A Convolution Neural Network-Based Representative Spatio-Temporal Documents Classification for Big Text Data

Round 1

Reviewer 1 Report

There are massive text message data generated by social medial users, most of them are related to spatio-temporal information. This paper defined the Representative Spatio-Temporal Documents (RepSTDoc) and proposed a CNN-based representative spatio-temporal document classification model. The methods and case studies presented in this paper show that it is a meaningful solution on big text data classification.

 

Although the paper defined the RepSTDoc, I cannot know the details of the RepSTDoc, and what is the difference between the RepSTDoc and the common documents? And how does the difference help increase the classification accuracy ratio?  Please define RepSTDoc in a formal and present an example to show how its internal information is organized, especially spatiotemporal information.

 

Please list the methods for text classification in a table.

 

The result should include the time efficiency comparison of the algorithms.

 

What is the character-level? And please describe how it work in your model in details.

 

The abstract is recommended to rewrite.  

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

1. Why the classifying representative spatio-temporal documents is difficult­­? Expand it.
2. How much longer is the RepSTDoc_ConvNet method compared to the existing ones?
3. In what areas is your method applicable?
4. Expand the literature review, as many articles appear on the Convolution Neuron Network and their scope is very different.
5. Make a comparison with other models, and show it in Fig. 3, 4
6. Figure 4 RepSTDoc_ConvNet batch size larger than 64 does not exist?
7. Update the literature, you write about the CNN model, and the literature of 2010, 2014

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

Dear Authors,

Proposing a character-level CNN based document classifier to classify representative spatio-temporal document; the idea lacks in terms of novelty. I suggest authors to review the latest work done by other researchers and come up with something new and better. I believe, readers are interested in a new ideas or anything that improves current problems.

Also to train the proposed CNN model, 3,500 training data were used; the data size is small and seems to create an overfitting module.

It would also be better if author are clear on how parameters for the CNN were defined. What was considered? and how authors came up with this model and parameters? Please include such details.

I suggest authors to compare with the work done by other researchers.

Also did authors try varying the % of training, validation, and test data and see that had impact on accuracy?

 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

The authors took the recommendations of the reviewers into account and improved the paper.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

Other comments have been addressed properly and I am ok with it; except for the overfitting module. I still believe the model is overfitted.

Author Response

Please see the attachment.

Author Response File: Author Response.docx