Discovering Key Successful Factors of Mobile Advertisements Using Feature Selection Approaches

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The manuscript identifies major factors of mobile advertisements based on SVM-RFE, correlation-based selection, and consistency based selection methods. The performance is also evaluated, analyzed and presented for readers.

Some comments and suggestions are listed in the following.

1. It is believed that machine learning or deep learning algorithms are increasingly employed in the mobile advertising domain. Please introduce the status quo, the limitation or remaining challenges of previous studies. Thus, the innovation or improvement of the proposed method can be highlighted.

2. The SVM-RFE is used to select major factors in the study. It is believed that alternative feature selection approaches are also available. Please elaborate on the reason why SVM-RFE is employed. Furthermore, the advantages of SVM-RFE and limitation can be presented for readers.

3. Is there any new insight identified by SVM-RFE? In other words, if similar result can be obtained by statistical methods, what benefit can be brought by SVM-RFE?

4. In the pre-test phase, the subjects are adolescents (15 to 29 years old). Please elaborate on the reason for selecting this particular age range. Is it significant and representative? Will it cause potential bias to the study?

5. The manuscript states that 1) the average monthly income of samples is primarily less that 20000 NTD; 2) the number of mobile device purchases in a half-year of samples is primarily less than five. Is it significant and representative? Will it cause potential bias to the study?

6. In Section 4.3, the manuscript states that "Table 8 lists the classification performance of SVM by using the original feature set which means without implementing feature selection, and the results of using three feature subsets by three feature selection methods." It seems the performance of three algorithms is similar. Thus, the benefit of using SVM-RFE remains unclear.

7. Please include more recent studies (only one study from 2024 and none from 2023).

Author Response

Dear Reviewer,

Thank you for your comments regarding the submitted paper. The response to reviewer's comment is as attached. The corrections have been made according to your suggestions. All correction parts have been in red color in the manuscript. Please check the revised manuscript.

Author Response File: Author Response.docx

Reviewer 2 Report

Comments and Suggestions for Authors

1. Introduction and Theoretical Grounding

• The introduction lacks a clear research gap and theoretical framework. The discussion on mobile advertising and programmatic buying is repetitive and should be more concise.
• Theoretical constructs such as "preference," "credibility," and "informativeness" should be grounded in existing models (e.g., TAM, TRA, or advertising response models).

2. Methodology Concerns

• The use of SVM-RFE, correlation-based, and consistency-based feature selection is insufficiently justified. These are typically used for high-dimensional numeric datasets (e.g., gene expression), not for Likert-scale survey responses.
• No preprocessing details are given (e.g., how ordinal data was treated), and the questionnaire in the appendix is missing.

3. Sampling and Validity

• The sample is highly skewed (young, Taiwanese respondents, mostly students), limiting the generalizability of the findings.
• Snowball sampling introduces bias, which is not acknowledged or mitigated in the analysis.

4. Presentation of Results

• Tables are overly complex and lack visualization. Consider using bar charts or Venn diagrams to highlight overlapping features selected by different methods.
• The performance gains after feature selection are marginal, and no statistical significance testing was applied to validate improvements.

5. Conclusion and Implications

• While the conclusion summarizes the findings, it lacks depth and fails to discuss practical implications for advertisers or mobile developers.
• No limitations or directions for future research are discussed.

6. Language and Structure

• The manuscript suffers from grammar and style issues, awkward sentence construction, and repetition. A thorough language revision is recommended.
• Some sections are verbose and would benefit from clearer structure and tighter wording.

 

Comments on the Quality of English Language

The manuscript would benefit significantly from thorough English language editing. While the overall meaning is generally understandable, there are numerous issues related to:

• Grammar and syntax errors: Frequent problems with verb tense agreement, article usage ("a"/"the"), and incorrect prepositions.
• Awkward or unclear sentence structure: Many sentences are verbose, redundant, or poorly organized, making them difficult to follow.
• Repetitive phrasing: Several ideas, especially in the Introduction and Literature Review, are repeated with only minor rewording.
• Improper academic tone: Informal phrasing and imprecise word choices detract from the scholarly quality expected in a research article.

A comprehensive proofreading by a native English speaker or a professional scientific editing service is strongly advised to improve clarity, flow, and academic quality.

Author Response

Dear Reviewer,

Thank you for your comments regarding the submitted paper. The response to reviewer's comment is as attached. The corrections have been made according to your suggestions. All correction parts have been in red color in the manuscript. Please check the revised manuscript.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

All the comments and suggestions are carefully addressed in the revised manuscript. The manuscript is in good shape for publication.

Author Response

Comment 1: All the comments and suggestions are carefully addressed in the revised manuscript. The manuscript is in good shape for publication.

Response: Many thanks for your kind assistance and valuable comments.

Reviewer 2 Report

Comments and Suggestions for Authors

Thank you for your careful and thorough revision. Your manuscript has improved substantially, but I offer the following comments and suggestions to help you refine it further:

Major Comments

1. Theoretical Grounding of Constructs

While you’ve added a clearer research gap in the introduction (“academic research on programmatic advertising, especially within mobile contexts, remains limited”), the key constructs—preference, credibility, informativeness, etc.—would benefit from explicit linkage to established theories (e.g., TAM, TRA, or the Elaboration Likelihood Model). A brief paragraph anchoring each construct in its original theoretical model would strengthen the conceptual foundation and increase the theoretical contribution of the paper.

2. Treatment of Ordinal Data

You describe encoding 1–5 Likert-scale items and normalizing them to the [0,1] range for feature selection. However, since ordinal scales do not strictly meet interval assumptions, please clarify whether you tested for approximate interval behavior (e.g., via distributional checks or nonparametric alternatives) or supplemented with techniques like Spearman correlation to ensure robustness.

3. Sample Bias and Mitigation

You appropriately acknowledge the demographic skew toward 15–29-year-old Taiwanese respondents and the use of snowball sampling. To strengthen this section, consider reporting any additional checks (e.g., early vs. late respondent comparisons or subgroup factor loadings) to assess homogeneity of variance or detect potential response bias within the sample.

4. Effect Size Reporting for Statistical Tests

The t-tests showing no significant difference (p > 0.05) between models using reduced vs. full feature sets are informative. However, reporting standardized effect sizes such as Cohen’s d for both classification accuracy and training time would quantify the practical significance of these findings and better support the claim of improved efficiency.

5. Visualization of Overlapping Features

Figures 2–4 effectively illustrate the selected features from each method. Consider adding a simple Venn diagram to visually highlight overlapping features—especially “Price” and “Preference,” which were selected by all three methods. This would enhance clarity and emphasize key takeaways.

 

Minor Comments

• Appendix Labeling: Ensure that the questionnaire items (pp. 25–27) are properly referenced as Appendix A in both the manuscript text and the Table of Contents for easy navigation.
• Active Voice & Terminology Consistency: Replace passive constructions (e.g., “It can be found that…”) with active voice (e.g., “We find that…”). Also, clearly distinguish between “repurchase intention” (your operational dependent variable) and “customer loyalty” (your broader conceptual construct) to avoid interpretive ambiguity.
• Practical Examples in Implications: In Section 5.2, where you recommend tactics such as “leveraging location-based services” and “device-optimized formats,” consider including a concrete example—for instance: “a push notification offering a 10% discount on items left in cart.” This would make your implications more actionable for practitioners.
• Future Methodological Suggestions: In addition to recommending stratified sampling and cross-cultural replication, consider suggesting model-agnostic interpretability techniques (e.g., SHAP values). For example, recent work such as:

Tanone, L.-H. Li, and S. Saifullah, “ViT-CB: Integrating hybrid Vision Transformer and CatBoost to enhance brain tumor detection with SHAP,” Biomed. Signal Process. Control, vol. 100, p. 107027, Feb. 2025, doi: [10.1016/j.bspc.2024.107027].

can serve as a reference model for validating feature importances beyond SVM-RFE or correlation filters.

These suggestions are intended to further elevate an already well-developed manuscript. Best wishes as you move toward final acceptance.

Comments on the Quality of English Language

Overall, the manuscript is intelligible and flows reasonably well, but it would benefit from more consistent and polished use of academic English. I noticed some shifting between present and past tenses—particularly in the Methods and Results sections. Standardizing on past tense for completed actions (e.g., “we collected,” “we applied,” “we conducted”) would improve clarity and maintain conventional scientific reporting style.

There are occasional issues with article usage (“a,” “an,” “the”) and subject-verb agreement—for instance, phrases like “data is” should be revised to “the data are,” and “questionnaire was distributed” should be “the questionnaire was distributed.” Addressing these will enhance grammatical precision.

Some sentences are structurally awkward or overly verbose. For example, “This step was began designing” should be revised to “This step began with the design of,” and “Using the 5-fold cross-validation method can be used in the results…” could be simplified to “We applied 5-fold cross-validation to enhance the robustness of our feature-selection results.” Breaking up long sentences, using parallel structure in lists, and applying consistent hyphenation (e.g., “feature-selection methods,” “data-driven approach”) will improve overall readability.

A final, careful copy-edit—either by a professional proofreader or through a dedicated language-editing tool—would help address these residual issues and ensure the English reads smoothly and precisely throughout.

Author Response

Many thanks for your valuable suggestions. All corrections have made according to your comments. Please check the attached file.

Author Response File: Author Response.docx