A Clustered Federated Learning Method of User Behavior Analysis Based on Non-IID Data

Round 1

Reviewer 1 Report

1. Please consider to revise the title of this manuscript so that the contributions can be better reflected.

2. The problem statements and research gaps that motivated the current works are not evident. Further elaborations are needed to justify the motivations of current work.

3. Please elaborate the technical contributions and novelty of current work.

4. More critical analyses is needed for literature review in Section 2. A table can be used to summarize the works.

5. Nomenclature table is needed to summarize the definition of mathematical symbols and abbreviation used in this manuscript.

6. What are the loss function used for the training process? Please explain and include in the revised manuscript.

7. The concepts of cluster mediator and two-tier cache mechanism are introduced into the proposed method. How do these modifications can affect the computational time? Please discuss this issue.

8. What are the limitations and future works of current study? Please discuss.

9. It will be good if the authors can share the source code of FedTCM for verification purpose. 

Author Response

Please check the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

This paper proposed a new federated learning algorithm – FedTCM – to mitigate the impact of Non-IID data on the training process of federated learning models. Overall, the topic is of great interest to the field of machine learning, but the authors need to address the following concerns:

1. The literature review in section 2: Related Work can be more comprehensive and organized. Listing the references is not a good way to make the reader understand the work’s contribution. The authors also need to better clarify the main contribution of this work following the literature review.

2. In Figure 1 and Figure 3, the annotations of first-tier cache and second tier cache are misplaced. Please correct this and read through the paper to correct similar typos.

3. One of my big concern is the accuracy of FedTCM is on average 1.1% higher than the typical federal method FedAvg. The 1.1% sounds suspicious to me, as it might be just because of data bias or computational error. I would suggest the authors add one more experimental study besides the user shopping behavior case, possibly on video-click behavoir to further check your model performance.

Author Response

Please check the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

The authors have addressed all comments given. However, it is observed that the responses provided for Comments #7 and #8 are not reflected in the revised manuscript. Please include these response to enhance the clarity of manuscript.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf