Coupling Hyperbolic GCN with Graph Generation for Spatial Community Detection and Dynamic Evolution Analysis
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThe manuscript addresses a very interesting subject with various beneficial real-world applications. However, unfortunately, I must admit that the current version of the manuscript has several fundamental flaws, and only after a major revision and one more round of review, I can suggest a decision about its acceptance or rejection.
Below, I listed only my major concerns, and if the authors address them in their revision I will provide more detailed issues in my next round of review.
1. Lack of clarity is one of the main issues of the manuscript that shows itself in various sections of the manuscript, such as notation, introduction, and even in the literature review.
2. Not accessible text: Section 3 is not accessible at all. I recommend rewriting this section and inserting all missing descriptions of the mathematical equations.
3. Ambiguity: For instance, in subsection 3.2, they introduced the notation of ordinary networks; however, at the end of this subsection, they talked about the possibility of adding attributes: which changes their subject for the node-attributed network or feature-rich network. Moreover, although they were concerned about the evolution of networks (over time) there is no sign of "time" in their introduced notation until subsection 3.5.
4. Lack of justification: why Euclidean distance, especially in their case, for which their data points is a vector of six elements (pickup longitude, pickup latitude, pickup time, drop-off longitude, drop-off latitude, 367 drop-off time) is not applicable, and why do they need embedding? Moreover, what does "sequentially" (mentioned in line 250) mean? Why do we need hyperbolic embedding, not a convolutional neural network?
5. Authors should apply their proposed method to a similar well-known benchmark data set (with ground truth) to prove the effectiveness of their proposed method.
6. The literature review must be extended.
Comments on the Quality of English LanguageThe English of the manuscript needs professional revision.
Author Response
Please see the attachment.
Author Response File: 
Author Response.docx
Reviewer 2 Report
Comments and Suggestions for AuthorsThis study aims to analyze urban spatial interaction community detection as well as the temporal evolution. Specifically, this study propose a method for detecting community by using hyperbolic GCN. The manuscript is well organized and the content is novel in method. I have the follow comments:
1. How to quantify and assess the effectiveness of community detection among different detection methods (such as proposed method vs other complex methods like infomap, Leiden)，Although the Figure 4 shows the visualization of proposed method and Leiden，how to quantify the effectiveness and demonstrates that your proposed method is better than other detection methods? Currently, there also have many excellent complex detection methods. And the efficiency of your proposed method vs existing methods?
2. A discussion is need to be added to describe the practical significance of the research results. How these dynamic evolution of communities is meaningful for urban management or planning?
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Reviewer 3 Report
Comments and Suggestions for Authors-
The paper considers community detection and the evolution of dynamic communities for geospatial analysis. Community detection in urban studies is widely used and it is meaningful an innovative to study the group evolution as well. Therefore the reserach topic of the paper is relevant and tits into the scope of IJGI.
The main problem with the paper is that no clear reserach question with respect to Geo-information research is identified, In the introduction the objective of the paper is very generally described with „to understand the spatiotemporal dynamic structure of urban spaces”. Principally it is important and for policy making and urban management elements of this concept can be exploited, but the way of that and the specific problems to be solved need to be defined. The use case and its discussion does not help either to answer this question for the reader – community detection and its primer analysis can be interesting as itself, but this methodlogy should be used to support problem solution in a secondary manner. It is not established either in the paper what the advantage of graph embeddding is compared to traditional methods. In the test case the comparison to the Leiden method does not justify this either. The paper needs a very careful revision in which the clear objective and contribution are clarified.
A few specific comments are also listed below.
- In the introduction the authors write about the „static” properrty of spatial structures, but it is unclear what it mean.
- In the related work of community detection a short review about deep learning based methods is missing.
- In 3.2 it remains unclear how exactly the networks is constructed. An example would be beneficial.
- In 4.3 why the Leiden method is chosen for comparision? In 2.2 it is emphasized that the special geospatial community detection algorithms perform better than the general purpose methods. Therefore comparisions to methods from this category would be meaningful.
English is Ok
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Reviewer 4 Report
Comments and Suggestions for Authors1. The authors have modified mostly it. The article can be understood easily by the readers.
2. cite the figure 1 from where it was taken.
3. Figure 2 was not visible. Redraw by using high resolution.
4. Equations 1 to 5 have to be described briefly. explain each parameter by considering the parameters.
5. Explain the steps regarding the GED algorithm.
6. The GCN model has already used by so many authors. Can the authors justify if there is a novelty in their model?
7. Results should be compared with the existing models. Compare each snapshot.
8. Separate the section 5. Maintain the Discussion section separately. Conclusion and future work in another section.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Round 2
Reviewer 1 Report
Comments and Suggestions for Authors The manuscript has significantly been extended and improved. The majority of my previous concerns have been addressed satisfactorily. While reading the previous version of the manuscript was rather unpleasant and confusing to me, I did enjoy the current version. I would like to congratulate the authors: they proposed an interesting method and managed to explain it nicely.Nevertheless, I think the authors need to take care of a couple more comments, which are mostly minor, and after that, the manuscript will be publishable. Below, I listed my (major and minor) comments. Additionally, I marked some of them with asterisk (*). The comments with an asterisk must be addressed, while the others can be considered for future work.
Major:
1*) In subsection 3.2, the authors formulated the problem using the data structure which in the literature is referred to as feature-rich (attributed) networks. So they should either justify why they did not review the related literature or add a subsection in the literature review, and review a couple of those methods, including, the methods mentioned below:
[1] Chunaev P. Community detection in node-attributed social networks: a survey.
Computer Science Review. 2020 Aug 1;37:100286.
[2] Shalileh S, Mirkin B. Community partitioning over feature-rich networks using an extended k-means method. Entropy. 2022 Apr 29;24(5):626.
[3] Shalileh S, Mirkin B. Community Detection in Feature-Rich Networks Using Gradient Descent Approach. In International Conference on Complex Networks and Their Applications 2023 Nov 28 (pp. 185-196). Cham: Springer Nature Switzerland.
[4] Rostami M, Oussalah M, Berahmand K, Farrahi V. Community detection
algorithms in healthcare applications: a systematic review. IEEE Access. 2023 Mar
[5] Tsitsulin A, Palowitch J, Perozzi B, M ̈uller E. Graph clustering with graph neural networks. Journal of Machine Learning Research. 2023;24(127):1-21
2*) they did not explain how they proposed to construct the spatial interaction networks (sec, 3.2). Fig 3. Does not provide enough information.
3) clarifying the range of evaluation metrics will help the readers to have a better understanding of the quality of the proposed method.
4*) As far as I understand, the number of clusters is defined in the membership vectors and it should be a hyper-parameter. This needs to be clarified in the text. If I am not mistaken in my understanding, then the question arises how did they decide the number of clusters while they compared their proposed method with two previous works?
5) I recommend authors to repeat their experiments,
reported in Tables 3-6, and report the average and standard deviation of their evaluation metrics. Or at least, in their future work, do not merely report the value of the metrics for one run. (I usually reject papers without reporting the average and standard deviation; however, for the current paper, due to the other set of experiments, I do not have an intention to reject the paper for such an issue)
6*) Fig. 8 (and wherever different numbers of clusters were used in the experiments) the authors should clarify how the number of clusters was determined for each method.
Minor:
- Line 80, which network was used,
- Line 82, 83: clarify how the mapping from spatial interaction network to hyperbolic space was done.
- Line 115, what is GED?
- Line 158 what is GN? Please avoid using abbreviations without introducing the full name before it.
- Fig -c is not mentioned in the text.
- to increase the readability I recommend subsections at the beginning of subsection 3.1
- Line 316, clarify what is nose representation vectors (by providing an example)
-* Eqn. (10): typo: GCN should be HGCN (if i am wrong clarify)
- It is recommended to mention the disadvantages of the proposed method.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Reviewer 2 Report
Comments and Suggestions for AuthorsI have no comments.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Reviewer 3 Report
Comments and Suggestions for AuthorsThe paper improved significantly and considered my comments. Nevertheless, I still have a few minor comments.
1. In the reviewer answer document the authors gave appropriate answer to my comment identified as Comment no 8. However it is not described in the paper. Please also involve an appropriate explanation in the paper
2. In the abstract the new part starting with "Most existing methods..." is still not clear enough. Please clarify and reword it.
3. My comment identified as Comment no 2 in the answer is appropriately answered in Section 4.4. On the other this potential application to justify the background of the research and the choice of the use case should be mentioned/highlighted as early as in the Introduction.
4. The review paper of Rossetti, G., & Cazabet, R. (Community Discovery in Dynamic Networks. ACM Computing Surveys (CSUR), 51 (2017), 1 - 37 .) should be also cited.
5. The citation no 47 has some editing error.
Comments on the Quality of English LanguageThe English is Ok, some proofreading is needed
Author Response
Please see the attachment.
Author Response File: Author Response.docx
