All reviews of published articles are made public. This includes manuscript files, peer review comments, author rebuttals and revised materials. Note: This was optional for articles submitted before 13 February 2023.
Peer reviewers are encouraged (but not required) to provide their names to the authors when submitting their peer review. If they agree to provide their name, then their personal profile page will reflect a public acknowledgment that they performed a review (even if the article is rejected). If the article is accepted, then reviewers who provided their name will be associated with the article itself.
I am pleased to inform you that your work has now been accepted for publication in PeerJ Computer Science.
Please be advised that you cannot add or remove authors or references post-acceptance, regardless of the reviewers' request(s).
Thank you for submitting your work to this journal. I look forward to your continued contributions on behalf of the Editors of PeerJ Computer Science.
With kind regards,
[# PeerJ Staff Note - this decision was reviewed and approved by Sedat Akleylek, a PeerJ Section Editor covering this Section #]
**PeerJ Staff Note:** Although the Academic and Section Editors are happy to accept your article as being scientifically sound, a final check of the manuscript shows that it would benefit from further English editing. Therefore, please identify necessary edits and address these while in proof stage.
The paper is clear and easy to understand. However, there are a few typos in the revised revision that need to be fixed, e.g. "EEthereum" at the beginning of the second paragraph. Figure 6 and Table 4 both show the results of the abalation study, whose different ways of presentation makes it difficult to compare the numbers in them. Combining them together in a unified table can be more clear.
no comment
no comment
All concerns raised by the reviewers have been satisfactorily addressed; however, the paper still requires further proofreading by a native English speaker.
Additionally, some figures are unclear, and further statistical analysis is needed.
These issues require a minor revision. If you are willing to make the necessary changes, I would be happy to reconsider my decision. Please submit a list of changes or a rebuttal addressing each point raised when you submit your revised manuscript.
**Language Note:** The Academic Editor has identified that the English language must be improved. PeerJ can provide language editing services - please contact us at [email protected] for pricing (be sure to provide your manuscript number and title). Alternatively, you should make your own arrangements to improve the language quality and provide details in your response letter. – PeerJ Staff
This paper's writing is generally easy to understsand. However, there are some places that need improvement on grammar and clarity. For example, line 113-114 (Wang learned embeddings...) seems gramatically weird (it should be Wang et al's work that learns embeddings). Also, Figure 4 is confusing, as the paragraph of line 481 states that two partial datasets of 30% and 60% are used respectively, but the texts or the caption do not show which dataset's results are presented in Figure 4.
The ablation study measures the contribution of each feature by only enabling one of them at a time for comparison. However, this approach is less convincing than the typical approach to ablation study, i.e., studying the impact of removing one or more parts of the model. An experiment that measures the performance of the models with each of the proposed features removed is more reliable to prove that each feature has overall positive impact.
Figure 4 shows high variance in several bars, especially for GAT. This needs explanation and statistic analylsis to ensure that the methodology is correct and the conclusion is sound.
**PeerJ Staff Note:** Please ensure that all review, editorial, and staff comments are addressed in a response letter and that any edits or clarifications mentioned in the letter are also inserted into the revised manuscript where appropriate.
**Language Note:** The review process has identified that the English language must be improved. PeerJ can provide language editing services - please contact us at [email protected] for pricing (be sure to provide your manuscript number and title). Alternatively, you should make your own arrangements to improve the language quality and provide details in your response letter. – PeerJ Staff
1. English is understandable but verbose; the Abstract, Conclusion, and Sections 3.1–3.3 contain long, multi-clause sentences and some repetition. Please tighten wording, split long sentences, and keep verb tenses consistent.
2. Figures 1–3 lack legends explaining node/edge colours; Tables 2–4 do not mark best / second-best values. Add full captions, unify font size, and ensure equation numbering is global and continuous.
3. Related Work stops at early 2024. Include key 2024–2025 papers on graph contrastive learning and blockchain fraud detection to strengthen context and motive.
4. The current GitHub repo is empty or missing a README and dependency list. Upload full code, scripts for random subsets, and tag the release (or mint a DOI) to meet the journal’s raw-data policy.
1. R-NTN is trained on author-generated 2-hop subgraphs, while most baselines use the original graph; this skews results. Re-evaluate all methods on identical graph structures and train/test splits, or show that each model benefits equally from the subgraph strategy.
2. Results are from a single random split. Run at least five different seeds or K-fold cross-validation and report mean ± SD.
3. Bias walk parameters (τ, T, L, etc.) are given without justification. Provide selection rationale and a sensitivity plot.
R-NTN looks stronger than the baselines, but the evidence is not yet solid. Results come from one random split and a special subgraph that only R-NTN uses, so the gain may be due to the data setup. No p-values or confidence intervals are given, and the code plus subset-generation scripts are still missing, making it hard for others to repeat the test. Until these points are fixed, the findings remain suggestive rather than conclusive.
Beyond the core issues, a few practical tweaks would greatly lift the paper’s utility: add clear legends to all figures and highlight top values in tables, release a fully working GitHub repo (code, environment file, and random-subset scripts), and include a short case study that visualises one true-positive and one false-positive account to show how each feature group drives the prediction. These additions will make the work easier to read, reproduce, and apply in real-world settings.
-
-
-
1. The level of English used appears to be appropriate, but it is wise to consider the perspectives of other reviewers or a native speaker. Nevertheless, there are punctuation issues or words that need to be addressed, including:
• Line 15: Write complete words for R-NTN.
• Line 148: Replace “.however” with “. However”
• Lines 291-294: Rather than commencing a sentence with “n”, one should write 'term n'. Other similar cases in this paragraph should also be corrected.
• Line 415: Delete the double space that comes before “The”.
• Line 501: Add a space before “It”.
• Line 502: Delete the double space that comes before “This”.
• Line 510: Delete the double space that comes before “The”.
• Line 522: Delete the double space that comes before “The”.
• Kindly review the entire manuscript for similar instances.
2. The arrangement of Section 1, known as 'Related Work', is effective; nonetheless, it would improve the quality if the references in Sections 1.1 and 1.2 were updated to reflect more recent information. Moreover, not all references, tables, and figures are clickable. To facilitate access, please utilize hyperlinks.
3. The manuscript's complete structure seems to offer significant advantages. Specifically, the authors have emphasized three sub-structures that focus on node behavioral features, graph-based transaction features, and graph-based network features.
4. In my view, the explanation of your anomaly detection strategy in this article is insufficient. It is essential to start with a definition of what constitutes an anomaly and the techniques for detecting them. You should also clarify the specific type of anomaly you intend to identify, such as point, collective, or conditional anomalies. Additionally, please address whether the dataset included labels for anomalies and describe the methodology you employed to label some instances as anomalies. I understand that you have marked 1154 nodes (line 401) as anomalies.
5. A further critical consideration is the challenge of imbalanced data. I am unclear about how you have managed this issue. The techniques for addressing imbalanced datasets are apparent, including data-level, algorithmic-level, and hybrid solutions. Which strategy did you choose in this case? If you have implemented a strategy that considers the impact of each feature in the domain, please describe your approach in detail, step by step.
6. Lines 72-73: Was your motivation for introducing your proposed model as a robust model solely based on overcoming the imbalanced data issue? Utilizing a machine learning model to forecast security vulnerabilities in Ethereum can significantly improve efficiency over traditional vulnerability detection methods.
7. My assessment of other elements, including the quality of figures, tables, datasets, code, methods, and the research gap, is positive.
This article, in my opinion, remains unacceptable until items 4-6 are thoroughly clarified.
All text and materials provided via this peer-review history page are made available under a Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.