Generative Adversarial Framework with Composite Discriminator for Organization and Process Modelling—Smart City Cases
Round 1
Reviewer 1 Report (Previous Reviewer 3)
Comments and Suggestions for AuthorsThe paper is improved considering given comments and I recommend its publishing.
Author Response
Response to Reviewer
We sincerely thank the reviewer for your valuable time and efforts in reviewing our manuscript!
Comment 1: The paper is improved considering given comments and I recommend its publishing.
Reply: Thank you very much for the positive evaluation of our work!
Author Response File: Author Response.pdf
Reviewer 2 Report (New Reviewer)
Comments and Suggestions for AuthorsThe manuscript cannot be accepted for publication as Figures 7-11 appear to be raw software outputs rather than professionally prepared scientific figures meeting academic publication standards.
Author Response
Response to Reviewer
We sincerely thank the reviewer for your valuable time and efforts in reviewing our manuscript! Those comments are all valuable and very helpful for revising and improving our article, as well as the important guiding significance to our researches. We have studied comments carefully and have made correction which we hope meet with approval.
The description of different fonts used in this document are as follows:
- Reviewers’ original comments are reproduced in red-colored
- Plain fonts are our answers to Reviewers’ comments.
- Revised and added material is highlighted in the article with yellow color.
Comment 1: The manuscript cannot be accepted for publication as Figures 7-11 appear to be raw software outputs rather than professionally prepared scientific figures meeting academic publication standards.
Reply: Thank you for noticing this issue! In order to follow academic publication standards we have formatted the software outputs as listings (Listings 1-6). Besides, we have included a figure illustrating the potentially selected process model (Figure 7).
Author Response File: Author Response.pdf
Reviewer 3 Report (New Reviewer)
Comments and Suggestions for AuthorsThe paper aims to present an approach to generating and complementing graph-based organization and process models using GAN and the examples of its use in smart logistics and smart tourism, especially when the dataset is small. The main contribution is to propose a solution that can be used to generate models that would be helpful in further research in different scientific fields. The developed framework was put in a publicly available open-source library and can be used by other modeling software developers.
The paper is well-written, the structure is clear and understandable, and each research step is clearly explained.
The topic of the paper is very interesting, as it shows the use of AI in generating models that are as far created by modelers, mainly using standard software. The solution may be very helpful in research as it may save time for the researchers to focus on the problem and not waste time building the process model.
Specific comments:
Why is some text highlighted in yellow? Please remove that.
Fig. 1 – the text that is placed at the sides of the figure should be moved to the text of the paper. The descriptions in the figure should be short, and the explanation should be in the text.
The same for Fig. 2.
The language needs some corrections; the sentences are sometimes built in an incorrect order.
In my opinion, the discussion section should discuss the authors’ results in context with the results of other researchers, especially those mentioned in section 2.
The discussion section should contain the main findings, the practical implications (as given in the highlights), and the research’s limitations (as described in section 7).
Author Response
Response to Reviewer
We sincerely thank the reviewer for your valuable time and efforts in reviewing our manuscript! Those comments are all valuable and very helpful for revising and improving our article, as well as the important guiding significance to our researches. We have studied comments carefully and have made correction which we hope meet with approval.
The description of different fonts used in this document are as follows:
- Reviewers’ original comments are reproduced in red-colored
- Plain fonts are our answers to Reviewers’ comments.
- Revised and added material is highlighted in the article with yellow color.
Comment 0: The paper aims to present an approach to generating and complementing graph-based organization and process models using GAN and the examples of its use in smart logistics and smart tourism, especially when the dataset is small. The main contribution is to propose a solution that can be used to generate models that would be helpful in further research in different scientific fields. The developed framework was put in a publicly available open-source library and can be used by other modeling software developers.
The paper is well-written, the structure is clear and understandable, and each research step is clearly explained.
The topic of the paper is very interesting, as it shows the use of AI in generating models that are as far created by modelers, mainly using standard software. The solution may be very helpful in research as it may save time for the researchers to focus on the problem and not waste time building the process model.
Reply: Thank you very much for the positive evaluation of the paper!
Comment 1: Specific comments:
Why is some text highlighted in yellow? Please remove that.
Reply: The paper is a resubmission of a previously selected version with yellow colors highlighting the changes compared to the previous version (as required by the editors).
In this revised version we have removed the previous highlightings but highlighted the newly introduced modifications. These will be removed in the final version of the paper.
Comment 2: Fig. 1 – the text that is placed at the sides of the figure should be moved to the text of the paper. The descriptions in the figure should be short, and the explanation should be in the text.
The same for Fig. 2.
Reply: Thank you for identifying this issue! Both Fig. 1 and Fig. 2 have been updated to remove the explanations, and the explanations have been included into the paper text describing the figures.
Comment 3: The language needs some corrections; the sentences are sometimes built in an incorrect order.
Reply: The paper has been proof-read and a number of language issues have been fixed.
Comment 4: In my opinion, the discussion section should discuss the authors’ results in context with the results of other researchers, especially those mentioned in section 2.
The discussion section should contain the main findings, the practical implications (as given in the highlights), and the research’s limitations (as described in section 7).
Reply: The discussion section has been modified and extended according to the provided comment. It now includes a comparison of the developed approach with a conventional GAN as well as main findings and practical implications.
Author Response File: Author Response.pdf
Reviewer 4 Report (New Reviewer)
Comments and Suggestions for AuthorsProblem 1: The effectiveness analysis of the composite discriminator
The composite discriminator is the core innovation of this paper, but its effectiveness is only validated through the constraint satisfaction accuracy of the generated models, lacking comparative experiments with traditional GAN models. It is recommended to add comparative experiments to evaluate the performance of the composite discriminator against traditional single discriminators, particularly in terms of model diversity, convergence speed, and constraint satisfaction accuracy.
Problem 2: Insufficient experiments
The experimental section validates the method's effectiveness using two cases: smart logistics and smart tourism. However, the dataset scale is relatively small, especially with only 20 samples for the logistics system model, which may affect the generalizability of the results. It is recommended to expand the dataset size by incorporating more real-world data or generating more diverse training samples through data augmentation techniques. Additionally, while the introduction mentions the method's potential applicability to other domains, no related evidence is provided. It is suggested to include cross-domain validation to demonstrate the method's effectiveness.
Problem 3: Lack of diversity evaluation for generated models
The paper does not evaluate the diversity of the generated models. Diversity is a crucial metric for GANs, especially in organization and process modeling, where the generated diverse models can help address varying business needs. It is recommended to add an evaluation of diversity the of generated models.
Problem 4: Insufficient discussion regarding practical application scenarios.
The experimental scenarios in the paper are somewhat idealized and do not fully consider the complexities and uncertainties of real-world applications (e.g., model adaptability in dynamic environments). It is recommended to real-world application scenarios, particularly focusing on model adaptability in dynamic environments. For instance, this paper can explore how to quickly adjust generated models in response to changing requirements or how to handle modeling under incomplete information.
Problem 5: Future work outlook
The future work section of this paper is relatively brief. It is recommended to expand this section, particularly in areas such as the scalability of GAN models, cross-domain applications, and integration with other machine learning methods.
Comments on the Quality of English LanguageThis paper requires a comprehensive review to address numerous issues in grammar, syntax, and phrasing.
Author Response
Response to Reviewer
We sincerely thank the reviewer for your valuable time and efforts in reviewing our manuscript! Those comments are all valuable and very helpful for revising and improving our article, as well as the important guiding significance to our researches. We have studied comments carefully and have made correction which we hope meet with approval.
The description of different fonts used in this document are as follows:
- Reviewers’ original comments are reproduced in red-colored
- Plain fonts are our answers to Reviewers’ comments.
- Revised and added material is highlighted in the article with yellow color.
Comment 1: Problem 1: The effectiveness analysis of the composite discriminator
The composite discriminator is the core innovation of this paper, but its effectiveness is only validated through the constraint satisfaction accuracy of the generated models, lacking comparative experiments with traditional GAN models. It is recommended to add comparative experiments to evaluate the performance of the composite discriminator against traditional single discriminators, particularly in terms of model diversity, convergence speed, and constraint satisfaction accuracy.
Reply: Thank you very much for deep points!
Unfortunately, there are no direct competitive models to compare with. One of the closest models is MolGAN (referred in the paper), but it does not address generation of numerical parameters of graph nodes and partial definition of generated graphs.
We have included a comparison of the GAN training process using the developed composite discriminator with the conventional GAN to the Discussion section. It is based on the training of simplified Smart Logistics System Modelling Use Case with application of only structural constraints. It can be seen that the application of the composite discriminator enabled reduction of the incorrect generated models for 36.4%: the achieved accuracies after 22000 epochs for the composite discriminator and the conventional discriminator are 88.3% and 81.6% respectively.
Comment 2: Problem 2: Insufficient experiments
The experimental section validates the method's effectiveness using two cases: smart logistics and smart tourism. However, the dataset scale is relatively small, especially with only 20 samples for the logistics system model, which may affect the generalizability of the results. It is recommended to expand the dataset size by incorporating more real-world data or generating more diverse training samples through data augmentation techniques. Additionally, while the introduction mentions the method's potential applicability to other domains, no related evidence is provided. It is suggested to include cross-domain validation to demonstrate the method's effectiveness.
Reply: As it was mentioned in the paper, today there are no high quality datasets for the considered domain (enterprise and process modelling). That was the main reason to develop the presented in the paper augmentation procedure.
For example, the presented process modeling case demonstrates training on a subset of the SAP-SAM subset, which is itself a big dataset but very diverse (even in terms of model types and languages used) and unfiltered dataset. So that only 29 applicable diverse high-quality models from the same area (tourist trip booking) have been found.
That is the main reason for reporting only two experiments in the paper.
The corresponding explanation has been added to the Discussion section.
Comment 3: Problem 3: Lack of diversity evaluation for generated models
The paper does not evaluate the diversity of the generated models. Diversity is a crucial metric for GANs, especially in organization and process modeling, where the generated diverse models can help address varying business needs. It is recommended to add an evaluation of diversity the of generated models.
Reply: We completely agree that the diversity of the generated models is one of the crucial metric for GANs. However, in the presented scenarios this is not the case. Good specification of the being designed enterprise or process model does not allow for a wide range of diverse models. In fact, what is needed is several similar (but different in few aspects) models that meet the given requirements and the modeler can adjust what is needed.
This is the reason why the diversity analysis has not been performed. This indeed opens a discussion that GAN might be not an optimal mechanism for the given task, however, our approach illustrates that it is still capable of solving it. We have included this point into the Discussion section.
Besides, one can observe the diversity of generated models in the process model scenario. For this reason, we presented 5 various generated models.
Comment 4: Problem 4: Insufficient discussion regarding practical application scenarios.
The experimental scenarios in the paper are somewhat idealized and do not fully consider the complexities and uncertainties of real-world applications (e.g., model adaptability in dynamic environments). It is recommended to real-world application scenarios, particularly focusing on model adaptability in dynamic environments. For instance, this paper can explore how to quickly adjust generated models in response to changing requirements or how to handle modeling under incomplete information.
Reply: Thank you for suggesting an interesting scenario! The suggested framework does not support adjustment of the models, instead, the modeler can modify the input parameters for the model to generate a new one that would meet the changed requirements. The input parameters in this case can be not only numeric ones, but also a partially defined model – so the modeler can “freeze” the fragments of the model, that have to be preserved, update the input parameters and generate new models. In this case, the generation would take the same time as the generation of the new models.
Comment 5: Problem 5: Future work outlook
The future work section of this paper is relatively brief. It is recommended to expand this section, particularly in areas such as the scalability of GAN models, cross-domain applications, and integration with other machine learning methods.
Reply: Thak you for the recommendation! We have extended the future work description in the paper! Unfortunately, integration with other machine learning methods is not currently a topic for the nearest future, but we will keep this in mind for the more remote future since this is indeed a very interesting research topic.
Author Response File: Author Response.pdf
Round 2
Reviewer 4 Report (New Reviewer)
Comments and Suggestions for AuthorsNo other comments.
This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThe paper use of GANs for smart city applications, but it has several technical limitations and potential flaws:
- The approach may face challenges when applied to large-scale or highly complex organizational systems due to the computational intensity of GAN training, especially with composite discriminators​​.
- Validation of differentiable and non-differentiable constraints in the composite discriminator might not always converge efficiently, potentially affecting the accuracy and speed of model generation​​.
- The mechanism relies on meaningful data augmentation to handle limited datasets. However, the augmentation strategy might introduce biases or fail to adequately represent real-world complexities​​.
- The proposed AMGS algorithm for multi-level model generation assumes smooth parameter transfer between levels. However, real-world systems often exhibit non-linear interactions and feedback loops that are not accounted for​​.
Reviewer 2 Report
Comments and Suggestions for AuthorsThe article addresses a highly interesting, timely, technically and scientifically significant topic. In particular, the proposed article is well-suited for publication in a reputable international scientific journal.
The paper discusses an algorithm developed and proposed by the authors, focusing on the application of generative AI techniques to accelerate and streamline the development of organizational structures and processes. The topic is well related with smart cities, a crucial aspect in consideration of the rapid growth of smart systems, smart infrastuctures and digital twins of engineering applications. The article, therefore, creates a link between these concepts, the use of specific alghorithms of machine and deep learning and suggest a new method for the case of smart turism.
The article is overall well-written, well-organized and allows readers to clearly follow the logical process developed by the authors. Furthermore, the novelty of the article is effectively summarized both in the paper itself and in the highlights and abstract. The authors well introduece the topic with a literature reviwe well bakanced and with appropriate references. The conclusions well resume the most significant results achieved.
However, before final publication, the authors are requested to make a few minor revisions:
- Some minor typographical mistakes are present in the text, such as on page 3, line 108. A general and comprehensive review of the text is recommended, although the language used is already good and satisfactory.
- In the literature review, some cited papers are not very clearly introduced, particularly those between references 26 and 30. For these articles, the term “authors” is often used. For references where the contribution is more significant, it is recommended to use the expression “name of first author et al.” to cite them.
- On page 15, several variants are presented as figures of lines of code directly in the text. It is recommended to designate these as proper figures with captions, while explaining the various modifications more thoroughly in the text.
- In Figure 6, add axis titles. Moreover, it is suggested to adequate the figure dimensions to the tempalte margins. You can consider the possibility of using two lines for a better comprehension of the graphs.
Reviewer 3 Report
Comments and Suggestions for AuthorsThe paper presents an architecture of generative adversarial network (GAN) in support of creation organizational and process models. The proposed approach is described in detail and it is demonstrated through examples. The results of experimentations are disccused as some limitations of the approach are mentioned.
The paper could be improved in the following way:
- In the paper title is mentioned just one case in smart tourism, bur in the content there are examined two cases: logistic system and smart tourism. The title and content of the article should be aligned.
- The aim of the paper should be very precisely formulated at the end of the Introduction, as well as the main contributions.
- It is not entirely clear whether the developed library is the purpose of presenting it in this article or whether this could be done in a previous article. This should be made very clear.
- Subheadings 2.1 and 2.2 should be different. Please rephrase at least one of them.
- It is not clear whether the used algorithms in section 4 (ATGM, ATCD and AMGS) are developed by authors. If they were developed by the authors, this should be stated. If they are not, this should also be stated and the relevant literature sources should be cited.
- Formulas should be numbered and cited if not created by the authors.
- There are two subsections 5.1. Please correct this.
- The section 6. Programming Library is redundant since its development is presented in [53]. Here you can only mention and indicate the literature source.