Multiscale Feature Fusion and Graph Convolutional Network for Detecting Ethereum Phishing Scams
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThe paper is well written and has a good narrative. The effort in making the paper understandable for a general Computer Science audience is evident. The idea is novel and topical. The authors show significant results to support their design. Overall, good paper.
Author Response
The paper is well written and has a good narrative. The effort in making the paper understandable for a general Computer Science audience is evident. The idea is novel and topical. The authors show significant results to support their design. Overall, good paper.
Response： Thank you for your review.
Reviewer 2 Report
Comments and Suggestions for AuthorsThis paper proposes using GNN+GRU with time trading features in scam detection in Ethereum. The authors claim GNN is suited for analysis of time sequence because it extracts the topological information in the target graph.
The proposed method is tested and compared with other typical ML methods, and showed better performance. Finally, the ablation study is done, which indicated that adding the time trading features improves performance.
The structure of this paper is good, and comparison is exhaustive. It looks that the ablation study is rather negative, but it often happens when proposing a new network.
I believe this paper is worth publishing as is (of course, please correct typos and grammatical errors. e.g. Fig 10 statisic --> statistic)
Comments on the Quality of English Languagenone
Author Response
This paper proposes using GNN+GRU with time trading features in scam detection in Ethereum. The authors claim GNN is suited for analysis of time sequence because it extracts the topological information in the target graph.
Response： Thank you for your review.
The proposed method is tested and compared with other typical ML methods, and showed better performance. Finally, the ablation study is done, which indicated that adding the time trading features improves performance.
Response： Thank you for your review.
The structure of this paper is good, and comparison is exhaustive. It looks that the ablation study is rather negative, but it often happens when proposing a new network.
Response： Thank you for your review.
I believe this paper is worth publishing as is (of course, please correct typos and grammatical errors. e.g. Fig 10 statisic --> statistic)
Response：The typo in Figure 10 has been corrected.
Reviewer 3 Report
Comments and Suggestions for AuthorsThe authors of this paper propose a multi-scale feature fusion and graph convolutional network for detecting ethereum phishing scams. The manuscript is quite well structured and written, and fits the scope of the journal. However, I have some suggestions to improve its overall quality. They are listed below:
- Within the abstract, I would suggest to highlight better the research gap you aim to fill, then the innovative contribution of your paper.
- In the introduction, I would suggest to describe a little bit the success of the blockchain technology in several sectors, by citing the following literature review papers: manufacturing [R1], agri-food [R2], healthcare [R3], energy [R4]. This will improve the trust by the reader, considering that this technology is quite new.
[R1] Leng et al. (2020). Blockchain-empowered sustainable manufacturing and product lifecycle management in industry 4.0: A survey. Renewable and sustainable energy reviews, 132, 110112.
[R2] Mirabelli et al. (2021). Blockchain-based solutions for agri-food supply chains: A survey. International Journal of Simulation and Process Modelling, 17(1), 1-15.
[R3] Hasselgren et al. (2020). Blockchain in healthcare and health sciences—A scoping review. International Journal of Medical Informatics, 134, 104040.
[R4] Andoni et al. (2019). Blockchain technology in the energy sector: A systematic review of challenges and opportunities. Renewable and sustainable energy reviews, 100, 143-174.
- At the end of the introduction, I would suggest to introduce the rest of the paper. I mean "Section 2 shows...", "Section 3 is about...", "In Section 4, we do...", etc.
- Please, add more details about Figure 1. I would like to understand better the different blocks and arrows.
- In case equations in 3.2 are derived from the scientific literature, add the related references. Vice versa, please specify clearly that they a novelty in your manuscript.
- Please, describe the reasons behind the choice to use the dataset by Lin et al. [25].
- In the conclusions, the limitations of your study should be better highlighted.
Author Response
- Within the abstract, I would suggest to highlight better the research gap you aim to fill, then the innovative contribution of your paper.
Response：Thank you for your suggestion, the abstract has been revised.
- In the introduction, I would suggest to describe a little bit the success of the blockchain technology in several sectors, by citing the following literature review papers: manufacturing [R1], agri-food [R2], healthcare [R3], energy [R4]. This will improve the trust by the reader, considering that this technology is quite new.
[R1] Leng et al. (2020). Blockchain-empowered sustainable manufacturing and product lifecycle management in industry 4.0: A survey. Renewable and sustainable energy reviews, 132, 110112.
[R2] Mirabelli et al. (2021). Blockchain-based solutions for agri-food supply chains: A survey. International Journal of Simulation and Process Modelling, 17(1), 1-15.
[R3] Hasselgren et al. (2020). Blockchain in healthcare and health sciences—A scoping review. International Journal of Medical Informatics, 134, 104040.
[R4] Andoni et al. (2019). Blockchain technology in the energy sector: A systematic review of challenges and opportunities. Renewable and sustainable energy reviews, 100, 143-174.
Response：Thank you for your suggestion. The above references have been added to the introduction of the article.
- At the end of the introduction, I would suggest to introduce the rest of the paper. I mean "Section 2 shows...", "Section 3 is about...", "In Section 4, we do...", etc.
Response：Thanks for the suggestion, I have added an overview of the rest of the paper at the end of the Introduction section.
- Please, add more details about Figure 1. I would like to understand better the different blocks and arrows.
Response：Thanks for your suggestion, more details were added in the introductory Figure 1 section of Section 3.
- In case equations in 3.2 are derived from the scientific literature, add the related references. Vice versa, please specify clearly that they a novelty in your manuscript.
Response：Thank you for your suggestion. Formula (1) and formula (6) in 3.2 are our own definitions and have been marked in the article. Other formulas come from reference [31] and have also been marked in the text.
- Please, describe the reasons behind the choice to use the dataset by Lin et al. [25].
Response：Because the amount of data in Ethereum is very large, the resulting transaction graph is very complex. Therefore, it is very important to use an effective subgraph sampling method. After reviewing the relevant literature in this field, we found that most of the subgraphs are generated using random walk methods. The subgraphs obtained in this way are not representative and may affect the final result. test results. In the process of reviewing the literature, we found that Lin et al. [25] used a K-order subgraph sampling method to obtain the local structure of the target account. K-order sampling generates a directed K-order subgraph centered on each target account. Considering the real Ethereum phishing detection scenario, there may be illegal transactions in the node before the target node and the next three nodes. The data set obtained by this method can more effectively reflect the characteristic information of the target node while reducing the transaction graph, so we choose this data set.
Due to changes in the reference, Lin et al. [25] has been changed to Lin et al. [29].
- In the conclusions, the limitations of your study should be better highlighted.
Response：Thank you for your suggestion, we have modified it in the conclusion.
