Eye Movement Analysis and Usability Assessment on Affective Computing Combined with Intelligent Tutoring System

Abstract

Education is the key to achieving sustainable development goals in the future, and quality education is the basis for improving the quality of human life and achieving sustainable development. In addition to quality education, emotions are an important factor to knowledge acquisition and skill training. Affective computing makes computers more humane and intelligent, and good emotional performance can create successful learning. In this study, affective computing is combined with an intelligent tutoring system to achieve relevant and effective learning results through affective intelligent learning. The system aims to change negative emotions into positive ones of learning to improve students’ interest in learning. With a total of 30 participants, this study adopts quantitative research design to explore the learning situations. We adopt the System Usability Scale (SUS) to evaluate overall availability of the system and use the Scan Path to explore if the subject stays longer in learning the course. This study found that both availability and satisfaction of affective tutoring system are high. The emotional feedback mechanism of the system can help users in transforming negative emotions into positive ones. In addition, the system is able to increase the learning duration the user spends on learning the course as well.

1. Introduction

1.1. Research Background and Motivation

Due to vigorous development of information technology, digital learning development and applied education are very common. Emotions are an important part of everyone and can affect behavior, thinking skills, decision-making, resilience, well-being, and the way human beings communicate with each other [1]. Based on the discussions, emotion not only is the driving factor that promotes learning but also is the primary factor that hinders the learning process as well. Hence, it is crucial to have reliable methods of emotion recognition in academic contexts [2]. The term “affective computing” was proposed by Professor Rosalind Picard in 1997 [3], and it has been guiding computers to identify and express emotions and respond intelligently to human emotions [4]. Owing to this, a trend has developed that applies emotional lenses to emerging academic research and positions emotion at the core of learning [5,6]. Kort and other scholars have proposed an emotion conceptualization module (2001) to combine affective computing with intelligent tutoring system for the purpose of identifying learners’ emotions, responding to affections of current learners and promoting learning effectiveness [7]. Moreover, it can also identify the user’s emotions and corresponding system responses as well as guide how to keep positive emotions via “mood proxy” and promote the achievement of learning objectives [8,9]. As Chen Huang Cheng mentioned in this book (2006), the best way to relieve negative emotions is to shift your thinking direction and appropriately reduce the expansion of negative emotions. Impacted with positive emotions, determination to solve problems will be generated when encountering difficulties [10]. In many studies on the affective tutoring system, the experimental results mainly discuss the usability and interactivity of the system. In this study, we use Scan Path to understand the subject’s eye movement trajectory and observe his/her eye movement fixation duration in each ROI block to understand if the subject is more willing to stay in the digital course after combining the tutoring system with the affective computing module.

1.2. Research Purpose

With the advancement of technology, diversified learning modes can create richer teaching materials and learning methods. Based on the aforementioned research background and motivation, this study aimed to combine affective computing with intelligent tutoring system to understand the user’s learning emotions, convert the user’s negative emotions into positive ones and improve his/her interest of learning. Moreover, we used an eye tracker to perform eye movement analysis and understand if the user is able to increase the course learning duration by using the affective tutoring system. Based on the research background and motivation as well as the above discussions, the purpose of this study was set as follows: 1. Exploring the usability assessment of the affective tutoring system. 2. Understanding the fixation duration when using the affective tutoring system through eye movement analysis. 3. Understanding if the user is able to increase the course learning duration by using the affective tutoring system through eye movement analysis. Eye movement analysis was conducted to explore if the affective tutoring system can improve learners’ attention and increase the course learning duration. The study themes are as follows:

• How satisfied are the users with the affective tutoring system?
• Is the user able to effectively increase the course learning duration by using the affective tutoring system?

2. Literature Review

2.1. Affective Computing

Affective computing methods have been applied to many fields such as training and learning environments [11]. In learning environments, the most important emotions to be dealt with are those associated with the teaching process (such as boredom, frustration, confusion, and engagement). Picard (1997) proposed four levels of affective computing: Recognize Emotion, Understand Emotion, Express Emotion and Emotion Intelligence. Affective Computing aims to detect signals caused by emotions and affections, such as language, physiological changes and body movements through various sensors. The computers will analyze these signals and make appropriate responses to current emotions. Emotion recognition can be detected by heartbeat, skin potential difference and facial emotion expression [12].

2.2. Affective Tutoring System

Emotions play a significant role in human behaviors in individual and social communities. This can happen in any kind of human activity such as learning online [13]. Recently, researchers have acknowledged the role of emotions in online learning in improving learning outcomes and enhancing students’ experience [14,15,16]. The significance of incorporating emotional states with the learning process has necessitated the development of ATSs, which is the extended research of ITSs and with the ability to adapt to the learner’s adverse emotion effectively to spark the learner’s motivation to learn [17]. The affective tutoring system is based on the intelligent tutoring system, combining with affective computing and featuring the ability to detect the learners’ emotions when they are learning [18]. The affective computer-based digital learning system proposed by Duo and Song (2012) aims to simulate the traditional teaching mode to analyze and recognize learners’ emotions and improve learners’ moods with virtual agents [19]. Mao and Li (2010) proposed that success in teaching lay in the ability to quickly identify the learners’ emotional state, timely adjust the learners’ emotions and enhance the learners’ learning motivation. Ammar et al. (2010) added a facial expression detection module to the affective tutoring system to boost the learners’ moods and improve the emotional communication between the system and the learner. The final results indicate that affective computing can effectively monitor the learners’ emotions, appropriately lead to positive emotions and thus improve the learning motivation [20]. As Gerald (2004) said, learners may greatly reduce their learning motivation due to negative emotions, but positive emotions can effectively improve the learners’ learning willingness. Graesser et al. (2004) used natural language to set up an emotion module for about 1000 subjects who are students in computer or physics majors. As the test results showed, such an emotion module has significantly improved the learning effects on both basic knowledge learning and in-depth research and discussion [21]. The indicators for measuring the satisfaction of using the affective tutoring system include learners’ attitude and affective computing, the performance ability of the tutoring system, the accuracy of emotion recognition, the quantity of emotion recognition, the teaching course activities and the availability of the system etc. [17]. This study will therefore take the above characteristics into account in the design of the system.

2.3. Intelligent Virtual Agents

Intelligent virtual agents (IVAs) powered by artificial intelligence (AI) are prevalent in our daily lives. Amazon’s Alexa, Apple’s Siri, and Microsoft’s Cortana are all IVAs that search information in real-time and verbally communicate the results to human beings [22]. Intelligent Virtual Agents are programs that can be assigned to perform user-specified jobs in a way that is comparable to human beings, featuring reactivity, positivity [23], autonomy [24], social ability and veracity [25], etc. which is different from traditional software. It is personalized, autonomous, proactive, adaptive and continuously running.

2.4. Emotion and Learning

The research carried out in education and psychology shows that emotion and learning have a hidden correlation, which will ultimately enhance learning performance [26]. Russell (1980) has built a 2D emotion model diagram and the eight emotions are divided into four quadrants, namely, happiness, surprise, boredom, fear, calmness, sadness, disgust and anger [27]. Ekman and Friesen (1971) defined six facial expressions: happiness, anger, fear, disgust, surprise and sadness [28]. Because emotions may affect learning effectiveness and the learning process may also influence the emotions, the correlation has therefore led many studies to focus on emotions and teaching. As Guo Shuzhen pointed out (2010), tutors need to be able to guide students to think in a right direction, influence students in a practical way, cultivate positive thinking and problem-solving ability, trigger students’ potential abilities and improve learning effects in a happy learning way [29]. As Wen-Tzu Chiang (2004) proposed, the best way to adjust your emotion is to shift your attention away from your current mood and change the way you think and the point of view with which you view things [30]. Ying-Chun Sun (2009) pointed out that the cognitive style can generate different learning effects for understanding different emotions during the learning process. Therefore, in order to improve learning motivation, students shall be guided to maintain positive thinking over time to achieve optimal effect of learning [31].

2.5. Eye Movement Analysis

In recent years, eye tracking has been widely used as a research tool in many fields [32,33]. By tracking the user’s fixation and analyzing eye movements and pupil size, his/her cognitive state can be studied, such as attention pattern and learning preferences [34,35]. Eye movement analysis mainly seeks to detect the learner’s distraction process and understand the learner’s attention on news reading through fixation points and eye movement trajectories [36]. As for eye movements upon reading, as Rayner (2009) said, tracking eye movements is regarded as an intuitive and effective approach to explore people’s cognitive process in reading, scene perception and visual search [37]. The average fixation duration is 260–330 ms when the reader views images and 225–250 ms when reading words. The movement of fixation is to follow the texts from left to right or up to down. About 10–15% of fixation is associated with looking back, moving in an opposite direction of the text order. The situation is due to the fact that the reader spends a longer time reading the text and has to look back. If more than ten words are looked back by the reader, it indicates that the reader does not understand the current texts. Attention is a very important step for learning education because you shall first attract the attentions of learners before starting to tutor knowledge [38]. Many researchers agree that the direction of the head in interpersonal interaction is the easiest way to know the direction of the individual’s fixation. When the head is fixed, the eye direction is almost the same as that of the individual’s attention, except when the subject suffers from severe strabismus [39]. When the head is fixed, the eye direction is the same as that of the individual’s fixation [36].

3. Methods

The objects of this study included 30 college students or above aged 20 to 30 who are divided into an experiment group and a control group, each with 15 subjects. The former conducted the experiment with the affective tutoring system and the latter conducted the experiment with the traditional tutoring system. The eye tracker was applied to perform eye movement analysis and tracking on both groups, and the whole experiment took about 15 min. At the end of the experiment, the subjects were asked to fill in the System Usability Scale, which took 10 min.

3.1. Interface Design

The system interface is divided into Interfaces A, B and C. A is an intelligent virtual agent, B is a course module and C is a course menu, as shown in Figure 1. The specific functions of the three interfaces are as follows:

A.

Intelligent Virtual Agent: The intelligent virtual agent gives feedback to the user by combining Chinese semantic emotion recognition. When the user types in text sentences in the text input box, the system will identify the emotional keywords contained in the sentences and timely respond to the user’s emotional results. The agent will interact with the user with different emotions and the system will finally propose new emotional questions to achieve real-time interaction and communication between the user and the system.

B.

Course Module: The experimental textbook contains information about interactive technology describing the main development of interactive technology and other related technologies in recent years (such as wearable device, interactive technology, somatosensory and five-sense experience etc.). It helps the user to know more about the current course by combining with relevant online videos.

C.

Course Menu: The menu displays each chapter in the course, allowing the user to control the reading time for each chapter and then select the next chapter.

3.2. Course Model

The course refers to digital art. In addition to the textual description of the course, picture and video examples are used to help the users understand the course contents. It takes 15 min for the subject to watch the course. Interactive design-related technologies are added in the learning process to improve the user’s interest of learning.

3.3. Agent Model

The agent model can affect the learner’s learning emotion, perceive the learner’s learning situation and give corresponding emotional guidance. In this study, the semantic analysis of affective computing was used to judge the semantics of dialogue between the user and the agent and provide appropriate emotional feedback to the agent to adjust the user’s learning emotion and enhance his/her learning willingness. Therefore, the system set up an agent model as a bridge between the learner and the system. During the recognition, the system will respond to corresponding sentences and change the graphs of the agent model to enhance the learner’s willingness to operate. For example, when any positive emotion is recognized in the learner’s sentences by the system, the agent module will display a graph of positive emotion, and when any negative emotion is identified from the learner’s sentences by the system, the emotion agent model will display a graph of negative emotion. If no emotional keywords are contained in the sentence or the system is unable to recognize the input sentence, a dynamic confusion graph without emotional feedback will be displayed. After the user’s emotions are recognized by the system, the agent’s emotional feedback will be given to the user in the set emotional performance state, and eight kinds of agent emotion feedback will be set up, including happiness, sadness, fear, frustration, anger, surprise, disgust, and doubt, as shown in Figure 2 below.

3.4. Analysis of Eye Movement Statistics

The proportion and duration of time spent on the ROI blocks of the subject. In this study, an eye tracker gifted by Professor He Hongfa of National Taiwan Normal University was used. It is a research outcome from developing the eye tracker hardware and software in a five-year project for the Top University Project sponsored by the Ministry of Education of the Republic of China. The statistics analysis software developed by this study was analyzed according to the calculation formula in the eye tracker’s manual, as shown below.

• Total Time in Zone (ms)
  [Formula Definition: Total Time in ROI (fix.txt; duration totaling in ROI)]
  [Meaning: Total fixation time in zone]
• Total Fixation Duration (ms)
  [Formula Definition: Total fixation duration of fix.txt]
  [Meaning: Total fixation duration of the subject during the experiment]
• Average Fixation Duration (ms)
  [Formula Definition: Average fixation duration of fix.txt]
  [Meaning: Average fixation duration of the subject during the experiment]
• Fixation Counts
  [Formula Definition: Total fixation counts of fix.txt]
  [Meaning: Total fixation counts of the subject during the experiment]
• Percent Time Fixated Related to Total Fixation Duration (%)
  [Formula Definition: fix.txt; duration totaling/fixation duration in ROI (duration totaling of fix.txt)]
  [Meaning: Fixation duration in the zone to total fixation duration]

3.5. System Usability Scale

In this study, the System Usability Scale (SUS) was used. It was developed by Equipment Co., Ltd. in 1986 and mainly used to evaluate the usability of the system. The System Usability Scale is a low-cost, reliable and rapid method to effectively assess the user’s subjective feelings toward the system [40]. The Likert scale was used to measure the scale from point 1 to 5 from strongly disagree to strongly agree. The formula was, however, divided into even numbers and odd numbers. For even-numbered questions, the numerical value 5 was subtracted from the original scores of the question to be the score available. For odd-numbered questions, the original scores were subtracted from the numerical value 1 to be the score available. Finally, the scores of the even and odd-numbered questions were added and multiplied by 2.5 to obtain the final satisfied scores [41]. The scale was analyzed by the formula, and the statistics selected by the subjects according to the statistics scale were scored with a total of 100 points for evaluation and analysis. The higher the score, the higher the degree of satisfaction on the system evaluated.

4. Data Analysis and Results

4.1. System Usability Analysis

In order to analyze the user’s usability of the system, the statistical analysis was conducted with the System Usability Scale. A total of 15 participants took part in the experiment and all of them filled in the scale, which means that all the 15 scales are effective.

4.1.1. System Usability Scale—Reliability Analysis

Some of questions in the System Usability Scale are roundabout questions. Therefore, it is required to invert these roundabout questions first, and then to perform the statistical analysis according to the scale statistics filled in by the users to determine the user’s usability of the system. From the overall usability scale, the average value is 4.01 with a standard deviation of 0.512. This means that the internal consistency reliability of the scale is high. Figure 3 shows the histogram of the overall usability scale.

4.1.2. System Usability Scale—Descriptive Statistics

This refers to the statistical result of each question of the System Usability Scale.

Table 1. Analysis Statistics of System Usability Scale (Percentage of five-point scale).

	1	2	3	4	5
Q1	0%	0%	6.7%	53.3%	40.0%
Q2	0%	13.3%	40%	33.4%	13.3%
Q3	0%	0%	26.7%	40%	33%
Q4	0%	13.3%	46.7%	26.7%	13.3%
Q5	0%	0%	40%	40%	20%
Q6	0%	0%	26.7%	46.7%	26.6%
Q7	0%	0%	6.7%	46.7%	46.6%
Q8	0%	0%	6.7%	46.7%	46.6%
Q9	0%	0%	20%	26.7%	53.3%
Q10	0%	13.3%	26.7%	20%	40%

shows the analysis statistics of the System Usability Scale (percentage of five-point scale) and

Table 2. Results of Analysis of Each Question of the Usability Scale.

	N	Minimum Value	Maximum Value	Summation	Average Number	Standard Deviation	Variance
Q1	15	3	5	65	4.33	0.617	0.381
Q2	15	2	5	52	3.47	0.915	0.838
Q3	15	3	5	61	4.07	0.799	0.638
Q4	15	2	5	51	3.40	0.910	0.829
Q5	15	3	5	57	3.80	0.775	0.600
Q6	15	3	5	60	4.00	0.756	0.571
Q7	15	3	5	66	4.40	0.632	0.400
Q8	15	3	5	66	4.40	0.632	0.400
Q9	15	3	5	65	4.33	0.816	0.667
Q10	15	2	5	58	3.87	1.125	1.267

shows the results of the analysis of each question of the Usability Scale.

The sum of the highest and second highest percentages for each question on a 5-point Likert scale represents agreement. The third highest percentages for each question on a 5-point Likert scale represent neutrality and the second lowest and lowest percentages for each question on 5-point Likert scale represent disagreement. Analysis results are as follows.

Q1: I think I would often use the system. This is a straightforward question with an average number of 4.33 and standard deviation of 0.617. In all, 40% of users are very willing to use the affective tutoring system often for learning, 53.3% of users would like to use the affective tutoring system for learning often and 6.7% of users are generally willing to use the system. As the system is a combination of emotion recognition and the course, emotion feedback is given to users in the aspect of emotion recognition, which has increased the enjoyment of the users. Therefore, it is inferred that most users are satisfied with this system.

Q2: I think the system is too complicated. This is a roundabout question with an average number of 3.47 and standard deviation of 0.915. In all, 13.3% of users strongly disagree that the system is too complicated, 33.4% of users disagree that the system is too complicated, 40% of users are neutral about the complexity of the system and 13.3% of users think that the system is too complicated.

Q3: I think the system is easy to use. This is a straightforward question with an average number of 4.07 and standard deviation of 0.799. In all, 33% of users think the system is very easy to use, 40% of users think the system is easy to use and 26.7% of users are neutral about the complexity of the system.

Q4: I think I need a technician’s help to use the system. This is a roundabout question with an average number of 3.40 and a standard deviation of 0.910. In all, 13.30% of users think that it is absolutely necessary to have a technician to help use the system, 26.7% of users think that it is necessary to have a technician to help use the system, 46.70% of users are neutral about whether it is necessary to have a technician to help use the system and 13.3% of users think that it is unnecessary to have a technician to help use the system. Therefore, we are getting to a conclusion that the users do not have sufficient knowledge of affective computing and the functions of the agents need to be introduced. Human resources are required to be of assistance to guide the users.

Q5: I think all functions of the system are integrated well. This is a straightforward question with an average number of 3.80 and standard deviation of 0.775. In all, 20% of users strongly agree that all functions of the system are integrated well, 40% of users agree that all functions of the system are integrated well and 40% of users are neutral about whether all functions of the system are integrated well.

Q6: I think there is too much contradiction in the system. This is a roundabout question with an average number of 4.00 and standard deviation of 0.756. In all, 26.60% of users strongly disagree that there is too much contradiction in the system, 46.70% of users disagree that there is too much contradiction in the system and 26.70% of users are neutral about whether there is too much contradiction in the system.

Q7: I think most people could learn how to use the system fast. This is a straightforward question with an average number of 4.40 and standard deviation of 0.632. In all, 46.60% of users strongly agree that most people could learn how to use the system fast; 46.70% of users agree that most people could learn how to use the system fast and 6.70% of users are neutral about whether most people could learn how to use the system fast.

Q8: I think the system is very difficult to use. This is a roundabout question with an average number of 4.40 and standard deviation of 0.632. In all, 46.60% of users strongly disagree that the system is very difficult to use, 46.70% of users disagree that the system is very difficult to use and 6.70% of users are neutral about whether the system is very difficult to use.

Q9: I think I am very confident of using the system. This is a straightforward question with an average number of 4.33 and standard deviation of 0.816. In all, 53.30% of users strongly agree that they are very confident about using the system, 26.70% of users agree that they are very confident about using the system and 20% of users are neutral about whether they are very confident about using the system.

Q10: I think I have to learn something to use the system. This is a roundabout question with an average number of 3.87 and standard deviation of 1.125. In all, 40% of users strongly agree that they have to learn something to use the system, 20% of users agree that they have to learn something to use the system, 26.70% of users are neutral about whether they have to learn something to use the system and 40% of users disagree that they have to learn something to use the system. The researcher arrived at a conclusion that he/she does not know about affective computing and needs the help of technicians; he/she therefore thinks he/she needs to learn relevant knowledge before using the system.

4.1.3. Users’ Satisfaction Analysis

For the score of each question, the odd-numbered questions are straightforward, with the score of each question being subtracted by the numerical value 1 to get final score; for example, if the score is 3, the calculated result will be 2. The even-numbered questions are roundabout, with the numerical value of 5 being subtracted by the score of each question to get final score; for example, if the score is 4, the calculated results will be 1. After the scores of each question are calculated, all the scores are added and multiplied by 2.5 to get the total score. The statistics obtained after the score conversion are shown in

Table 3. Calculated Results of SUS Formula.

	Sample Size	Average Number	Median	Maximum Value	Minimum Value	Standard Deviation
Overall	15	81.5	82.5	95	67.5	7.245688

. According to the calculated results of the SUS formula, the average score of the subject for the system is 81.5. Figure 4 shows how the adjective ratings compare to both the school grading scale and the acceptability ranges [41]. The statistical results show that the users are satisfied with the usability of the system.

4.2. Eye Movement Analysis

The 30 objects of this study were divided into an experiment group and a control group with each 15 subjects. The former conducted the experiment by using the affective tutoring system, while the latter conducted the experiment by using the traditional tutoring system. The eye tracker was applied to both groups in order to perform the eye movement analysis and tracking. The whole experiment took about 15 min.

4.2.1. Eye Movement Fixation Analysis

The teaching contents of the system were divided into four sessions. The control group was compared with the experiment group in each session to understand the fixation duration of the subjects using the affective tutoring system. After finishing the experiment, the aggregate and average data of both the control group and the experiment group, as shown in

Table 4. Sessions one to four—Overall Average Table.

	Control Group	Experiment Group
Total Average Fixation Duration	80,889.27	120,477.2
Course Contents Total Average Fixation Duration	80,687.73	86,751.6
Course Contents Total Average Fixation Counts	796,133	803,533
Course Contents Single Average Fixation Duration	101,349	107,962

, were calculated to obtain the total average fixation duration: 80,889.27 for the control group and 120,477.2 for the experiment group. The total average fixation duration at the course contents was 80,687.73 for the control group and 86,751.6 for the experiment group. The total average fixation count on the course contents was 796,133 for the control group and 803,533 for the experiment group. The single average fixation duration was 101,349 for the control group and 107,962 for the experiment group. In the analysis of eye movement statistics, the results of the experiment group were better than those of the control group, which indicates that using the affective tutoring system increased the course learning fixation duration.

4.2.2. Learning Duration Analysis of Eye Movement Course

The teaching contents of the system were divided into four sessions. The control group was compared with the experiment group in each session to understand if the affective tutoring system can effectively increase the learners’ course learning duration. According to

Table 5. T Verification—Group Statistical Analysis (Total Course Learning Duration).

	Average Number	Person(s)	Standard Deviation	Standard Error Mean Value
Control Group	851,582.27	15	109,821.034	28,355.669
Experiment Group	1,084,836.47	15	171,437.643	44,265.009

, the average number of the experiment group and the control group was 1,084,836.47 and 851,582.27 respectively. The former was higher than the latter, which indicates that the affective tutoring system lengthened the learning duration compared with the traditional tutoring system. According to the significance judgment and analysis results of the data in

Table 6. T Verification—Independent Sample Verification (Total Course Learning Duration).

	Levene Variance Equality Test		Test (T) to Check if the Average Value Is Equal
F	Significance	T	df	Significance (two-tailed)
0.881	0.356	−4.437	23.834	0.000

, the significance was 0.000 < 0.05. We can say that the total course learning duration of the experiment group using the affective tutoring system is significantly different from that of the control group using the traditional tutoring system.

According to

Table 7. Test (T)—Group Statistics (Fixation duration at the course contents).

	Average Number	Person(s)	Standard Deviation	Standard Error Mean Value
Control Group	80,687.73	15	12,762.089	3295.157
Experiment Group	86,751.60	15	17,452.989	4506.342

below, the average number of the experiment group and control group is 86,751.60 and 80,687.73 respectively. Based on the significance judgment and analysis results of the data in

Table 8. Test (T)—Independent sample test (course content gaze time).

		Levene Variance Equality Test		Test (T) to Check if the Average Value Is Equal
	F	Significance	T	df	Significance (Two-Tailed)
Fixation at the course contents	1.427	0.242	−1.086	25.643	0.287

, the significance is 0.242 > 0.05. According to Table 7 and Table 8, the Test (T) table shows that there is no significant difference in the total fixation duration of the control group, but the average duration is increased.

According to

Table 9. Test (T)—Group Statistics (Total fixation duration).

	Average Number	Person(s)	Standard Deviation	Standard Error Mean Value
Control Group	80,889.27	15	12,842.757	3315.986
Experiment Group	120,477.20	15	24,068.766	3315.986

below, the average number of the experiment group and the control group is 1,084,836.47 and 851,582.27, respectively. Based on the significance judgment and analysis results of data in

Table 10. Test (T)—Independent Sample Test (Total fixation duration).

		Levene Variance Equality Test		Test (T) to Check if the Average Value Is Equal
	F	Significance	T	df	Significance (Two-Tailed)
Fixation at the course contents	10.209	0.003	−5.620	28	0.000

, the significance is 0.003 < 0.05. According to Table 9 and Table 10, it can be found that there is a significant difference in the total fixation duration of the experiment group using the affective tutoring system compared with the control group using the general tutoring system. It shows that the fixation duration of the users with the affective tutoring system is longer than that of the users with the general tutoring system.

4.2.3. Eye Movement ROI Block Analysis

The teaching contents of the system were divided into four sessions. The control group was compared with the experiment group in respect of the time before the first visit of ROI block and the counts of fixation before the first visit.

Time before the first visit of the ROI block, the time before the first visit of the ROI block from sessions one to four of both the control group and the experiment group is zero. In this study, the course contents were in the middle position of the page layout, and after the nine-point correction, all subjects looked straight ahead and waited for the experiment to start, so they had already looked into the ROI block of the course contents at the very beginning of the first time, and thus the statistics show zero.

Fixation counts before the first visit, the fixation counts before the first visit from sessions one to four of both the control group and the experiment group are zero. In this study, the course contents were in the middle position of the page layout, and after the nine-point correction, all subjects looked straight ahead and waited for the experiment to start, and the standard fixation duration is 80 ms and the count statistics show zero.

4.2.4. Analysis of Eye Movement Hot Zone

The teaching contents of the system were divided into four sessions for analysis and comparison of hot zones to understand the hot zones of learners’ attention to the contents of the textbook in the course of study. Figure 5 and Figure 6 are the first session of the course contents of the control group and the experiment group. Figure 7 and Figure 8 are the hot zones for the course contents of the control group and the experiment group in the second session. Figure 9 and Figure 10 are the course contents of the control group and the experiment group in the third session of the fixation hot zone. Figure 11 and Figure 12 show the course contents in the fourth session of the control group and the experiment group. The experiment shows that both the control group and the experiment group were completely fixated on the course contents in the course content fixation hot zone.

5. Discussion and Conclusions

In this study, the affective computing was built on the intelligent tutoring system, which not only enhanced the learning interest of the users, but also provided a deeper learning experience. The real-time interaction between the users and the emotional agents provided emotional feedback and guidance that makes the users turn negative emotions into positive ones during the learning process and improve the learning interest of the users. SUS was used to understand the usability of the system, and eye movement statistics were analyzed to explore the fixation duration at the system. The analysis shows that the average score of the users for the usability of the system is 81.5, and the overall satisfaction is very high. As shown in the descriptive statistics, the users found it easy to use the system and the learning process was attractive, which increased the willingness to learn. Therefore, the statistics show that the design of the system can be adopted, and its usability is quite good. In the analysis of the eye movement statistics, the use of affective tutoring system increased the learning duration of the course. The eye movement statistics show that the fixation duration of the users from sessions one to four of the learning and the length and position of the fixation duration could represent the distribution and preference of personal attention. In this study, the fixation duration of the experiment group was longer than that of the control group, and the eye movement statistics analysis divided the experiment group into the two ROI blocks, the agent block and the course block, while there was only the course block in the ROI block of the control group. To further confirm whether the subjects are interested in the course contents, remove the fixation duration of the agent block in the experiment group and intercept the course block of both the experiment group and the control group to compare the course learning duration. The statistics show that the total learning duration of the course contents of the experiment group was better than that of the control group. The research proved that the system can increase the learning duration of users’ courses. Through our experiments, the following items were found: (1) For the semantic emotion recognition module, the system was unable to recognize some fashionable terms used by young people according to the users’ feedback. It was expected to collect some popular social platform, advertising works and even some emojis that can express emotions. (2) For the sentiment analysis and judgment of words and sentences, more considerations are needed because, for some popular words used by young people, the meaning is usually not the meaning of emotions, and there is some irony or implied meaning. It is easy to be confused with the original words. It is expected that, in the future, the system will be more fashionable to match social trends and be closer to the users. (3) For the technology of the agent module, it is expected to perform deep learning to train the machine to become a ChatBot, which can interact with users more smoothly and naturally. For this part, we have planned them into the work for the future.

6. Future Prospects

In recent years, the intelligent tutoring system has integrated users’ emotional state into the intelligent tutoring system, which has led to the transformation of the intelligent tutoring system to the affective tutoring system. Therefore, affective computing is undoubtedly important in the education environment. In the future, multi-media contents can be applied in the design and development of the affective tutoring system. In addition, we expect to apply deep learning to the affective tutoring system and make the system feature a ChatBot message function. Message function has become an indispensable part of contemporary daily life. Under this premise, it is an inevitable trend to be equipped with a good ChatBot. Chatbots are no longer seen as mere assistants, for they interact in a way that brings them closer to the users as friendly companions [42]. Based on this, we will explore the user experience and evaluate the learning effectiveness in the subsequent study.