Discovering Critical Factors in the Content of Crowdfunding Projects

Round 1

Reviewer 1 Report

Dear Authors,

the topic of your research is very interesting.  My comments relate to three issues:

1. in verses 61-62 you wrote:

"In addition, previous research mainly relied on qualitative research based on questionnaires". In my opinion, it is a mistake. Did you mean quantitative? The qualitative methodology is based on interviews and focus groups mostly. A questionnaire is linked to quantitative methodology. 

2. Discussion part- in the paper I couldn't find the discussion part.  How does your research fit in with previous research findings, what are the differences, what are the similarities? 

There are several references to previous studies in the text in the conclusion, but these should be placed in a separate section. The conclusions are a summary, while the discussion should refer to the literature cited earlier/previous studies.

3.The text should also include some of the limitations of the research and directions for future research.

I hope that my comments will help you to improve the article.

My kind regards, 

Reviewer

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

The authors propose to detect the critical factors in crowdfunding projects using machine learning, PLN and data preprocessing techniques. The article presents experimental results on two real data sets. However, the paper illustrates various issues that should be addressed or clarified.

1. The authors have used the following words Relief F, Relief-F. However, the correct term is ReliefF.

2. To comply with reproducibility, the authors should add the parameters used in the algorithms; for example, the type of kernel used in SVM, Pruning in DT, etc. All these data do not appear in the article; therefore, it will be difficult for readers to replicate your experiments using the same parameters and/or conditions.

3. Why is there no description of ReliefF in section 2.3? It should exist just like Decision Tree did.

4. Using a decision tree as a feature selection tool is very debatable since there are methods mainly focused on this problem. So why didn't they use other forms of feature selection?

5. The results may be biased and therefore, the conclusions may not be valid. This is because the authors have yet to consider the problem of unbalanced classes, which has been shown to affect various classifiers and feature selection algorithms. So, the results obtained could be strongly influenced by the bias present in the data. The very low results in the evaluation metrics suggest that in addition to the imbalance, there are other complexities that the authors still need to address or analyze. In this sense, the authors should analyze what problems the data they use to have (unbalanced classes, overlapping) and how they affect their results.

5. Determining the most critical features using accuracy when the set is unbalanced is not recommended since you would choose those features that benefit the predominant class. Review and discuss this.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Dear Authors,

Thank you for your revisions. Now all data is clear and the literature/discussion section have a good scientific soundness.

My kind regards and wishes of many successes,

The Reviewer