30 M. Leontidis, C. Halatsis, and M. Grigoriadou 2.2 Emotions, Mood and the OCC Model Although many efforts have taken place there is not an explicit definition for the emo- tion. It is easy to feel, but it is hard to describe it. According to Scherer [19], emotion is the synchronized response for all or most organic systems to the evaluation of an external or internal event. Nevertheless, various attempts have been made, but the cognitive theory of emotions, known as OCC model, which formulated by Ortony, Clore and Collins [16], keeps a distinctive position among them. The three authors constructed a cognitive theory of emotion that explains the origins of emotions, de- scribing the cognitive processes that elicit them. The OCC model provides a classifi- cation scheme for 22 emotions based on a valence reaction to events, objects and agents. Events are situations which are interpreted by people in a certain way. Objects are material or abstract constructions. Agents can be human beings, animals, artificial entities which represent humans or animals and software components which act in a specific way. The origin of emotions relate to the subject’s perspective against Goals, Standards, and Attitudes. The events are evaluated in terms of their desirability, ac- cording to the goals of the subject. Standards are used to evaluate actions of a subject and objects are evaluated as appealing depending on the compatibility of their attrib- utes with subject’s attitudes. Emotion is analogous to a state of mind that is only momentary. Mood is a pro- longed state of mind, resulting from a cumulative effect of emotions [19]. Mood dif- fers from the emotion because it has lower intensity and longer duration. It can be consequently considered that mood is an emotional situation more stable than emo- tions and more volatile than personality. Based on this definition we categorize mood into two categories named, positive and negative. We consider that the student has either a positive mood when he feels emotions like joy, pride, hope, satisfaction, grati- fication, love, or a negative mood when feels emotions like sadness, fear, shame, frustration, anger, disappointment, anxiety. Depending on this mood we speculate the possible emotions of the student. In our work we adopt the OCC model, because it elicits the origin of emotions un- der a cognitive aspect and it is possible to be computerized. So, based on this model we are able to classify and interpret a student’s emotions in the learning process. The authors of the OCC model consider that it could be computationally implemented and help us to understand which are the emotions that the human beings feel, and under which conditions. Furthermore, they believe that relying on this model we could pre- dict and explain human reactions to the events and objects. This is the main reason we use the OCC model in our study. The perspective by which, we construct the follow- ing component is interdisciplinary and focuses in the intersection of Artificial Intelli- gence and Cognitive Psychology. 3 The Architecture of the MENTOR The Adaptive Educational Systems (AES) are intelligent systems that improve a stu- dent’s performance by adapting their operation according to his needs and interests and by supporting them with the appropriate learning strategy. An AES interacts dynami- cally with the student, using adaptation techniques like adaptability and adaptivity. It e-Learning Issues under an Affective Perspective 31 uses knowledge about the student (user model), in combination with specific knowl- edge (domain knowledge), to achieve through a set of pedagogical rules (teaching model), the adaptation of the system via the adaptive engine [3]. Thereby, an AES determines the educational content and the teaching process in a way that it appertains to teaching in a real classroom. In the real educational process, the teacher takes into consideration the emotional state of his student by motivating him effectively and achieving thus, the desirable learning goals. Consequently, the investment in individual differences and the emo- tional “potential” of the student in combination with his cognitive abilities could be a significant factor, so that the learning goals can be achieved more efficiently, from a pedagogical aspect of view. Many researchers have demonstrated the pedagogical value of emotions and personality and have incorporated this perception in their edu- cational systems [2], [5], [7]. Fig. 1. The architecture of the Mentor MENTOR is an “affective” module which aims to recognize the emotions of the student during his interaction within an educational environment and thereafter to provide him with a suitable learning strategy. The operation of MENTOR is based on the FFM [14] and the OCC model [16]. The module is being attached to an Educa- tional System providing the system with the essential “emotional” information in order to determine the strategy of learning in collaboration with the cognitive infor- mation. The architecture of MENTOR is presented in Figure 1. The MENTOR has three main components: The Emotional Component (EC), the Teacher Component (TC) and the Visualization Component (VC), which are respec- tively responsible for: a) the recognition of student’s personality, mood and emotions during the learning process, b) the selection of the suitable teaching and pedagogical strategy and c) the appropriate visualization of the educational environment. The com- bined function of these components “feeds” the AES with the affective dimension optimizing the effectiveness of the learning process and enhancing the personalized teaching. The main purpose of MENTOR is to create the appropriate learning envi- ronment for the student, taking into account particular affective factors in combination 32 M. Leontidis, C. Halatsis, and M. Grigoriadou with cognitive abilities of the student offering in this way personalized learning. In the next two sub-sections the Emotional and the Teacher Components are being analyzed in more details. The analysis of the operation of the Visual Component is beyond the scope of this paper. 3.1 Recognizing the Emotions of the Student The necessity of recognizing the student’s emotion during the learning process, espe- cially in distant learning environments is crucial and has been pointed out by many researchers in the e-learning field. Because of this need, many methods have been proposed with the aim of recognizing or predicting a student’s emotions. Some of them are based on the detection of physical and biological signs [18] and others are based on AI techniques like Dynamic Decision Networks (DNNs), Machine Learning Techniques or Transition Networks [5], [15], [17]. Inferring student’s emotions in an on-line educational environment is a multi-parameter and highly demanding task closely related to the current mood and the personality of the student. Concerning the MENTOR, responsible for the recognition of the student’s emo- tions is the Emotional Component. This component (Figure 1) is composed by three subcomponents, the Personality Recognizer (PR), the Mood Recognizer (MR) and the Emotion Recognizer (ER), which are responsible for the recognition of the personal- ity, mood and emotions of the student. As it has been already mentioned, there are five personality types. When the student uses the system for the first time, the PR subcomponent selects a suitable dialogue specified by the FFM to assess the type of a student's personality. The dialogue is articulated in accordance to Goldberg's ques- tionnaire [8]. As a result, the student's traits are being recognized and are being used by the Teacher Component for the suitable selection of pedagogical and teaching strategy. For example, a student that has been recognized as Openess, according to FFM is imaginative, creative, explorative and aesthetic [6]. These characteristics are evaluated by the TC providing the system with an exploratory learning strategy, giv- ing more autonomy of learning to the student and limiting the guidance of the teacher. The MR subcomponent provides the system with a dialogue that can elicit emotions depending upon the semantics and its context. This dialogue is used in every new session and defines the current student's mood. Based on this dialogue the student's mood is recognized either as positive or as negative. In our approach, good mood consists of emotions like joy, satisfaction, pride, hope, gratification and bad mood consists of emotions like distress, disappointment, shame, fear, reproach. As a result, we have an initial evaluation of the current emotions of the student. Thus, if the stu- dent is unhappy for some reason, the MR recognizes it and in collaboration with TC, it defines the suitable pedagogical actions that decrease this negative mood and try to change it into a positive one. Finally, the ER subcomponent is in every moment aware of the student's emotions during the learning process, following the forthcoming method. So as to deal effectively with the emotions elicitation process, the Emotional Com- ponent has an affective student model where the affective information is stored. On- tology of emotions is used for the formal representation of emotions. Ontology is a technique of describing formally and explicitly the vocabulary of a domain in terms of concepts, classes, instances, relations, axioms, constraints and inference rules [23]. It e-Learning Issues under an Affective Perspective 33 is a formal way to represent the specific knowledge of a domain, providing an explicit and extensive framework to describe it. Lastly, except form AI, a lot of fields in In- formation Science like knowledge engineering and management, education, applica- tions related to Semantic Web, Bio-informatics make use of ontologies [21]. Our ontology has been built to recognize 10 emotions which are: joy, satisfaction, pride, hope, gratification, distress, disappointment, shame, fear, reproach. The former five emotions compose the classification of positive emotions and are related to the posi- tive student’s emotional state. The latter five emotions compose the classification of negative emotions and are related to the negative student’s emotional state. The con- struction of the ontology was based on the OCC cognitive theory of emotions. Thus, the concepts of the ontology are defined in terms with this theory. For instance, the positive student’s emotional state is described as follows: (POSITIVE-EMOTIONAL-STATE (SUBCLASSES (VALUE (JOY, SATISFACTION, PRIDE, HOPE, GRATIFICATION))) (IS-A (VALUE (EMOTIONAL-EVENT))) (DEFINITION (VALUE ("emotions or states, regarded as positive, such as joy, satis- faction, pride, hope, gratification")))) We use the DL-OWL (Description Logic – Ontology Web Language) as a reason- ing and inference mechanism to acquire the essential production rules, as well as to analyse the domain knowledge and interaction data. For instance, the emotion of fear is represented as: fear ti (P ,¬G) means that the student who is performing a plan P, feels the fear the particular period of time t i that will not accomplish his learning goal G. (1) In this way, the formal and flexible representation of an emotion can be efficiently achieved in relation to the learning goal of a student. The proposed ontology of emo- tions was implemented with the Protégé tool. Furthermore, we adopt a decision tree approach, an AI technique (C4.5 algorithm [22]) to extract information from the proposed “emotional” ontology and to make inferences about the emotions of the student. This process comprises three steps which respectively are the following: 1. The creation of the decision tree 2. The extraction of the rules from the decision tree 3. The triggering of the extracted rules to infer student’s emotions This approach, which is used for carrying out the representation and the inference of emotions is based on the OCC model which combines the appraisal of an Event with the Intentions and Desires of a subject. Thus, taking advantage of this model, MENTOR infers about the student’s emotions after the occurrence of an educational event which is related to his learning goal. 34 M. Leontidis, C. Halatsis, and M. Grigoriadou 3.2 Providing the Student with the Appropriate Affective Tactic As it has already been stated, the objective of the MENTOR is to foster the appropri- ate affective conditions, since these are a crucial factor for the learning process and to obtain the student with the suitable learning method. The latter goal is achieved by the Teaching Component which is responsible for providing the student with the appro- priate affective tactic considering his emotional state. It consists of two subcompo- nents, the Teaching Generator and the Pedagogical Generator, which are responsible respectively for the appropriate teaching and pedagogical strategy as illustrated in Figure 1. The Teaching Generator is a sub-component which is responsible for the selection and the presentation of the suitable educational material, according to the student model. The student model provides information about the cognitive status of the stu- dent such as his learning style, the knowledge that has already been acquired and his learning preferences and goals. Evaluating this information, the Teaching Generator decides about the sequence of the educational material, if a theoretical or practical subject will be presented next to the student and what kind would this be, for example a more or less detailed theoretical topic or an easier or a trickier exercise. The Pedagogical Generator is a sub-component which is responsible for the forma- tion of the pedagogical actions which will be taken into account during the learning process. Once the recognition of the student’s emotions and his emotional state has been stored in the affective student model, the Pedagogical Generator has all the nec- essary information in order to support and motivate the student to the direction of the achievement of his learning goals. As a teacher does in the real class [12], the Peda- gogical Generator encourages the student, gives him positive feedback, congratulates him when he achieves a goal, and keeps him always in a positive mood, with the view of engaging him effectively in the learning process. Combining the interaction of its two sub-components, the Mentor Component forms the appropriate affective tactic for the student. In this way, a traditional instruc- tional tactic is enhanced with a motivational one and this would be proved beneficial to the student from two aspects [20]. The first concerns the planning of the teaching strategy and the educational content, which and what topic will be taught to the stu- dent next and which method will be used for it. The second is more related to the delivery planning, how this topic will be taught. At this point, it should be noted, that the outputs of the two sub-components might be contradictory. For example, the Teaching Generator evaluates the current knowledge state of the student and suggests a difficult exercise. On the other hand, relying on his current emotional state, the Pedagogical Generator recommends an easier one, because it judges that the student’s confidence is low. So, resolving an easier exercise, it esti- mates that his confidence will be reinforced. In that case, the Mentor Component is designed so that, it would rather promote its Pedagogical Generator recommendation. Let us examine the reverse case, where the Teaching Generator suggests a trivial problem to a confident openness student. This suggestion might be considered as mo- tiveless by the Pedagogical Generator, compared to the student’s current emotional state. To tackle with this conflict, a more difficult problem is presented by the Mentor Component, demanding the student’s harder effort and challenging his interest further. . motivating him effectively and achieving thus, the desirable learning goals. Consequently, the investment in individual differences and the emo- tional “potential” of the student in combination. AES determines the educational content and the teaching process in a way that it appertains to teaching in a real classroom. In the real educational process, the teacher takes into consideration. we use the OCC model in our study. The perspective by which, we construct the follow- ing component is interdisciplinary and focuses in the intersection of Artificial Intelli- gence and Cognitive