Attention-Based 1D CNN-BiLSTM Hybrid Model Enhanced with FastText Word Embedding for Korean Voice Phishing Detectionâ€
Round 1
Reviewer 1 Report
The authors have adopted a hybrid artificial neural network architecture that is composed of three layers:
· 1-dimensional Convolutional Neural 3 Network (CNN) layer
· Bidirectional Long Short-Term Memory (BiLSTM) layer
· Hierarchical 4 Attention Networks (HANs)
However the following suggestions need to add in the manuscript:
· Authors must explain unique contribution to the adopted application
· Why authors concentrated on a Convolutional Neural Network (CNN), a Bidirectional Long Short-term Memory 66 (BiLSTM) architecture, and a Hierarchical Attention Network (HAN) for detecting 67 Korean voice phishing attacks. Of course, there are many recent models have been applied on Korean Call Content Vishing (KorCCVi) dataset.
· The abstract should be improved. The contributions of the work should be reflected in the abstract in a better way.
· Authors need to justify why 1-dimensional Convolutional Neural 3 Network (CNN) layer was adopted in the proposed model
· The authors must include the recent publications contributions in the survey section
· Authors need to include table of survey with clear points which are base for the manuscript.
· Since, authors have not implemented and small size dataset is considered. Hence, there is a lot of scope to further to tune the results. Therefore, authors are advised to exhibit the results with large and real dataset.
· In the introduction it is necessary in a paragraph briefly talk about the importance of DL, ML and Classification models.
· The given model for phishing classification, how many classes has? It is necessary to show some ground truth voce or text of each class.
· Experimental result: the experimental results are good. However, there are not enough contrast experiments to show the special advantages of the proposed method. Please add at least two more existing methods.
· Equation and mathematic quality: Some mathematical symbols are not professional.
· The figures in the results section can be plotted with different means for better clarity of concepts.
· Overall English grammar need to checked
· The discussion part is not adequate. It should be enhanced especially on the validation of results.
· The comparison in discussion section is not explained properly. The basis of the comparison is not clear. Whether the same dataset was adopted for other models?
Overall English grammar need to checked
Author Response
Dear Reviewer,
Thank you for your time and detailed review of our work. We appreciate the thoughtful comments and insights you have provided, which have helped us broaden our perspective and strengthen the quality of our paper. Please, see the attachment, where we provide a point-by-point response to your comments.
We hope all the modifications in our revised manuscript have addressed your concerns effectively, and we look forward to any additional feedback you may have.
Best regards.
Author Response File: Author Response.pdf
Reviewer 2 Report
The paper addresses an attention-Based 1D CNN-BiLSTM hybrid model enhanced with FastText word embedding for Korean voice phishing detection.
1- The abstract should be modified. There is not enough information about methodology, proposed work, conclusion in this part. I suggest you structure your abstract as presented in:
https://www.principiae.be/pdfs/UGent-X-003-slideshow.pdf
2- Figure 1 has some datils where had not been explained well in the text of the article. I recommend explaining well each section of this figure.
3- Simulation conditions are not well discussed. The proposed approach was illustrated only on some specific simulations, which is not enough to draw a complete and accurate conclusion about the proposed approach.
4- This method should be compared with more famous methods to determine the superiority of the proposed method. The evaluations in not enough.
5- Please, do not forget that the clarity and the good structure of an article are important factors in the review decision. Please read the paper carefully (again) and correct it in English.
Minor editing of English language required
Author Response
Dear Reviewer,
Thank you for your time and detailed review of our work. We appreciate the thoughtful comments and insights you have provided, which have helped us broaden our perspective and strengthen the quality of our paper. Please, see the attachment, where we provide a point-by-point response to your comments.
We hope all the modifications in our revised manuscript have addressed your concerns effectively, and we look forward to any additional feedback you may have.
Best regards.
Author Response File: Author Response.pdf
Reviewer 3 Report
Dear Authors,
Thank you for letting me know about your research.
The authors of “Attention-Based 1D CNN-BiLSTM Hybrid Model enhanced with FastText Word Embedding for Korean Voice Phishing Detection” extend a work which they have previously presented at KCC 2022 (reference number 6 in the list of references). In the current work, they build an artificial neural network (ANN) model by combining deep learning (DL) techniques that are typically used in the detection of phishing attacks, with a convolutional neural network (CNN) and a bidirectional long-short term memory (BiLSTM) architecture. They train and evaluate the prediction performance of their model on the latest version of the Korean Call Content Vishing (KorCCVi) dataset.
Your work is of scientific interest and I have no objections accepting its publication in the MDPI-Mathematics, in particular as it is a good example of the recently proposed approach of combining both data-centric and model-centric AI, which provides a more beneficial way to have a compact and robust solutions. Following are minor issues to be resolved:
Minor Issues:
1. You proposed a hybrid ANN model that combines several widely accepted techniques, in order to resolve the phishing attack problem. In light of the ongoing debate surrounding whether AI research should shift from a model-centric approach to a data-centric approach, your work successfully combines both data-centric and the model-centric AI methodologies, confirming the complementary nature of optimal AI solutions. In particular, the work reveals how the smart use of existing approaches, in the face of relatively small datasets (Section 4.1 “Dataset Details”), still leads to a smart solution. This is in full compliance with the idea that was recently proposed as a way to benefit from the advantages of both data-centric AI and model-centric AI. You need to highlight this additional positive point as a further contribution within the “Introduction” section, somewhere on page 2, lines 62 to 73. Here is a suggestion of how to do this as one of the contribution points:
“In light of the ongoing debate surrounding the selection of data-centric AI versus model-centric AI and taking into account the limited availability of datasets in case of Korea, our research highlights the effectiveness of adopting a complementary approach that combines both data-centric and model-centric AI methodologies (Hamid, 2023).”
Hamid, O. H. (2023). Data-Centric and Model-Centric AI: Twin Drivers of Compact and Robust Industry 4.0 Solutions. Applied Sciences, 13(5), 2753.
2. In some positions along the manuscripts, there are statements that require more consideration. For instance, in the “Related Work” Section, on page 2, lines 84 to 85, you write: “A review of the literature reveals that AI is the most commonly employed approach for detecting various types of phishing attacks.” This suggests, there is a predominant trend, or even consensus among researchers in the field of cybersecurity, that AI is a highly effective technique for detecting phishing attacks. But it leaves the readers curious about what other approaches might be there. Therefore, I suggest that you reformulate this sentence by referring to the other approaches concretely. For instance, you may write:
“Among various techniques including rule-based and reputation-based approaches, content analysis, user awareness, and the use of anti-phishing toolbars and browser extension, AI is commonly the most employed approach for detecting various types of phishing attacks.”
There is something that needs reformulation: Table 5 (page 12) compares the performance of the proposed hybrid model with other base line. The results clearly show that some other models have higher accuracy and F1 scores. However, you write that your model “outperforms” the other models, for example in the “Abstract section” and in the “Conclusion and Future Work” section. This must be reformulated in order to reflect the results presented in Table 5. You may use expressions like
“Our results show that the performance of our proposed model is comparable to that of other baseline models.”
Author Response
Dear Reviewer,
Thank you for your time and detailed review of our work. We appreciate the thoughtful comments and insights you have provided, which have helped us broaden our perspective and strengthen the quality of our paper. Please, see the attachment, where we provide a point-by-point response to your comments.
We hope all the modifications in our revised manuscript have addressed your concerns effectively, and we look forward to any additional feedback you may have.
Best regards.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Authors have addressed all the suggestions provided during the review
Authors have addressed all the suggestions provided during the review