AddAG-AE: Anomaly Detection in Dynamic Attributed Graph Based on Graph Attention Network and LSTM Autoencoder

Round 1

Reviewer 1 Report

1. For readers convenience please collect all notation in a separate table.

2. Page 7, Eqn. (13). Please specify lower bounds for parameters $i$ and $t$ in summing up operations $\sum.$ The same comment relates to Eqn (8) and Eqn (9) on Page 6.

3. Please check all the formulas throughout all the paper, especially, correctness of using indexes. Say, in Eqn (12), $\hat{Z}^{t-1}_t$, meanwhile right on the next line 238  it is already $\hat{Z}^{t-1}_i$

4. For readers convenience please use all the notations in uniform style. Say, in the left side of Eqn (14), $t$ is superscript while in its right side $t$ already is subscript.It might be very confusing for the readers.

5. Line 153. Presumably $\vert A^t\vert $ should be $A^t$ since right on the same line $\vert \cdot\ vert$ is used to define the number of elements.

My main worry for a while is about notations used throughout the paper. In the present form they might be confusing for the readers. Please see comments in the box above.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

•          The whole paper has to be revised.

•          The main contribution is unclear.

• Authors should list the benefits and limitations of the various anomaly detection techniques in a table.

•  Fig. 1 is blurry and unclear.

•   To make it simpler to comprehend the inputs and the output after embedding, authors should include an example of graph embedding.

•          How many LSTM layers are there?

•          It should include a sample of data sets.

•          Measures of precision and recall verifying the effectiveness of the model

•           AUC results for training or testing?

•          A graph illustrating the relationship between the number of epochs and loss is crucial to demonstrating the significance of the proposed model.

•          Authors need to interpret and explain the model results for performance.

the whole document should be revised. 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

I think the revised paper might be published as it is.

Reviewer 2 Report

authors addressed all my comments