Emotion Recognition in Affective Tutoring Systems Collection of Ground truth Data Procedia Computer Science 104 ( 2017 ) 437 – 444 Available online at www sciencedirect com 1877 0509 © 2017 The Author[.]
Available online at www.sciencedirect.com ScienceDirect Procedia Computer Science 104 (2017) 437 – 444 ICTE 2016, December 2016, Riga, Latvia Emotion Recognition in Affective Tutoring Systems: Collection of Ground-Truth Data Sintija Petrovicaa,*, Alla Anohina-Naumecaa, Hazım Kemal Ekenelb a Riga Technical University, Kalku street 1, Riga, LV-1658, Latvia b Istanbul Technical University, 34469 Maslak, Istanbul, Turkey Abstract For the last 50 years, intelligent tutoring systems have been developed with the aim to supporting one of the most successful educational forms – individual teaching Recent research has shown that emotions can influence human behavior and learning abilities, as a result developers of tutoring systems have also started to follow these ideas by creating affective tutoring systems However, adaptation skills of the mentioned type of systems are still imperfect The paper presents an analysis of emotion recognition methods used in existing systems to enhance ongoing research on the improvement of tutoring adaptation Regardless of the method chosen, the achievement of accurate emotion recognition requires collecting ground-truth data To provide groundtruth data for emotional states, the authors have implemented a self-assessment method based on Self-Assessment Manikin © 2017 by by Elsevier B.V.B.V This is an open access article under the CC BY-NC-ND license © 2016 The TheAuthors Authors.Published Published Elsevier (http://creativecommons.org/licenses/by-nc-nd/4.0/) Peer-review underresponsibility responsibility of organizing committee the scientific committee of the international conference; ICTE 2016 Peer-review under of organizing committee of theofscientific committee of the international conference; ICTE 2016 Keywords: Affective computing; Intelligent tutoring systems; Emotion recognition; Ground-truth data; Self-Assessment Manikin Introduction Progress in the field of affective computing and research carried out in education and psychology, which uncovers a close relationship between emotions and learning, have led to the emergence of a new generation of intelligent tutoring systems (ITSs) – affective tutoring systems (ATSs) In general, understanding emotions is a quite complicated process even for humans because each emotional state can have its own possible causes and it may * Corresponding author Tel.: +371-26569654; fax: +371-67089584 E-mail address: sintija.petrovica@rtu.lv 1877-0509 © 2017 The Authors Published by Elsevier B.V This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/) Peer-review under responsibility of organizing committee of the scientific committee of the international conference; ICTE 2016 doi:10.1016/j.procs.2017.01.157 438 Sintija Petrovica et al / Procedia Computer Science 104 (2017) 437 – 444 influence differently person's behavior However, teachers can evaluate student's emotions with a rather high reliability based on facial expressions, body language, and speech Consequently, experienced teachers can make changes in the teaching process evaluating not only the student's knowledge level but also observing other student's characteristics, including emotional state1 Similarly, computerized tutoring systems must be capable of assessing students' emotions and using this information, promote learning and achievement of better learning outcomes2 However, there still exists a drawback of these systems comparing their adaptation skills with the same skills possessed by human-teachers There is still a lack of an emotional intelligence3 Since there is a close correlation between the classification of various student's states (both knowledge and emotional) and appropriate adaption of the tutoring process (selection of tutoring strategies and tactics), each of the classifiers involved in the process must be highly accurate and must work in real time to manage student's emotions Only in this way, it will be possible to model an individual student and to provide a truly customized tutoring process for him/her4 Emotional intelligence and intelligent tutoring systems This section is divided into two parts The first part reviews the concept of ITSs and their evolution to systems possessing emotional intelligence The second part provides the analysis of ATSs and their operating principles, as well as describes differences between these two types of tutoring systems from the architectural point of view 2.1 Intelligent tutoring systems and emotions ITSs are a generation of software systems, which aim to support and improve the teaching and learning process in a certain domain ITSs simulate human-tutors and provide benefits of the individual teaching by exploiting methods of artificial intelligence to provide a learning environment adapted (personalized content, feedback, navigation, etc.) to characteristics of an individual student2 Adaptation is possible because of special types of knowledge integrated into the traditional architecture of such systems (see Fig 1), which includes: a) a student diagnosis module collecting and processing information about the student (his/her learning progress, behavior, psychological characteristics, etc.) and a student model storing this information; b) a pedagogical module responsible for the implementation of the teaching process and a pedagogical model storing pedagogical knowledge; c) a problem domain module able to generate and solve problems in the domain and a domain model storing knowledge intended to be taught; d) an interface module managing interaction between ITS and students through various input/output devices Fig The traditional architecture of ITSs For decades, the field of ITSs inherited ideas from such learning theories as cognitivism and constructivism focusing more on student's cognitive processes However, in recent years researchers have shifted their emphasis from cognitive processes to emotionally-cognitive processes5 These changes can mostly be explained by the increased attention paid to the relationship between emotions and learning Research results show that emotions are a significant factor in the learning process and they can affect student's motivation and learning abilities Studies demonstrate that students experience a wide diversity of positive and negative emotions during the learning process, e.g., interest, flow, surprise, and pride, as well as anger, boredom, frustration, anxiety, confusion, and shame6,7 This implies that more attention should also be given to these emotions in the development process of ITSs Sintija Petrovica et al / Procedia Computer Science 104 (2017) 437 – 444 2.2 Affective tutoring systems The direction of affective computing (AC) started to evolve at the end of 1990-ties when the book “Affective Computing” was published by Rosalind Picard8 AC is a branch of artificial intelligence which focuses on the design of systems and tools able to process, recognize, and explain human emotions Around 10 years ago, ideas from AC came also in the development of ITSs and, as a result, a new generation of educational systems appeared – ATSs Different reasons are found in the literature in relation to the necessity to consider emotions as one of the parameters in computerized tutoring systems Some of these reasons are, for example, a possibility to ensure an optimal emotional state for the achievement of better learning results9,10, reduced risks of uncertainty and disturbing interventions of ITSs and improved system's adaption abilities11, a more effective imitation of pedagogical decisions made by human-teachers12, a timely recognition of negative emotions and minimization of their negative effect with the aim to increase students' motivation13, promotion of the students' involvement and confidence14 Taking into account the previously mentioned information about ATSs and their development purposes, an ATS can be defined as follows: an ATS is an ITS that imitates a human-teacher and his/her abilities of adapting not only to student's knowledge but also to his/her emotional state with the aim to intervene (react accordingly) only in those situations when an emotional state can become a threat to student's willingness to engage in the learning process, as a result leaving a negative impact on knowledge acquisition and learning outcomes Supporting functionality of ATSs requires the extension of the traditional architecture of ITSs (see Fig 1) by adding additional components Commonly, three additional parts are incorporated into the architecture to form the so called affective behavior model that allows providing appropriate responses considering both student's knowledge and emotions13,15 The first component is usually responsible for the automatic identification of a student's emotional state15 Emotion recognition is carried out by detecting and analyzing different features (e.g facial expressions, body gestures, speech, physiological characteristics, etc.) and applying various classifiers to identify student's emotions15,16 Typically, the acquired emotional state is stored in the student model, thus expanding the available information about the student and forming the so called affective student model An emotion response module or an affective (behavior) pedagogical model is often defined as the second component It can be considered as an extension of the pedagogical module13,17 This component provides reasoning on a current tutoring situation and allows for further adaption of the tutoring process on the basis not only on the student's current knowledge level and learning characteristics but also on the student's emotional state18 Therefore, the main task of the affective pedagogical model is to match data about student's emotions and tutoring situation acquired from the student model with appropriate responses of ATS12,17 By analyzing variations of the architecture of different existing ATSs, an emotion expression module can be found as the third component This module can be referred as an extension of the interface module that allows ATS to express its own emotions as a response on student's actions and emotions Usually, this component is represented as a virtual tutor or a pedagogical agent (PA) with its own emotions18 Related work ATSs would not be able to respond to a student's emotions if they were unable to determine it Consequently, most of the research carried out so far is directed to the identification of emotions19 This section provides an analysis of different ATSs integrating emotion recognition Several aspects are examined, e.g., sensors used for the acquisition of emotional data, methods used for emotion classification, and the most commonly modeled emotions 3.1 Review of affective tutoring systems For this research, different ATSs were selected to cover various problem domains – both from “hard” sciences (for example, mathematics, physics, natural sciences, computer science) and “soft” sciences or humanities (e.g study of languages) However, it must be noted that most of the ITSs have been developed for well-defined problem domains, where more rules exist regarding task generation and solving, whereas development of ITSs for ill-defined domains still remains a challenge20 This is also the case for ATSs: most part of such systems has been mainly developed for well-defined domains This can also be explained by the fact that during the learning process in these areas, students most likely encounter difficulties and experience emotions inherent in the learning process such as 439 440 Sintija Petrovica et al / Procedia Computer Science 104 (2017) 437 – 444 confusion or frustration21 A summary of analyzed ATSs is provided in Table 1, reviewing such systems as MathSpring22, Prime Climb23, WaLLis24, Easy with Eve25, FERMAT26, Cognitive Tutor Algebra27 and PAT2Math21 intended for teaching mathematics, ITSPOKE28 and AutoTutor7 used for physics, EER-Tutor29 developed for computer science field, CRYSTAL ISLAND30 and Guru Tutor31 intended for biology, Inq-ITS32 developed for natural sciences, INES33 and MetaTutor34 for teaching topics related to medicine and VALERIE35 used for French language Table The summary of different emotion recognition aspects in the existent ATSs ATS Sensors Detection of emotional data and emotion classification Recognized emotions AutoTutor Video camera Pressure sensitive chair Posture and eye pattern extraction, analysis of log files Classifiers: Naïve Bayes, neural networks, logistic regression, nearest neighbour, C4.5 decision trees Flow, confused, bored, frustrated, eureka, neutral Cognitive Tutor Algebra Not used Analysis of log files recording features related to the student's behaviour, event and activity history in the learning process Classifiers: J48 decision trees, K* algorithm, step regression, JRip, Naïve Bayes, REP-Trees Bored, concentrated, frustrated, confused CRYSTAL ISLAND Not used Analysis of surveys, interviews, and log files Emotions are modelled using a Dynamic Bayesian Network Anxious, bored, confused, curious, enthusiastic, focused, frustrated Easy with Eve Video camera Facial feature extraction Classifier: support vector machines (SVM) Smiling, laughing, surprised, angry, scared, sad, disgusted, neutral EER-Tutor Video camera Facial feature (eyes, eyebrows, lips) tracking and extraction Features are classified by analyzing the calculated distances for each of facial features comparing to the neutral face Happy, smiling, angry, frustrated, neutral emotions FERMAT Video camera Extraction of facial feature points and regions of interest Classifiers: neural network, a fuzzy expert system Angry, disgusted, scared, happy, sad, surprised, neutral Guru Tutor Eye tracker Video camera Eye tracking and gaze pattern extraction, analysis of log files Analysis of the attention time paid to the screen Disinterested, bored Inq-ITS Not used Analysis of log files Classifiers: J48 decision trees, step regression, JRip Bored, confused, frustrated, concentrated INES Not used Analysis of the student’s activity level, difficulty of the task, previous progress, number of errors, severity of the error Emotions are predicted by appraisal rules Worried, confident, depressed, enthusiastic ITSPOKE Microphone Negative, positive and neutral emotions MathSpring Not used Extraction of acoustic-prosodic, lexical features (speech intensity, energy, volume, duration, and pauses) and dialogue features (e.g the accuracy of the answer) Semantic analysis is used for the assessment of answer accuracy and linear regression for confidence evaluation Analysis of log files, self-assessment reports, behaviour patterns, etc Classifier: linear regression MetaTutor Eye tracker Extraction of gaze data features and features related to areas of interest within system’s interface Classifiers: random forests, Naïve Bayes, logistic regression, SVMs Bored, curious, interested PAT2Math Video camera Analysis of log files and extraction of facial feature points Emotions are identified based on Facial Action Coding System and psychological model of emotions (OCC model) Satisfied, disappointed, happy, sad, grateful, angry, ashamed Prime Climb Physiological (phys.) sensors Determination of skin conductivity, heart rate, muscle activity, and analysis of log files Biometrical data is analyzed via unsupervised clustering Happy, sad (for the game), admiration, criticism (for PA), pride, shame (for himself) VALERIE Video camera Microphone Mouse Phys sensors Determination of skin conductivity, heart rate, extraction of facial and speech features, analysis of mouse movement Classifiers: nearest neighbour, discriminant function analysis, Marquardt Back-propagation algorithm Sad, angry, surprised, scared, frustrated, amused WaLLis Not used Analysis of log files Classifier: J4.8 decision tree algorithm Frustrated, confused, bored, confident, happy, enthusiastic Confident, worried, excited, inactive, satisfied, frustrated, interested, bored Sintija Petrovica et al / Procedia Computer Science 104 (2017) 437 – 444 3.2 Analysis: sensors, emotional features, and emotion classification In general, the determination of a student's emotional state is implemented by analyzing various data sources providing features that can give information about emotions Ideally, a quantitative and continuous measurement of emotional experience is required in an objective and unobtrusive manner, e.g analysis of interactional content6 Two most commonly used categories of features for the identification of emotions are facial features and features acquired from log files Regarding facial features, mostly patterns of different facial features (eyes, eyebrows, and lips) are extracted25,29 Eye movement is tracked and gaze patterns are acquired indicating on regions of interest to which the student is paying attention26,34 Features recorded in log files are related mainly to student's interaction with the system27,33 Acquired features include both information linked to the student (e.g., behavior patterns, action history, activity level, etc.) and data characterizing a current tutoring situation (e.g., progress, content difficulty, made errors, etc.) This category of features can be considered as the least disturbing one from the student's point of view, because it does not require additional actions from the student However, the selection of features can be a challenging task for developers in relation to the achievement of sufficient emotion classification results Besides these two most common categories of features, other characteristics are also acquired for emotion classification, e.g., body language7, speech features (intensity, volume, duration)28, physiological signals (heart rate, muscle movement, skin conductance)23 , and usage patterns of input devices, e.g., mouse35 To perceive these features, various sensors are used Cooper et al.36 have grouped sensors in three possible categories considering the level of discomfort they cause to the student Physiological sensors (e.g., skin conductivity sensor, heart rate sensor, electromyograph) cause the greatest discomfort because they require a direct contact with certain parts of the body and they should be connected to the computer, which registers all the data Touch or haptic sensors (e.g., pressure-sensitive mouse or chair) induce less discomfort and very often students not notice them; however, the usage of such sensors for emotion recognition requires a student to touch them in such a way limiting his/her movement freedom Observational sensors (e.g., video cameras, eye trackers, microphones) are not physically intrusive; however, they can distract students and make them feel uncomfortable knowing that all actions are recorded Besides usage of sensors, emotion identification in some ATSs is based on results of students’ filled surveys or self-assessment reports, where students report their own feelings, emotions, or mood in a particular learning situation22,30 This can be considered as an "accurate" method for the emotion acquisition, if students are aware of their emotions; however, a possibility exists that students will consider such surveys as redundant and they will not provide correct information about their emotions, e.g., if their goal is to complete these surveys faster In general, from the developers' point of view this can be regarded as one of the less time consuming methods since its implementation does not require the use of sensors or the application of classification algorithms Other method used for emotion prediction is based on the analysis of causes of emotions21 For this purpose, the OCC emotion model37 is used, which includes appraisal of the world in self-relevant terms, mental representations, and factors The classification of emotions and used classifiers mainly depend on whether sensors are used for feature extraction or not If emotions are recognized on the basis of data received from a video camera, then algorithms are used to analyze the distances between features in relation to a neutral facial expression Neural networks, Naïve Bayes, logistic and linear regression, SVMs, nearest neighbor algorithm, various types of decision trees, and other methods are being used for feature classification and emotion recognition In most cases, more than one classifier is applied because some of classifiers provides higher classification results for specific emotions7 Regarding the most commonly modeled student’s emotions, a part (although minor) of existing ATSs carry out the recognition of facial expressions to identify basic emotions (anger, disgust, happiness, fear, sadness, and surprise) that mostly are not experienced in the learning process25,26,35 In addition, it should be noted that most of these emotions (except happiness and sadness, which are directed towards learning outcomes) rarely appear in the learning process, consequently the identification of basic emotions is largely insignificant for the adaption of the tutoring process However, it is only a small part of ATSs and emotion modeling trends are improving and developers have started focusing their attention on emotions that are felt during the learning process and can directly influence it Therefore, most of the analyzed ATSs are aimed at learning-specific emotions and are able to determine, whether the student is, for example, concentrated (interested or in flow state), confused, bored, frustrated, anxious, ashamed, etc7,22,24,27,30 441 442 Sintija Petrovica et al / Procedia Computer Science 104 (2017) 437 – 444 Determination of an affective state through self-assessment This section is divided into two parts The first part reviews Self-Assessment Manikin (SAM) used for acquiring three emotional dimensions and the second part describes the implementation of SAM for collecting ground-truths 4.1 Self-assessment Manikin Currently, a popular direction regarding the emotion recognition is the analysis of log files, which record interaction between students and tutoring system and allow identifying behavior patterns in a particular learning situation and linking them with potential students’ emotions This approach is considered as a sensor-free approach27,32 Mainly, the development of new ATSs or modification of existing ATSs with the sensor-free approach can be explained by the limited availability of sensors in real learning conditions (in the best case, computer classes or students' laptops are equipped with microphones and video cameras but not with physiological sensors22) Since the sensor-free approach does not provide very high accuracy of emotion recognition and can crucially decrease accuracy of adaptation of the tutoring process, one of the possible solutions to overcome this problem is the so called a sensor-lite approach requiring a minimal use of available sensors, e.g., built-in cameras or microphones19 Whatever approach is chosen, to achieve the emotion recognition as accurate as possible, the first step is collecting a ground-truth emotion data set, which can be later used for training and comparing results of automatic measurement of affect38 Regarding this issue, SAM as one of the most popular self-assessment methods is analyzed It can be used independently from the sensor-based approaches (e.g., without requiring video cameras) The method allows getting from students themselves their feelings using a graphic representation of three fundamental emotional dimensions – pleasure, arousal and dominance (PAD)39 Pleasure indicates how pleasant person feels about something; arousal describes the level of mobilization or energy for the person; and dominance symbolizes an ability to cope with the situation After the self-assessment, it is possible to represent all three emotional dimensions in the PAD space, where each graphic depiction can have its own value in the range [–1 1] By combining values from all three dimensions in the PAD emotion space, the determination of emotions can be done Russell and Mehrabian40 have published a complete list of emotions and their corresponding PAD values In this research, it was decided that an initial step for emotion recognition is the implementation of SAM, which will be used as an independent method for the acquisition of emotional data to identify students’ emotions while they are going through various instructional activities (e.g., starting new topic, solving tasks, receiving feedback, etc.) The collected data will serve as ground-truth for sensor-based emotion classification studies and will be applied to testing the functionality of the pedagogical module in relation to the adaption of the tutoring process 4.2 Emotion identification In relation to the SAM, possible solutions were analyzed that could be adopted for research purposes One of the existing SAM implementations is AffectButton tool, which is freely available and can be customized and used in other research projects to acquire emotional data from systems’ users41 The AffectButton is a measurement tool that enables a user to give a detailed emotional feedback about his/her feelings, mood, and attitudes towards different objects After clicking the button, three values are generated corresponding to each of the emotional dimensions in the PAD emotional model Currently, the source code of AffectButton tool is already adapted and integrated in the environment for research requirements Since this method provided only PAD values characterizing specific emotions but not emotions themselves, a discrete emotion calculation based on acquired PAD values is implemented to determine to which of learning-specific emotions the acquired PAD values correspond the most (see Fig 2) Formula (1) is applied to determine a distance “d” between the acquired PAD values for emotion ej and the defined learning-specific emotion ei The less is the distance value, the more similar emotions are In total, 15 different emotions are incorporated (angry, anxious, bored, concentrated, confused, curious, excited, fearful, frustrated, happy, helpless, interested, relaxed, sad, surprised) but only the closest five are represented on the screen: d ( ei , e j ) (e Pi e Pj ) (e Ai e A j ) (e Di e D j ) (1) Sintija Petrovica et al / Procedia Computer Science 104 (2017) 437 – 444 where Pi, Ai, Di corresponding PAD values for emotion ei and Pj, Aj, Dj corresponding PAD values for emotion ej Fig AffectButton and calculation of emotions based on the generated PAD values Emotion self-assessment can be carried out during the whole learning process allowing students to report about their emotional changes at any time when they prefer to this or when the ATS itself prompts them to this while they are going through learning activities Despite possible inconveniences, which this method can cause, it will allow identifying emotions during the learning process Conclusion The more detailed analysis of applied emotion recognition methods in existing ATSs is performed covering sensors used for the acquisition of features characterizing specific emotions, most often extracted features, methods used for feature classification and most commonly identified emotions To provide ground-truths for automatic emotion identification, SAM is analyzed and its available developments are examined The existing SAM solution, AffectButton, has been adopted for research purposes and an additional functionality is implemented to calculate 15 possible discrete emotions based on acquired PAD values However, one of the problems is related to close PAD values for some of emotions One of the possible solutions could be to identify “mood type” or one of eight octants in the PAD space to which the emotion belongs42 This could narrow a range of possible similar emotions, as well as could reduce calculation time Acknowledgements This work was supported by the COST Action IC1303 Algorithms, Architectures and Platforms for Enhanced Living Environments Short-Term Scientific Mission grant No IC1303-161115-068104, by TUBITAK Project No 113E067, and by a Marie Curie FP7 Integration Grant within the 7th EU Framework Programme References Meyer DK, Turner JC Scaffolding Emotions in Classrooms In: Emotions in Education; 2007 p 243-258 Petrovica S, Pudane M Simulation of Affective Student Int J Educ Learn Syst 1; 2016 p 99-108 Zakharov K et al Towards Emotionally-Intelligent Pedagogical Agents In: Proceedings of the ITS’2008; 2008 p 19-28 Sottilare R et al A Guide to Instructional Techniques, Strategies and Tactics to Manage Learner Affect, Engagement, and Grit In Design Recommendations for Intelligent Tutoring Systems Vol U.S Army Research Laboratory; 2014 p 7-33 Petrovica S Tutoring Process in Emotionally Intelligent Tutoring Systems Int J Technol Educ Mark (1); 2014 p 72-85 Afzal S, Robinson P Modelling Affect in Learning Environments In: Proceedings of ICALT'2010; 2010 p 438-442 D’Mello SK, Picard RW, Graesser AC Toward an Affect-Sensitive AutoTutor IEEE Intell Syst 22 (4); 2007 p 53-61 Picard RW Affective Computing Cambridge, MA: MIT Press; 1997 Ochs M et al Optimal Emotional Conditions for Learning with an Intelligent Tutoring System In: Proc of the ITS’2004 p 845-847 443 444 Sintija Petrovica et al / Procedia Computer Science 104 (2017) 437 – 444 10 Hickey TJ, Tarimo WT The Affective Tutor J Comput Sci Coll 29 (6); 2014 p 50-56 11 Landowska A Affect-awareness Framework for Intelligent Tutoring Systems In: Proceedings of the HIS'2013; 2013 p 540-547 12 Lin HCK, Wu CH, Hsueh YP The Influence of Using Affective Tutoring System in Accounting Remedial Instruction on Learning Performance and Usability Comput Hum Behav 41; 2014 p 514-522 13 Kaklauskas A et al Affective Tutoring System for Built Environment Management Comp Educ 82; 2015 p 202-216 14 Spaulding SL Developing Affect-Aware Robot Tutors Master Thesis USA: Massachusetts Institute of Technology; 2015 15 Malekzadeh M et al A Review of Emotion Regulation in Intelligent Tutoring Systems Educ Technol Soc 18 (4); 2015 p 435-445 16 Sarrafzadeh A et al E-learning with Affective Tutoring Systems In: Intelligent Tutoring Systems in E-learning Environments IGI Global; 2011 p 129-140 17 Hernández Y, Sucar E, Conati C An Affective Behavior Model for Intelligent Tutors In: Proc of the ITS'2008; 2008 p 819-821 18 Gu X et al Design of Emotional Intelligent Tutor System based on HMM In: Proc of the ICNC'2010; 2010 p 1984-1988 19 D’Mello SK, Graesser AC Feeling, Thinking, and Computing with Affect-Aware Learning Technologies In: Calvo RA, D'Mello SK, Gratch J, Kappas A (eds.) The Oxford Handbook of Affective Computing UK: Oxford University Press; 2015 p 419-434 20 Nye BD, Goldberg B, Hu X Generalizing the Genres for ITS: Authoring Considerations for Representative Learning Tasks In: Design Recommendations for Intelligent Tutoring Systems Vol U.S Army Research Laboratory; 2015 p 47-64 21 Jaque PA et al Rule-Based Expert Systems to Support Step-by-Step Guidance in Algebraic Problem Solving: The Case of the Tutor PAT2Math Expert Syst with Appl 40 (14); 2013 p 5456-5465 22 Wixon M et al The Opportunities and Limitations of Scaling Up Sensor-Free Affect Detection In: Proc of the EDM'2014 p.145-152 23 Amershi S et al Using Feature Selection and Unsupervised Clustering to Identify Affective Expressions in Educational Games In: Ikeda M, Ashlay KD, Chan TW (eds.) Proceedings of the ITS'2006 Springer-Verlag; 2006 p 21-28 24 Porayska-Pomsta K et al Diagnosing and Acting on Student Affect User Model User-Adap 18 (1-2); 2008 p 125-173 25 Sarrafzadeh A et al How Do You Know that I Don’t Understand? Comput Hum Behav 24; 2008 p 1342-1363 26 Zataraín-Cabada R et al Affective Tutoring System for Android Mobiles In: Proceedings of the ICIC'2014; 2014 p 1-10 27 Baker RSJ et al Towards Sensor-free Affect Detection in Cognitive Tutor Algebra In: Proc of the EDM’2012 p 126-133 28 Litman D, Forbes-Riley K, Silliman S Towards Emotion Prediction in Spoken Tutoring Dialogues In: Hearst M, Ostendorf M (eds.) Proc of the HLT/NAACL'2003 USA: Association for Computational Linguistics; 2003 p 52-54 29 Zakharov K Affect Recognition and Support in Intelligent Tutoring Systems Master thesis New Zealand: University of Canterbury; 2007 30 Sabourin JL et al Considering Alternate Futures to Classify Off-Task Behavior as Emotion Self-Regulation: A Supervised Learning Approach J Educ Data Mining (9); 2013 p 9-38 31 Olney A et al Guru: A Computer Tutor that Models Expert Human Tutors In: Proceedings of the ITS'2012 p 256-261 32 Paquette L et al Sensor-Free Affect Detection for a Simulation-Based Science Inquiry Learning Environment In: Trausan-Matu S, Boyer KE, Crosby M, Panourgia K (eds.) Proc of the ITS'2014 Switzerland: Springer International Publishing; 2014 p 1-10 33 Heylen D, Nijholt A, Akker HJ Affect in Tutoring Dialogues J Appl AI 19 (3-4); 2005 p 287-310 34 Jaques N et al Predicting Affect from Gaze Data during Interaction with an Intelligent Tutoring System In: Trausan-Matu S, Boyer KE, Crosby M, Panourgia K (eds.) Proceedings of the ITS'2014 Switzerland: Springer International Publishing; 2014 p 29-38 35 Paleari M, Lisetti C, Lethonen M VALERIE: Virtual Agent for Learning Environment Reacting and Interacting Emotionally In: Proceedings of the AIED'2005 Amsterdam: IOS Press; 2005 36 Cooper D, Arroyo I, Woolf BP Actionable Affective Processing for Automatic Tutor Interventions In: Calvo RA, D'Mello SK (eds.) New Perspectives on Affect and Learning Technologies New York: Springer-Verlag; 2011 p 127-140 37 Ortony A, Clores GL, Collins A The Cognitive Structure of Emotions MA: Cambridge University Press; 1988 38 Gunes H, Nicolaou MA, Pantic M Continuous Analysis of Affect from Voice and Face In: Salah AA, Gevers T (eds.) Computer Analysis of Human Behaviour UK: Springer-Verlag London; 2011 p 255-291 39 Bradley M et al Measuring Emotion: Self-Assessment Manikin and the Semantic Differential J Behav Ther Exp Psy; 1994 p 49-59 40 Russell JA, Mehrabian A Evidence for a Three-Factor Theory of Emotions J Res Pers 11 (3); 1977 p 273-294 41 Broekens J, Brinkman WP AffectButton: A Method for Reliable and Valid Affective Self-Report Int J Hum-Comput St 71 (6); 2013 p 641-667 42 Mehrabian A Pleasure-Arousal-Dominance: A General Framework for Describing and Measuring Individual Differences in Temperament Curr Psychol 14(4); 1996 p 261-292 Sintija Petrovica obtained Master’s degree in Computer Systems in 2011 at Riga Technical University (RTU), Latvia In 2011, she started to work as a scientific assistant in Department of Artificial Intelligence and System Engineering at RTU She is a PhD student of the study program “Computer Systems” at RTU She is developing her PhD thesis related with the development of pedagogical module for affective tutoring systems to adapt tutoring process to student’s emotions Her research interests include intelligent tutoring systems and affective computing Contact her at sintija.petrovica@rtu.lv ... Review of Emotion Regulation in Intelligent Tutoring Systems Educ Technol Soc 18 (4); 2015 p 435-445 16 Sarrafzadeh A et al E-learning with Affective Tutoring Systems In: Intelligent Tutoring Systems... step is collecting a ground- truth emotion data set, which can be later used for training and comparing results of automatic measurement of affect38 Regarding this issue, SAM as one of the most popular... and Grit In Design Recommendations for Intelligent Tutoring Systems Vol U.S Army Research Laboratory; 2014 p 7-33 Petrovica S Tutoring Process in Emotionally Intelligent Tutoring Systems Int J Technol