The Effect of the Integration of Metacognitive and Motivation Scaffolding Through a Pedagogical Agent on Self- and Co-Regulation Learning

Nowadays, online learning has become commonplace in higher education. Various factors influence the success of online learning. Factors such as low self-regulation and co-regulation of learning skills can affect student engagement and motivation in online learning activities. Therefore, it is essential to provide external support in the online learning process. Pedagogical agents are one solution to increasing self-regulation and co-regulation learning in online learning environments. This study aims to determine the effect of integrating metacognitive and motivation scaffolding support provided through pedagogical agents on self-regulation and co-regulation learning skills. This study uses a mixed-method explanatory sequential approach. A quasi-experimental method of the nonequivalent type (pretest and posttest) control group is used for quantitative data collection, and an in-depth survey is used to analyze qualitative data. The results showed that the learning support provided by pedagogical agents who act as scaffolders using metacognitive and motivation scaffolding integration has a significant impact on increasing students' self-regulation and co-regulation learning skills. Also, the results of qualitative data analysis supported the findings obtained from quantitative data. Research limitations and suggestions for further research are presented at the end of this article.

. Pedagogical agents (PAs) are one of the tools to increase student engagement and motivation. PAs in online learning environments are shapes like anthropomorphic virtual characters, which help learners to overcome these obstacles [7]. Previous PA research found that students learn better from systems with agents than without agents and can facilitate more effective learning [8].
PAs use instructional strategies to provide learning support to students. A meta-analysis of the effectiveness of scaffolding conducted by Doo et al. [9] found that scaffolding can improve the quality of learning and learning outcomes in online learning in universities. PAs with a scaffolding learning strategy approach can potentially improve learning outcomes. A PA that uses a scaffolding approach effectively encourages complex learning in a relatively short time [10].
On the other hand, students need to organize themselves and others' learning in the group discussion [11]. Self-regulation learning (SRL) skills are essential skills for students to be engaged and motivated in the learning process individually [12], [13], [14]. According to Pintrich [15], self-regulation refers to students' active and constructive processes to set goals, monitor, and evaluate their cognition, influence, and behavior. In collaborative learning activities, co-regulation learning (CRL) skills are essential so students can solve problems and build collective knowledge in groups [16]. CRL extends self-regulation, including cognitive and social dimensions [11].
In contrast to SRL, CRL is a skill in which each group member supports each SRL member by paying attention to other group members whether they deviate from their task objectives [16]. SRL and CRL are skills that have a reciprocal relationship. CRL refers to the transition process of acquiring SRL, where individual learners gradually adjust SRL through interactions with others in solving the same problem [17]. The CRL can be formed by the following [18].
1) The regulator requests regulatory support (e.g., asking someone to ascribe the task criteria). 2) Others by prompting someone to engage in regulatory processes or practices (e.g., cueing someone to examine the results of their work). 3) Technology (e.g., reminders to check the time). Previous research on PAs has focused mainly on using metacognitive scaffolds. Previous studies investigated the ability of PAs to help regulate CRL [16] and SRL [19], [20], [21], [22], This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/ [23], [24], [25], [26] externally. However, studies of PAs using motivational scaffolds are rare [27], [28], [29]. In addition, the effectiveness of motivational scaffolding on students' SRL and CRL skills is not yet known.
This study aims to develop PAs by integrating two types of scaffolds: metacognitive and motivation. There are reasons why combining the two types of scaffolding can increase student engagement and motivation in online learning. SRL is the highest form of the interdependence relationship between cognitive engagement and motivation [30], [31], [32]. SRL is a learner's cognitive process driven by metacognitive components [30]. Based on the SRL perspective, the success of online learning is influenced by two contributing components: metacognition and motivation [33]. Metacognition and motivation are key factors in the learning process, so using metacognition and motivation is interrelated and has a reciprocal relationship [34]. Therefore, instructional scaffolding in the metacognitive and motivational domains plays an essential role in online learning at the college level.
Currently, the use of the scaffolding approach in PA research is focused mainly on metacognitive scaffolding. Past research assumed that most scaffolds designed for metacognitive support indirectly increase learner motivation by increasing their expectancy of success. This shows that motivation needs to be managed optimally so that it can trigger a decrease in the level of student engagement and motivation. In addition, several attempts were made to design motivational interventions not linked to motivation theory. Online learning often takes place over a long period, and it is essential to develop scaffolding with motivational support. With motivational support, scaffolding can attract learner interest, maintain learning goals, and control frustration [35].
This study contributes to addressing previous research gaps by extending the theoretical studies used to understand the effect of metacognitive and motivation scaffolding on the interaction of PAs with students. This study investigates the effect of metacognitive and motivational scaffolding integration on SRL and CRL skills at the higher education level. In the next section, we present a theoretical framework and a summary of work related to PA research, followed by an explanation of the intervention of PAs on the individual and group learning examined in this study. After that, we present the research objectives, questions, and hypotheses.

A. Scaffolding
Instructional scaffolding is guidance or support provided by teachers, instructors, or agents who can help students achieve their learning goals. Scaffolding serves as a support in completing tasks in the learning process. The support provides students with information on completing assignments and encourages peer interaction [36]. The statement identifies that individual learning activities and collaborative learning can use scaffolding. Using scaffolding with a combined analysis of individual and collaborative activities can produce intelligent and productive support [37].
Scaffolding can be provided by teachers/instructors (oneto-one scaffolding), peers (peer scaffolding), and computers (computer-based scaffolding) [38]. Belland [38] reviewed forms of scaffolding and found that computer-based scaffolding was no better than one-to-one scaffolding. However, computerbased scaffolds embedded in intelligent agents provide more changes than other computer-based scaffolds.

B. Pedagogical Agent With the Support of Metacognitive Scaffolding in Individual and Collaborative Learning
Metacognitive refers to awareness of cognitive processes that exist in oneself. Using metacognitive strategies in scaffolding will help learners to plan and analyze, monitor [39], [40], and evaluate [39], [41] the concepts that have been studied [38], [42]. Metacognitive scaffolding does not affect the quantity of domain knowledge acquired but its quality [43], [44], [45]. Quintana et al. [41] provided a framework to support metacognitive scaffolding. The framework [41] consists of three components. First, task understanding and planning, the scaffolding feature can provide a clear task structure so that students can find out the material related to the task. Second, monitoring and regulation, scaffolding can provide information explicitly needed by learners to help learners monitor and organize their learning. Third, reflection. Scaffolding can help learners by providing explicit opportunities for reflection, such as providing criticism that can help them assess their quality.
Currently, research on PAs focuses more on metacognitive scaffolding in its approach to learners [16], [21], [22], [23], [24], [25], [26]. The use of metacognitive scaffolding aims to improve higher-order thinking skills. Improving students' higher-order thinking skills is a top priority during the 21st century [46]. Research on PAs with metacognitive scaffolding has mostly examined their effect on SRL skills [19], [21], [22], [23], [24], [25], [26]. Only one study of PAs examined their effect on CRL skills [16]. Research on CRL in collaborative learning with computer support is relatively new [11]. Previous studies of PAs using metacognitive scaffolding found that metacognition is crucial in developing learner-learning effectiveness. Bannert [44] analyzed the effects of metacognitive scaffolding and various forms of scaffolding on learning outcomes at the group and individual levels. The results of the experimental study of Molenaar et al. [47] confirmed that metacognitive scaffolding could function in collaborative settings to improve learning outcomes.

C. Pedagogical Agent With the Support of Motivation Scaffolding in Individual and Collaborative Learning
The use of motivation theory support for learning in computerbased scaffolding can increase motivation [35]. Belland et al. [35] proposed a motivation scaffolding framework, which Low and Robinson [48] elaborated further. The framework [35], [48] consists of six components. First, establish task value. The task's value refers to Wigfield and Eccles' expectancy-value theory, learners' perception of the importance of completing learning tasks. Second, promoting mastery goals refers to the goal-orientation theory. Mastery goal orientation causes learners to engage in more challenging learning tasks to develop their understanding [35].
Third, promoting belonging refers to self-determination (intrinsic/extrinsic) theory. According to Deci and Ryan in Belland et al. [35], learners who have a sense of belonging allow to feel intrinsic motivation so that the desire to be involved in the task is higher. In group activities, individuals have different personal interests and goals, so it is crucial to use scaffolding to outline different aspects of the problem allowing group members to choose a common goal.
Fourth, promoting emotional regulation refers to the goalorientation theory. Kim and Pekrun, in Belland et al. [35], stated that academic emotions arise in the context of school/campus, which generates from the affective reactions of learners to school/campus assignments. Negative emotions are related to performance goals, and positive emotions are related to mastery goals, so scaffolding should promote adaptive emotional regulation. Fifth, expectancy for success refers to expectancy-value theory, social cognition, and attribution. The motivation literature states that learners interested in learning tasks will not engage in the learning process if they do not expect success [35]. The hope for success will be greater if scaffolding provides reliable strategies for task achievement. Sixth, promoting autonomy refers to self-determination (intrinsic/extrinsic) theory. Scaffolding needs to provide the learner with a choice of opportunities to have ownership of the learning assignment. Scaffolding can outline and detail different aspects, thus providing options for learners to choose their preferred learning assignments.
Motivational scaffolding is still rare in computer-based scaffolding, often only mentioned in the suggestions for further research. In addition, if applied in computer-based scaffolding, researchers often only use one theory of motivation [35]. Belland et al. [35] also emphasized that research and scaffold design using motivational scaffolding has excellent potential to expand knowledge about supporting learning and engagement. With the support of motivational scaffolding in its approach, research on PAs found that it can motivate students in the learning process [29], [49]. However, how effective motivational scaffolding is on SRL and CRL skills is still unknown.

D. Shared Metacognition Questionnaire
This study uses the shared metacognition questionnaire developed by Garrison and Akyol [50]. Garrison and Akyol developed and confirmed shared metacognitive constructs that focused on self-and co-regulation based on research on metacognition and the Community of Inquiry framework [50]. The metacognition construct is built based on understanding the role of students in the learning process, both as individuals and as members of groups [50], [51]. This questionnaire has been further validated through confirmatory factor analysis [52], an infit/outfit mean square, and standardized fit statistics [51] so that it can be a stable and helpful tool for researching the dynamics of shared metacognition [53].
This study chose the shared metacognition questionnaire because the items represent monitoring/awareness and management/strategic action skills when students are involved in individual SRL and CRL group settings [50]. There are 26 items with 2 dimensions: 13 SRL dimensions and 13 CRL dimensions [50]. The following items represent monitoring and management functions [53], [54].

III. CURRENT STUDY: POSITIONING AND RESEARCH QUESTIONS
Based on a literature review, students with high SRL skills in online learning in universities can achieve better learning outcomes than students with low SRL skills [49], [55], [56], and CRL skills can improve student performance in small group contexts [57], [58].
We found the same limitations in several previous studies of PAs in the previous section. Investigating the effect of the role of PAs as scaffolders on students' SRL and CRL skills, individually and collaboratively, is not yet known. Based on this point of view, this study examines the impact of integrating metacognitive and motivational scaffolding through PAs in an online learning environment on students' SRL and CRL in higher education. This study also examined students' opinions regarding PA support in individual and group learning processes. The test was carried out using a PA named MeMo Tutor.
MeMo Tutor is a PA that integrates metacognitive and motivational scaffolding to support online student learning in higher education [6]. Fig. 1 illustrates the interaction between PAs, students, and lecturers using the web platform in this study. Lecturers enter questions and trigger discussions through the Moodle learning management system (LMS) [59]. The PA model framework was built using Node.js (Version 12.14.1) [60] and NeDB (Version 1.8.0) [61]. Both client and server use Node.js with object-oriented programming. Main.js is the main file on the server, which is used to retrieve Moodle data using the web service API and respond to the client using a webhook. The hardware environment used to implement the PA consists of a processor of at least 800 MHz, an internal memory of at least 512 Mb, and an internet connection.
We investigated the research objectives with the following research questions: 1) RQ1: Does integrating metacognitive scaffolding and motivation in pedagogical agents affect students' selfregulated learning and co-regulated learning skills? Based on RQ1, we hypothesize that at least two interactions of PAs will be identified based on two experimental conditions, namely experiments on individual and group learning. The two  Table I). This is consistent with previous studies of PAs, where PAs with metacognitive scaffolding affect CRL skills [16] and SRL skills [19], [21], [22], [23], [24], [25], [26]. Meanwhile, research on PAs with motivational scaffolding found that PAs can motivate students in the learning process [29], [62]. However, it is not yet known how effective motivational scaffolding is on SRL and group CRL skills.

1) RQ2: Does integrating metacognitive scaffolding and motivation in pedagogical agents affect the online learning process individually and in groups?
In response to RQ2, we hypothesized that PA interactions with individual and group learners would result in better learning outcomes. We believe that if the interaction of PAs affects students' SRL and CRL skills (RQ1), it will provide the possibility that students will benefit from individual and group learning.

A. Participants
The participants were 92, 1st-year undergraduate students majoring in a 4-year bachelor's degree in computer science at a private university in Bandung, Indonesia, with a strong e-learning culture. All participants (between 17 and 20 years, 79% male and 21% female) took the Design Thinking course, one of the compulsory subjects in the curriculum package offered by the study program. They participated in the experiment voluntarily (via a consent form) and were randomly assigned to one of two conditions. The first condition used PAs in the learning process (getting feedback with metacognitive scaffolding and motivational scaffolding approaches), and the second condition did not use PAs (does not get feedback). Table I provides examples of interventions given in the first condition in individual and group learning. A total of 44 students with 11 small groups were assigned to the first condition (experimental), whereas 48 students with 12 small groups were assigned to the second condition (control). Based on the pretest scores, there was no significant difference in SRL levels between the two experimental conditions (see Section V).

B. Experimental Procedure
MeMo Tutor is a web-based PA represented by text and static images. Cognitive engagement and student motivation are processes within students that cannot be seen explicitly. Using a text-based approach encourages students to write down their understanding so that students' cognitive involvement and motivation become more visible. The MeMo Tutor version used in this study consisted of two learning activities (individual and group) on the topic of Web and Mobile Application Design (see Fig. 2), using the integration of metacognitive and motivational scaffolding (see Fig. 3). MeMo Tutor functions as a scaffolder. MeMo Tutor prompts students to do something (e.g., asking students to add arguments, asking students to check the contributions of friends in discussions) and performing regulation (e.g., giving examples of how to give opinions, providing checklists of discussions that students need to complete).
The observation lasted for six weeks (half-semester). In the first week, students in the experimental and control classes were assigned to randomly form groups of 4-5 people. Through this group, students must complete project assignments. Project assignments are tasks that implement the understanding of the Design Thinking method that has been given for the previous five weeks. Students in the experimental class were taught about MeMo Tutor and informed that learning activities would use MeMo Tutor.
In the second week, students in the experimental and control classes were asked to fill out a pretest metacognition questionnaire. It also explained the topic of the task to be carried out. The task asked each group to find a website or mobile application regarding tourist villages or wedding organizers. Then, ask each group to evaluate using a questionnaire that measures the quality of the website based on user perceptions. Each group can learn the advantages and disadvantages of designing a website or mobile application as a basis for initial knowledge regarding the next task.
Students in the experimental and control classes discussed group assignments in the third to the fifth week. The lecturer explains the topics that need to be discussed at the beginning of each meeting. Lecturers also review material related to design principles in the Design Thinking method according to the topic of discussion. The group discussion was conducted online for 90 min.
In group learning activities, the MeMo Tutor will display triggers for group discussions according to the topic. If students understand the sequence of discussions that need to be discussed and know how to post opinions, then students can immediately provide comments/opinions. However, if the student needs a discussion checklist, the student can request a discussion checklist. Next, MeMo Tutor will provide a discussion checklist. Students can also study examples of how to post opinions/comments given by the MeMo Tutor whenever the student wants to learn. After students provide comments/opinions, MeMo Tutors will give appreciation and remind them to check whether other members have contributed or not. The discussion continues until all discussion checklists are completed. Students should conclude the discussion results when all the discussion checklists have been covered. Fig. 4 shows a flowchart of group learning activities. The control class did not get this assistance, but the instructor provided directions on discussion procedures and material access to work on individual assignments.
After the discussion, students are given individual assignments with a deadline of three days. Individual assignments contain questions about the theory and application of the Design Thinking method according to the topic of discussion.
In individual learning activities, the MeMo Tutor will display questions according to the topic. If the learner already knows and believes in answering the question, the learner can immediately answer. However, if students need help, then students can ask for instructions. Furthermore, MeMo Tutor will provide work instructions based on the questions being worked on, provide modules so that students can review the material related to the question, and provide literature links related to the material. After the learner provides an answer, the MeMo Tutor will display the next question. If all the questions have been answered, MeMo Tutor will appreciate the students who have completed the assignment. Fig. 5 shows a flowchart of individual learning activities.
In week six, students in the experimental and control classes were asked to complete the posttest using a metacognition questionnaire.

C. Measures
This study uses a mixed-methods with a sequential explanatory approach. According to Creswell [63], the explanatory sequential method is first carried out by quantitative methods, followed by qualitative methods. Qualitative analysis can compensate for the small sample size in quantitative studies [64]. This study's experimental design used a quasi-experimental method of the nonequivalent type (pretest and posttest) control group.
RQ1: Investigate students' SRL and CRL skills. To answer this research question, we used a quantitative method. The questionnaire used in this study's pretest and posttest is the shared metacognition questionnaire [50]. We used a 5-point Likert scale to respond to the questionnaire, from very incorrect (1) to correct (5). Data were analyzed using Version 23 of SPSS [65] and Winsteps (Version 3.90.2) [66] to analyze the Rasch model. The Rasch model can provide relevant information about people's abilities and problem difficulties on a test [67]. In this study, the Cronbach alpha values of the questionnaire for all dimensions were 0.82, for the experimental class was 0.89, and for the control class was 0.86. The Cronbach alpha value for the SRL subdimension for the whole class is 0.84, for the experimental class is 0.82, and for the control class is 0.79. The Cronbach alpha value for the CRL subdimension for the whole class is 0.81, for the experimental class is 0.80, and for the control class is 0.82. The overall Cronbach alpha value is above 0.70, which means the instrument has high reliability.
This study also analyzed data on the LMS Moodle to support the data collected through the questionnaire. In individual learning activities, the data analyzed are the answers' accuracy, the students' arguments on the answers, and the number of assignments collected. The accuracy of the answers can provide information on the number of answers that are answered correctly by students. Answer arguments can provide information on the number of students who answer questions by including arguments. The number of assignments submitted can provide information on the number of students who submitted assignments.
In group learning activities, the data analyzed are the active participation of students in discussions. The number of students considered active in this analysis is students who post meaningful comments more than two times. We excluded meaningless comments, such as "Ok," "Good," or "Agree/Disagree" in the calculation. Meaningful comments contain student understanding or comments that respond to friends by adding arguments, such as: "I agree. It's following the theory …. (etc.)." RQ2: Examine the relationship between PAs and learner benefits. To answer the second research question, we use the conceptual content analysis to analyze qualitative data. This qualitative data were collected through an in-depth surveys on students' opinions regarding the individual and group learning process by MeMo Tutor. We asked students to answer their opinion about MeMo Tutor and the benefits and barriers they felt when using MeMo Tutor in learning. There are four questions posed as follows: "Does MeMo Tutor help you guide your online group discussions? Give the reason," "Does MeMo Tutor help you give instructions on online self-study (Quiz)? Give the reason," "Do you feel helped by the presence of the MeMo Tutor application? Give the reason," and "Mention the obstacles you feel (if any) in using MeMo Tutor." From students' responses through an in-depth survey, we developed four coding categories: individual learning process, group learning process, benefits, and obstacles.

V. RESULTS
We present the results of this study following the two research questions previously described.

A. RQ1: Does Integrating Metacognitive Scaffolding and Motivation in Pedagogical Agents Affect Students' Self-Regulated Learning and Co-Regulated Learning Skills?
To answer the first research question, we observed learning activities between MeMo Tutors and learners recorded in log files and Moodle activity logs. PA involves several steps to influence learners' SRL and CRL skills. In collaborative learning activities, PA influences learners as follows.
1) PA encourages learners to make comments on the discussion triggers that have been given. 2) Learners make comments.
3) PA asks learners to check other members in the group. 4) Learners check other members. If they have made comments, learners need to provide other comments or conclude. However, if other members have not commented, learners remind (call) these members using tags/notifications. 5) PA asks learners to draw conclusions. 6) Learners draw conclusions. 7) PA provides feedback and appreciation. Meanwhile, in individual learning activities, PA influences learners as follows.
1) PA invites learners to answer questions.
3) PA provides feedback on the answers to questions. 4) Learners respond to feedback. Then, we observed the pretest and posttest results of the questionnaire. Table II shows the descriptive statistics of the students' pretest and posttest SRL and CRL. From Table II, the mean scores of the posttest increased compared to the pretest scores for both groups. The results indicate that online learning activities can improve students' SRL dan CRL skills.
Kolmogorov-Smirnov test was used to determine normality distribution from this study's pretest and posttest data. The overall Kolmogorov-Smirnov test results are interpreted in Table III.  The normality test results indicate that there were normally   TABLE II  DESCRIPTIVE STATISTICS OF LEARNERS' LEARNING SKILLS   TABLE III  RESULTS OF THE NORMALITY TEST FOR THE EXPERIMENTAL AND CONTROL  CLASSES distributed data and not, so the test analysis was carried out using nonparametric statistics. The difference analysis of the pretest and posttest in the experimental class was carried out using the Wilcoxon signed rank. The analysis of posttest results between experimental and control classes using the Mann-Whitney U test. The level of significance (0.05) was taken as the basis.
The following hypothesis-based testing criteria were used to see whether there is a significant difference in the results of the pretest-posttest SRL skills between the experimental and control classes. H1: The experimental class's level of self-regulated learning skills was the same (there was no significant difference in the level of ability before and after the intervention).

H2: The level of self-regulation learning skills of the experimental and control classes is the same (there is no significant difference in ability levels between classes).
The Wilcoxon signed-rank test was performed to test the first hypothesis (H1). The Wilcoxon test showed no statistically significant difference (p > 0.05) in the improvement of individual learning abilities in the experimental class (Z = −1.520 and p = 0.128). The Mann-Whitney U test was conducted to test the second hypothesis (H2). The Mann-Whitney U test showed that the experimental class showed a statistically significant difference (p < 0.05) in the improvement of individual learning abilities compared to the control class (U = 715, Z = −2.672, p = 0.008) with a moderate effect size (r = 0.28) [68], [69].
The hypothesis-based testing criteria were used to see whether there is a significant difference in the pretest and posttest results of the CRL ability between the experimental and control classes.
H3: The level of co-regulated learning in the experimental class is the same (there is no significant difference in ability levels before and after the intervention).

H4: The level of co-regulated learning in the experimental and control groups is the same (there is no significant difference in ability levels between classes).
The Wilcoxon signed-rank test was performed to test the third hypothesis (H3). The Wilcoxon test showed that there was a statistically significant difference (p < 0.05) in the improvement of group learning ability in the experimental class (Z = −2.283 and p = 0.022) with a moderate effect size (r = 0.24) [64], [65]. The Mann-Whitney U test was conducted to test the fourth hypothesis (H4). The Mann-Whitney U test showed that the experimental class showed a statistically significant difference (p < 0.05) in the improvement of learning ability as a group compared to the control class (U = 717, Z = −2.658, p = 0.008) with a moderate effect size (r = 0.28) [68], [69].
To follow up on the results of the hypothesis, we performed a Rasch analysis. The Rasch model can provide objective measurements in various settings [70]. Rasch analysis used in this study is the average person measure, which can show the tendency of students' SRL and CRL skills based on responses to the questionnaire. Fig. 6 shows the results of the average person's measure in this study.
Based on Fig. 6, students' average SRL and CRL skills have a positive value, which means that students tend to have high skills. However, the posttest scores of the experimental class showed a relatively high increase in skills compared to the control class.
This study also analyzed data on the Moodle LMS to support the data collected through the questionnaire. Based on the graph of the analysis of individual learning activities in Fig. 7, the experimental class shows a good trend. The criteria for answers that were answered correctly by the experimental class were an average of 86%-93% of the total number of students, compared to the average in the control class, which is 69%-88%. The answer criteria by adding arguments to the experimental class is an average of 80% compared to the control class, 63%-79%. Likewise, with the criteria for collecting assignments, the experimental class average collected more assignments than the control class. The results indicate that students in the experimental class used instructions from the MeMo Tutor to reopen the material  related to the assignment and add arguments so that students were more engaged and motivated in collecting assignments.
Based on the graph analysis of group learning activities in Fig. 8, the experimental class showed a stable discussion activity compared to the control class, which tended to decrease. The results indicate that MeMo Tutor can guide students in the experimental class to be involved and motivated in group discussions.

B. RQ2: Does Integrating Metacognitive and Motivation Scaffolding in Pedagogical Agents Affect the Online Learning Process Individually and in Groups?
The qualitative data analysis of the student opinion from the experimental class regarding the MeMo Tutor's use was carried out using the content analysis method. We found both positive and negative responses arising from this content analysis. A summary of student opinions regarding learning activities with PAs is presented in Table IV.
Students in the experimental class gave their opinion regarding the use of PAs in individual learning. In general, students feel helped by the presence of PAs, especially in providing clear instructions for completing tasks (f = 23), helping learners to learn independently (f = 6), and enabling learners to relearn the assigned material (f = 3). Encourage learners to reevaluate the answers to the given assignment (f = 2), encourage students to look for other literature related to the assigned material (f = 2), and help learners focus on the given topic (f = 1). Train students to provide arguments for each question in the assignment (f = 1). However, a student had a negative opinion about PAs. According to him, the application runs slower, especially when sending answers or displaying questions. Some of the opinions given by students regarding the individual learning process are as follows: MeMo Tutor is evident in giving instructions so I can fill in answers quickly; In MeMo Tutor, there is a command about the material link that has been submitted by the lecturer so that we can understand more before continuing the material before doing the assignment; making me recheck the answer until I am sure of my answer.
Students feel helped by PAs' presence in group learning, especially in providing clear instructions to discuss well (f = 19) and encouraging learners to be involved in discussion activities (f = 13). It was encouraging learners to appreciate other people's opinions by doing praise or rebuttal well (f = 5), encouraging students to give opinions (f = 3), and encouraging students to discuss sincerely (f = 1). On the other hand, two students have negative opinions about PAs. According to them, online discussion activities are less fun because they have to reciprocate writing and prefer direct (face-to-face) discussion activities. Some of the opinions given by learners regarding the group learning process are as follows: I am immensely helped in the process of arguing; MeMo Tutor gives me the confidence to communicate with friends I don't know; I learned a lot about how to express opinions and respect opinions; With the how-to and how-to features, we can better discuss our work.
We also asked students for their opinions regarding the benefits and constraints they felt after using PAs. Table V presents a summary of student opinions regarding the benefits and constraints.
In Table V, there is an analysis of students' opinions regarding the benefits and constraints of using a PA (MeMo Tutor). When asked about MeMo Tutor's benefits in general, students stated  = 19), and discussion activities are well documented (f = 4). It helps students identify problems in doing individual tasks (f = 4) and encourage them to become active learners (f = 3). Regarding the obstacles to MeMo Tutor in general, 13 students stated that there were no obstacles. However, some students stated that the page needs to be updated when they want to see new comments/opinions (f = 16). The web feels slow (f = 4). Sometimes there is a long lag when posting comments/answers (f = 3), and there is no facility to add comments in the form of images (f = 2). Some opinions of students regarding the benefits and constraints of PAs are as follows: I find it very helpful, especially in discussing with friends, identifying problems together feels easier; Students become more active in sharing their opinions and ideas in discussions; Discussion activities are more fun, but must refresh every time a friend give updates; Hopefully, we can add a picture in the discussion column.

VI. DISCUSSION AND CONCLUSION
This study investigates the effect of metacognitive and motivational scaffolding integration on SRL and CRL skills at the higher education level. This study found that metacognitive and motivation scaffolding integration can improve students' self-regulation and CRL abilities based on the quantitative data analysis results. Also, the results of qualitative data analysis supported the findings obtained from quantitative data.
In individual learning, the test results reject the first null hypothesis and support the second null hypothesis. The results show that the integration of scaffolding positively affects SRL skills in students who study with agents compared to students who learn without agents. These results are consistent with previous research that students learn more from systems with agents than systems without agents and show that scaffolding-based PA can support SRL [21], [26], [71]. However, there was no significant effect between learning before and after using agents on the experimental class learners. It can be because learners are experienced and accustomed to individual learning in online learning activities. It knew that the COVID-19 pandemic that has occurred since March 2020 in Indonesia had caused learning activities at all levels of education to be carried out online [72]. Several studies in Indonesia show that even though they are full of challenges in online learning activities, students can adapt and try to learn [73], [74], [75]. From the results of qualitative data analysis, students feel helped by PAs' presence in individual learning activities, especially in providing instructions for completing tasks and helping learners learn independently. It enabled learners to re-study the assigned material, encourage learners to reevaluate answers from the assignment given, encourage learners to look for other literature related to the assigned material, help learners focus on the given topic, and train them to provide arguments on each question in the assignment.
In group learning, the test results reject the third and fourth null hypotheses. This test shows that PAs positively affect CRL skills in students who study with agents compared to students who learn without agents. The experimental class also showed a positive increase in CRL skills before and after using PAs. From the results of qualitative data analysis, students feel helped by PAs' presence in individual learning activities, especially in providing clear instructions for good discussion. It encourages learners to be involved in discussion activities, encourages students to respect other people's opinions by giving praise or rebuttal well, encourages students to give opinions, and encourages students to discuss sincerely.
Overall, the PA (using metacognitive and motivation scaffolding integration) can support students' learning process in higher education, individually and collaboratively. Also, as many as 86% of students stated that group learning activities with group discussions (4-5 people) were meaningful learning activities. This is in line with the study of Haugland et al. [76], who found that collaborative learning in small groups can promote in-depth learning. This study also revealed that learners' obstacles in discussing were not knowing discussion partners, not knowing how to start discussions (embarrassed/confused), and not understanding how to give opinions. The presence of a PA can help learners to overcome these obstacles.
These results provide preliminary evidence that integrating metacognitive and motivational scaffolds fosters cognitive engagement and manages learner motivation. Increased cognitive engagement and motivation are evident through the efforts made by students in individual and group learning activities. The PA is crucial in providing assistance to upgrade SRL and CRL skills. Increasing students' SRL and CRL skills can encourage students to increase cognitive engagement and motivation, resulting in a better understanding. This is because metacognitive scaffolding leads to deeper conceptual understanding, as evidenced by the ability of students to answer and argue. On the other hand, motivational scaffolding manages students' extrinsic motivation, as evidenced by their activeness in discussion and assignment collection.
However, this research also encountered some limitations. First, the PA prototype is still simple. The existence of technical constraints in the implementation of PA can affect the results. For example, slow web access when students work on individual assignments/group discussions can affect student motivation. Therefore, further studies are needed to develop prototypes and discuss technical influences on learning activities.
Second, experimental activities are carried out only during the semester. The effect of scaffolding integration may affect student achievement. Further research by adding various tasks/discussion topics for students can help understand the effect of scaffolding integration more broadly. Third, experimental studies can expand control conditions, such as groups that only use metacognitive or motivational scaffolding. It is necessary to examine whether the prompt or feedback in PA is sufficient to trigger metacognitive and motivational processes as expected.