A Machine-Learning-Inspired Opinion Extraction Mechanism for Classifying Customer Reviews on Social Media

Round 1

Reviewer 1 Report

First of all, I would like to thank you for the possibility of reviewing this interesting paper that I have read with great interest.

The paper may have a clear interest associated to researchers from different scientific disciplines and, therefore, could have a notable repercussion in specialized scientific literature.

Why is this study necessary? should make clear arguments to explain what the originality and value of the proposed model is. This should be stated in the final paragraphs of introduction and conclusion sections.

Please review this to paragraphs:

he task of image classification requires a robust and distinctive feature set 63
that remains invariant in the presence of variations in view angle, scale, and illumination 64
[21 ]. The characters within the image have various shapes and textures, which can be 65
described using texture and shape descriptors [21]. Intra-class variations in color, shape, 66
and illumination of the image make image identification quite challenging. Therefore, we 67
require stable attributes in the image that remain invariant despite changes in orientation, 68
scale, and illumination of the image objects. Various types of neural networks have been 69
developed for the classification of the ImagNet dataset. The most popular 2D DNN 70
models include AlexNet, ResNet, DagNet, Vgg-16, InceptionV3, GoogleNet, and Yolo 71
[22].

 The iterative 195
process continues until the difference between the centroid in the ith and i − 1 iteration 196
reaches a certain threshold or limit. The cluster centroids for each class describe the 197
vocabulary set of the dataset. The final histogram is the frequencies of occurrence of each 198
vocabulary centroid in the raw feature data representing the final feature vector.

I would like to suggest the following references:

Fülöp, M. T., Topor, D. I., Căpușneanu, S., Ionescu, C. A., & Akram, U. (2023). Utilitarian and Hedonic Motivation in E-Commerce Online Purchasing Intentions. Eastern European Economics, 1-23.

Akram, U., Fülöp, M. T., Tiron-Tudor, A., Topor, D. I., & Căpușneanu, S. (2021). Impact of digitalization on customers’ well-being in the pandemic period: Challenges and opportunities for the retail industry. International Journal of Environmental Research and Public Health, 18(14), 7533.

Conclusions: pleas add theoretical, managerial, and practical implications, limitation and further research. Some parts are included but must be extended.

However, I hope that all these comments will serve the author to improve the quality of the paper. Finally, I hope that the comments will be understood positively by the authors of this interesting paper.

Good luck!

Author Response

Thank you very much for your encouraging comment. We have answered, edited, and adjusted our manuscript according to your queries and suggestions. The response to reviewer comments is attache as .docx documents here:

Author Response File: Author Response.docx

Reviewer 2 Report

The combination and coordination between BoF and DNN seem to be successful, which can be seen from your experimental results. But only one dataset, AliExpress, is involved for performance evaluation, which is insufficient to give us a general instruction whether the proposed framewwork behaviors well in other datasets or scenarios. For reviews on social media, which are generally very short, does the feature extraction mechanism proposed in this paper still work? Have the authors estimated the average length of 3512 reviews in AliExpress?

Meanwhile, the manuscript has a lot of syntactic or other errors, including:

Why does the paper starts from "2. Introduction" instead of "1. Introduction"?

Page1, line 26, in the sentence "classification performance R3", what does R3 mean?

Page3, line 94, in the sentence "from the set of true and predicated labels", please make sure that predicated is not predicted.

Page3, line 96, in the sentence "Existing customer reviews classification approaches suffer due to the complex nature of human language" is not complete, suffer what?

Page4, line 188, in the sentence "The Bof clusters the keywords in k different sets to generate" of the last paragraph, the Bof should be BoF. Does the description of this paragraph actually corresponds to the k-means algorithm? Why not describe briefly by specifying the k-means?

Page5, line 194, in the sentence "The χ in Eq. 1 denote the ith keyword belonging to cluster j", does χ should be Xi? And Eq. 1 should be Eq.(2)?

Page5, line 197, "The cluster centroids for each class" should be "The cluster centroid for each class". 

Page6, line 211, "drop out" should be "dropout".

Page6, line 216, "vanish" should be "vanishes".

Page6, line 217, "ReLu" should be "ReLU" (also in line 219, line 223), "convert" should be "converts".

Page6, line 218, "preserve" should be "preserves".

Page7, the tense is inconsistent. For example, "The framework is trained on a labeled data" in line 237 uses the simple present tense, while "The performance of the proposed model was assessed" in line 246 uses the past tense.

•  

Author Response

Thank you very much for your encouraging comment. We have answered, edited, and adjusted our manuscript according to your queries and suggestions. The response to reviewer comments is attache as .docx documents here:

Author Response File: Author Response.docx

Reviewer 3 Report

In this paper, the author proposed a novel machine learning-based framework for categorizing customer reviews that uses a bag-of-feature approach for feature extraction and a hybrid DNN framework for robust classification. The performance of proposed framework is assessed by using AliExpress product review data provided by customers. Experimental results show a classification 10 accuracy of 91.5% with only 8.46% fallout is achieved. There are following suggestions to be incorporated in the manuscript

1. Analysis the results shown in Table 1, in depth. Why the proposed model's performance is better than the other models?

2. Also rewrite the conclusion section by including the comparison results with the existing models.

3. Also show the results by changing the size of training, testing and validation data. 

English is OK.

Author Response

Thank you very much for your encouraging comment. We have answered, edited, and adjusted our manuscript according to your queries and suggestions. The response to reviewer comments is attache as .docx documents here:

Author Response File: Author Response.docx

Reviewer 4 Report

Contribution can be written specifically in bullet points.

Practical implications of this work should be written in introduction.

Authors have presented related work for machine learning based methods and deep learning based methods individually in different paragraphs. It is advised to give title to these paragraph to identify easily and improve readability.

Equations 13 to 14, full form of abbreviations needs to write.

In line 247, “Insta360 x3” what is this? It seems some typos.

Caption of figure 3 has typos. Please correct.

Discussion of the results is missing. It should be added.

Conclusion can be improved and add future work.

In the proposed work, figure 2, why 1D CNN layer is used? As it is text, it can be given to LSTM. What is the significance to add 1D CNN?

Software and hardware details needs to be mentioned.

Hyper-parameter values need to be mentioned. Like learning rate, epochs, batch size etc…

Authors have presented result by taking different algorithms and shown the proposed approach is performing best. However, comparisons with SOTA methods is required which show how better your approach compared to others.

English is ok. spell check and normal editing is required.

Author Response

Thank you very much for your encouraging comment. We have answered, edited, and adjusted our manuscript according to your queries and suggestions. The response to reviewer comments is attache as .docx documents here:

Author Response File: Author Response.docx

Round 2

Reviewer 3 Report

The paper has been improved significantly and is acceptable in current form.

Author Response

Hi,

Thanks for the valueable feedback, i am thankfull for reviewing my manuscript by providing such valueable comments.

Author Response File: Author Response.pdf

Reviewer 4 Report

Authors have addressed the given comments.

Future work should be added.

English is fine.

Author Response

Hi, Respected reviewer,

 

I have address the comments here, and added the ROC curves. I am thankfull to you for providing valueable comments to improve the quality of our manuscript.

Author Response File: Author Response.pdf