The Use of Deep Learning to Improve Player Engagement in a Video Game through a Dynamic Difficulty Adjustment Based on Skills Classification

Round 1

Reviewer 1 Report

Statistical validation can be done to handle the randomness of neural network algorithms.

Error plot can be shown, and error values can be tabulated.

Training and testing process with cross validation of the datasets can be elaborated.

Author Response

Thank you for your insightful comments on handling randomness in neural network algorithms. We agree that the inherent randomness of these algorithms can introduce some variability in the model performance. To address this issue and provide a more robust estimation of the model's performance, we have added a discussion section on implementing k-fold cross-validation in the revised manuscript. Specifically, we have explained how cross-validation can manage the uncertainty in our Feedforward Neural Network (FNN), offering a more reliable estimate of the model's ability to classify player skill levels. This method would also help identify any overfitting in our model, which is crucial when dealing with complex models like FNNs. We appreciate your suggestion and believe that this addition strengthens our study.

Reviewer 2 Report

The manuscript entitled " Use of Deep Learning to improve player engagement in a video 2 game through a dynamic difficulty adjustment based on skills 3 classification ", Should be improved before the final submission. The flow of manuscript and the way of citing the reference is not up to the expectation of the readers. A strong proof read is mandatory,use of appropriate capitalization is needed.  The reviewer was wondering why the author have not elaborated the pseudo code / algorithm in detail.  In addition the following comments to be addressed.

Comment #1: Abstract is unnecessarily wordy. Make it brief and concise. Also, Conclusion should clearly state the outcome.
Comment #2: Discuss the stability of the system in terms of complexity.
Comment #3: Authors should provide the comments of the cited papers after introducing each relevant work. What readers require is, by convinced literature review, to understand the clear thinking/consideration why the proposed approach can reach more convinced results. In addition, authors also should provide more sufficient critical literature review to indicate the drawbacks of existed approaches, then, well define the main stream of research direction, how did those previous studies perform? Employ which methodologies? Which problem still requires to be solved? Why is the proposed approach suitable to be used to solve the critical problem? We need more convinced literature reviews to indicate clearly the state-of-the-art development. The existing references are not satisfactory.
Comment #4: Novelty is confusing. A highlight is required.
Comment #5: The authors should introduce their proposed research framework more effective, i.e., some essential brief explanation vis-à-vis the text with a total research flowchart or framework diagram for each proposed algorithm to indicate how these employed models are working to receive the experimental results. It is difficult to understand how the proposed approaches are working.

Comment #6: Mention the limitations and future works of the developed system elaborately.
Comment #7: Nothing is mentioned about the implementation challenges?
Comment #8: What type of computing environment is required to implement ?
Comment #9: It would be interesting if the authors report the trade-off compared to other methods especially the computational complexity of the models. Some techniques require more memory space and take longer time, please elaborate on that,if needed can use graphs.
Comment #11: References are not promising,use latest research works(many articles are available), and cite them in chronological order.

Across the script tenses,typos,use of we and our should be edited.
Language in the paper should be thoroughly checked with the help of English professionals and should be proof read before resubmitting.

Author Response

Thank you for the comprehensive feedback you have provided on our manuscript. We appreciate the time you have taken to offer such detailed suggestions for improvement. We have grouped your comments into categories to address them in a structured way and make the necessary improvements to our manuscript.

• Clarity and structure of the paper.
• Literature review and justification of the research.
• Technical implementation and comparison with other methods.
• Limitations and future works.

We hope that addressing your feedback in this manner will significantly improve the manuscript, and we look forward to your re-evaluation of our work.

 

Clarity and structure of the paper.

 

Comment #1: Abstract is unnecessarily wordy. Make it brief and concise. Also, Conclusion should clearly state the outcome.

 

We agree that the clarity of the abstract and conclusion is of utmost importance. In response to your suggestions, we have revised both sections to make them more concise and direct.

 

Comment #4: Novelty is confusing. A highlight is required.

 

In order to make the unique contributions of our research more prominent, we have added more details of the novelty in the introduction.

 

Comment #5: The authors should introduce their proposed research framework more effective, i.e., some essential brief explanation vis-à-vis the text with a total research flowchart or framework diagram for each proposed algorithm to indicate how these employed models are working to receive the experimental results. It is difficult to understand how the proposed approaches are working.

 

We have also updated the manuscript to incorporate a flowchart diagram to further illustrate our research framework visually.

 

Literature review and justification of the research.

 

Comment #3: Authors should provide the comments of the cited papers after introducing each relevant work. What readers require is, by convinced literature review, to understand the clear thinking/consideration why the proposed approach can reach more convinced results. In addition, authors also should provide more sufficient critical literature review to indicate the drawbacks of existed approaches, then, well define the main stream of research direction, how did those previous studies perform? Employ which methodologies? Which problem still requires to be solved? Why is the proposed approach suitable to be used to solve the critical problem? We need more convinced literature reviews to indicate clearly the state-of-the-art development. The existing references are not satisfactory.

Comment #11: References are not promising, use latest research works(many articles are available), and cite them in chronological order.

 

We added new references.

 

Technical implementation and comparison with other methods.

 

Comment #2: Discuss the stability of the system in terms of complexity.

 

Comment #7: Nothing is mentioned about the implementation challenges?

Comment #8: What type of computing environment is required to implement ?

Comment #9: It would be interesting if the authors report the trade-off compared to other methods especially the computational complexity of the models. Some techniques require more memory space and take longer time, please elaborate on that,if needed can use graphs.

 

We concur with your observations and have expanded our model validation section and added the implementation challenges in the revised manuscript. Specifically, we have delved deeper into how the complexity of the Feedforward Neural Network (FNN) contributes to the system's robustness and adaptability.

 

 

Limitations and future works.

 

Comment #6: Mention the limitations and future works of the developed system elaborately.

 

We recognize the importance of discussing in more detail the limitations of the system developed and potential directions for future work. We incorporated a new section into our revised manuscript.

 

Across the script tenses,typos,use of we and our should be edited.

Language in the paper should be thoroughly checked with the help of English professionals and should be proof read before resubmitting.

 

Thank you for pointing out the need for language and grammatical refinements. We have made comprehensive language edits to enhance the clarity and flow of the text.

Reviewer 3 Report

Abstract has some odd grammar choices - it could do with some more polish. It also only mentions the gender of a player when they are playing poorly which you ascribe as female. That is odd.

ln 53 - how is that crucial? Odd phrasing.

Starting a sentence (and indeed paragraph) with a citation (i.e [13]) is rather strange, and a bit lazy. Mention the author name and then put [13] after it, integrate it better into the sentence structure. You have done this multiple times.

Odd to note that UT is a registered trademark, but to not provide citation that is a web reference to the company that owns it...

I suggest restating / mentioning the problem (or project aim) at the start of section 3 before stating the methodology used. This refreshes the readers mind as to what the paper is doing, but then enables you to lead it into a justification of why that method is the appropriate method for this task. Randomly stating it leaves it hanging out there as to why that method, what is the point??

Why no mention that Astro is (or inspired by) space invaders - a game with  over 40 years of history. 

3.2 is very stilted and strangely phrased.

What is Path 1, Path 2 and Path 3 - mentioned in text, but not in figure. Left guessing as to what is what. why not use coloured arrows, or more detail in diagram.

Within 4.2 it doesn't seem to mention any details on how many of the participants had their game adjusted. Did you not record whether or not the difficulty was adjusted by the ML model during their play? i.e it would be good to know if it was lowered for some, or whether it raised to max for most of them.

The comparison between whether it was actually adjusted and their reported satisfaction would also be interesting. Is there a placebo effect in them knowing it "can" adjust, but not whether it actually did.

It would also be good to see, as it is a game with a score, what scores were achieved, and whether or not those that were perceived as being "middling" actually got a higher score by lasting longer with a lower difficulty. A harder game, could result in it being shorter, and thus less high scores, the relationship between these at a quantitative level, as separate to the survey focusing on the experience would be valuable to include.

Ethical question:

Did the informed consent include the photographs taken of them included in the paper? Or are the photos of participants of the researchers posing as participants??

 

 

 

 

Author Response

Abstract has some odd grammar choices - it could do with some more polish. It also only mentions the gender of a player when they are playing poorly which you ascribe as female. That is odd.

 

We have revised the abstract section to make them more polished.

 

ln 53 - how is that crucial? Odd phrasing.

 

We have revised the mentioned section to better articulate the importance of the Flow concept in understanding player-game dynamics. We believe this revised phrasing better communicates the significance of Flow in our research context.

 

Starting a sentence (and indeed paragraph) with a citation (i.e [13]) is rather strange, and a bit lazy. Mention the author name and then put [13] after it, integrate it better into the sentence structure. You have done this multiple times.

 

Amended the mentioned.

 

Odd to note that UT is a registered trademark, but to not provide citation that is a web reference to the company that owns it...

 

Thank you, the reference has been added.

 

I suggest restating / mentioning the problem (or project aim) at the start of section 3 before stating the methodology used. This refreshes the readers mind as to what the paper is doing, but then enables you to lead it into a justification of why that method is the appropriate method for this task. Randomly stating it leaves it hanging out there as to why that method, what is the point??

 

Thanks for the suggestion, the text has been added.

 

Why no mention that Astro is (or inspired by) space invaders - a game with  over 40 years of history.

 

Thank you for the suggestion, we added the mention to the manuscript.

 

3.2 is very stilted and strangely phrased.

 

The title was rephrased.

 

What is Path 1, Path 2 and Path 3 - mentioned in text, but not in figure. Left guessing as to what is what. why not use coloured arrows, or more detail in diagram.

 

We added colors to the figure

 

Within 4.2 it doesn't seem to mention any details on how many of the participants had their game adjusted. Did you not record whether or not the difficulty was adjusted by the ML model during their play? i.e it would be good to know if it was lowered for some, or whether it raised to max for most of them.

 

The comparison between whether it was actually adjusted and their reported satisfaction would also be interesting. Is there a placebo effect in them knowing it "can" adjust, but not whether it actually did.

 

To address the reviewer's concerns, we generated a table detailing selected individual game sessions and how the FNN machine learning model adjusted game difficulty based on player performance.

 

It would also be good to see, as it is a game with a score, what scores were achieved, and whether or not those that were perceived as being "middling" actually got a higher score by lasting longer with a lower difficulty. A harder game, could result in it being shorter, and thus less high scores, the relationship between these at a quantitative level, as separate to the survey focusing on the experience would be valuable to include.

 

We appreciate your suggestion to include a quantitative analysis of the scores achieved concerning the game duration and adjusted difficulty. However, in this study, we focused primarily on evaluating the gameplay experience and player satisfaction rather than analyzing the specific scores achieved.

Since our primary goal was to investigate how dynamically adjusting difficulty based on player skills affects the gameplay experience and player engagement, we focused on collecting qualitative data.

A more detailed analysis of the scores obtained and their relationship to game length and adjusted difficulty could be valuable. We will consider this suggestion for future research, where a more comprehensive quantitative analysis of the scores achieved, and their correlation with adjusted difficulty can be performed.

We appreciate your comments and will continue to improve our work based on the suggestions received.

 

Ethical question:

 

Did the informed consent include the photographs taken of them included in the paper? Or are the photos of participants of the researchers posing as participants??

 

We appreciate your inquiry about the photographs included in the article. We want to clarify that all photographs used in the study were captured with the informed consent of the participants. The informed consent form included taking photographs during the game sessions and their use in research reports.

Reviewer 4 Report

 

Introduction: It could be useful to highlight that DL loosely models the neurons in a biological brain. Regarding the topic of Flow, in my opinion, it would be advisable to explain further the "relationship" between challenges and skills, which lies undefined (line 69), and quote Csikszentmihalyi's works. Unfortunately, he is not cited, although he recognized and named the very concept of "Flow".

Related Works: It would be advisable to state whether the DL is simulating a player or an NPC more clearly. Especially for reference [13], where an unspecified "robot" is suddenly introduced, and [14], where the reader can guess the intelligent agent is a simulated player only by knowing the game Snake.

Evaluation: The procedure does not seem to involve a control group where the DL was not used while playing Astro. Therefore, positive results related to the engagement level during gameplay could be simply due to Astro being a great game per se. Further analysis should be conducted to investigate whether there are noticeable and statistically significant differences between standard Astro gameplay and Astro gameplay with complexity adjustment.

Additional notes: The manuscript states that some participants perceived the complexity adjustment. It would be interesting to explore this topic further, as some players may feel inadequate if they realize the game is lowering the difficulty due to their lack of skills. Also, the gender of the participants is not specified, and it might be interesting to explore possible differences in reactions, as well as potential differences in DL training related to different spatial skills between genders.

Finally, I noticed some typos in the manuscript and some issues regarding verb accordance.

Author Response

REVIEWER 4

Introduction: It could be useful to highlight that DL loosely models the neurons in a biological brain. Regarding the topic of Flow, in my opinion, it would be advisable to explain further the "relationship" between challenges and skills, which lies undefined (line 69), and quote Csikszentmihalyi's works. Unfortunately, he is not cited, although he recognized and named the very concept of "Flow".

Thank you for the suggestion, the reference was added and the DL concepts reviewed.

Related Works: It would be advisable to state whether the DL is simulating a player or an NPC more clearly. Especially for reference [13], where an unspecified "robot" is suddenly introduced, and [14], where the reader can guess the intelligent agent is a simulated player only by knowing the game Snake.

 

We apologize for any confusion. Our proposal is not about simulating an NPC but dynamically adjusting the game's difficulty based on the actual player's skills using DL. Instead of introducing a simulated intelligent agent, our approach focuses on understanding and classifying the player's skills to adapt to the game's difficulty in a personalized way.

Evaluation: The procedure does not seem to involve a control group where the DL was not used while playing Astro. Therefore, positive results related to the engagement level during gameplay could be simply due to Astro being a great game per se. Further analysis should be conducted to investigate whether there are noticeable and statistically significant differences between standard Astro gameplay and Astro gameplay with complexity adjustment.

Additional notes: The manuscript states that some participants perceived the complexity adjustment. It would be interesting to explore this topic further, as some players may feel inadequate if they realize the game is lowering the difficulty due to their lack of skills. Also, the gender of the participants is not specified, and it might be interesting to explore possible differences in reactions, as well as potential differences in DL training related to different spatial skills between genders.

 

In the limitations and future work, we address the reviewer's concerns.

Finally, I noticed some typos in the manuscript and some issues regarding verb accordance.

 

Thank you for pointing out the need for language and grammatical refinements. We have made comprehensive language edits to enhance the clarity and flow of the text.

 

 

Round 2

Reviewer 2 Report

AUTHORS HAVE ADDRESSED THE GIVEN QUERIES. CAN BE ACCEPTED.

Author Response

 

Thank you for your insightful comments on our paper.

Reviewer 4 Report

Csikszentmihalyi's works, the founder of the concept of Flow, are still not included in bibliography.

In Lines 493- 497 it is not appropriate to state that "Dynamic game complexity adjustment based on the skill level significantly improved player engagement and satisfaction" because we do not have a baseline. From what level of engagement are we "improving"?
A more honest statement would be that players perceived a high level of engagement and satisfaction attributed to the introduction of Dynamic game complexity adjustment.

Author Response

 

We apologize for the omission, the reference has been added. Reference 10.

 

In Lines 493- 497 it is not appropriate to state that "Dynamic game complexity adjustment based on the skill level significantly improved player engagement and satisfaction" because we do not have a baseline. From what level of engagement are we "improving"?

A more honest statement would be that players perceived a high level of engagement and satisfaction attributed to the introduction of Dynamic game complexity adjustment.

 

Thank you for bringing this to our attention. We have made the necessary text edits to the manuscript regarding that issue.