Research on Short Video Hotspot Classification Based on LDA Feature Fusion and Improved BiLSTM
Round 1
Reviewer 1 Report
Review of Applied Sciences Manuscript ID: 2028270
Research on Short Video Hotspot Classification Based on LDA Feature Fusion and Improved BiLSTM
This reviewer finds the paper to be aimed at solving the important problem of classification of short video hotspots. This reviewer recommends that the paper be revised by the authors to address the couple of concerns of this reviewer.
Strengths:
1. The authors propose details of a method that uses LDA and other ideas from topic modeling, self-supervised word vector representations, and BiLSTM with some improvements to perform classification.
2. Adequate literature review has been performed.
3. The paper contains good diagrams for explainability.
4. The details of the proposed method has been provided.
Comments and Concerns:
1. One major concern this reviewer has with the paper is that they provide no comparison with other methods in the literature. The comparison they provide can be considered more of an ablation study where they add/remove different components but they don’t compare to other methods for video classification from literature.
2. Are the authors planning to release the dataset and code?
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Reviewer 2 Report
Authors present some significant work related to Short Video Hotspot Classification Based on LDA Feature Fusion and Improved BiLSTM. After reading manuscript, some technical queries need to be address by authors and make necessary changes in manuscript :
1.Any specific reason why authors have chosen LDA only for classification?
2. There are certain shortcomings of LDA like " LDA-based models are criticized for commonly neglecting co-occurrence relations" .Do it affects the research defination of authors and how it is mitigated.
3. Authors mentioned that the proposed LBSA model reached 91.52% precision which i personally believe is very less ? It should be more.Kindly address with suitable justification.Author may include recently published literature.
4. Further it is not clearly mentioned results are obtained through training or testing.Further training and testing results are not reliable. 10-fold cross validation technique is a reliable technique which gives unbiased prediction results.Refer recently published literature in which authors have applied 10-fold cross validation that too on fusion feature vector.
a. http://nopr.niscpr.res.in/handle/123456789/55680
b. https://ieeexplore.ieee.org/document/8359052
Authors are advised to include suitable justification with inclusion of literature.
5. Quality of figures need to be improved.
6. How the effect of noise will be eliminated as a pre-processing stage. There are higher chances that it will affect the prediction results.
Author Response
Please see the attachment
Author Response File: Author Response.docx
Round 2
Reviewer 1 Report
This reviewer has carefully reviewed the changes the authors have made to the paper. This reviewer strongly recommends adding the reference numbers to the table 4 and 5 associated with the models they are comparing to. That way, the authors of other models can see an honest comparison of their earlier work with this work. Once the authors make the change, this paper can be accepted for publication.
Author Response
Please see the attachment
Author Response File: Author Response.docx
Reviewer 2 Report
Authors have modified manuscript as per suggestion.However, suggested references not included in revised manuscript as per author comments in point 4.
Author may include suggested references during proofreading
Author Response
Please see the attachment
Author Response File: Author Response.docx