Personalized Feedback in Massive Open Online Courses: Harnessing the Power of LangChain and OpenAI API

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The paper is interesting and highly up-to-date.

The methodology is adequate to the study aim.

The context of the research is well presented.

 

I have the following critical remarks:

 

1) At row 33, several language models are introduced without any reference.

 

2) At row 70, the LangChain framework is introduced without any reference.

 

3) At row 179, where you mention "conventional automated feedback systems, which frequently depend on inflexible algorithms founded on rules", a refernce to such a rule-based feedback system is needed, e.g. https://drops.dagstuhl.de/entities/document/10.4230/OASIcs.SLATE.2017.6

 

4) At row 192, where you write "An inherent strength of generative AI lies in its capacity to acquire knowledge and develop gradually", a reference supporting this statement is welcome.

 

5) Section 5. Discussion has no references to results of other authors. This kind of section should put your results in the context of prior work.

 

6) Section 6. Conclusions and Future work is missing the description of study limitations (e.g., non-representative sample, etc.).

 

Author Response

Journal Electronics (ISSN 2079-9292)

Manuscript: ID electronics-2996457

Type: Article

Title: Personalized Feedback in Massive Open Online Courses: Harnessing the Power of LangChain and OpenAI API

Response to Reviewers

Dear Reviewers,

We appreciate the time and effort you have dedicated to reviewing our manuscript titled "Personalized Feedback in Massive Open Online Courses: Harnessing the Power of LangChain and OpenAI API." Your insightful comments and suggestions have been invaluable in helping us refine our paper and strengthen the presentation of our research. We are grateful for your recommendations and have addressed each point (in blue) in the following sections.

Regarding the question “Quality of English Language”. Two of three of the reviewers pointed the option: English language fine. No issues detected.

Below, we highlight the comment from one of the reviewers regarding the English language review, considering their observations.

Reviewer's Comment: The English language needs to be further checked, some grammatical errors in the paper.

Response: Thank you for pointing out the need for further language revisions. We recognize the importance of clear and correct language to effectively communicate our research findings. To address this, we have thoroughly reviewed the entire manuscript for grammatical accuracy to ensure the text meets academic standards.

Review #1

The paper is interesting and highly up-to-date.

R/ Thank you for recognizing the relevance and timeliness of our study.

The methodology is adequate to the study aim.

R/ We appreciate your affirmation of the methodology used in our research.

The context of the research is well presented.

R/ Thank you for your positive feedback on how we have presented the research context.

 Critical Remarks

At row 33, several language models are introduced without any reference.

R/ Thank you for pointing this out. We have now added the necessary references to support the introduction of the language models mentioned.

Row 33 was updated with: These advancements have enabled AI systems, such as OpenAI's GPT (Generative Pre-trained Transformer) series [1], Google's BERT (Bidirectional Encoder Representations from Transformers) [2], and DALL·E for image creation [3], to understand and produce human-like text, images, and even code [4].

At row 70, the LangChain framework is introduced without any reference.

R/ We appreciate this observation. References to the LangChain framework have been added to further substantiate the discussion.

Row 70 was updated with: This tool will integrate a linguistic model and the LangChain framework to leverage advanced AI capabilities, such as those found in the GPT series [7].

At row 179, where you mention "conventional automated feedback systems, which frequently depend on inflexible algorithms founded on rules", a reference to such a rule-based feedback system is needed, e.g. https://drops.dagstuhl.de/entities/document/10.4230/OASIcs.SLATE.2017.6

R/ Thank you for the suggestion. We have included this reference to enhance our discussion on conventional automated feedback systems.

Row 179, now 184 was updated with the suggested reference: In contrast to conventional automated feedback systems [22], which frequently depend on inflexible algorithms founded on rules, generative AI has the capability to customize its responses in order to accommodate the unique requirements and circumstances of individual learners.

At row 192, where you write "An inherent strength of generative AI lies in its capacity to acquire knowledge and develop gradually", a reference supporting this statement is welcome.

R/ We acknowledge this important remark and have added a reference that supports the statement about the capabilities of generative AI.

Row 192, now 218 was updated with one of the new references supporting the statement as follows: An inherent strength of generative AI lies in its capacity to acquire knowledge and develop gradually [28].

Section 5. Discussion has no references to results of other authors. This kind of section should put your results in the context of prior work.

R/ Thank you for your constructive feedback regarding the lack of references to results of other authors in the discussion section of our manuscript. Initially, our literature review did not identify similar studies that specifically addressed the issue of using AI for feedback enhancement in educational settings. Consequently, our focus was primarily on potential problems such as the lack of engagement. As you noted in line 98-100, "This lack of engagement and interaction can lead to a sense of isolation and decreased motivation, prompting students to abandon their courses" [8-10].

Following your recommendation, we have revised the discussion section to better contextualize our results within the framework of existing research. Specifically, we have included references to previous studies [8-10] that elaborate on engagement issues in educational contexts. These references are now detailed in lines 424 to 432 of the revised manuscript, highlighting how our findings on AI-generated feedback align with and extend the current understanding of engagement and retention in MOOCs.

Now, lines 424 to 432 read as follows: The positive feedback received from students regarding the sufficiency, timeliness, helpfulness, appropriateness, and necessity of AI-generated feedback, as measured by the Likert scale ratings, substantiates the potential of leveraging ChatGPT for educational purposes. This approach not only enhances the learning experience but also significantly boosts engagement, especially in MOOC environments where high dropout rates are a concern, as indicated by studies [8-10].

Section 6. Conclusions and Future work is missing the description of study limitations (e.g., non-representative sample, etc.).

R/ We appreciate your feedback on this. We have updated the Conclusions and Future Work section to include a discussion of the study's limitations.

Now, lines 536 to 545 read as follows: This study has several areas for further exploration. First, it was conducted with a relatively small group of students within a MOOC, suggesting that future research could benefit from a larger sample size to enhance the generalizability of the findings. Additionally, there is potential to further investigate how the use of generative AI impacts skill acquisition, which could provide deeper insights into the effectiveness of AI-driven educational tools. Lastly, the surveys in this study were administered at the end of the course, which may reflect participants' overall satisfaction with the course. Future studies might consider implementing longitudinal or cross-sectional research designs to isolate the effects of generative AI more accurately on learning outcomes and satisfaction over time.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

Thanks for authors effort on investigating the application of GenAI in educational feedback.  My concerns and suggestions are listed below:

1. the title is personalized feedback, but I did not see the formal definition of personalization in the paper. What did authors mean "personalization"? What's kind of personalized feedback?

2. The paper is about providing feedback for learners. However, there is no description about the existing gaps about feedback in MOOC in the Introduction section. The research motivation is not clear to me.

3. Section 2, Literature Review, this section is really unsystematic and missed lots of significant literature in feedback and automated feedback systems. The feedback framework proposed by Hattie and Timperley is classic and showing significant definition about feedback. However, many recent feedback literature describes that feedback is shifting to learner-oriented process [1]. It's ok to use feedback model by Hattie and Timperley but authors should also discuss the latest feedback framework in their paper. Then, section 2.2, authors missed lots of significant research about evaluating the potentials of GPT models on providing feedback. Dai et al (2023) [2] first investigated the feedback generated by ChatGPT for students' essay and further evaluated the feedback by GPT-4 [3]. These papers demonstrated the potentials of GPT models in generating feedback, but still insufficient, please check the details in these papers.

4. I saw authors used "learning styles" in this paper. I have to say many researchers from learning science demonstrated the evidence that "learning styles" is pseudo science. I would suggest authors to rephrase it in the paper

5. The research method is unclear. How did you design the experiment. Is this randomized control study? How did authors design the prompt? The prompt engineering step is important to be presented in the paper since the quality of prompt significantly influence the quality of generated text.

6. The result is not convincible to me. I did not find the comparison group. Authors only showed the results for the group of students who used GPT-generated feedback. Where is the control group? any statistical difference between control and experimental groups? 

 

 

[1] Henderson, M., Ajjawi, R., Boud, D., & Molloy, E. (2019). Identifying feedback that has impact. In The impact of feedback in higher education: Improving assessment outcomes for learners (pp. 15-34). Cham: Springer International Publishing.

[2] Dai, W., Lin, J., Jin, H., Li, T., Tsai, Y. S., Gašević, D., & Chen, G. (2023, July). Can large language models provide feedback to students? A case study on ChatGPT. In 2023 IEEE International Conference on Advanced Learning Technologies (ICALT) (pp. 323-325). IEEE.

[3] Dai, W., Tsai, Y. S., Lin, J., Aldino, A., Jin, F., Li, T., & Gasevic, D. Assessing the Proficiency of Large Language Models in Automatic Feedback Generation: An Evaluation Study.

Comments on the Quality of English Language

The English language needs to be further checked, some grammatical errors in the paper

Author Response

Review #2

 Reviewer's Comment: Thanks for authors effort on investigating the application of GenAI in educational feedback. My concerns and suggestions are listed below:

R/ Thank you for acknowledging our efforts in exploring the application of generative AI in educational feedback. We value your feedback and have carefully considered your concerns and suggestions to improve our manuscript as detailed below.

Reviewer's Concern: The title is personalized feedback, but I did not see the formal definition of personalization in the paper. What did authors mean by "personalization"? What kind of personalized feedback?

R/ We appreciate your observation regarding the absence of a formal definition of personalization. In response, we have now added a clear definition of "personalization" in educational feedback, explaining that it involves tailoring feedback based on individual learner characteristics, learning progress, and specific educational needs. This enhances the understanding of what we refer to as personalized feedback in our study and we included more references to expand the explanation.

Now, lines 102 to 112 read as follows: Educational research continuously emphasizes the crucial role of timely and personalized feedback in improving learning outcomes and student involvement, which has the potential to address some of the problems that contribute to high dropout rates. Personalized feedback involves tailoring answers based on individual learner characteristics, learning progress, and specific educational needs [11]. However, providing personalized feedback to a large number of students is a challenging task, especially when there is a limited number of instructors available in MOOC contexts. This discrepancy greatly impedes the ability to provide individualized feedback, further aggravated by prevailing evaluation methods in MOOCs that prioritize memorization and often yield evaluations with limited constructive input [12,13].

Reviewer's Concern: The paper is about providing feedback for learners. However, there is no description about the existing gaps about feedback in MOOCs in the Introduction section. The research motivation is not clear to me.

R/ Thank you fr pointing this out. We have revised the Introduction to clearly articulate the existing gaps in feedback mechanisms within MOOCs, which include the lack of personalized and timely feedback. This revision aims to better clarify the motivation and significance of our research. The text in the introduction was improved in lines 68-80 as follows:

 In light of the identified challenges, this project aimed to conduct an exploratory study to assess the degree of student satisfaction with personalized and automated feedback for learning activities within MOOCs. For this purpose, a tool was developed that enables the generation of personalized and automated feedback for learning activities within MOOCs. This tool integrated the LangChain framework to leverage advanced AI capabilities, such as those found in the GPT series [7].

The methodology involves using a structured evaluation rubric, defined by teaching staff, to categorize learner responses and generate context-specific feedback. This system capitalizes on the data-aware and agentive properties of LangChain, designed to enhance the learning experience by providing timely and individualized feedback crucial for student engagement and success in online educational environments. The tool’s effectiveness in delivering relevant feedback will be assessed to support students' autonomous learning in MOOCs.

 

Reviewer's Concern: Section 2, Literature Review, this section is really unsystematic and missed lots of significant literature in feedback and automated feedback systems. The feedback framework proposed by Hattie and Timperley is classic and shows significant definition about feedback. However, many recent feedback literature describes that feedback is shifting to a learner-oriented process [1]. It's okay to use the feedback model by Hattie and Timperley but authors should also discuss the latest feedback framework in their paper. Then, section 2.2, authors missed lots of significant research about evaluating the potentials of GPT models on providing feedback. Dai et al (2023) [2] first investigated the feedback generated by ChatGPT for students' essay and further evaluated the feedback by GPT-4 [3]. These papers demonstrated the potentials of GPT models in generating feedback, but still insufficient, please check the details in these papers.

R/ Thank you for your constructive feedback on the Literature Review section of our manuscript. We value your detailed suggestions, particularly your emphasis on the need to incorporate a discussion on the recent paradigm shifts in feedback processes toward a learner-oriented approach and the specific studies you have recommended. Acknowledging the importance of these insights, we have thoroughly revised this section to include a broader discussion on contemporary feedback frameworks, aligning with the recent literature [1] you pointed out.

Additionally, we have enriched our review by integrating significant research regarding the potential of GPT models for providing feedback. This includes the studies from Dai et al. (2023) [2] and their evaluations of GPT-4 [3] that you specifically mentioned. These amendments ensure that our literature review comprehensively covers both classic frameworks and the latest advancements in the field of automated feedback systems.

Section 2.2 was improved, in lines 197-217 as follows: Recent literature in the field of educational feedback emphasizes a significant shift towards learner-oriented processes [25]. This evolution reflects a departure from traditional feedback models, such as those proposed by Hattie and Timperley [14], which have been foundational but may not fully address the dynamic needs of today's diverse learner populations. In response to these developments, our review incorporates a discussion on how current feedback frameworks are adapting to be more learner-centered, facilitating a more personalized learning experience that actively engages students in their educational journeys.

Significant research has also been conducted on the potentials of GPT models in providing feedback, notably by Dai et al. [26]. Their study, "Can large language models provide feedback to students? A case study on ChatGPT," explores the capabilities of ChatGPT in delivering detailed and coherent feedback that not only aligns closely with instructors' assessments but also enhances the feedback by detailing the process of task completion, thus supporting the development of learning skills. The findings indicate that ChatGPT can generate feedback that is often more detailed than that provided by human instructors and exhibits a high degree of agreement with the instructors on the subjects of student assignments [27]. These insights have spurred further investigations into the practical applications of GPT models, leading this work to the conceptualization of a tool that utilizes a rubric defined by educators to generate personalized feedback automatically. This tool aims to assess student perceptions and explore the scalability of the solution, considering the economic model of token-based GPT tools.

Reviewer's Concern: I saw authors used "learning styles" in this paper. I have to say many researchers from learning science demonstrated the evidence that "learning styles" is pseudo science. I would suggest authors to rephrase it in the paper.

R/ We appreciate your critical feedback on this issue. In light of the current consensus in educational research, we have removed references to "learning styles" and have instead focused their preferences. More information on the topic https://fee.org/articles/learning-styles-don-t-actually-exist-studies-show/.

 

Reviewer's Concern: The research method is unclear. How did you design the experiment? Is this a randomized control study? How did authors design the prompt? The prompt engineering step is important to be presented in the paper since the quality of prompt significantly influences the quality of generated text.

R/ Your feedback has helped us realize the need for clearer methodological details. We have now thoroughly described the experimental design, confirming that it was a randomized controlled trial. Additionally, we elaborated on the prompt engineering process and complemented with references on the importance of proper prompt design by the teachers, which is crucial for the quality of AI-generated feedback, to ensure the rigor and reproducibility of our methods.

The quality and relevance of the prompts used can have a significant impact on the effectiveness of the response of the AI-based tool. In the context of instructional design for MOOCs, prompt engineering is a critical aspect of working with generative AI tools such as ChatGPT. To effectively communicate with AI-based tools, instructional designers must design prompts that are clear, concise, and relevant to the learning objectives of the course. In this context, one of the most common types of dialogue generation prompts is the professional perspective prompt, which requires the AI to assume the role of a particular person or profession and describe a topic within a given context. It has been demonstrated that using professional perspective prompts improves the quality of responses generated by language model-based dialogue systems. The suggested structure for professional perspective prompts is "Act as [author or profession] and describe [topic] + context," allowing ChatGPT to assume a particular role and provide a more detailed, objective, and structured response. And for this, examples were provided to the teachers for the evaluation rubric, with this the feedback was produced based on the answer from the student. As an example the following rubric prompt:

In the next paragraph the prompt generated by the instructors team is presented:

"Act as an expert teacher in artificial intelligence in education, with extensive experience in virtual education. Provide effective feedback to the student who responded to the question posed as a learning activity in Lesson 1 of the course 'Transforming Education with AI: Chat GPT.' Activity: 'Considering the exposed features and limitations of ChatGPT, how do you think this tool could transform teaching and learning in the classroom?' Student's response: [XXXXXXX].

For the feedback provided to the student, an empathetic and motivating tone should be used, evaluating the student's response based on the assessment criteria: (a) Understanding the potential of ChatGPT in education, (b) Consideration of ChatGPT's limitations, and (c) Clarity and coherence of the response. When generating the feedback, you should not show the student the evaluation criteria. Only generate effective feedback highlighting strengths and areas for improvement. And it should be presented in the format of an email."

 

Reviewer's Concern: The result is not convincible to me. I did not find the comparison group. Authors only showed the results for the group of students who used GPT-generated feedback. Where is the control group? any statistical difference between control and experimental groups?

R/. Thank you for your insightful comments and for highlighting concerns regarding the experimental design of our study, particularly the absence of a control group and the statistical analysis between control and experimental groups. We appreciate your feedback as it has helped clarify the context and objectives of our article concerning the implementation of a tool and its practical application.

In the current phase of our research, we focused on developing and testing a tool that utilizes a rubric-based approach to generate personalized feedback via a GPT model. We gathered qualitative feedback from students who interacted with the tool, specifically assessing their perceptions regarding the quality of the feedback generated by the rubric and the tool itself. These initial findings are crucial as they provide foundational insights into the usability and effectiveness of the AI-generated feedback in an educational setting.

We acknowledge the limitation mentioned regarding the lack of a control group. At this preliminary stage, our study was designed to explore the feasibility and acceptance of the tool. As such, a comparative analysis involving a control group was not conducted. However, we recognize the importance of this aspect for substantiating the results and have outlined plans for future research in the Conclusions and Future Work section of our manuscript. This future research will involve a more rigorous experimental design, including control and experimental groups, to enable a quantitative evaluation of the tool’s impact on learning outcomes compared to traditional feedback methods.

Additionally, we have expanded the discussion on the limitations of the current study in the manuscript’s conclusions. This includes detailing how these limitations will guide the subsequent phases of our research, ensuring that the next steps are clearly defined and aligned with best practices in educational research.

Lines 538 to 541 read now as follows: This study is not without limitations. Firstly, it was conducted with a relatively small group of students within a MOOC, which suggests that future research could benefit from a larger sample size to enhance the generalizability of the findings, as well as conduct a comparative study between an experimental and a control group.

 

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

The manuscript is quite briefly for the topic to be discussed and lacked a lot of details. We hope that the authors can provide more details. Here are some of our comments:

1. Could the authors describe what training data has been added to the LLM of GPT 4 through using LangChain ?

2. There are many ways to customize LLM. Please describe the specific reasons why the authors select LangChain, and describe the reasons why other LLM customization methods are not suitable, such as fine tuning, prompt tuning, and PEFT,... etc.

3. As far as we know, using LangChain may cause longer interactive response times. Please describe the impact of these additional delays caused by LangChain on the quality of interactions with students.

4. The authors did not mention in the manuscript what quality was achieved through the customization of LLM using LangChain, so that the growth of students' learning satisfaction in the manuscript can be achieved. The authors are encourage to provide supplemental information to this aspect.

5. Likert scale rating of 4.32 out of 5 (86%) can prove students' satisfaction with the use of generative AI, but it seems not able to specifically verify the effectiveness of  using of generative AI in improving learning quality. Is it possible to provide more specific experimental data related to improving learning effectiveness ?

Author Response

Review #3

Thank you for your constructive feedback on our manuscript. We acknowledge the need for more detailed information to fully convey the scope and results of our study. Here are the responses to each of your concerns:

Comment: Could the authors describe what training data has been added to the LLM of GPT-4 through using LangChain?

R/ Thank you for this question. In our study, we did not directly add training data to the GPT-4 model. LangChain was utilized to integrate GPT-4 with specific educational data sources and interfaces, which allowed the model to dynamically reference up-to-date educational materials and student feedback during interactions. The training data is primarily base on the prompt engineering and tuning. The quality and relevance of the prompts used can have a significant impact on the effectiveness of the response of the AI-based tool. In the context of instructional design for MOOCs, prompt engineering is a critical aspect of working with generative AI tools such as ChatGPT. To effectively communicate with AI-based tools, instructional designers must design prompts that are clear, concise, and relevant to the learning objectives of the course. In this context, one of the most common types of dialogue generation prompts is the professional perspective prompt, which requires the AI to assume the role of a particular person or profession and describe a topic within a given context. It has been demonstrated that using professional perspective prompts improves the quality of responses generated by language model-based dialogue systems. The suggested structure for professional perspective prompts is "Act as [author or profession] and describe [topic] + context," allowing ChatGPT to assume a particular role and provide a more detailed, objective, and structured response. And for this, examples were provided to the teachers for the evaluation rubric, with this the feedback was produced based on the answer from the student. As an example the following rubric prompt:

In the next paragraph one the prompt generated by the instructors team is presented:

"Act as an expert teacher in artificial intelligence in education, with extensive experience in virtual education. Provide effective feedback to the student who responded to the question posed as a learning activity in Lesson 1 of the course 'Transforming Education with AI: Chat GPT.' Activity: 'Considering the exposed features and limitations of ChatGPT, how do you think this tool could transform teaching and learning in the classroom?' Student's response: [XXXXXXX].

For the feedback provided to the student, an empathetic and motivating tone should be used, evaluating the student's response based on the assessment criteria: (a) Understanding the potential of ChatGPT in education, (b) Consideration of ChatGPT's limitations, and (c) Clarity and coherence of the response. When generating the feedback, you should not show the student the evaluation criteria. Only generate effective feedback highlighting strengths and areas for improvement. And it should be presented in the format of an email."

 

 

Comment: There are many ways to customize LLM. Please describe the specific reasons why the authors select LangChain, and describe the reasons why other LLM customization methods are not suitable, such as fine tuning, prompt tuning, and PEFT, etc.

R/ We appreciate your interest in our choice of customization technique. We selected LangChain due to its ability to seamlessly integrate with existing educational platforms and its flexibility in managing dynamic data sources without the need for retraining the model. Compared to fine-tuning and prompt tuning, LangChain offers a non-intrusive customization approach that maintains the generalizability of GPT-4 while tailoring its responses to specific educational contexts. We have expanded this discussion in the manuscript, the main objective was to prepare a first working tool.

 

Comment: As far as we know, using LangChain may cause longer interactive response times. Please describe the impact of these additional delays caused by LangChain on the quality of interactions with students.

R/ Thank you for your valuable comment regarding the potential impact of using the LangChain framework on interactive response times in our tool. We appreciate your insight, as it allows us to address an important aspect of our system's performance.

We acknowledge that integrating external frameworks like LangChain can indeed introduce longer response times due to the additional processing required to interface with OpenAI’s API. However, it is important to clarify that our tool has been designed primarily as a proof of concept rather than as a real-time chat or instant response tool. It is used specifically for the evaluation of student responses and formative activity within educational settings, where immediate interaction speed is less critical compared to the accuracy and relevance of the feedback provided.

These aspects of response time and system scalability are crucial and warrant long-term evaluation, particularly in relation to the economic model of connecting to OpenAI’s API. As such, we have included a discussion of these concerns in the Conclusions and Future Work section of our manuscript. This inclusion outlines our plans to investigate the scalability of the solution and the potential economic implications of API usage as we further develop and refine the tool.

 

Comment: The authors did not mention in the manuscript what quality was achieved through the customization of LLM using LangChain, so that the growth of students' learning satisfaction in the manuscript can be achieved. The authors are encouraged to provide supplemental information to this aspect.

R/ This is an excellent point. To address this, we have added details about the specific improvements in learning satisfaction observed in our study and as complemented in the response to comment #1. Customization through LangChain led to feedback that was more aligned with students' immediate learning contexts and the rubric by the teacher, which significantly contributed to enhanced satisfaction and perceived value of the feedback received. These results have now been detailed in the Results section of our manuscript.

 

Comment: Likert scale rating of 4.32 out of 5 (86%) can prove students' satisfaction with the use of generative AI, but it seems not able to specifically verify the effectiveness of using generative AI in improving learning quality. Is it possible to provide more specific experimental data related to improving learning effectiveness?

R/ We appreciate your suggestion to provide more specific experimental data on learning effectiveness. While our current study primarily focused on measuring satisfaction with the AI-generated feedback, we recognize the importance of directly assessing learning outcomes. At this stage, we do not have the additional data required but agree that this is a critical area for future research. We have added a section in the manuscript outlining potential studies that could be undertaken to rigorously evaluate the impact of generative AI on learning quality. These future studies could include control group comparisons and pre-post tests specifically designed to measure learning gains and the effectiveness of AI interventions in educational settings.

Lines 538 to 541 read now as follows: This study is not without limitations. Firstly, it was conducted with a relatively small group of students within a MOOC, which suggests that future research could benefit from a larger sample size to enhance the generalizability of the findings, as well as conduct a comparative study between an experimental and a control group.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

The revisions have significantly improved the clarity, depth, and overall quality of the manuscript, making it a valuable contribution to the field. I appreciate the authors' efforts in revising the paper to enhancing its quality. 

Reviewer 3 Report

Comments and Suggestions for Authors

We recommend to accept this manuscript based on its content and the replied comments.