PAD-MPFN: Dynamic Fusion with Popularity Decay for News Recommendation

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The paper proposes PAD-MPFN, a news recommendation system that integrates dynamic user interest modeling, popularity bias mitigation, and cold-start handling through a multi-perspective fusion network.

Pros:
1. The paper is well-organized. 2. The authors conduct extensive experiments on multiple datasets. 3. The paper tackles important issues in news recommendation, such as dynamic user interests and cold-start problems.

Cons:

1. The integration of sequential, graph-based, and popularity-based components is complex. It's unclear if simpler models could achieve similar performance.
2. The choice of parameters (e.g., time saturation factor τ) is not well justified. How sensitive is the model to these parameters?
3. The construction of user-news interaction graphs involves co-occurrence frequency. How does the choice of this metric affect the graph structure and subsequent recommendations?
4. The article mainly discusses recommendation. It is recommended to introduce related methods: DisCo: graph-based disentangled contrastive learning for cold-start cross-domain recommendation
5. The paper claims improvements in cold-start scenarios, but the evaluation is limited to users with very few clicks. How does the model perform for users with zero historical data?
6. The training involves multiple steps, including negative sampling and batch processing. How does the choice of batch size and sampling ratio affect convergence?

Author Response

We sincerely appreciate the reviewer's valuable comments and suggestions. We have carefully addressed all points raised by Reviewer #1 in our detailed response document titled "PAD_MPFN_MDPI-response to reviewers.pdf", which has been submitted as a separate attachment. This document contains our point-by-point responses to each comment, along with explanations of the corresponding revisions made to the manuscript.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The article addresses the question of how to effectively integrate dynamic user interests, social influence, and popularity trends into a single news recommendation framework. The authors present a model that uses adaptive subspace projections, a novel time-decay mechanism based on logarithmic transformations, and a dynamic gating strategy to combine different user representations.

The topic is relevant and timely within the domain of recommendation systems. The research fills a gap by jointly modeling multiple dimensions of user preferences - sequential behavior, social relationships, and time-aware popularity. This aspect could have been emphasized more as a core innovation, as currently the novelty of the fusion method may seem incremental compared to earlier works.

Compared to other published material, the paper offers a unified and dynamically adaptive model that shows consistent improvements over strong baselines in empirical evaluations. The authors evaluate their approach on the widely used MIND dataset, using standard metrics such as AUC, MRR, and nDCG, and include thorough analyses such as ablation studies and cold-start performance breakdowns. The evidence shows that PAD-MPFN achieves superior accuracy and diversity of recommendations, especially under cold-start conditions.

Nevertheless, the methodology, while solid overall, would benefit from several improvements. First, the explanation of the architecture and its components, though detailed, suffers from issues with clarity due to stylistic and grammatical problems throughout the paper. The manuscript needs careful proofreading to correct numerous issues with spacing, punctuation, article use, and sentence structure. For example, in the conclusion section, the word "text" appears out of context and should be removed. Additionally, the caption under Figure 1 is poorly worded and includes a double period; it should be rewritten for clarity and correctness. The most important improvement would be to significantly expand the experimental evaluation. The current results are based on a single dataset, which is insufficient to draw generalizable conclusions. Additional experiments on diverse and representative datasets are necessary to validate the robustness and applicability of the proposed method.

The conclusions of the paper are mostly consistent with the presented evidence and well-aligned with the central research question. The experimental results support the model’s effectiveness, and the authors provide a convincing case for their approach through both quantitative and qualitative analysis, including a user-level case study. However, the conclusions would benefit from a clearer summary of what distinguishes PAD-MPFN from closely related methods. More emphasis on the interpretability and efficiency of the popularity decay modeling would help strengthen the perceived novelty of the work.

The references included in the manuscript are generally appropriate, comprehensive, and up to date. The authors cite a wide range of relevant studies from both sequence-based and graph-based recommendation literature, as well as works on popularity modeling and news personalization. One minor point is that the placement and consistency of references could be better formatted in some places, but this is a minor editorial issue.

With regard to the figures and tables, most of them are relevant and help to support the text. However, some figures, such as Figure 1, are densely labeled and could be made clearer by improving their resolution and simplifying their structure.

Author Response

We sincerely appreciate the reviewer's valuable comments and suggestions. We have carefully addressed all points raised by Reviewer #2 in our detailed response document titled "PAD_MPFN_MDPI-response to reviewers.pdf", which has been submitted as a separate attachment. This document contains our point-by-point responses to each comment, along with explanations of the corresponding revisions made to the manuscript.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

1. Clarify the choice of mathematical modeling for popularity decay – why not try non-parametric or kernel-based methods?
2. Improve English clarity, especially in complex methodological sections (e.g., Section 4.3 on popularity modeling).
3. Add statistical significance analysis to support comparative claims in performance tables.
4. Consider including runtime analysis or complexity assessment to gauge scalability for production systems.
5. Discuss the computational cost of training and inference in PAD-MPFN.
6. Visual aids (Figures 1 and 3) could be simplified and enhanced with color for improved accessibility.
7. Suggest including a link to the code or model repository for reproducibility.

Comments on the Quality of English Language

The English could be improved to convey the research more clearly.

Author Response

We sincerely appreciate the reviewer's valuable comments and suggestions. We have carefully addressed all points raised by Reviewer #3 in our detailed response document titled "PAD_MPFN_MDPI-response to reviewers.pdf", which has been submitted as a separate attachment. This document contains our point-by-point responses to each comment, along with explanations of the corresponding revisions made to the manuscript.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The author addressed my concerns and I was inclined to accept

Reviewer 2 Report

Comments and Suggestions for Authors

The authors revised the article and took my comments into account correctly.