A Review of Emotion Regulation in Intelligent Tutoring Systems

Malekzadeh, M., Mustafa, M. B., & Lahsasna, A. (2015). A Review of Emotion Regulation in Intelligent Tutoring Systems. Educational Technology & Society, 18 (4), 435–445.

A Review of Emotion Regulation in Intelligent Tutoring Systems Mehdi Malekzadeh1*, Mumtaz Begum Mustafa1 and Adel Lahsasna2 1

Multimodal Interaction Research Lab, Faculty of Computer Science& Information Technology, University of Malaya, Kuala Lumpur, Malaysia // 2Artificial Intelligence Department, Faculty of Computer Science& Information Technology, University of Malaya, Kuala Lumpur, Malaysia // [email protected] // [email protected] // [email protected] * Corresponding author (Submitted September 11, 2014; Revised January 26, 2015; Accepted March 18, 2015) ABSTRACT Having improved emotional (affective) state may have several benefits on learners, such as promoting higher cognitive flexibility and opens the learner to discovery of new ideas and possibilities. On other side, negative emotional states like boredom and frustration have been linked with less use of self-regulation and cognitive strategies for learning as well as increases in disengaged and disturbing behavior during learning. In the area of computerised learning, several researchers strongly agree that intelligent tutoring systems (ITSs) would significantly improve its performance if it can adapt to the affective state (emotional state) of the learners. This idea has spawned an important trend in the development of ITSs, which are systems with the ability to regulate a learner’s adverse emotions. In the present study, we discuss the existing studies that have implemented different emotion regulation strategies such as coping strategies and implementation of these strategies in the domain of intelligent tutoring system (ITS). The results of the review show that applying emotion regulation strategies during computerised learning may produce more optimistic emotions as well as better learning gain.

Keywords Emotion, Emotion regulation, Emotion regulation strategy, Emotion coping strategy, Learning, Intelligent tutoring system

Introduction A common view of emotions is that they are generated as a results of human’s judgment about the world and initiated by individual’s appraisal in response to and interaction with stimulus, such as material that the individual is learning (Desmet, 2002; Lazarus, 1991). Recent findings in neuroscience and psychology found that emotions are widely related to cognition, influencing various behavioural and cognitive processes, such as attention, long-term memorizing, decision-making, and so on (Ahn & Picard, 2005). Researches on emotion and learning suggest that positive emotions (affects) have a vital influence on various cognitive processes relevant for learning, such as information processing, communication processing, decision-making processing, negotiation processing, category sorting tasks, and creative problem-solving processes (Erez & Isen, 2002). Positive emotions promote higher cognitive flexibility and allow the learner to discover new ideas and possibilities. In addition, as a function of positive emotion, cognitive processes may be more flexible that result in greater creativity and improved problemsolving (Isen et al., 1987). Emotion also influences memory, where positive emotional state improved recall and it served as effective retrieval cues for long-term memory in many experiments (Isen et al., 1978). Reciprocally, negative emotional states like boredom and frustration have been linked with less use of self-regulation and cognitive strategies for learning as well as increases in disengaged and disturbing behavior during learning in the class (Isen, 2001). Thus, emotions, governed by proper attention, self-regulation and motivational strategies result in positive effects on learning, and lead to better achievement among the learners (Pekrun, Goetz, Titz, & Perry, 2002). In traditional learning environment, a teacher maintains a sympathetic relationship with learners to facilitate the development of positive emotions. For instance, students who feel happy generally perform better than students who feel sad, angry, or scared (Connor & Davidson, 2003). This relationship also exists in a computerized learning environments and researchers of computer science in education field had studied techniques of artificial intelligence to make the educational systems more customized to the emotional state (affective states) of students (Jaques & Vicari, 2007). Intelligent tutoring system (ITS) is a computer-based educational system that provides individualised instructions similar to like a human tutor. Typical ITSs determine how and what to teach a student based on the learner’s pedagogical state to enhance learning. As experienced human tutor manages the emotional states of a learner to ISSN 1436-4522 (online) and 1176-3647 (print). This article of the Journal of Educational Technology & Society is available under Creative Commons CC-BY-ND-NC 3.0 license (https://creativecommons.org/licenses/by-nc-nd/3.0/). For further queries, please contact Journal Editors at [email protected].

435

motivate him or her and to improve the learning process, researchers also have augment the learner model structure in ITSs to determine the emotional state of learners (Neji, Ben Ammar, Alimi, & Gouardères, 2008). Researchers endow ITSs with the ability to detect learners’ unpleasant emotional states (e.g., confusion, frustration, etc.), respond to these states, and generate appropriate tutoring strategies as well as emotional expressions by embodied pedagogical agents. These emotion-sensitive ITSs aspire to narrow the interaction bandwidth between computer tutors and human tutors with the hope that this will lead to an improved user experience and enhanced learning gains (Aghaei Pour, Hussain, AlZoubi, D’Mello, & Calvo, 2010; Klein, Moon, & Picard, 2002). In embedding emotional state reasoning into ITSs and intelligent learning environments, there are two main issues that are faced by the developers. First is determining the emotional states of the target learners, and second is determining factors that causes those states as well as how to respond and regulate negative emotional state (Avramides & Du Boulay, 2009; Du Boulay, Rebolledo Méndez, Luckin, & Martínez-Mirón, 2007). To deal with the first issue, researchers (e.g., Aghaei Pour et al., 2010) paid attention to the determination of students' emotions. Despite the complexity associated with real-time emotion detection, several researches have embarked on learner’s emotion detection. However, not many researches that focused on the causes of favorable or adverse emotional state of learners and strategies for regulating them. If the ITS design or the feedback offered were not suited to individual user needs and character, the learner can be frustrated or bored. The challenge is therefore to help learners to regulate their emotional states so that positive states such as flow/engagement persevere, while negative states such as frustration and boredom are prevented or regulated (Zakharov, Mitrovic, & Johnston, 2008). In this paper, we have reviewed the efforts taken in existing researches that are related to regulation of negative emotional states in users’ learning process using emotion-sensitive intelligent tutoring systems (EITSs). We discuss researches that apply the emotion regulation methods in emotion-sensitive intelligent tutoring environment. The findings of this review reveal that utilization of emotion regulation strategies benefits the learners during the learning process.

Materials and methods For purpose of this study, we have searched the electronic databases that are relevant to education, psychology, information technology and social science: (a) IEEE XPLORE, (b) ACM Digital Library, (c) Science Direct, (d) Springer Link, (e) ERIC (Educational Resources Information Center) and (f) Web of Science. Searches were restricted to peer-reviewed articles, written in English, and published between 2008 and 2014 (research over the last six years). We only included original articles pertaining to empirical research that focuses on applying different techniques for managing the negative emotional state of user such as boredom, anxiety, and sadness to improve learning productivity of the learner during learning episode with computerized learning system. The exclusion criteria include articles published in languages other than English and the research studies that do not meet the inclusion criteria.

The framework of Emotion-Sensitive ITS (EITS) The interest in Intelligent Tutoring Systems began in the late 1970s, where these systems employs effective intelligent algorithms that would optimally conform to the learner and formulate strategy that optimizes the learning. The late 1990s and the early 2000s witnessed an exciting infusion of ITSs that incorporated tutoring strategies such as error identification and correction, frontier learning (expanding on what the student already knows), student modeling (inferring what the student knows and using that information to guide tutoring), and natural language dialogs (Aleven & Koedinger, 2002; Anderson, Douglass, & Qin, 2005; Woolf et al., 2009). Around the same time affective computing was beginning to rise as a new and exciting research area. Affective computing is about creating technologies that can monitor and appropriately respond to the emotional states of the user (Picard, 2010). Affective computing is a sub area of Human-Computer Interaction (HCI), where the emotional states of a user (feelings, moods, emotions) are incorporated into the decision cycle of the interface to develop more influential, user-friendly and natural applications (Picard, 1999).Throughout the last decade, several ITSs (Woolf et al., 2009; Zakharov et al., 2008) have been developed to incorporate assessments of students’ cognitive and emotional states into its educational 436

and motivational strategies to manage student’s engagement, self-confidence, regulate negative emotions, and maximizes learning (Calvo & Mello, 2011). An EITS is generally divided into two main components. The first component is automatic identification of a student’s emotional states. There are several methods of emotion recognition proposed in the literatures such as through facial expression, body gesture, speech, text, and physiological measurements. The learners’ emotions are modeled in the valence arousal space, which is a 2D model for emotion modeling. Arousal describes the physical activation, varying from low to high, while valence refer to pleasantness or hedonic value, varying from negative to positive. Emotion such as stress, for instance, is modeled as high arousal and low valence, while joy and elation would be high arousal and also high valence (Schlosberg, 1954).The emotion classifier is trained on the recorded data using various sensors as student interact with ITS. Recorded data is annotated by multiple human judges including student (self-judgments) and trained judges. Labelling is required for supervised learning systems. In multimodal emotion detection, the emotion recognition component uses a decision-level fusion algorithm where each channel (conversational cues, face, posture, etc.) independently provides its own recognition of the learner’s emotional state. These individual recognitions are infused with algorithm that selects a single emotional state and a confidence value of the detection. The second module responds to a user’s actions by adapting a teaching strategy based on pedagogical state (e.g., knowledge level, learning speed) and emotional state of learner. In EITSs, agents continuously track student cognition, behaviour, or emotion and offer students with support based on individual differences along these parameters (Baker et al., 2006; D’mello & Graesser, 2012; Rebolledo-Mendez, du Boulay, & Luckin, 2006; Woolf et al., 2009). The agent behaviour and responses can be considered as a type of formative feedback to learners, and agents often propose a diversity of formative feedback strategies (Rebolledo-Mendez et al., 2006). Generally, the feedback strategy for emotion management can be domain dependent (e.g., providing hints and definition related to the course content) and domain independent (e.g., providing empathy or encouragement or requests to stop undesired behaviour).

Figure 1. The architecture of EITS In EITSs, agents simulate human tutors by synthesizing emotional elements through the generation of speech, facial expressions, and other gestures. Using animated pedagogical agents allow EITS to offer sophisticated, real-time problem-solving advice and active emotional support with solid visual appeal. In addition, agents can motivate learners to interact more regularly with agent-based EITS and consequently increases the quality of a learner’s training over periods of months and years (Lester et al., 1997; Zakharov et al., 2008). In EITS, agent responses to the student’s action based on the learner’s cognitive and emotional state using a set of rules. Each rule has a set of feedback messages determining the agent’s verbal response. In addition, each rule includes a numeric value which triggers a change in the agent’s emotional appearance. For instance, when the learner answer correctly, the agent 437

responds with a joyful smile together with an admiring message (Zakharov et al., 2008). Figure 1, depicts the architecture of EITS.

Emotion coping and regulation In psychology, the concepts of emotion coping and emotion regulation are addressed to manage user emotional states. The emotion coping expends conscious effort to solve personal and interpersonal problems such as stress and conflict, and seeks to master, minimise or tolerate them (Lazarus & Folkman, 1984). Based on the work of Gross (1998), emotion regulation concerns with the ability to reduce high levels of emotion arousal and the capacity to change user’s feelings. Emotion coping focuses on decreasing negative emotion experience, whereas emotion regulation addresses increasing and decreasing both positive and negative emotions (Gross, 1998). Therefore, emotion can be regulated by using emotion coping as well as emotion regulation strategies by focusing on reducing negative emotional experiences. Lazarus (1991) classified emotion coping strategy in two different categories: • Problem-focused coping strategy: Solving of the problem that causes the emotional situation such as providing definitions and examples related to the course content to learner during learning. • Emotion-focused coping strategy: Reduction and management of the intensity of negative emotions caused by a stressful situation such as tutor provides encouraging statements during learning. • Gross (1998), divides emotion regulation strategies into two categories: antecedent-focused and responsefocused. Antecedent-focused strategies (i.e., situation selection, situation modification, attentional deployment, and cognitive change) occur before an emotional response is fully generated to influence an emotional state. Response-focused strategies (i.e., response modulation) occur after an emotional response is fully generated. The following statements describe these strategies (Gross, 1998). • Situation selection: Avoids or creates an emotionally relevant situation. • Situation modification: Modifies a situation to change the emotional state. • Attentional deployment: Distracts one’s attention away from a state. • Cognitive change (Reappraisal): Reinterprets the meaning of an event. • Response modulation: Attempts to directly influence experiential, behavioural, and physiological response systems.

Review of related works We have reviewed prominent research studies in the area of ITSs based on the selection criteria stated earlier. It is worth noting that in most studies, researchers did not specifically state the strategies they have applied in designing the feedback component for managing user negative emotions. However, these strategies can be categorised as emotion regulation strategies and coping strategies. D’mello and Graesser (2012) designed evaluated two systems; AutoTutor and Affective AutoTutor. AutoTutor is an ITS that helps students to learn complex technical content in Newtonian physics, computer literacy, and critical thinking. AutoTutor is quite effective in helping students learn by holding a conversation in natural language, simulating the pedagogical and motivational strategies of human tutors and modeling and responding to their cognitive states. The affect-sensitive versions of AutoTutor, called the Supportive and Shakeup tutors, are collectively referred to as Affective AutoTutor were also developed.The emotional sensitive version of AutoTutor is capable in detecting learner’s emotional states, regulating negative emotional states, and synthesize emotions of the animated pedagogical agent. The agent’s feedback has been designed based on reactions to the emotional states of boredom, frustration, and confusion. The agent’s action to students’ negative emotions were derived from two sources, which are theoretical foundation (attribution theory and cognitive disequilibrium during learning (Craig, Graesser, Sullins, & Gholson, 2004) and recommendation by pedagogical experts. The attribution theory addresses boredom and frustration using empathetic responses from the tutor. The cognitive disequilibrium theory is also applied to address confusion, when a learner enters a state of confusion. Staying in a state of cognitive disequilibrium for too long is not recommended and the tutor should display empathy to acknowledge the learner’s attempts and lead the learner out of the state of confusion.

438

D’mello and Graesser (2012) have experimented with 36 undergraduate students from a university in the U.S using three computer literacy applications. Proportional learning gains were computed with different type of AutoTutor (regular, Supportive, Shakeup). The ANOVA analysis did not show any significant influence of tutor type or tutor. However, there was a 0.18 sigma trend in favor of the Supportive tutor compared to the regular tutor and a 0.28 sigma trend for the Supportive tutor over the Shakeup tutor. The results also have shown that, Supportive AutoTutor was more effective than the regular tutor for low-domain knowledge students in some sessions and the students with more knowledge never benefited from Supportive AutoTutor. They advise that, system should not be supportive until the students need support. In the Wayang intelligent tutor system proposed by Woolf et al. (2009), variety of heuristic policies to respond to a learner’s emotions (providing text messages, mirroring student actions) were used. They investigated five independent emotional variables, including frustration, motivation, self-confidence, boredom and fatigue. The tutor responded to these emotional states by providing empathetic responses, agent change voice and gesture, presenting graphs and hints, giving encouragement, attributing failure to external factors, and changing the scenario. These types of responses are considered as problem-focus coping strategy (providing graphs and hints) or emotional-focus coping strategy (empathy messages) and emotion regulation strategies like situation modification (change the scenario) and cognitive reappraisal (attribute failure to external factors). They have measured interventions in relation to their impact on a student’s affect, behaviour and learning. Chaffar et al. (2009) recognised a learner’s emotional responses after some tutoring of data structure web courses. They simulated two situations for the users. For the first situation, the tutor used problem-focused actions (using examples or definitions to change the situation that causes the negative emotion) and emotion-focused actions (helping participants to change their way of sensing the situation) to alleviate the effects of any negative emotion produced. For the second situation, after providing evaluation marks to students, the tutor used three emotionfocused actions, including encouragement, recommendation and congratulation as a way to encourage students to improve their marks and their knowledge in the future. The results of the ANOVA test showed that learners need help in understanding instead of encouragement when they did not understand the course. Hence, using a problemfocused action during learning was proposed. The results revealed that recommendation and encouragement actions have positive effects on the emotional states of weak learners after receiving their marks. Strain and D‘Mello (2011), have analysed the effects of cognitive reappraisal (an emotion regulation strategy) on learners’ emotional states and comprehension scores during a reading comprehension task. First, they injected negative emotions to participants. Next, they manage their negative emotions using two forms of cognitive reappraisal (deep and shallow reappraisal conditions). Subsequently, in a web-based learning session, participants were asked to learn about the U.S. Constitution and Bill of Rights and then answer questions about what they had learned. The results show that the utilisation of cognitive reappraisal as an emotion regulation strategy lead to more positive activating emotions and better reading comprehension. Zakharov et al. (2008), used agent in their ITS to respond to students’ actions. Agent’s response is managed by a set of rules made in relation to the students’ cognitive states and emotional states. Each rule determines the agent’s verbal responses as well as changes to the agent’s emotional appearances. For example, when a learner has submitted a wrong answer several times, the agent’s verbal responses include a list of errors, with the appropriate emotional facial expression. Making the student conscious of their negative states may distract them from their negative feelings and move them towards their goal. Zakharov et al. (2008) used emotion coping strategies and regulation in designing feedback to reduce the negative emotions of learners. To evaluate the effectiveness of using the emotional agent in EITS, they performed an experiment in an introductory database course, with the experimental group that uses the emotion-aware version of the agent, while the control group had the emotion-unaware version of the agent. Since the learning sessions with ITSs were short, the researchers did not expect to observe significant difference in learning performance between experimental and control group. The comparison among different conditions was made based on the questionnaire responses. In general, the findings supported the presence of emotional educational agents, with the emotion aware agent having advantages over its non-emotional counterpart. Mao and Li (2009), proposed “Alice” an IETS with an emotion agent tutor. Alice was capable of recognising emotional states of a learner through facial expression, speech and text, and could adapt to emotional states of the learner with synthesized facial expression (providing empathy), emotional speech synthesis and text produced by the Artificial Intelligence Markup Language (AIML) Retrieval Mechanism. They consulted human teachers’ on suitable 439

educational and emotional actions for each scenario that can be applied by the agent in different learning situations. Mao and Li (2009), believed that emotional-aware agents incorporated in ITSs can optimise the learner behaviour towards learners’ enjoyment of the learning situation, though they did not report the result of any type of evaluation on their proposed ITS system. However, Mao and Li (2010) have conducted a pilot study, where 100 students used their proposed system to investigate the critical factors that impact learners’ satisfaction when using EITSs. It was found that the agent tutor’s pedagogical action and expressiveness of the emotion expression) are two of the significant factors in learners’ satisfaction from EITSs. Tian et al. (2014) have proposed architecture of interactive text-oriented emotion compensation mechanism in eLearning to compensate the lack of emotion interaction between teachers and students in e-Learning systems. Their framework, based on affective computing and active listening strategy, recognizes and regulates the e-learner’s emotions based on interactive Chinese texts. They analyses the textual interaction data such as chartrooms for courses, online Q&A, and group discussion for emotion recognition. Tian et al. (2014) introduced emotion regulation based on active listening, which is a non-judgmental feedback to an emotionally distressed individual, focuses on providing feedback of the emotional content itself. Active listening is an effective approach to regulate one’s emotion in real life (McNaughton, Hamlin, McCarthy, Head-Reeves, & Schreiner, 2008; Nugent & Halvorson, 1995). Tian et al. (2014) have applied a text-oriented emotion classification method to identify the e-learner’s emotion after he/she types a sentence, which is the listening step in active listening strategy. Furthermore, the case-based reasoning algorithm is adopted to recommend a similar emotion regulation such as a text-based advice, when the e-learner’s emotion is in negative states such as boredom, frustration, and fury. Tian et al. (2014) designed an Emotion Regulation Agent (ERA) which is the core of the whole emotion compensation mechanism in the proposed framework. It analyses the topics, speakers’ roles and the interaction features of user input texts for the purpose of emotion detection from the text and predicts each learner’s emotion trend. In addition, it decides on the preference of emotion regulation strategies, so that e-Learners’ negative emotions can be regulated. The emotion regulation strategy library was constructed by including successful emotion regulation case base, according to the classification of Gross’s emotion regulation strategies (Gross, 2001). The success cases and the corpus were collected and labelled manually. The successful emotion regulation cases and some typical cases offered by psychologists are standardized into structured case templates and stored in the case bases. Finally, the computation of similarity between short sentences and emotion regulation instances is presented. As an example, in a scenario of student group discussion, three-learner writes in the chat-box that he/she is depressed because his/her teacher criticized his/her report. Topic detection and tracking method is used to identify whether this is a new event or a recipient’s input. If the sentence was identified as new event, the sentence will be labelled and the emotional state and event content will be extracted. Every successful emotion regulation instance is shown in a format such as Event_set, Event_type, Emotion_category, Regulation_strategy, in emotion strategy database. The former two elements of an EES (Event Emotion regulation-in-Stance) are formed as (‘‘his teacher criticized his report,” ‘‘frustration’’). They used this feature to match the former element of EES of instances in the emotion regulation strategy database (Tian et al., 2014). Based on the recommended successful instances, a listener who communicates with depressed e-learner will suggest him/her or say something like “communicate with the teacher”. This will comfort/relax the e-learner’s negative emotions. They have not reported evaluation of a comprehensive experiment on the real learners using their proposed framework. Rodrigo et al. (2012), studied the emotional states of students while using ITS for Scatterplots with and without an interactive software agent (embodied conversational agent (ECA). Scatterplot Tutor (Baker et al., 2006) is an ITS that teach learners how to make and interpret scatterplots. This tutor was originally designed as part of the Middle School Mathematics Tutor and previously found to lead more learning gains (Baker, Corbett, Koedinger, Evenson, 2005). Scooter the Tutor was added to a Cognitive Tutor for Scatterplots, Scooter (which is an agent) reacted to gaming behavior with a combination of metacognitive messages (including requests to stop gaming), expressions of positive and negative emotion, and supplementary exercises covering the material the student avoided through gaming. Scooter successfully lessened gaming behavior of learner and significantly influence on students’ learning as compared to the same tutoring system with no agent. Baker et al. (2005), repeated the same experiment with students from a high school in Quezon City, Philippines. The participants are between the ages of 12 to 14. Data were collected from 126 students (64 in experimental condition and 62 in control condition), using a quantitative field observation method originally proposed by Pekrun et al. (2010). The findings from Rodrigo et al. (2012), were not the same as in Baker et al. (2005). Although Scooter is well favored by students and improves student learning 440

outcomes relative to the original tutor, emotional states and transitions between emotional states (emotion dynamics) were very similar between students in both conditions. In other word, Scooter did not have a significant influence on students’ emotional states or their dynamics. Boredom, confusion, and engaged concentration persisted in both conditions almost equally. One of the main finding of their research which was confirmed by other researches in emotion dynamics (Baker, D’Mello, Rodrigo, & Graesser, 2010; Kort, Reilly, & Picard, 2001), is that student emotion within learning software is quite stable as a cycle, regardless of whether or not software agents are present; the same student often is experiencing the same affect in following observations (180 seconds apart). For this reason, they believe that the first educational intervention that fundamentally modify students’ moment-to-moment emotion during learning, especially avoiding vicious cycles and creating and fortify virtuous cycles, will have made a major contribution and a major difference to learners. Table 1 shows the summary of comparison among these studies mainly based on emotion regulation strategies that were used as well as their results. Table 1. Summary of comparison among different using emotion management strategies in ITSs Citation Application Regulation strategy Results Strain & D‘Mello Web-based Applying cognitive reappraisal Leads to more positive emotions 2011 learning system (deep and shallow reappraisal ) as and better reading comprehension emotion regulation strategies score. Chaffar et al., 2009 A virtual tutor that Problem-focused actions Problem-focused action leads to teaches data (providing an example or a induce positive emotion during structure definition) and emotion-focused the comprehension task. actions (change its way of Recommendation and perceiving the situation) encouragement actions have positive effects on the learners' emotions after receiving their marks. Woolf et al., 2009 Wayang Intelligent Emotion-focused coping strategies The interventions are measured in Tutor (teaching (providing empathetic responses, relation to their impact on student mathematic) agent change voice and gesture emotion, behaviour and learning. and encouragement). Problemfocused coping strategies (present graphs & hints) and Emotion regulation strategies (attribute failure to external, change the scenario) D’mello & AutoTutor An animated pedagogical agent Supportive (emotion sensitive) Graesser, 2012 was used to regulate negative AutoTutor was more effective emotional states such as than the regular tutor for lowfrustration and boredom based on domain knowledge students in Attribution theory , Cognitive some sessions and the students disequilibrium and experts with more knowledge have not recommendation benefited from Supportive AutoTutor. Learning gains for the Shakeup and regular tutors were almost the same. Zakharov et al., Intelligent Tutor Problem-focused coping strategies Based on learner’s opinions, ITS 2008 system (teaching (presenting the list of learner equipped with emotion-aware database design errors) and Emotional-focused version of the agent has skill) coping strategies (change in the advantage over its non-emotional agent’s emotional appearance to counterpart. However, there is no empathise with the learner) expectation for observing significant difference in learning performance measures because of short learning session. 441

Mao & Li, 2009

Intelligent elearning system (Teaching concept of affective computing

Emotional-focused coping strategies (adapt to emotional state of learner with facial expression generation (providing empathy), emotional speech synthesis)

Tian et al., 2014

Chinese TextBased e-Learning system

Using emotion regulation based on active listening to provide sincere, non-judgmental emotional content feedback to an emotionally distressed individual

Rodrigo et al., 2012

Intelligent tutoring system for Scatterplots

An interactive software agent (Scooter) reacted to gaming behaviour with a combination of Emotion-focused coping strategies (metacognitive messages including requests to stop gaming, expressions of positive and negative emotion)and problem focused strategy(supplementary exercises)

Emotional-aware agent in ITS influence the mood states of the learner, or create positive impression. Agent tutor’s pedagogical action and agent tutor’s expressiveness are two important factors in learner satisfaction from using EITSs. A depressed e-learner will receive some sincere feedback according to the recommended successful instances. This will comfort/relax the e-learner’s negative emotions. However, They have not reported evaluation of a comprehensive experiment on the real learners using their proposed framework. The students are attracted to the agent (Scooter) and it enhances student learning outcomes relative to the original tutor, however the Scooter did not have a significant influence on students’ emotional states or their dynamics. Boredom, confusion, and engaged concentration persisted in both conditions (with and without embedded agent for emotion regulation) quite equally.

Discussion, conclusion and future work Arousing negative emotional states such as anxiety, frustration, and boredom in learning environments as well as computerised learning such as ITS is inevitable. This situation may be due to the mismatch between the learner’s character and needs as well as the available functionality in ITS, inadequate interface implementations, system limitations, lack of flexibility, occurrences of errors and crashes. These factors contribute to the learner’s emotional state and filtering out negative emotion is difficult. These negative emotions can have severe consequences on students’ metacognitive and cognitive processes and their learning gains. Thus, ITSs should be equipped with the ability of guiding learners to regulate their negative emotions to achieve positive learning outcomes. Hence, in this paper, we investigated the findings of eight prominent researches related to emotion-sensitive computerized learning systems by applying different emotion coping and regulation strategies on the feedback provided to learners. We were interested to identify the strategies that are successfully and effectively implemented into EITSs to regulate negative emotional state of learners. We also study potential emotion regulation strategies that can be effectively implemented in ITSs. Based on our findings and learner’s opinions, ITS equipped with emotion-aware agent has advantage over its nonemotional counterpart. Emotional-aware agent in ITS may influence the mood of the learner, or create positive impression. Agent tutor’s pedagogical action and agent tutor’s expressiveness are two important factors in learner motivation and satisfaction from using EITSs. The students are attracted to the agent and it enhances student learning outcomes relative to the original tutor. However, there are a few cases where the agent (e.g., Scooter) did not have a significant influence on students’ emotional states. Researches also show that emotion sensitive ITS is usually more effective than the non-emotionally sensitive for low-domain knowledge students compared to high-knowledge students. In addition, providing feedback in EITS based on emotion coping strategies were influential in regulating negative emotion of learners. 442

In most of the existing studies, emotion coping strategies including problem focused and emotion focused strategies are widely and effectively used in regulating negative emotion of learners in EITS. However, emotion regulation strategies proposed by Gross (2001) are rarely used in ITSs. Among the reviewed studies, Strain and D‘Mello (2011), looks at the possibility of applying emotion regulation strategies such as cognitive reappraisal in intelligent tutoring system. As appraisal plays a serious role in the generation and experience of emotion, reappraisal changes the emotion experienced by the learner (Gross & Thompson, 2007). Findings from their experiments have demonstrated instructing learners to reappraise negative emotional states as they arise to help learners avoid negative affect and achieve better learning outcomes. In other words, instructed reappraisal (IR) involves instructing individuals to think of a negative situation positively to make the emotional experience less negative. Since IR strategies could help learners to become more engaged and improve their learning gains, such strategies could be implemented and employed in ITSs. Therefore, ITSs that are capable to recognize quickly the learners shift toward negative emotional state, can prompt learners to involve in IR strategies that will help them manage these negative emotional states and become affectively engaged. The limitation of IR strategies is that these methods have not been experimented on challenging topics such as mathematics or physics. Hence, effectiveness of instructing learners to reappraise negative emotional states during learning of challenging topics is still not known conclusively. Our future research is to explore other strategies of emotion regulation like response modulation (suppressing emotional responses to stimuli) or attentional deployment to determine if these strategies are effective in changing negative emotional states of learners and their learning outcomes. Then they have potential to be implemented in the computerised learning environments.

References Aghaei Pour, P., Hussain, M. S., AlZoubi, O., D’Mello, S., & Calvo, R. A. (2010). The Impact of system feedback on learners’ affective and physiological states. In V. Aleven, J. Kay, & J. Mostow (Eds.), Intelligent Tutoring Systems (Vol. 6094, pp. 264– 273). Springer Berlin Heidelberg. doi:10.1007/978-3-642-13388-6_31 Ahn, H., & Picard, R. W. (2005). Affective-cognitive learning and decision making: A Motivational reward framework for affective agents. In J. Tao, T. Tan, & R. W. Picard (Eds.), Proceeding of the 1st International Conference on Affective Computing & Intelligent Interaction (ACII 2005) (Vol. 3784 LNCS, pp. 866–873). doi:10.1007/11573548_111 Aleven, V. A., & Koedinger, K. R. (2002). An Effective metacognitive strategy: Learning by doing and explaining with a computer-based cognitive tutor. Cognitive Science, 26(2), 147–179. Anderson, J. R., Douglass, S., & Qin, Y. (2005). How should a theory of learning and cognition inform instruction?. In A. F. Healy (Ed.), Experimental cognitive psychology and its applications. (pp. 47–58). American Psychological Association. doi:10.1037/10895-004 Avramides, K., & Du Boulay, B. (2009). Motivational diagnosis in ITSs: Collaborative, reflective self-report. In Frontiers in Artificial Intelligence and Applications (Vol. 200, pp. 587–589). Amsterdam, The Netherlands: IOS Press. Baker, R. S. J. D., Corbett, A. T., Koedinger, K. R., Evenson, S., Roll, I., Wagner, A. Z., Naim, M., Raspat, J., Baker, D. J., Beck, J. E. (2006). Adapting to when students game an intelligent tutoring system. In M. Ikeda, K. D. Ashley, & T. W. Chan (Eds.), Proceedings of the 8th International Conference on Intelligent Tutoring Systems (pp. 392–401). Berlin, Germany: Springer. doi:10.1007/11774303_39 Baker, R. S. J. D., D’Mello, S. K., Rodrigo, M. M. T., & Graesser, A. C. (2010). Better to be frustrated than bored: The incidence, persistence, and impact of learners’ cognitive–affective states during interactions with three different computer-based learning environments. International Journal of Human-Computer Studies, 68(4), 223–241. doi:10.1016/j.ijhcs.2009.12.003 Baker, R. S., Corbett, A. T., Koedinger, K. R., & Evenson, S., & Mitchell, T. (2005). Designing intelligent tutors that adapt to when students game the system (Doctoral Dissertation, Carnegie Mellon University, Pittsburgh). Retrieved from http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.92.9222 Calvo, R. A., & Mello, S. K. D. (2011). Introduction – Theoretical perspectives. In R. A. Calvo & S. K. D’Mello (Eds.), New Perspectives on Affect and Learning Technologies (pp. 3–10). New York, NY: Springer New York. doi:10.1007/978-1-44199625-1 Chaffar, S., Derbali, L., & Frasson, C. (2009). Inducing positive emotional state in intelligent tutoring systems. In V. Dimitrova, R. Mizoguchi, B. Du Boulay, & A. Graesser (Eds.), Proceedings of 14th International Conference on Artificial Intelligence in Education (Vol. 2009, No, 200, pp. 716–718). Amsterdam, The Netherlands: IOS Press. doi:10.3233/978-1-60750-028-5-716 443

Connor, K. M., & Davidson, J. R. T. (2003). Development of a new resilience scale: The Connor-Davidson Resilience Scale (CDRISC). Depression and Anxiety, 18(2), 76–82. doi:10.1002/da.10113 Craig, S., Graesser, A., Sullins, J., & Gholson, B. (2004). Affect and learning: An Exploratory look into the role of affect in learning with AutoTutor. Journal of Educational Media, 29(3), 241–250. doi:10.1080/1358165042000283101 D’mello, S., & Graesser, A. (2012). AutoTutor and affective AutoTutor: Learning by talking with cognitively and emotionally intelligent computers that talk back. ACM Transactions on Interactive Intelligent Systems, 2(4), 1–39. doi:10.1145/2395123.2395128 Desmet, P. (2002). Designing emotions (Unpublished doctoral dissertation). Delft University of Technology, Delft, The Netherlands. Du Boulay, B., Rebolledo Méndez, G., Luckin, R., & Martínez-Mirón, E. (2007). Motivationally intelligent systems: Diagnosis and feedback. In Proceedings of the 2007 conference on Artificial Intelligence in Education: Building Technology Rich Learning Contexts That Work (pp. 563–565). Amesterdam, The Netherlands: IOS Press. Retrieved from http://dl.acm.org/citation.cfm?id=1563601.1563697 Erez, A., & Isen, A. M. (2002). The Influence of positive affect on the components of expectancy motivation. Journal of Applied Psychology, 87(6), 1055–1067. doi:10.1037/0021-9010.87.6.1055 Gross, J. J. (1998). The Emerging field of emotion regulation: An Integrative review. Review of General Psychology, 2(3), 271– 299. doi:10.1037/1089-2680.2.3.271 Gross, J. J. (2001). Emotion regulation in adulthood: Timing is everything. Current Directions in Psychological Science, 10(6), 214–219. doi:10.1111/1467-8721.00152 Gross, J. J., & Thompson, R. A. (2007). Emotion regulation: Conceptual foundations. In J. J. Gross (Ed.), Handbook of emotion regulation (pp. 3–24). New York, NY: Guilford Press. Isen, A. M. (2001). An Influence of positive affect on decision making in complex situations: Theoretical issues with practical implications. Journal of Consumer Psychology, 11(2), 75–85. doi:10.1207/S15327663JCP1102_01 Isen, A. M., Daubman, K. A., & Nowicki, G. P. (1987). Positive affect facilitates creative problem solving. Journal of Personality and Social Psychology, 52(6), 1122–1131. doi:10.1037/0022-3514.52.6.1122 Isen, A. M., Shalker, T. E., Clark, M., & Karp, L. (1978). Affect, accessibility of material in memory, and behavior: A cognitive loop?. Journal of Personality and Social Psychology, 36(1), 1–12. doi:10.1037/0022-3514.36.1.1 Jaques, P. A., & Vicari, R. M. (2007). A BDI approach to infer student’s emotions in an intelligent learning environment. Computers and Education, 49(2), 360–384. Klein, J., Moon, Y., & Picard, R. W. (2002). This Computer responds to user frustration—Theory, design, and results. Interacting with Computers, 14(2), 119–140. doi:10.1016/S0953-5438(01)00053-4 Kort, B., Reilly, R., & Picard, R. W. (2001). An Affective model of interplay between emotions and learning: Reengineering educational pedagogy-building a learning companion. In Proceedings of IEEE International Conference on Advanced Learning Technologies, ICALT 2001 (pp. 43–46). doi:10.1109/ICALT.2001.943850 Lazarus, R. S. (1991). Emotion and adaptation. New York, NY: Oxford University Press. Lazarus, R. S., & Folkman, S. (1984). Stress, appraisal, and coping. New York, NY: Springer Publishing Company. Lester, J. C., Converse, S. A., Kahler, S. E., Barlow, S. T., Stone, B. A., & Bhogal, R. S. (1997). The Persona Effect: Affective Impact of Animated Pedagogical Agents. In Proceedings of the SIGCHI conference on Human factors in computing systems CHI ’97 (pp. 359–366). New York, NY: ACM Press. doi:10.1145/258549.258797 Mao, X., & Li, Z. (2009). Implementing emotion-based user-aware e-learning. In Proceedings of the 27th international conference extended abstracts on Human factors in computing systems - CHI EA ’09 (pp. 3787-3792). New York, NY: ACM Press. doi:10.1145/1520340.1520572 Mao, X., & Li, Z. (2010). Agent based affective tutoring systems: A Pilot study. Computers & Education, 55(1), 202–208. doi:10.1016/j.compedu.2010.01.005 McNaughton, D., Hamlin, D., McCarthy, J., Head-Reeves, D., & Schreiner, M. (2008). Learning to listen: Teaching an active listening strategy to preservice education professionals. Topics in Early Childhood Special Education, 27(4), 223–231. doi:10.1177/0271121407311241

444

Neji, M., Ben Ammar, M., Alimi, A. M., & Gouardères, G. (2008). Agent-based framework for affective intelligent tutoring systems. In B. Woolf, E. Aïmeur, R. Nkambou, & S. Lajoie (Eds.), Intelligent Tutoring Systems SE - 71 (Vol. 5091, pp. 665–667). Springer Berlin Heidelberg. doi:10.1007/978-3-540-69132-7_71 Nugent, W. R., & Halvorson, H. (1995). Testing the effects of active listening. Research on Social Work Practice, 5(2), 152–175. doi:10.1177/104973159500500202 Pekrun, R., Goetz, T., Daniels, L. M., Stupnisky, R. H., & Perry, R. P. (2010). Boredom in achievement settings: Exploring control–value antecedents and performance outcomes of a neglected emotion. Journal of Educational Psychology, 102(3), 531– 549. doi:10.1037/a0019243 Pekrun, R., Goetz, T., Titz, W., & Perry, R. P. (2002). Academic emotions in students’ self-regulated learning and achievement: A Program of qualitative and quantitative research. Educational Psychologist, 37(2), 91–105. doi:10.1207/S15326985EP3702_4 Picard, R. W. (1999). Affective computing for HCI. In Proceedings of the 8th HCI International on Human-Computer Interaction: Ergonomics and User Interfaces (pp. 829–833). Mahwah, NJ: Lawrence Erlbaum Associates, Inc. Retrieved from http://dl.acm.org/citation.cfm?id=647943.742338 Picard, R. W. (2010). Affective computing: From laughter to IEEE. IEEE Transactions on Affective Computing, 1(1), 11–17. Rebolledo-Mendez, G., du Boulay, B., & Luckin, R. (2006). Motivating the learner: An Empirical evaluation. In M. Ikeda, K. Ashley, & T. W. Chan (Eds.), Intelligent Tutoring Systems SE - 54 (Vol. 4053, pp. 545–554). doi:10.1007/11774303_54 Rodrigo, M. M. T., Baker, R. S. J. D., Agapito, J., Nabos, J., Repalam, M. C., Reyes, S. S., & San Pedro, M. O. C. Z. (2012). The Effects of an interactive software agent on student affective dynamics while using ; An Intelligent tutoring system. IEEE Transactions on Affective Computing, 3(2), 224–236. doi:10.1109/T-AFFC.2011.41 Schlosberg, H. (1954). Three dimensions of emotion. Psychological Review, 61(2), 81–88. doi:10.1037/h0054570 Strain, A., & D‘Mello, S. (2011). Emotion regulation during learning. In G. Biswas, S. Bull, J. Kay, & A. Mitrovic (Eds.), Artificial Intelligence in Education SE - 103 (Vol. 6738, pp. 566–568). Springer Berlin Heidelberg. doi:10.1007/978-3-64221869-9_103 Tian, F., Gao, P., Li, L., Zhang, W., Liang, H., Qian, Y., & Zhao, R. (2014). Recognizing and regulating e-learners’ emotions based on interactive Chinese texts in e-learning systems. Knowledge-Based Systems, 55, 148–164. doi:10.1016/j.knosys.2013.10.019 Woolf, B., Burleson, W., Arroyo, I., Dragon, T., Cooper, D., & Picard, R. (2009). Affect-aware tutors: Recognising and responding to student affect. International Journal of Learning Technology, 4(3/4), 129. doi:10.1504/IJLT.2009.028804 Zakharov, K., Mitrovic, A., & Johnston, L. (2008). Towards emotionally-intelligent pedagogical agents. In B. P. Woolf, E. Aïmeur, R. Nkambou, & S. Lajoie (Eds.), Intelligent Tutoring Systems, 9th International Conference, ITS 2008 (Vol. 5091, pp. 19–28). doi:10.1007/978-3-540-69132-7_7

445