RLC-GNN: An Improved Deep Architecture for Spatial-Based Graph Neural Network with Application to Fraud Detection

Round 1

Reviewer 1 Report

This paper describes the novel method on the graph neural network called residual layered CARE-GNN (RLC-GNN).

According to their reports, the performances of their proposed algorithm look good on the experiments using Yelp and Amazon datasets. The paper illustrates good results of recall, AUC, and F1-score from their proposal methods.

The results are interesting and have some value to readers. However, it remains some room to improve this paper. Most of readers are unfamiliar with the concept of GNN. Although Section three exprains the basic concept of GNN, it has only one figure (Figure 1), and it does not clearly describe how the GNN works. The reviewer considers it will be more readable if some figures to illustrate the GNN from its essential idea.

Author Response

Dear reviewer,

      Thank you for your comments on our manuscript “RLC-GNN: An Improved Deep Architecture for Spatial-based Graph Neural Network with Application to Fraud Detection” (ID: applsci-1243084). Those comments are very helpful to revise and improve our paper. And our responses to the comments are as follows:

Point1: The results are interesting and have some value to readers. However, it remains some room to improve this paper. Most of readers are unfamiliar with the concept of GNN. Although Section three explains the basic concept of GNN, it has only one figure (Figure 1), and it does not clearly describe how the GNN works. The reviewer considers it will be more readable if some figures to illustrate the GNN from its essential idea.

Response1: GNN is a young extension branch of classic deep learning area. The fact that most readers are not familiar with this field are indeed not well considered before. Therefore, we give a detailed explanation following the logic below:

1. Introduce the concept of graph, and give a real-world example so that readers can intuitively feel the concept;
2. Explain the limitation of classis deep learning, and why we need graph learning;
3. Lead to the basic encoder-decoder framework for solving graph representation learning problem;
4. Present the general modern GNN framework.

In addition, we also add appropriate quantities of figures to supplement the explanation. All our modification is marked with blue color and you should receive the new manuscript later from dear editor.

Here are our responses to the comments. We sincerely wish for your approval. Thank you again for your good comments.

Best regards

Reviewer 2 Report

This manuscript presented an improved algorithm named Residual Layered CARE-GNN. There are several concerns should be considered.

1. In section 4, this manuscript should present an pseudo-code of the proposed algorithm and briefly describe.
2. This manuscript should compare the experimental results obtained using the proposed method with those obtained using alternative classifiers.

Author Response

Dear reviewer,

      Thank you for your comments on our manuscript “RLC-GNN: An Improved Deep Architecture for Spatial-based Graph Neural Network with Application to Fraud Detection” (ID: applsci-1243084). Those comments are very helpful to revise and improve our paper. And here are out responses to the comments:

Point1: In section 4, this manuscript should present a pseudo-code of the proposed algorithm and briefly describe.

Response1: The fact that no pseudo-code is provided for algorithm is indeed not well considered before. Therefore, after the describing of the mathematical principle, we provide the pseudo-code for RLC-GNN and give a brief description for it in section 4. From the perspective of a node, we show every key step and how the two introduced structures (layered structure and residual structure) work in the algorithm.

Point2: This manuscript should compare the experimental results obtained using the proposed method with those obtained using alternative classifiers.

Response2: For fairness, we should use the public results that are obtained on the same dataset. Therefore, we quote the performance of various GNNs from research [2] and descript how the performance is gradually improved.

All our modification is marked with blue color and you should receive the new manuscript later from dear editor.

Here are our responses to the comments. We sincerely wish for your approval. Thank you again for your good comments.

Best regards