Hybrid Multichannel-Based Deep Models Using Deep Features for Feature-Oriented Sentiment Analysis
, Tasswar Iqbal 1, Muhammad Attique Khan 2,*, Usman Tariq 3 and Jae-hyuk Cha 4
Round 1
Reviewer 1 Report
The authors need to address the following.
1. The abstract may be rewritten such that the main contribution should be written in first or second sentence.
2. What specific improvements should the authors consider regarding the methodology? What further controls should be considered?
3. The manuscript shows many results on different point of view.
4. References are relevant, referenced correctly, and included appropriate key studies.
5. The data results are relevant and presented clearly. However, it would
be better to provide the figures with high resolution.
Author Response
Please see the attachment.
Author Response File: 
Author Response.docx
Reviewer 2 Report
1. Re-write the abstract to highlight your contribution
2. Some abbreviation need to definition when first time appear in text.
3. I think some error in title of table 1.
4. Some equations need to reference.
5. In page '10' last 3 lines, in equations (9-10) the authors used activation equation is (Leaky ReLU ), and in equation (11) , they used the activation function is Tanh. Why?
6. I suggest the authors to add accuracy column to table 3.
7. The links of databases which used in manuscript must be put in the footer of paper and use the name of this database within text.
8. The authors may consider providing a graph showing the loss/objective as a function of training epoch, to demonstrate that the modl has been trained to convergence.
9. The authors must be explain how to choose the values of the parameters and hyper parameters such as (epoch number and batch size) and all other parameters which affect the performance and accuracy in table 4.
10. The authors may consider providing a graph showing the loss/objective as a function of training epoch, to demonstrate that the neural network has been trained to convergence.
11. The manuscript needs language editing to fix some spelling and grammar errors.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Reviewer 3 Report
After reviewing, the following suggestions are put forward:
(1) Should pay attention to the abbreviation of professional terms.
a. "Aspect Based Sentiment Analysis(ABSA)" in line 125 should appear in line 40.b. Nouns that first appear in the text need to be re-shortened. For example, "GRU" in line 75 should be "gated recurrent unit (GRU)". In addition, there are "NLP," "CNN," and "PoS." Please check and correct the full text.
(2) The title of Section 3.1.3 is the same as that of Section 3.1.4, which is "Attention layer". Please correct.
(3) In Section 3.1.4, "This section may be divided by subheadings. It should provide a concise and accurate description of the experimental results, their interpretation, and the experimental conclusions that can be drawn. "This content does not match. Please fix it.
Sections 5.3 and 5.4 also appear "This section may be divided by subheadings. It should provide a concise and accurate description of the experimental results, their interpretation, and the experimental conclusions that can be drawn." This comment statement can be omitted, please check the full text and correct.
(4) This paper integrates Att-MC-BiGRU and Att-MC-CNN together, and points out that the novelty of this method is that the two models share hidden layers for transfer learning. However, this article does not elaborate on the technical details of "Sharing hidden layers". Please add corresponding sections in the main text.
(5) This paper establishes Att-JM combining Att-MC-Bi-GRU and Att-MC-CNN to realize emotion classification. It is suggested to supplement the experimental results of a single Att-MC-Bi-GRU model and a single Att-MC-CNN model to further verify the advantages of the parallel fusion algorithm.
(6) This paper emphasizes many times that sharing hidden layer information among parallel combinatorial algorithms is the reason for achieving combinatorial capability benefits. It is suggested to supplement the comparison experiment without shared hidden layer information, so as to ensure the integrity of the experiment and the rationality of the conclusion.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
