Kort’s Learning Spiral Model 

Context in source publication

Context 1

... The core of the platform includes a number of "smart classrooms" distributed around Shanghai, the Yangtze River delta, and even in remote western regions of China such as Tibet, Yan’an, Xing Jiang, and Nin Xia. These classrooms are equipped with smart devices/sensors and specially developed software for distance learning. For example, the touch screen of the room displays presentations (e.g. PowerPoint), while also acts as a whiteboard for handwriting. The instructor can write on materials projected on the screen using a laser E-Pen. To optimize the video quality, a pan-camera can follow the instructor when he/she moves around in the classroom. RFID (Radio Frequency IDentifier) tags are used to identify and track students. Another tracking camera is mounted in the front of the classroom and it captures students’ attention status by recognizing the ‘blink frequency’ of their eyes. During the class session, instructors can load their pre-prepared PowerPoint and Word documents and write on the whiteboard (even when they are away from the whiteboard). The students can also write individual notes on the instructors’ handwriting window. All these classroom activities are being transmitted through various technologies to students at a distance. They are also recorded and archived for later review. Using this hi-tech environment, the teacher can move freely in the room, use his body language to communicate, and also interact with learners naturally and easily as in a traditional face-to-face classroom. This online college has about 17,000 Students, and 99% of them are working professionals who attend school part time. Therefore, their academic backgrounds, knowledge, and skills vary a great deal. Given such diversity, it is important to provide personalized learning services. The Shanghai system has harnessed data-mining technologies to organize learning communities and provide learning content recommendation based on student profiles (L. P. Shen & Shen, 2005; R. M. Shen, Yang, & Han, 2003). The large number of students in this college and its expansive course delivery systems make it an ideal place to test new and emerging technologies. The goal of this study is to understand how learners’ emotions evolve during learning process, so as to develop learning systems that recognize and respond appropriately to their emotional change. This paper also proposes a prototype of an affective e-Learning model that combines learner’s emotions with the Shanghai e-Learning platform. As Picard (2004) stated, theories of emotional impact on learning need to be tested and further developed. Until today, there is still a lack of comprehensive and empirically validated theories about emotion and learning. In this experiment, we examined several of the existing emotion theories in learning to help construct our affective e- Learning model. We used Russell’s ‘circumplex model’ to describe user’s emotion space. We then used the Kort’s ‘learning spiral model’ as the starting point to explore the affective evolution during learning. Following is the description of these models and our proposal for a prototype of an affective e-Learning model. In our search for an emotion theory we focused on dimensional models instead of cognitive appraisal model for user emotion modeling because they cover the feeling of emotional experience both on a low level and a higher, cognitive level. One well-established dimensional model is psychologist Russell’s circumplex model of affect (Russell, 1980) where emotions are seen as combinations of arousal and valence. In Russell’s model, emotions are distributed in a system of coordinates where the y-axis indicates the degree of arousal and the x-axis measures the valence, from negative to positive emotions. The Russell’s model is widely used in recent researches (Craig et al., 2004; Fagerberg et al., 2004; Kort et al., 2001; Leon et al., 2007; Picard et al., 2001). And most of these just explored from three to eight basic emotions. Though Kort et al. (2001) proposed five sets of about thirty emotions that may be relevant to learning, however, we believe that skilled human tutors and teachers react to assist students based on a few ‘least common set’ of affect as opposed to a large number of complex factors; thus, we carefully select a basic learning emotion set which we deem most important for shaping our affective learning model. The basic set includes the most important and frequently occurred emotions during learning, namely, interest, engagement, confusion, frustration, boredom, hopefulness, satisfaction and disappointment. They might not be placed exactly the same for all people when put in the Russell’s two-dimension emotion space, because this model focuses on subjective experiences. Figure 2 is an example of two-dimensional basic learning emotion space. We anticipate revising this emotion set when the study progresses. Kort, Reilly and Picard (2001) proposed a four quadrant learning spiral model (Figure 3) in which emotions change while the learner moves through quadrants and up the spiral. In quadrant I the learner is experiencing positive affect and constructing knowledge. At this point, the learner is working through the material with ease and has not experienced anything overly puzzling. Once discrepancies start to arise between the information and the learner’s knowledge structure, he/she moves to quadrant II, which consists of constructive learning and negative affect. Here the learner experiences affective states such as confusion. As the learner tries to sort out the puzzle but fails, he might move into quadrant III. This is the quadrant of unlearning and negative affect, when the learner experiences emotions such as frustration. After the misconceptions are discarded, the learner moves into quadrant IV, marked by ‘unlearning’ and positive affect”. While in this quadrant the learner is still not sure exactly how to move forward. However, he/she does acquire new insights and search for new ideas. Once the learner develops new ideas, he/she is propelled back into quadrant I; thus, concluding one cycle around the learning spiral of Kort’s model. As the learner move up the spiral, cycle after cycle, he/she become more competent and acquire more domain knowledge. Kort, Reilly and Picard (2001) also described the empirical research methods to validate this spiral model, and promised a future paper to report the results of the empirical research. There were some efforts of empirical research on this model, e.g. Kort & Reilly (2002) stated that “we are in the process of performing empirical research on this model”, and “ideally, the Learning Companion should observe and try to understand the processes a learner experiences during all of these quadrants; however, this is currently beyond the capabilities of the technology”(Kapoor, Mota, & Picard, 2001) , and “The process of cycling through these emotions during learning is currently being investigated in the present project”(D'Mello et al., 2005). But to our best knowledge, there have been no empirical validation report about this model. We used this ideal learning spiral model as the starting point to explore learner’s emotional evolution during the learning process. For our study, we used Russell’s ‘circumplex model’ to describe learner’s emotions detected from biophysical signals, and used the Kort’s ‘learning spiral model’ as the starting point to explore learners’ emotional evolution during the learning process. Finally, based on our previous work (L. P. Shen, Leon, Callaghan, & Shen, 2007; L. P. Shen & Shen, 2005), we proposed a model of affective learning (Figure 4) focusing on how we could make use of the information when we have got the learner’s emotion states and their evolution. This affective learning model considers the contextual information of the learner and the learning setting, and generates appropriate responses to the learner, based on his/her emotional states, cognitive abilities, and learning goals. This model can also be used to customize the interaction between the learner and learning system, to predict learner responses to system behavior, and to predict learner's future interaction with the learning system. The affective learning model used a combination of (a) a cognitive appraisal approach to affective user modeling (which inferences emotions according to situations experienced by the user as well as the observable behavior of the user) and (b) a physiological approach. Figure 4 shows a high-level description of this affective learning model, which only displays the general factors involved. The upper part of the model was modified OCC (Ortony et al., 1990) cognitive appraisal model for emotions, and the lower part of the model was the physiology recognition of emotions, where these two models converge One of the pattern recognition method-Bayesian Networks (Duda, Hart, & Stork, 2000) were employed to model the relations between the emotional states and their causal variables. This affective model indicated that the user’s emotional states were related to learner profiles (e.g., learning preferences, cognitive skills, knowledge structure), goals of learning, and learner interaction with the learning system. The user’s emotional states would in turn influence the measurements of the available sensors. The advantage of having a model based on a Bayesian Network was that it could leverage any evidence available on the variables related to emotional states to make predictions for any other variable in the model. For instance, the user’s emotional state could be assessed using existing information on learner profile and the learning context, even in absence of reliable sensors. Or, we could assess both emotional state and learner profile from reliable sensors, learner behavior and system reactions. The emotion detection system helped the system to provide timely help or adequate content based on the emotions the learner ...