Mood-Based Music Discovery: A System for Generating Personalized Thai Music Playlists Using Emotion Analysis

Response to Reviewer 1 Comments

Dear Reviewer

We are extremely grateful to the reviewers for their thorough and constructive reviews. Their detailed comments and insightful suggestions have been invaluable in improving the quality of our manuscript.

Please find the detailed responses below and the corresponding revisions/corrections highlighted/in track changes in the re-submitted files.

Best regards,

Porawat Visutsak (Corresponding author)

Point-by-point response to Comments and Suggestions for Authors (the line number in this col. referred to the previous format)

Reviewer’s Evaluation

Response and Revisions

(I have made the corrections according to the reviewer #1 by using blue highlighted text.)

(1) The authors provided a good literature review of the current research progress. It is suggested that the research gaps be summarized before presenting and introducing this paper's main contributions and novelty. In addition, it is advised to include some discussions on emerging areas of machine learning, especially in terms of multi-scale perception multi-level feature fusion image quadrant entropy, and a high-accuracy method using swarm intelligence optimization entropy.

Yes/Can be improved/Must be improved/Not applicable

We appreciate the reviewer's positive comments on our literature review and their insightful suggestions. We agree that summarizing the research gaps before presenting our contributions would enhance the clarity and impact of our paper. More discussion are also added in revised manuscript.

(2) The article is deficient in theoretical formulas.

Yes/Can be improved/Must be improved/Not applicable

We appreciate the reviewer's suggestion to provide theoretical formulas. We added section 2.1 the mathematics foundation to our paper (please see section 2.1 Mathematics Foundations and ref. no. 16-23).

(3) The authors should provide detailed parameter settings for the proposed model and the comparison method.

Yes/Can be improved/Must be improved/Not applicable

We appreciate the reviewer's suggestion to provide detailed parameter settings for our proposed model and the comparison methods. In response, we have included a comprehensive table outlining the specific parameter values used for each model (please see Table1).

(4) In the data-processing stage, the details of data preprocessing are lacking. For instance, the methods for handling missing values and the specific steps of data standardization are not provided.

Yes/Can be improved/Must be improved/Not applicable

The reviewer rightly pointed out the need for more detailed information about our data preprocessing steps. We apologize for the omission and appreciate the opportunity to clarify these aspects. We provide detailed information about how we handled missing values and standardized the data

(please see section 2.2 Data Preprocessing).

(5) The figures are blurry and should be replaced with high-resolution versions. Additionally, ensure uniform font size and format in all figures.

Yes/Can be improved/Must be improved/Not applicable

All figures are enhanced to 300 dpi (please see the supplementary files).

Response to Reviewer 2 Comments

Dear Reviewer

We are extremely grateful to the reviewers for their thorough and constructive reviews. Their detailed comments and insightful suggestions have been invaluable in improving the quality of our manuscript.

Please find the detailed responses below and the corresponding revisions/corrections highlighted/in track changes in the re-submitted files.

Best regards,

Porawat Visutsak (Corresponding author)

Point-by-point response to Comments and Suggestions for Authors (the line number in this col. referred to the previous format)

Reviewer’s Evaluation

Response and Revisions (I have made the corrections according to the reviewer #2 by using green highlighted text.)

According to the manuscript, this study built a system for generating personalized Thai music playlists using emotion analysis. But there are still the following contents that need to be supplemented to improve the research

(1) The clarity of the images in the article needs to be improved. Please provide high-definition resolution images (Figure 2).

Yes/Can be improved/Must be improved/Not applicable

All figures are enhanced to 300 dpi (please see the supplementary files).

(2) What is the author's basis for classifying different categories in the dataset? Detailed explanation is needed.

Yes/Can be improved/Must be improved/Not applicable

We manually labeled each song in our dataset with one of four mood classes: love, happy, sad, and angry. (please see section 2.3)

(3) The author needs to supplement the information on the latest relevant research internationally, and present it in the form of a table to visually demonstrate the differences with this study, compare and analyze the differences between various methods, and evaluate the performance of the method proposed by authors.

Yes/Can be improved/Must be improved/Not applicable

We add the recent works of “emotion detection/classification of thai songs” and compare to our study in the “Discussion” section (ref. no. 24-26).

(4) The author needs to provide detailed parameter configurations of each model before training.

Yes/Can be improved/Must be improved/Not applicable

We appreciate the reviewer's suggestion to provide detailed parameter settings for our proposed model and the comparison methods. In response, we have included a comprehensive table outlining the specific parameter values used for each model (please see Table1).

(5) The author needs to conduct a detailed and comprehensive analysis of the limitations of the methods proposed in order to objectively evaluate the value of the research conducted.

Yes/Can be improved/Must be improved/Not applicable

We modify the manuscript to include a discussion of the limitations in the "Discussion" section.

Response to Reviewer 3 Comments

Dear Reviewer

We are extremely grateful to the reviewers for their thorough and constructive reviews. Their detailed comments and insightful suggestions have been invaluable in improving the quality of our manuscript.

Please find the detailed responses below and the corresponding revisions/corrections highlighted/in track changes in the re-submitted files.

Best regards,

Porawat Visutsak (Corresponding author)

Point-by-point response to Comments and Suggestions for Authors (the line number in this col. referred to the previous format)

Reviewer’s Evaluation

Response and Revisions (I have made the corrections according to the reviewer #3 by using yellow highlighted text.)

(1) The authors should consider providing more details about the preprocessing techniques used.

Yes/Can be improved/Must be improved/Not applicable

The reviewer rightly pointed out the need for more detailed information about our data preprocessing steps. We provide more details about data preprocessing in section 2.2.

(2) The authors should consider providing more details about how features were extracted from the song dataset.

Yes/Can be improved/Must be improved/Not applicable

We provide more details about feature extraction in section 2.3.

(3) The authors should consider providing more details about the classifier selection as to why specific classifiers were chosen.

Yes/Can be improved/Must be improved/Not applicable

- Section 2.1 provides a detailed explanation of the mathematical foundations of the five classifiers chosen for the study: Random Forest, XGBoost, Decision Tree, Logistic Regression, and SVM.   

- Section 2.4 discusses the rationale for selecting these classifiers, highlighting their strengths and suitability for the task of music mood classification.   

- The manuscript mentions that Random Forest was chosen for its ability to handle high-dimensional data and its robustness to overfitting.   

- XGBoost was selected for its efficiency and strong performance in classification tasks, especially with larger datasets.   

- Decision Tree was included for its ease of interpretation, while Logistic Regression was adapted for its ability to predict probabilities for multi-class problems.   

- SVM was chosen for its ability to handle non-linearly separable data using kernel functions.

(4) The authors should consider providing more discussion about the potential bias due to class imbalance.

Yes/Can be improved/Must be improved/Not applicable

- In the Results section (Section 3), we explicitly mentions the class imbalance within the dataset:

The results also highlight a class imbalance within the dataset, with “Sad” being the most prevalent and “Angry” the least. This imbalance potentially impacts the model’s performance, especially for the less frequent classes.  

- We further elaborates on the potential impact of this imbalance on the classification of the "Angry" category:

The “Angry” class, in particular, presents challenges, having the lowest F1-score (0.71) on the test set. This suggests potential difficulties in accurately classifying songs within this category, possibly due to limited training examples or the inherent complexity of the emotion itself.

- We also write the future direction to handle the class imbalance.

(5) The authors should consider providing more details about the train-test split ratio.

Yes/Can be improved/Must be improved/Not applicable

Thank you for this comment. We completely forgot to mention the train-test split ration in the experiment setting. We used a consistent 80:20 split ratio for all five models in our experiment (please see section 2.4).

(6) The authors should consider providing more details about any statistical significance testing done to determine whether the classifier improvements were meaningful.

Yes/Can be improved/Must be improved/Not applicable

We appreciate the reviewer's suggestion to provide more details about statistical significance testing to determine whether the classifier improvements were meaningful. Unfortunately, the manuscript does not provide details about any statistical significance testing. But we discuss this issue in the future work section.

(7) The authors should consider providing more details about possible overfitting for Random Forest.

Yes/Can be improved/Must be improved/Not applicable

In section 2.4, where the rationale for selecting different classifiers is discussed, we mention that Random Forest was chosen for its ability to handle high-dimensional data and its robustness to overfitting. This implies that while Random Forest is generally less prone to overfitting than some other algorithms, the possibility of overfitting still exists.   

- Additionally, in section 2.1.1, where the mathematical foundations of Random Forest are explained, we discuss how the algorithm utilizes Bagging (Bootstrap Aggregating) and Feature Randomness to enhance tree diversity and reduce variance, which are key factors in mitigating overfitting.

(8) The authors should consider comparing the study's contributions more closely with existing works and providing more concrete future work recommendations.

Yes/Can be improved/Must be improved/Not applicable

We add the recent works of “emotion detection/classification of thai songs” and compare to our study in the “Discussion” section (ref. no. 24-26). And the recommendation of future is also added in this revised manuscript.