Research on Fraud Detection Method Based on Heterogeneous Graph Representation Learning

Round 1

Reviewer 1 Report

- The introduction section is more like a related works section. There is no introduction to the problem itself or the associated terminology and preliminary information the reader needs (e.g., heterogeneous graphs).

- Section 2 packs a lot of information in this one paragraph. The information the authors try to convey to the reader is densely presented and not clear. Due to the formatting of the paper, it is also difficult to see where section 2.1 starts.

- Section 3.2 is unclear in its contribution. Are some of the features only selected? Just the ones shown in the Tables? Because their captions state otherwise.

- The experimental evaluation is extensive, but there are many issues with it. For instance, what is graphSAGE or SemiGNN? Are these related works? Are they established graphs? Table 3 does not make sense with no further information. Is this a comparison study?

The entire manuscript needs proofreading for typos and mistakes. For instance, there are random words in the middle of sentences that start with capital letters, and some other words at the beginning of a sentence start with small letters.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

The manuscript : RESEARCH ON FRAUD DETECTION METHOD BASED ON

HETEROGENEOUS GRAPH REPRESENTATION LEARNING has a good number of references , as 25 , which are in the state of the art. Authors may consider to include a table that can summarize main findings wirh advantages of the proposed approach versus similar Works. For instance, table 3 shows experimental results of different models, but there is not a relation with the corresponding references. On the side of presentation of the manuscript, the captions of the tables must be placed on the top, not on the bottom, as done in the journal papers of the majority of editorials.

Authors should update the tense of the words in the abstract, as it is, some Works are in past tense that is more appropriate for the conclusión section. For example, the text in the Abstract: Detecting fraudulent users in social networks can reduce the occurrence of online fraud 1

and telecommunication fraud cases, which is important to protect the lives and properties of Internet 2

users and maintain social harmony and stability. We studied how to detect fraudulent users by using 3

heterogeneous graph representation learning, and proposed a heterogeneous graph representation 4

learning algorithm to learn user node embeddings so as to reduce human intervention, and the 5

experimental results showed good results… the words studied and proposed should be updated to study and propose.

Also, authors should pay attention on the type of manuscript they are preparing, for example, se last sentence in the abstract is mentioning a chapter instead of a paper work: This chapter continues to investigate how to better use 6

heterogeneous graph representation learning to detect fraudulent users in social networks and further 7

improve the detection accuracy… in this case, authors may also detail percentages of improvement with respect to the experiments performed in similar Works.

In section 1. Introduction, authors may summarize the main goals of every section, as done in the majority of papers.

In section 2. Materials and Methods 48

In this section, we introduce the similarity-based multi-view GCN [8] fraud detection 49

model [9] proposed in this chapter (Similarity based Multi-view GCN, SMGCN), the model 50

is divided into three modules: the similarity-based single-view graph convolution module, 51

the multi-view fusion module, and the fraud detection module, respectively…. Authors do not emphasize if figure 1 is new ori t has been already introduced by other authors, and again, the Word chapter must be changed by paper. As the model is divided into three modules, authors must discuss if they are updated in their work.

In subsection 2.3 Fraudulent user detection 156

We pass the features of the user nodes obtained from the graph representation learning 157

[23] model into a single-layer network to classify the user nodes in order to distinguish 158

between normal and fraudulent users… authors do not clarify if the procedure has already been detailed in [23], or authors are improving the work given in that reference [23].

From the paragraph in section 3. Results 170

3.1 Dataset 171

To validate the effectiveness of the proposed similarity-based multi-view GCN model 172

for fraud detection task in this chapter, we selected two real datasets, Amazon dataset and 173

MicroblogPCU dataset, with specific statistical information in Table 1… authors must include the respective references of the two dataset, i.e. [24] and [25], and again, the Word chapter must be changed by paper

In line 238, authors enphasize the main contribution as follows: The experimental results are shown in Table 3. From the table, we can see that our 238

heterogeneous graph representation learning model SMGCN outperforms the comparison 239

models in both F1 and AUC when compared with other graph representation learning 240

models, and outperforms the other three comparison models for fraud detection tasks… To appreciate the main contribution, authors should include the references of all methods given in Table 3. Experimental result. In addition, authors can discuss the main advantage of SMGCN with respect to the models listed in that Table 3.

The subsection 3.6 Summary of this chapter 279

This chapter proposes a fraud detection model based on similarity and multi-view 280

graph convolution (SMGCN) the problem of isolated nodes and data category imbalance in 281

the structure of heterogeneous graphs in the field of fraud detection…. It must be changed to a Section labeled as conclusions, and the tense is now in past, so that the firts sentence: This chapter proposes a fraud detection model … must be updated to something like this: This paper proposed a fraud detection model…

Moderate editing of English language required

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Introduction and related works are presented in a better manner with this revision. Please replace all instances of "chapter" with "section".

There is still a need for some proofreading. The syntax seems off in some sentences.

Author Response

We have corrected the errors of expression and tense in this paper.

Author Response File: Author Response.pdf

Reviewer 2 Report

Still you write chapter instead of section. A deep revision must be performed in your whole manuscript

Still you write chapter instead of section. A deep english revision must be performed in your whole manuscript

Author Response

We have corrected the errors of expression and tense in this paper.

Author Response File: Author Response.pdf