Emotion Prediction Student Module based on CVS-Emotional Agent for ATS platform Horatiu MOGA and Csaba ANTONYA Transilvania University of Brasov/AEC Department, Brasov, Romania moga.horatiu@unitbv.ro antonya@unitbv.ro Abstract — An affective tutoring system (ATS) is a software process with decision capabilities focused on student’s training and adapted to student’s skills and needs. Generally, ATS systems have the following components: a pedagogical module, a domain module, a dialog module and a student module. The student module tries to predict features about student’s behavior like reactions, emotions, etc. In this research we aimed to design an emotion prediction student module based on a new approach of an emotional agent. This agent is named Control-Value Simplified Emotional Agent (CVSEA) and is based on the Theory of Control-Value in Psychology. I. INTRODUCTION An affective tutoring system (ATS) is a software process with decision capabilities focused on student’s training and adapted to student’s skills and needs. An ATS platform is a multi-agent system that integrates multimedia features, facial image processing features and pedagogical teaching skills that support the student in the training process. ATS systems have usually the following components [1]: a pedagogical module, a domain module, a dialog module and a student module. The pedagogical module implements the pedagogical strategies. The domain module comprises all the lessons and assigns the work needed for teaching. This module is strongly related to the pedagogical module. The dialog module uses a graphical user interface (GUI). By means of this GUI, the user either receives data from ATS or introduces data in the system. The student module aims to predict student’s emotions. An emotional agent is a software process having the capability of receiving the inputs from outside and to generate its own emotions. In emotional artificial intelligence the outputs are emotions of negative or positive nature [2], caused by other agents or the human users. In this research the emotional agent is employed for the implementation of the student module. The interaction between an ATS platform and student is a typical cognitive information communication. In this study the intra-cognitive communication is investigated by emotional psychological questionnaires and role play scenarios in order to translate it in a future inter-cognitive communication for student-ATS interaction. A. Motivations and Objectives In order to build the ATS platform, we constructed a predictive student module. The student module is inspired by the Control-Value Theory of educational psychology [3]. Generally, ATS platforms of the last twenty years are based on the OCC model [2] of emotions. The limits of those platforms are the lack of evaluation for emotions that are frequent in the face to face teaching process of student. These emotions are boredom, frustration and hopelessness which will be considered in this research. The focus of Control-Value Theory in the psycho- pedagogical field oriented us to new patterns for the emotional agent. For the implementation of the emotional agent we used a stochastic Moore machine. In this machine the inputs are the teacher’s actions, while the outputs are student’s emotions computed as bits. The internal states are computed with hidden Markov models (HMM). The name of this emotional agent is Control- Value Simplified Emotional Agent (CVSEA). This emotional agent is considering only the teacher's action, without taking into account the student's actions outcomes. II. RELATED WORKS A. Emotional Agents In the field of intelligent emotional agents (or emotional agents), there are some important directions for research. The paradigm for constructing emotional agents is OCC, proposed in 1988 [2]. This one has twenty two emotions and was then extended to thirty emotions [3]. Another emotional agent is Cathexis, having its roots in psychology, ethnology and neurobiology [4, 5]. CLARION is a modeling framework [7, 8] that is proposing agents with two layers and four components based on action, non-action, motivation and meta- cognition. The CogAff model is based on the evolution of microorganisms and uses three levels of information processing [9, 10]. EMA model is a special case that builds up a relation of dynamical dependence between internal emotion and external environment [11]. Emotional agents can also be built using fuzzy cognitive maps [12], fuzzy logic (FLAME or GEMA [13, 14]) or neural networks [15]. B. Affective Tutoring Systems This section refers to the involvement of emotional agents in affective tutoring systems. In this field it is proposed a metric evaluation framework based on three dimensions: student, platform and technology of implementation [16]. The research topics in this field are oriented towards the application of hypermedia 453 978-1-4673-5188-1/12/$31.00 ©2012 IEEE CogInfoCom 2012 • 3rd IEEE International Conference on Cognitive Infocommunications • December 2-5, 2012, Kosice, Slovakia