The Detection of Spurious Correlations in Public Bidding and Contract Descriptions Using Explainable Artificial Intelligence and Unsupervised Learning
, Raimundo Santos Moura 2, VinÃcius Ponte Machado 2, Anselmo Paiva 3, Weslley Lima 2 and Rodrigo Veras 2
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsSee attachment for my review
Comments for author File: 
Comments.pdf
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsRelated work
Kindly provide a brief explanation of the concepts below.
1. Counterfactual Generation (line 92)
2. Data Perturbation (lines 96 -97)
3. XAI techniques LIME and SHAP (lines 100 - 102). Also, write out the full meaning of LIME and SHAP before abbreviating.
4. Other approaches – statistical reweighting and regularisation methods (lines 106 - 108).
Section 2 - There is need to critically review existing literature/methodologies (not just listed). This section requires more work.
Theoretical framework
There is no clear justification provided for selection of algorithms. Why did the study adopt logistic regression, support vector machine and BERT?
Why was TF-IDF used for encoding the text? Why not BERT + logistic regression as an example?
How have you utilised the BERT model? For word embedding and classification layer? This area needs clarification.
In line 135, Kindly provide adequate definition of unsupervised learning approach.
Again, why was K-means clustering chosen?
Methodology
In lines 172 -174, kindly correct the repetition.
Kindly present the experimental set-up of the logistic regression, support vector machines and BERT in a Table.
Results
Kindly present the results in lines 325 – 340 in a Table.
In lines 340 – 350, the logistic regression and SVM-SGD results are unsurprising as the experiment utilised conventional TF-IDF for encoding. To some extent the approach uses term frequency. Authors need to perform sophisticated experiments like Word2Vec + logistic regression, RoBERTa + logistic regression as an example. This will strengthen arguments and findings of your study.
Conclusion:
Kindly add a subsection to discuss the theoretical and practical implications of your research output.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 2 Report
Comments and Suggestions for AuthorsThe study requires improvement in the experimentation and presentation. There is need to make clear what the authors declare as Word Embedding (WE). Which type of word embedding is this? How has this been deployed? Similarly, how was BERT used for classification? There is no clear novelty in this work.
Comments on the Quality of English LanguageKindly improve on writing.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 3
Reviewer 2 Report
Comments and Suggestions for AuthorsWell done on the improvements made.
