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,
Thank you for your revised submission in light of the experts' comments. I am pleased to inform you that your manuscript is being recommended for publication. Congratulations!!
However, please make sure that during your final/proofread version submission , you must incorporate the suggestion of reviewer which is being reproduced below for your information.
"As a minor and easily addressable point, I suggest ensuring consistent capitalization of "BiLSTM" throughout the manuscript (e.g., "BILSTM" on page 14, line 258, should be "BiLSTM") to maintain a uniform stylistic presentation."
[# PeerJ Staff Note - this decision was reviewed and approved by Shawn Gomez, a PeerJ Section Editor covering this Section #]
**PeerJ Staff Note:** Although the Academic and Section Editors are happy to accept your article as being scientifically sound, a final check of the manuscript shows that it would benefit from further editing. Therefore, please identify necessary edits and address these while in proof stage.
The authors have comprehensively and satisfactorily addressed all points and suggestions raised in the previous round of review. The revisions have significantly strengthened the manuscript, with notable improvements in methodological clarity, a more detailed discussion of the experimental results and ablation studies, and enhanced descriptions of the data preprocessing and computational setup. These additions greatly improve the transparency and reproducibility of the study.
As a minor and easily addressable point, I suggest ensuring consistent capitalization of "BiLSTM" throughout the manuscript (e.g., "BILSTM" on page 14, line 258, should be "BiLSTM") to maintain a uniform stylistic presentation.
Revision is with good form.
Revision is with good form.
Revision is with good form.
Revision is with good form.
Thank you for your submission to our esteemed journal. We are writing to inform you that your manuscript received an expert opinion and you will see that a number of changes need to be addressed. Therefore, we invite you to carefully address these changes and resubmit a detailed response.
The technical language of the paper should also be improved in the revised version.
Thank you
**PeerJ Staff Note:** Please ensure that all review and editorial 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.
**Language Note:** The Academic Editor has identified that the English language must be improved. PeerJ can provide language editing services - please contact us at [email protected] for pricing (be sure to provide your manuscript number and title). Alternatively, you should make your own arrangements to improve the language quality and provide details in your response letter. – PeerJ Staff
The paper presents a potentially valuable hybrid model for predicting energy awareness in green supply chains by combining K-means clustering, Dempster–Shafer evidence theory, and BiLSTM. The research topic aligns well with the journal’s focus on intelligent systems and sustainable computing. However, several major revisions are required before the paper can be considered for publication.
The combination of clustering, evidence fusion, and BiLSTM has already been reported extensively in previous literature on hybrid deep learning forecasting models.
The DS fusion mechanism is used mechanically without explaining how conflicting evidence is resolved or how it interacts with the BiLSTM output layer. Furthermore, the paper’s “multi-agent” perspective is only conceptual—no agent-level interaction, reward, or policy mechanism is mathematically modeled. The model architecture lacks formal justification, and the derivations presented are generic and partially incomplete.
The paper presents a potentially valuable hybrid model for predicting energy awareness in green supply chains by combining K-means clustering, Dempster–Shafer evidence theory, and BiLSTM. The research topic aligns well with the journal’s focus on intelligent systems and sustainable computing. However, several major revisions are required before the paper can be considered for publication.
The paper repeatedly emphasizes multi-agent decision-making, yet no explicit agent-based formulation, decision rule, or inter-agent interaction is defined. A mathematical or simulation-based agent behavior model should be introduced.
The DS theory section needs clearer derivation and explanation of the conflict coefficient and normalization process. Include a numerical example demonstrating how conflicting evidence affects trust updates.
Currently, the combination appears sequential rather than integrated. Clarify how clustering results influence BiLSTM inputs (e.g., through feature selection, weight initialization, or sequence segmentation).
The experimental section should include comparisons with more advanced deep learning models (e.g., Transformer, Graph Neural Network, or hybrid reinforcement learning models).
The ablation study is a strong point, but it should also evaluate computational efficiency, convergence speed, and data sensitivity to strengthen the justification for each module.
Substantial English editing is necessary. Improve figure captions, correct numbering inconsistencies (e.g., “Fig.5–Fig.9”), and ensure that all equations are properly referenced and formatted.
The manuscript investigates a hybrid deep learning framework—K-BiLSTM—integrating K-means clustering, Dempster–Shafer (DS) evidence theory, and bidirectional LSTM to predict multi-agent energy awareness evolution within a green supply chain context. The research direction is relevant and timely, aligning with sustainability-driven decision intelligence and data-driven supply chain management under the “dual-carbon” strategy. The combination of clustering, evidence fusion, and temporal modeling is theoretically sound; however, the current manuscript requires substantial revisions to strengthen its methodological rigor, empirical analysis, and logical exposition.
The paper shows potential to make a meaningful contribution if major issues concerning theoretical grounding, algorithmic integration, and experimental validation are addressed as outlined below.
1. Conceptual Framework and Model Justification
The term “multi-agent” appears repeatedly throughout the manuscript, yet no explicit definition or mathematical modeling of agents, their interactions, or decision-making processes is provided. To justify the multi-agent premise, the authors should formulate a clear representation of agent entities (e.g., suppliers, manufacturers, consumers) and define interaction rules or feedback functions among them. Incorporating a multi-agent simulation layer or an agent-based utility function could substantially strengthen the theoretical foundation.
The current explanation of how Dempster–Shafer theory interacts with BiLSTM outputs is ambiguous. The paper should explicitly show where and how evidence fusion occurs in the computational pipeline—before, within, or after the BiLSTM network. For instance, if DS theory fuses multiple prediction signals or contextual indicators, a mathematical mapping from fused belief mass functions to BiLSTM input vectors must be provided. Without this, the combination remains conceptual.
Figure 4 presents the BiLSTM structure, but there is no unified diagram illustrating the complete K-BiLSTM pipeline (including clustering, evidence synthesis, and temporal prediction). The authors are advised to add a new figure (e.g., “Fig. 5 Overall Workflow of the K-BiLSTM Framework”) summarizing the sequence of data preprocessing, clustering, fusion, training, and inference, ensuring each stage is annotated with input/output representations and data flow direction.
2. Algorithmic Description and Mathematical Rigor
Several equations (especially in Sections 3.2 and 3.3) contain undefined variables or missing numbering. For example, in the utility transformation and evidence fusion parts, symbols such as θ,H,yare introduced without contextual definitions. The paper should ensure that every symbol is properly defined and consistent across sections.
Equation (4) regarding DS fusion lacks derivation and normalization terms. The authors should provide the complete Dempster combination formula, including the conflict coefficient Kand the normalization factor 1/(1-K), to improve mathematical rigor. A numerical example illustrating the evidence combination procedure would enhance clarity.
The connection between clustering and BiLSTM input remains underspecified. Does clustering operate on static features (e.g., average energy use per firm) or on temporal segments? Clarify whether cluster labels are used as categorical input features, or whether each cluster corresponds to a separate BiLSTM sub-model. Without this, the integration appears sequential rather than synergistic.
To improve reproducibility, the authors should include pseudocode or an algorithm block summarizing the key steps of the proposed K-BiLSTM method. Each step should specify data inputs, clustering procedure, fusion computation, BiLSTM training, and output evaluation.
3. Experimental Design and Data Interpretation
The manuscript mentions “region A” and 12 manufacturing enterprises but provides limited information on data type, time span, or collection frequency. The authors should specify:
The number of time steps per enterprise and total sample size;
The nature of features (quantitative indicators vs. survey-based qualitative scores);
The preprocessing logic for handling missing values, normalization, and categorical encoding.
Without these details, the experimental section lacks transparency.
While CNN-BiLSTM, BiLSTM-Attention, and CEEMDAN-BO-BiLSTM are used for comparison, the chosen baselines are relatively outdated. For a fair evaluation, the authors should include recent spatiotemporal or attention-based models, such as:
Transformer-based time series models (e.g., Informer, Autoformer);
Graph-based temporal networks (e.g., STGCN, DCRNN);
RL-enhanced hybrid predictors (e.g., DDPG-BiLSTM).
This inclusion would make the comparative analysis more convincing.
.
.
.
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.