All reviews of published articles are made public. This includes manuscript files, peer review comments, author rebuttals and revised materials. Note: This was optional for articles submitted before 13 February 2023.
Peer reviewers are encouraged (but not required) to provide their names to the authors when submitting their peer review. If they agree to provide their name, then their personal profile page will reflect a public acknowledgment that they performed a review (even if the article is rejected). If the article is accepted, then reviewers who provided their name will be associated with the article itself.
Dear Authors,
All of the reviewers agree that the main issues pointed out in previous reviews have been successfully addressed.
Best wishes,
[# PeerJ Staff Note - this decision was reviewed and approved by Carlos Fernandez-Lozano, a PeerJ Section Editor covering this Section #]
Thank you for the revision. The manuscript looks much better, and I agree to its publication in its current form.
Thank you for the revision. The manuscript looks much better, and I agree to its publication in its current form.
Thank you for the revision. The manuscript looks much better, and I agree to its publication in its current form.
Thank you for the revision. The manuscript looks much better, and I agree to its publication in its current form.
The authors have addressed successfully the main issues pointed out in previous reviews.
The authors have addressed successfully the main issues pointed out in previous reviews.
The authors have addressed successfully the main issues pointed out in previous reviews.
.
.
.
.
Dear authors,
Thank you for the submission. The reviewers’ comments are now available. It is not suggested that your article be published in its current format. We do, however, advise you to revise the paper in light of the reviewers’ comments and concerns before resubmitting it.
Best wishes,
**PeerJ Staff Note:** Please ensure that all review, editorial, and staff comments are addressed in a response letter and that any edits or clarifications mentioned in the letter are also inserted into the revised manuscript where appropriate.
This paper introduces a dynamic features-based method for image retrieval and sentiment polarity analysis within the context of digital media. The authors aim to address the challenges posed by the vast amount of digital images and the complexity of sentiment extraction. They propose an innovative multi-modal approach that integrates image captioning with visual features, allowing for more nuanced semantic and emotional analysis of the content. The method leverages deep learning techniques to extract dynamic features from both image and text, providing an effective way of improving both retrieval accuracy and emotion detection. The experimental results presented in the paper demonstrate the method's impressive performance, achieving high accuracy scores and maintaining operational efficiency. However I have few concens and you are advised to revise your paper.
1. The proposed method of combining image captioning with dynamic features is a novel approach. It would be helpful to provide more detailed comparisons with traditional methods that use only visual features or textual descriptions.
2. The accuracy values mentioned (0.951 at 1, 0.985 at 5, and 0.989 at 10) are impressive. Could the authors explain how these values compare with the current state-of-the-art in digital image retrieval and sentiment analysis?
3. The description of the image captioning model (SEIC) is well-detailed. It would be beneficial to include an example of the actual captions generated for a few images to better illustrate its effectiveness.
4. Dynamic feature extraction is mentioned as a key part of the model. Can the authors clarify how dynamic temporal changes in images are captured and whether this has an impact on image retrieval accuracy?
1. The accuracy values mentioned (0.951 at 1, 0.985 at 5, and 0.989 at 10) are impressive. Could the authors explain how these values compare with the current state-of-the-art in digital image retrieval and sentiment analysis?
2. The description of the image captioning model (SEIC) is well-detailed. It would be beneficial to include an example of the actual captions generated for a few images to better illustrate its effectiveness.
3. Dynamic feature extraction is mentioned as a key part of the model. Can the authors clarify how dynamic temporal changes in images are captured and whether this has an impact on image retrieval accuracy?
4. The captioning module relies heavily on the Transformer architecture. Could the authors compare its performance with other models such as BERT or GPT for better context?
5. The multi-modal feature fusion mechanism is interesting, but the paper does not provide detailed analysis of how the different weight adjustments are made dynamically. A brief discussion on this process would add value.
1. The paper would benefit from more in-depth analysis on how the method scales with larger datasets. Do the results hold when tested on datasets significantly larger than CIRCO?
2. In the results section, the authors mention using Accuracy at K for evaluation. It might be beneficial to provide additional metrics such as Precision, Recall, and F1-score to give a more complete assessment of the method’s performance.
3. In Figure 1 and 2, the architecture and formula decomposition are presented well. However, the exact role of each component (like weight matrices and their significance) is not entirely clear. A more detailed explanation could help readers understand their impact on the model.
4. The method’s real-time processing efficiency (112.4 ms) is impressive. However, how does this performance compare with similar models under high-load conditions or when applied to real-time video analysis?
5. In Section 5.3, the authors compare their method with others such as MD-SAN, DFT, and OSCARB. Could the authors include a discussion of the weaknesses of their method relative to these competitors, especially in handling highly dynamic content?
The paper mentions that the method could be extended to handle temporal data (video). It would be valuable to explore how video frame sequence analysis can be integrated into the current framework and how this might affect both retrieval accuracy and sentiment polarity prediction.
This paper introduces a dynamic feature-aware collaborative filtering model combining user time-sensitive preference vectors with matrix factorization. While the proposed approach is conceptually promising, the manuscript would benefit from clearer exposition of model components and stronger empirical validation.
L34–36: The problem statement should emphasize the novelty gap more clearly. What specific limitations of prior dynamic models (e.g., timeSVD++) are you addressing?
L72–79: The transition from static to dynamic features is abrupt. Include a conceptual diagram comparing both settings
L89–91: The term “temporal dynamic interest vector” is central but not clearly defined. Is it updated per session or per time window?
L170–178: Use a table to summarize all variables used in the model for clarity.
L102–104: Clarify how the combination weights α and β are optimized—are they learned parameters or manually tuned?
L117–124: Algorithm 1 lacks explanation of input/output variables. Consider rewriting the pseudocode with clearer notation or variable descriptions.
L126–129: The attention mechanism is briefly mentioned but not explained. How is attention computed in your architecture?
L145–151: Time complexity analysis should be expanded. Compare computational cost with baseline methods such as DeepFM or Transformer-based recommenders.
L160–165: The update strategy for user features via gradient descent is not clearly formulated. Include update rules or derivations.
.
.
.
The manuscript proposes a dynamic feature-enhanced recommendation framework, pre
senting a logical structure. However, the experimental evaluation lacks depth, and clarity
issues in the writing hinder comprehension. Below are specific comments to improve the
manuscript.
• Abstract (Lines 18–25): The abstract does not include quantitative results to
substantiate the framework’s contribution. Including metrics such as Root Mean
Square Error (RMSE)reduction or other performance improvements would strengthen
the abstract’s impact and provide a clearer overview of the work’s significance.
• Introduction (Lines 40–43): General statements such as “users’ needs vary greatly”
lack specificity and empirical support. To enhance credibility, these claims should
be substantiated with references to user behavior studies or data-driven trends from
relevant literature.
• Methodology (Lines 108–115): Equations (1)–(4) are presented without defining the
variables involved. To improve clarity and accessibility, include a brief description
of each variable immediately following the respective equation.
• Feature Combination (Lines 140–144): The rationale for combining user preference,
item similarity, and temporal features in a weighted formulation is not adequately
explained. Provide empirical or theoretical justification for this approach, such as
referencing studies that demonstrate the effectiveness of such combinations.
• Organization (Lines 174–178): The manuscript’s readability could be improved by
using more descriptive section headers and subheaders. For example, renaming
the “Model Description” section to “Dynamic Feature Learning Framework” would
better reflect the content and enhance clarity.
• Experimental Setup (Lines 219–221): The experimental settings lack critical details
necessary for reproducibility, such as latent dimensions, learning rate, and batch
size. Including these baseline hyperparameters would strengthen the experimental
design and facilitate replication.
• Evaluation Metrics (Lines 250–253): The absence of standard recommendation
system metrics, such as Precision@K and Recall@K, is notable. Justify the omission
of these metrics or consider incorporating them to align with common evaluation
practices in the field.
• Baseline Comparisons (Lines 256–258): The comparison methods used are not
state-of-the-art. Including stronger baselines, such as SASRec [1], BERT4Rec [2],
or DIN [3], would provide a more robust evaluation of the proposed framework’s
performance.
• Results Visualization (Lines 261–267): Figures in the results section lack confi
dence intervals or error bars, limiting the interpretability of the findings. Plotting
standard deviations or confidence intervals over multiple runs would enhance the
reliability and transparency of the reported results
All text and materials provided via this peer-review history page are made available under a Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.