Learning Platform Emotional Learning
Ana Raquel Faria
Constantino Martins
Computer Science Department GECAD-Knowledge Engineering and Decision Support Group Institute of Engineering – Polytechnic of Porto
Computer Science Department GECAD-Knowledge Engineering and Decision Support Group Institute of Engineering – Polytechnic of Porto
Porto, Portugal
[email protected]
Porto, Portugal
[email protected]
Ana Almeida Computer Science Department GECAD-Knowledge Engineering and Decision Support Group Institute of Engineering – Polytechnic of Porto Porto, Portugal
[email protected]
Abstract—The purpose of this paper is to propose new model for emotional interaction that uses learning styles and student emotional state to adapt the user interface and learning path. This aims to reduce the difficulty and emotional stain that students encounter while interacting with learning platforms. To this end will be used techniques of Affective Computer that can capture the student emotional state and base on that change the course parameters (flow, organization or difficulty) or even an emotional interaction in order to recapture the student attention. Keywords - Learning Styles; Adaptive Systems; Affective Computing.
I.
INTRODUCTION
Learning is the act of obtaining or adapting existing knowledge and behaviours which in turn helps develop the numerous capacities of an individual [6]. These capabilities make an individual able to create personal relations with the environment in which they are inserted. For the learning process to develop the individual has to use its sensory, motor, cognitive, affective and linguistic [1]. Thus, this work purposes the study of the emotional effect in the use in learning systems, analyzing the extent to which emotional state can influence the thinking, decision making and learning process. In order to conduct this study of the effect of emotions in learning platforms, we must be able to capture the student's emotional state. The capture of the emotional state cannot be evasive therefore Affective Computing technologies was used. The technique chosen to be use was facial expression recognition, video is capture of the students and analyzed for its emotional content. Anticipating the creation of a model that combines one or more means of Affective Computing and correctly identifying a greater degree of reliability the emotional states of a student. This process will also include the study of emotional states that can be recognized, its duration in time, what is its influence on teaching and learning.
Ramiro Gonçalves Universidade de Trás-os-Montes e Alto Douro Vila Real, Portugal
[email protected]
Recognizing that the emotional state that provides a positive learning and what incentives can be introduced to reverse a negative emotional state [2] [3]. II.
BACKGROUND RESEARCH
This background research toke us to wide spread of areas, from the different branches of phycology to technical advances in computers and computers programs, passing by education and learning areas. The education, training, skills development, learning is something that we do constantly almost from the second we are born, and it is these characteristics, the ability to learn and teach, that allowed us to evolve as a person [4]. Thus learning is inevitably linked to History of Man, to its construction as a social being capable of adapting to new situations [5]. The concept of learning is described as the act of acquiring knowledge, behaviors, abilities, standards or preferences and the study of learning was been closely linked to the development of psychology as a science [6]. A growing amount of studies [7][8][9] also support the connection between emotion and the learning process, although this connection is a long way from being simple and direct. It is accepted that positive and negative emotion states can cause different kinds of thinking and can have an effect in the education perspective [2][3]. From the ample variety of human emotion we are forced to ask the question. What emotions are associated with e-learning? Previous study [10] show that the emotion could be deduced to five groups of emotions: frustration, fear, anxiety, apprehension, shame/embarrassment, enthusiasm/excitement and pride. Affective computer technologies have advanced throughout the years and these advances are direct linked to the same of most significant emotion theories that try to explain what is emotion and what cause emotion. These several viewpoints can be grouped into three main categories:
physiological, neurological, and cognitive. Physiological theories advocate that responses of the body are responsible for emotions. Neurological theories propose that activity within the brain leads to emotional reaction. Lastly, cognitive theories say that thoughts and other mental activity play an important role in the construction of emotions [11]. According to recent definitions, “Affective learning involves the melding of thinking and feeling in how people learn. Importance is placed on social learning environments for knowledge construction and application wherein deeper awareness and understanding of the role played by mental dispositions in how a person views, engages, and values learning can result in better understanding and use of knowledge and skills. Learning outcomes are focused on enculturation of norms, values, skillful practices, and dispositions for lifelong learning.” [12]. There are several Affective learning models, the use of a model is to understand how the emotions are evolving in the learning process like: • Russell’s Circumplex model [13] to describe user’s emotion space; •
Kort’s learning spiral model [14] to explore the affective evolution during learning process.
The subject of personality and learning styles was also explored for this work. The personality research aims to study what distinguish one individual from another [15]. Personality research depends on quantifiable concrete data that can be used to comprehend what people are like. Also the relationship between personality and learning is largely accepted [16][17][18]. One of the models largely use is the Big Five Model [19][20][21]. The Big Five Model dimensions are Openness, Conscientiousness, Extraversion, Agreeableness, and Neuroticism (OCEAN) each of its dimension has its own tendencies and motivation. In the learning style subject we can define a learning style as the strategies that allows an individual to learn best [22]. Different people learn in different ways and each one preferring a different learning style. Everyone has a mix of learning style, but some people may find that they have a dominate style of learning. Models that seems to be more adequate to apply in engineering process are: Learning Style Inventory model, propose by educational theorist David A. Kolb [23] and the Felder-Silverman model [24]. In this area of affective computing we have several uses and applications that can interact with our day to day life. These areas can be divided in five categories: security, health and medicine, military, home and education [4] [25]. In each category everyday are appearing new developments.
The architecture proposed for this prototype is composed of 4 major models: the Application Model, Emotive Pedagogical Model, Student Model, and Emotional Model. The student model consists in the user information and characteristics. This includes personal information (name, email, telephone, etc.), demographic data (gender, race, age, etc.), knowledge, deficiencies, learning styles, emotion profile, personality traits, etc. This information is use by the student model to better adapt the prototype to student. The emotion model gathers all the information the facial emotion recognition software and feedback of the students. The application model is compose by a series of modules contain different subjects. The subject consist in a number steps that the student has to pass in order to complete is learning program. Usually each subject is composed by a diagnostic test in order to access and update the student level of knowledge. Followed by the subject content in which the subject is explained and follow by the subject exercises and final test. The last model is the emotive pedagogical model that is composed by three sub-models: the rules of emotion adaptability, the emotional interaction mechanisms and the graph of concepts in case of failure. The rules of emotion adaptability consist in the way the subject content is presented. The subject content and subject exercises are presented according the learning style and personality of the student. The emotional interaction mechanisms consist in the trigger of an emotion interaction when is captured an emotion that need to be contradicted in order to facilitate the learning process. The emotions to be contradicted are: anger, sadness, confusion and disgusted. The interaction can depend on the personality and on the learning style of the student. Finally the graph of concepts in case of failure this indicates the steps to be taken when a student fails to pass a subject.
Application Model
Rules of emotion adaptability
Graph of concepts
III.
DEVELOPMENT
In order to prove that emotion can have influence in the learning process. A prototype was developed, a learning platform that takes into account the emotional aspect, the learning style and the personality traits, adapting the course (content and context) to the student needs.
Emotive Pedagogical Model
Emotional Interaction Mechanisms
Graph of concepts in case of failure
Student Model
Domain Dependent Data
Emotion Model
Facial Expressions Recognition
Domain Independent Data •Personality traits •Learning Styles •Emotional Profile
Figure 1. Architecture
Feedback
IV.
DATA ANALYSIS
We began our research based on the principal that emotion can influence several aspects of our lives. Emotion affects the decision process and knowledge acquisition of an individual. Therefore, they directly influence our perception, learning process, the way we communicate, and the way we make rational decisions. Our research goal is to see that if a learning platform that take into account the emotional profile of a student can improve the learning process, than a learning platform with no emotional interaction. A prototype was developed to test our assumption. This prototype takes into account the student’s personality, learning style and the emotional profile. A. Participants The participants in this prototype evaluation were the 1st year students of Higher Education Establishment of Computer Engineering of Oporto of two courses: Informatics Engineering and Systems Engineering. The total of students involved in these testes was 115 students with ages between 17 and 42 years old. This group of students was compose of 20% female (N=23) and 80% male (N=92) participants mainly from the districts of Oporto, Aveiro and Braga. B. Method To evaluate the prototype were conducted 2 pre-tests and a final test. The student’s participants in this prototype evaluation did not have any prior knowledge of the content of the subject approach by the prototype. For the first pre-test e students were group randomly in 2 groups as we can see in the table below. TABLE I. Groups Group v1 Group v2
GROUPS FOR THE FIRST PRE-TEST
Description Test the prototype with the emotional interaction and learning style with a high level of difficulty Test the prototype without the emotional interaction and learning style with a high level of difficulty
The Group v2 had to do diagnostic test (in paper) to help grade the initial knowledge of the student. Following the testing of the prototype and this time the students tested the prototype with same subject content as group v1 but without any emotional integration. After this test the students had to do a final test (in paper) to help grade the final knowledge of the student. The evaluation by this group also ended with the answer of the Acceptability questioner the same questionnaire given to group v1. Due to restrictions of schedule of the students we only had an hour to conduct the pre-test for each group, this proved to be inefficient because the students didn’t have enough time to complete the tasks required. Also was observed that due to the high level of the of difficulty of the application (80% to pass to the next task), the student that didn’t get that grade and had to start again, this proved to led the students to frustration and to give up the completion of the group task. Another problem was the capture of emotion that did not work properly. For this reasons was conducted a second pre-test for Groups v1 and Groups v2 with the same evaluation protocol but with more time, roughly 1½ hours and the application level of difficulty cut by half and this second pre-test was not conducted with the same students. These changes originated Groups v3 a Group v4 as we can see in the table below. TABLE II. Groups Group v3 Group v4
Description Test the prototype with the emotional interaction and learning style with a low level of difficulty Test the prototype without the emotional interaction and learning style with a low level of difficulty
The first and second pre-test were made by the students of the course of Informatics Engineering. For the final test, the experience was repeated with the students of the course of Systems Engineering, but only were tested groups v3 and v4 with a time increase, two hours to conduct the test. The following table shows the division of students between the groups of the two courses. TABLE III.
In each group evaluation process was different: The Group v1 had to do Diagnostic test (in paper) to help grade the initial knowledge of the student. Followed by the test of the prototype with the emotional interaction and learning style. This comprehends the completism of subject module composed by a diagnostic test in order to access and update the student level of knowledge. Followed by the subject content in which the subject is explained and follow by the subject exercises and final test. After this test the students had to do a final test (in paper) to help grade the final knowledge of the student. The evaluation by this group ended with the answer of the Acceptability questioner to determine the acceptability of the prototype. This questionnaire consists in first determine the acceptance of the prototype and second the degree of difficulty of use of each feature of the prototype.
GROUPS FOR THE SECOND PRE-TESEST
Groups Group v1 Group v2 Group v3 Group v4
GROUPS AND PARTICIPANTS COURSE OF INFORMATICS ENGINEERING Number of participants 31 31 14 11
TABLE IV. Groups Group v4 Group v5
Average Age
Female
Gender Male
18,3 18,4 21,29 18
10 9 2 0
20 22 12 11
GROUPS AND PARTICIPANTS COURSE OF SYSTEMS ENGINEERING Number participants 14 14
of
Average Age
Gender Female
19,2 19,4
1 1
Male 13 13
Form the data gather it was concluded that the distributions are not normal after having applied the Kolmogorov-Smirnov test shown in the following table. TABLE V.
TESTS OF NORMALITY
Groups Group v1
Kolmogorov-Smirnova Statistic df P 0,416 30 0,000 0,424 31 0,000
in Group v2 Group v3 0,349 Group v4 0,164 Group v1 0,238 Group v2 0,257 Final Test in paper Group v3 0,219 Group v4 0,227 *. This is a lower bound of the true significance. a. Lilliefors Significance Correction Diagnostic paper
Test
14 0,000 11 0,200* 30 0,000 31 0,000 14 0,046 11 0,118
Analyzing the results shown in Table V for group v4 for the diagnostic test has a p value with low bound but of true significance for the final test we also have p-value with some significance. All the other groups for the p values are not statistically significant. So we can conclude that the groups, with the exception of group v4, do not have normal distributions. The analyses of the student’s grades showed the following. The next table shows the descriptive statistics of the diagnostic test and final test across the groups.
TABLE VI. Groups Diagnostic Test in paper Final Test in paper Diagnostic Test in paper Group v2 Final Test in paper Diagnostic Test in paper Group v3 Final Test in paper Diagnostic Test in paper Group v4 Final Test in paper
DESCRIPTIVE STATISTICS N Minimum Maximum Mean
Std. Deviation
31
0
100
12.26%
22,9
31
0
100
36,00%
36,5
31
0
60
11,61%
20,5
31
0
100
34,19%
34,7
14
0
80
28,57%
37,4
14
0
100
67,14%
33,8
11
0
100
49,09%
36,2
11
0
100
56,36%
39,8
Group v1
As data gather does not have a normal distribution, so the two pairs of groups were compared using a non-parametric test Mann-Whitney Analyzing the data gather we found for the first pre-test due to its high level of difficulty and short time to perform the tasks relatively low mean scores. For group v1 for diagnostic test we have a mean of 12,26% (Standard Deviation(SD) = 22,9) and for the final test a mean of 36,00% (SD=36,5). For group v2 for diagnostic test we have a mean of 11,61% (SD =
20,5) and for the final test a mean of 34,19% (SD=34,7). The observed differences between these two groups are not statistically significant not only for the diagnostic test but also for the final test. For the second pre-test the means were a little higher than the previous test. For group v3 for diagnostic test we have a mean of 28,57% (SD = 37,4) and for the final test a mean of 67,14% (SD=36,2) and for group v4 for diagnostic test we have a mean of 49,09% (SD = 36,2) and for the final test a mean of 56,36% (SD=39,8). The observed differences between these two groups are also not statistically significant. Due the to the problems encounter in the two pre-tests the experience was repeated for the final test with an increase of time. Form the data gather form the final teste it was concluded that the distributions are not normal after having applied the Kolmogorov-Smirnov test shown in the following table. TABLE VII.
TESTS OF NORMALITY
Kolmogorov-Smirnova Statistic df P Group v4 0,229 14 0,046 Diagnostic Test in paper Group v5 0,184 14 0,200* Group v4 0,323 14 0,000 Final Test in paper Group v5 0,281 14 0,004 *. This is a lower bound of the true significance. a. Lilliefors Significance Correction Groups
Analyzing the results shown in Table VII only group v5 for the diagnostic test has a p value with low bound but of true significance. All the other groups for the p values are not statistically significant. So we can conclude that the groups, with the exception of group v5 in the diagnostic test, do not have normal distributions. The analyses of the student’s grades showed the following. The next table shows the descriptive statistics of the diagnostic test and final test across the groups. TABLE VIII. Groups
Group v4
Group v5
Diagnostic Test in paper Final Test in paper Diagnostic Test in paper Final Test in paper
N
DESCRIPTIVE STATISTICS Minimum Maximum Mean
Std. Deviation
14
0
100
45,7%
40,3
14
60
100
85,7%
12,2
14
0
80
37,1%
29,2
14
0
100
61,4%
33,7
As data does not have a normal distribution for the two groups they were compared using a non-parametric test MannWhitney. For group v4 for diagnostic test we have a mean of 45,7% (SD =40,3 ) and for the final test a mean of 85,7% (SD=12,2). For group v5 for diagnostic test we have a mean of 37,1% (SD
= 29,2) and for the final test a mean of 61,4% (SD=33,7). For the diagnostic test we have Mann–Whitney U = 83,0 and for a sample size of 14 students. For this analysis we found a P value of 0,479 which indicates that we don’t have any statistical difference which is understandable because it was assumed that all students had more or less the same level of knowledge. For the final test we have Mann–Whitney U =54,0 and for an equal sample size of the diagnostic test. For this analysis we found a P value of 0,029 in this case the differences observed are statistical different. Beside of the overall analyses of the performance of the students, a similar study was conducted to the data gather in each concept that the students had to learn. First we started analyze the data to find out its distribution and for that was applied the test Kolmogorov-Smirnov as we can see in the following table. TABLE IX.
TESTS OF NORMALITY Group
Kolmogorov-Smirnova Statistic df P Group v4 0,218 14 0,069 Binary Group v5 0,181 14 0,200* Group v4 0,290 14 0,002 Octal Group v5 0,319 14 0,000 Group v4 0,357 14 0,000 Hexadecimal Group v5 0,326 14 0,000 *. This is a lower bound of the true significance. a. Lilliefors Significance Correction
Analyzing the data for the subject of binary for group v5 we have a p value with low bound but of true significance a group v4 also have some significance. All the other groups the p values are not statistically significant. So we can conclude that the groups, with the exception of group v4 and group v5 in binary subject, do not have normal distributions. The next table shows the descriptive statistics of the subject result across the each group test. TABLE X. Group Binary Group Octal v4 Hexadecimal Binary Group Octal v5 Hexadecimal
DESCRIPTIVE STATISTICS
N Minimum Maximum Mean Std. Deviation 14 40 100 75,7 19,5 14 40 100 75,7 22,4 14 0 100 82,9 30,2 14 0 100 62,9 34,1 14 0 100 62,9 37,5 14 0 100 42,9 45,7
For group v4 for binary subject we have a mean of 75,7% (SD =19,5), for octal subject we have a mean of 75,7% (SD =22,4) and for hexadecimal we have a mean of 82,9% (SD =30,2). For group v5 for binary subject we have a mean of 62,9% (SD =34,1), for octal subject we have a mean of 62,9% (SD =37,5) and for hexadecimal we have a mean of 42,9% (SD =45,7). As the data does not have a normal distribution so the two pairs of groups were compared using a non-parametric test Mann-Whitney.
TABLE XI. Subject
Group Group v4 Binary Group v5 Group v4 Octal Group v5 Group v4 Hexadecimal Group v5
STATISTICAL COMPARISON USING MANN-WHITNEY TEST N Mean Mean Rank Mann–Whitney U P 14 75,7 15,86 79,0 0,336 14 62,9 13,14 14 75,7 15,57 83,0 0,468 14 62,9 13,43 14 82,9 18,18 46,5 0,012 14 42,9 10,82
Analyzing the results we have for the binary subject results have p-value of 0,336 in this case the result that the data is not statistical significant if higher then p≤ 0.05. The same occurs for the octal subject with a p value of 0.468. Analyzing the means of these two subjects they were both positive which leads to conclude that these two topics were fairly easy for the students to comprehend for both group tests. The same does not happened for the hexadecimal subject. In this case we have a p value of 0,012, which is significant at p≤ 0.05. Analyzing the means we can see that group v4 (application without emotional integration) struggled with this topic achieving a negative mean. Although the group v5 had no problem with this topic achieving an above average positive mean. V.
CONCLUSION
The current work aimed reduce the difficulty and emotional stain that student face while interacting with a learning platform in order to improve efficiency of the learning process. This led to propose a new approach for an adaptive learning system. This model will try to capture the emotional state of the student and together with his learning style and cognitive profile, will adapt the learning content and context to the learning requirements of the student. The prototype was called Emotion Test. The Emotion Test simulated the entire learning process, from the explanation of the subject, to exercises and test. Through the entire process, the emotional state, the personality traits and learning preferences were taken into account. The architecture proposed for this prototype is composed of 4 major models: the Student Model, Emotional Model, the Application Model and the Emotive Pedagogical Model. The prototype developed was tested by the 1st year students of Higher Education Establishment of Computer Engineering of Oporto of two courses: Informatics Engineering and Systems Engineering. The total of students involved in these tests was 115 students with ages between 17 and 42 years old. The first goal was to see that if a learning platform that takes into account the emotional profile of a student can improve the learning results, than a learning platform with no emotional interaction. To demonstrated and evaluate the new prototype that was developed, there were conducted two pretests and a final test. With the two pre-test we hoped to learn what worked and especially what did not work to apply corrective measures to the final test. For the first pre-test was observed that time stipulated (one hour) was insufficient to
complete the evaluation, and also the prototype high level of difficulty (80% to pass to the next task), the student that didn’t get that grade, had to start again. This proved to lead the students to frustration and to give up the completion of the group task. For the second pre-test was used the same evaluation protocol but with more time, roughly 1½ hours and the application level of difficulty cut by half. For the final test, again the evaluation time was increased to two hours. This measure enabled the students to get better results. In the final test we were able to perceived that the differences observed were statistical different. Thus enable us to have indicators that this prototype that takes into account the emotional profile of a student can indeed improve the learning results.
[9]
[10] [11]
[12] [13] [14]
[15]
ACKNOWLEDGMENT This work is supported by FEDER Funds through the “Programa Operacional Factores de Competitividade COMPETE” program and by National Funds through FCT “Fundação para a Ciência e a Tecnologia” under the project: FCOMP-01-0124-FEDER- PEst-OE/EEI/UI0760/2014.
[16]
[17]
[18]
REFERENCES [1] [2]
[3]
[4]
[5] [6] [7] [8]
R. P. Githens, “Quarterly Review of Distance Education, 8(4),” no. 2005, 2007. S. Park, “Building Bridge between Learning and Positive Emotion : How to Apply Emotional Factor in Instructional Designing Process ?,” pp. 679–687, 2003. G. L. Clore and J. R. Huntsinger, “How emotions inform judgment and regulate thought,” Trends Cogn. Sci., vol. 11, no. 9, pp. 393–399, Sep. 2007. R. W. Picard, S. Papert, W. Bender, B. Blumberg, C. Breazeal, D. Cavallo, T. Machover, M. Resnick, D. Roy, and C. Strohecker, “Affective learning - a manifesto,” BT Technol. J., vol. 22, no. 4, pp. 253–268, Oct. 2004. U. Hess and P. Thibault, “Darwin and emotion expression,” Am. Psychol., vol. 64, no. 2, pp. 120–8, 2009. M. L. Batshaw, e-Study Guide for: Children with Disabilities. Cram101, 2014, p. 92. C. Ingleton, “Emotion in learning : a neglected dynamic,” no. July, pp. 12–15, 1999. K. Oertel, R. Kaiser, J. Voskamp, and B. Urban, “AFFectIX – An Affective Component as Part of an E-Learning-System,” Emotion, no. 1, pp. 385–393, 2007.
[19]
[20]
[21]
[22]
[23] [24] [25]
M. Ben Ammar and M. Neji, “A multi-agent based system for affective peer-e-learning,” Proc. 2nd Int. Conf. Mob. Multimed. Commun. MobiMedia 06, p. 1, 2006. K. O. Regan, “EMOTION AND E-LEARNING,” Darwin, vol. 7, no. 3, pp. 78–92, 2003. R. A. Calvo and S. D. Mello, “Affect Detection : An Interdisciplinary Review of Models , Methods , and Their Applications,” Rev. Lit. Arts Am., vol. 1, no. 1, pp. 18–37, 2010. A. G. Stricker and P. D, “Why Affective Learning in a Situated Place Matters for the Millennial Generation,” 2009. J. A. Russell, “A Circumplex Model of Affect,” pp. 1161–1178, 1980. B. Kort, R. Reilly, and R. W. Picard, “An affective model of interplay between emotions and learning: reengineering educational pedagogybuilding a learning companion,” Proc. IEEE Int. Conf. Adv. Learn. Technol., pp. 43–46, 2001. R. Jorge, “Sistema de Apoio à Argumentação em Grupo em Ambientes Inteligentes e Ubíquos considerando Aspectos Emocionais e de Personalidade,” 2010. T. Chamorro-Premuzic and A. Furnham, “Personality, intelligence and approaches to learning as predictors of academic performance,” Pers. Individ. Dif., vol. 44, no. 7, pp. 1596–1603, 2008. V. V Busato, F. J. Prins, J. J. Elshout, and C. Hamaker, “The relation between learning styles, the Big Five personality traits and achievement motivation in higher education,” Personality and Individual Differences, vol. 26, no. 1. pp. 129–140, 1998. N. Ibrahimoglu, I. Unaldi, M. Samancioglu, and M. Baglibel, “the Relationship Between Personality Traits and Learning Styles : a Cluster Analysis,” vol. 2, no. 3, 2013. K. Kumar, a Bakhshi, and E. Rani, “Linking the Big Five personality domains to organizational citizenship behavior,” Int. J. Psychol. Stud., no. 2, pp. 73–82, 2009. M. Komarraju, S. J. Karau, and R. R. Schmeck, “Role of the Big Five personality traits in predicting college students’ academic motivation and achievement,” Learn. Individ. Differ., vol. 19, no. 1, pp. 47–52, Jan. 2009. M. Komarraju, S. J. Karau, R. R. Schmeck, and A. Avdic, “The Big Five personality traits, learning styles, and academic achievement,” Pers. Individ. Dif., vol. 51, no. 4, pp. 472–477, Sep. 2011. R. Morgan and F. Baker, “A Comparison of VARK Analysis Studies and Recommendations for Teacher Educators Ron Morgan and Fred Baker University of South Alabama,” 2012. D. a. Kolb, “Experiential learning: experience as the source of learning and development,” J. Organ. Behav., vol. 8, no. 1984, pp. 19–38, 1984. R. M. Felder, “Learning and Teaching Styles In Engineering Education (updated 2002),” Eng. Educ., vol. 78, pp. 674–681, 1988. J. Bullington, “‘ Affective ’ computing and emotion recognition systems : The future of biometric surveillance ?,” Public Policy, pp. 95– 99, 2002.
