend on the praiseworthiness of the
action of an agent as the result for an effected standard (pride,
shame, admiration, reproach).
 Attitude-based emotions: Depend on the appealing ness of an
object (love, hate).
The OCC Model
Virtual Reality Game
 The popularity of software games among children and
adolescents may be exploited for educational purposes.
Virtual reality games constitute a very promising mean of
developing more attractive educational applications.
 Such gaming applications, which typically provoke a
wealth of emotions to users, can become an advanced test
bed for affective states.
 The test bed for our research is an ITS for teaching English
orthography and grammatical rules. This ITS operates as a
virtual reality game, and is called VIRGE (Virtual Reality
Game for English).
Virtual Reality Game
 Students have the opportunity to play a 3D game, similar
to the commercial ones, which enables them to learn while
playing. The student must fight his/her way through a
maze by using his/her domain knowledge.
 It is an educational application that models aspects of
student behaviour, by combining evidence from students’
cognitive and detectable behavioural characteristics.
 Recognises evidence about important students’ emotions
and provides appropriate feedback.
Animated agents
 When users interact with a computer, they provide a great
deal of information about themselves.
 Agents have been quite successful at observing users’
behavior and they have been used in learning
environments in order to capture the users’ characteristics
and perform user-modeling tasks.
 VIRGE communicates with the student via three types of
animated agent, the virtual enemy, the virtual advisor and
the virtual companion.
The Virtual Enemy of the Game
The Virtual Advisor of the Game
Animated agents
 The animated agent who acts as an advisor, has the form of
a female angel, and appears in situations where the student
has to read new parts of the theory or has to repeat parts
that s/he appears not to know well.
 The virtual enemy agent is a dragon guard outside every
door of the game, and opposes himself as an obstacle on the
student’s course during the game, by asking questions.
 The virtual companion appears, and makes some remarks in
a casual way as if a friend was talking to the student. He is
responsible for showing empathy to the students and help
them in managing their emotions while playing.
Students’ Goals and Standards while
playing the game
At first we were able to record some of the students goals and standards
while using the virtual reality educational game:
 Goals depending on:
Prospect Relevant Events:
- Being Correct
- Avoid Mistakes
- Finish Quickly
Prospect Irrelevant Events:
- Have Fun
- Avoid Confusion
- Learn
 Standards depending on :
Self Agent Actions:
Other Agent Actions :
- Be a very good student
- Being Helped
- Not being a bad student
- Not Being Disturbed
Connection of Students’ characteristics
with their Goals and Standards
 The student characteristics that are being modelled concern
cognitive characteristics of students (answers’ results - errors) as
well as behavioural actions while learning (user actions), that are
connected to their goals and standards, and can lead to emotions.
 Behavioural characteristics: that are reactions to events or
actions of agents and can be connected to the students’ goals and
standards for the game.
 Cognitive characteristics: that is evidence on the students’ degree
and quality of knowledge of the parts of the lessons that are
examined during the game, and are related to the students’ goals
and standards for the game.
Example of a behavioural characteristic

Speed of answering: The time that it takes to the student to answer a
question. In the case this time is very small it might mean that the student
is having a good time playing and answers quickly to see more, so is
connected to the Prospect Irrelevant goal of the student “Have Fun” and
can lead to the emotion of Joy. For the same case the time is connected to
the Prospect Relevant goal of the student “Finish Quickly” and can lead
to the emotion of Satisfaction. On the opposite case that this time is
really high, then this can reveal the student’s hesitation to answer and is
connected to the Prospect Relevant goal of the student “Avoid Mistakes”
and can lead to the emotion of Fear.

There are more such behavioural characteristics that can be connected to
student goals and standards, and eventually lead to emotions.
Examples of cognitive characteristics
 Error frequency: The percentage of errors among the answers given to
all the questions so far. If the student has made a high degree of errors
for his answers so far, then this characteristic is related to the standard
of the student for Self Agent Actions “Not being a bad student” and
can lead to the emotion of Shame.
 Error persistent occurrence: Consecutive errors for the last questions
or for a specific orthography or grammatical rule. This characteristic
is connected to the Prospect Relevant goal of the student “Being
Correct” and can lead to the emotion of Disappointment.
Adaptation of the OCC Theory
 We were able to identify students’ goals and standards during their
interaction with our educational software.
 Observable behavioural characteristics and measurable cognitive
characteristics were tracked down, that are reactions to events and
actions of agents, either for self-agent or the application’s agents, and are
connected to students’ goals and standards .
 These characteristics are the intensity variables for the desirability of an
event and the praise-worthiness of an action of an agent, so are linked to
the intensity of the corresponding emotions according to the OCC
cognitive theory model.
Adaptation of the OCC Theory
The subsection chosen from the OCC model focuses on the Prospect
Based, Well-Being, and Attribution emotional categories of the original
OCC model. This model refers to 12 of the 22 OCC emotion categories:
 Reactions of the subject to an event which is relevant to his/hers goals
depending on the desirability of the unconfirmed, confirmed or
disconfirmed event, and whether it was anticipated, (joy, distress, hope,
fear, satisfaction, disappointment, relief, fears-confirmed).
 Reactions of the subject to an action of an accountable agent (including
itself) depending on the praiseworthiness of the action of the agent as
the result for an effected standard (pride, shame, admiration, reproach).
Intensity of an Emotion
 The intensity of an emotional state is very important for the
selection of the appropriate advice for the user.
 The combination of such emotional states, either positive or
negative, can provide information about the general mood
of the student (Happiness, Sadness) and lead to affective
computing.
Adaptation of the OCC Model
OCC Theory of emotions
 The OCC theory of emotions suggests that for the purpose
of finding out if an emotion really occurred to an individual
there is a need for the specification of a specific value that
is called the threshold value.
 If the potential value of an emotion is lower than the
threshold value then the individual is not considered to
experience the emotion. Otherwise the intensity of the
emotion experienced is the difference between the potential
value of the emotion and the threshold value.
Using OCC Theory
 We have used the OCC theory to find out which of the
characteristics of the student that are the intensity variables
for the emotional states of each student, have a value that is
significant enough for the emotion to occur.
 The threshold value of each intensity variable is calculated,
by taking into account the mean value and the standard
deviation value of each intensity variable for each
individual student.
 This decision was based on the fact that it is important to
know if an intensity variable (cognitive or behavioural
characteristic) is out of its usual bounds for the student.
Example of the intensity of an intensity
variable leading to an Emotional state
 For example if the value of the behavioural characteristic of speed of
answering in a particular answer is greater than its threshold value
(outside its usual bounds) then the intensity for this intensity variable is:
 Intensity (speed of answering) :=
Value – Threshold_Value (speed of answering).
 The Speed of answering characteristic, which in this case is related to
the goals “Having Fun” and “Finish Quickly”, is linked to the emotional
categories of Joy and Satisfaction. So the intensity of these emotions is
equal to the intensity of the intensity variable.
An Example of advice
 If a student is hasty, which is a result of its Speed of
answering characteristic characteristic intensity, and makes
a lot of errors receives the following advice from the
companion agent:
 “ You seem to be quite anxious about answering. There is
no need to feel that way. Take your time to think before
giving an answer “.
Conclusions
 We have described how evidence from the students’
behavioural and cognitive characteristics may be combined
with the OCC theory of emotions, for drawing inferences for
the student’s emotional state while interacting with the
educational application.
 It is very important for a intelligent tutoring system to be
flexible, depending not only on the learning model of a student
but also on the affective model.
 Affective computing needs a lot of work and evaluations as to
be able to provide safe conclusions.