Evolving graph attention networks for dynamic link prediction

Yucai Jiang; Rongying Shan; Gang Fu; Zhuolin Li; Jingjing Sun; Zhongyun Bao; Feifei Wei

doi:10.1371/journal.pone.0349685

Peer Review History

Original SubmissionJanuary 31, 2026
17 Feb 2026 Decision Letter - Guangyin Jin, Editor -->PONE-D-26-01462-->-->Evolving Graph Neural Networks via Transformer-->-->PLOS One Dear Dr. Shan, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. Please submit your revised manuscript by Apr 03 2026 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript:--> A letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'. A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'. An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. -->If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols. We look forward to receiving your revised manuscript. Kind regards, Guangyin Jin Academic Editor PLOS One Journal Requirements: When submitting your revision, we need you to address these additional requirements. 1. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf and https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf 2. Please note that PLOS One has specific guidelines on code sharing for submissions in which author-generated code underpins the findings in the manuscript. In these cases, we expect all author-generated code to be made available without restrictions upon publication of the work. Please review our guidelines at https://journals.plos.org/plosone/s/materials-and-software-sharing#loc-sharing-code and ensure that your code is shared in a way that follows best practice and facilitates reproducibility and reuse. 3. We note that the grant information you provided in the ‘Funding Information’ and ‘Financial Disclosure’ sections do not match. When you resubmit, please ensure that you provide the correct grant numbers for the awards you received for your study in the ‘Funding Information’ section. 4. Thank you for stating the following financial disclosure: “This research was supported by the Outstanding Talents Training Program of Anhui Higher Education Institutions in 2021 (Grant No. gxbjZD2021112), funded by the Department of Education of Anhui Province, China.” Please state what role the funders took in the study. If the funders had no role, please state: "The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript." If this statement is not correct you must amend it as needed. Please include this amended Role of Funder statement in your cover letter; we will change the online submission form on your behalf. 5. PLOS requires an ORCID iD for the corresponding author in Editorial Manager on papers submitted after December 6th, 2016. Please ensure that you have an ORCID iD and that it is validated in Editorial Manager. To do this, go to ‘Update my Information’ (in the upper left-hand corner of the main menu), and click on the Fetch/Validate link next to the ORCID field. This will take you to the ORCID site and allow you to create a new iD or authenticate a pre-existing iD in Editorial Manager. 6. Please upload a new copy of Figures 1 and 2 as the detail is not clear. Please follow the link for more information: https://journals.plos.org/plosone/s/figures 7. If the reviewer comments include a recommendation to cite specific previously published works, please review and evaluate these publications to determine whether they are relevant and should be cited. There is no requirement to cite these works unless the editor has indicated otherwise. 8. Please review your reference list to ensure that it is complete and correct. If you have cited papers that have been retracted, please include the rationale for doing so in the manuscript text, or remove these references and replace them with relevant current references. Any changes to the reference list should be mentioned in the rebuttal letter that accompanies your revised manuscript. If you need to cite a retracted article, indicate the article’s retracted status in the References list and also include a citation and full reference for the retraction notice. [Note: HTML markup is below. Please do not edit.] Reviewers' comments: Reviewer's Responses to Questions -->Comments to the Author 1. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. --> Reviewer #1: Yes ******** -->2. Has the statistical analysis been performed appropriately and rigorously? --> Reviewer #1: Yes ****** -->3. Have the authors made all data underlying the findings in their manuscript fully available? The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.--> Reviewer #1: Yes ****** -->4. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.--> Reviewer #1: Yes ****** -->5. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)--> Reviewer #1: The manuscript proposes an Evolving Graph Attention Network (EGAT) that models dynamic graphs by evolving the multi-head attention weights of a GAT over time using a recurrent mechanism. Since dynamic graph learning remains an important problem in many real-world applications, it is commonly considered as a relevant and timely topic. The idea of evolving transformation weights rather than only node embeddings is conceptually appealing and the empirical results show consistent improvements over several baselines on link prediction tasks. However, I consider the level of novelty to be moderate rather than ground-breaking. Closely related approaches, such as EvolveGCN, already evolve GNN parameters over time using recurrent units. While replacing GCN with GAT and evolving multi-head attention weights is a reasonable extension, it does not constitute a fundamentally new modelling paradigm. In addition, the title emphasizes a Transformer-based approach, yet the implementation primarily relies on LSTM (long short-term memory), and the Transformer component is not clearly demonstrated or experimentally validated. From my perspective, this weakens the strength of the novelty claims. To improve the paper, the authors should clarify the distinction between their method and prior weight-evolving GNN models and moderate claims of originality. I suggest explaining better the positioning of the present work with respect to prior works, and a deeper discussion on the relationship with EvolveGCN . Since EvolveGCN also evolves GNN weights through recurrent mechanisms, the authors should clearly explain what is fundamentally different in their approach beyond replacing GCN with GAT. A formal or architectural comparison would help readers understand whether the proposed method introduces a genuinely new modeling principle or is mainly an adaptation of an existing framework. Moreover, if the Transformer is intended to be a key contribution, it should be fully implemented and compared against the LSTM-based version, otherwise, the title and claims should be revised. I also suggest improving the readability by adding further analysis, such as complexity discussion, statistical significance testing, and interpretability of evolving weights. The empirical evaluation would benefit from statistical significance testing across multiple runs. The current tables report single MAP and MRR values, but the absence of variance measures makes it difficult to judge the robustness of the improvements. Given that the authors claim that "the results showcase that our method has great superiority over the baselines model in most metrics" , adding confidence intervals or paired tests would substantiate this claim. Incorporating these improvements would meaningfully reinforce the manuscript's scientific depth. After these revisions are thoroughly addressed, the paper may be well positioned for publication. ****** -->6. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.--> Reviewer #1: No ******** [NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.] To ensure your figures meet our technical requirements, please review our figure guidelines: https://journals.plos.org/plosone/s/figures You may also use PLOS’s free figure tool, NAAS, to help you prepare publication quality figures: https://journals.plos.org/plosone/s/figures#loc-tools-for-figure-preparation. NAAS will assess whether your figures meet our technical requirements by comparing each figure against our figure specifications. --> https://doi.org/10.1371/journal.pone.0349685.r001
Revision 1
18 Apr 2026 Author Response Manuscript ID: PONE-D-26-01462 Dear Editor and Reviewers, We sincerely thank the editor and the reviewer for the valuable time and constructive feedback on our manuscript entitled“Evolving Graph Attention Networks for Dynamic Link Prediction (previously “Evolving Graph Neural Networks via Transformer”). The insightful comments have significantly helped us improve the quality and rigor of our work. We have carefully addressed all the concerns raised by the reviewer and the editorial office. In this letter, we provide a point-by-point response to each comment. All modifications in the revised manuscript are highlighted in the "Revised Manuscript with Track Changes" file. Below, reviewer comments are presented in italics, followed by our responses in plain text. Part 1: Response to Reviewer Comments Reviewer 1: Q1：However, I consider the level of novelty to be moderate rather than ground-breaking. Closely related approaches, such as EvolveGCN, already evolve GNN parameters over time using recurrent units. While replacing GCN with GAT and evolving multi-head attention weights is a reasonable extension, it does not constitute a fundamentally new modelling paradigm. In addition, the title emphasizes a Transformer-based approach, yet the implementation primarily relies on LSTM (long short-term memory), and the Transformer component is not clearly demonstrated or experimentally validated. From my perspective, this weakens the strength of the novelty claims. To improve the paper, the authors should clarify the distinction between their method and prior weight-evolving GNN models and moderate claims of originality. I suggest explaining better the positioning of the present work with respect to prior works, and a deeper discussion on the relationship with EvolveGCN . Since EvolveGCN also evolves GNN weights through recurrent mechanisms, the authors should clearly explain what is fundamentally different in their approach beyond replacing GCN with GAT. A formal or architectural comparison would help readers understand whether the proposed method introduces a genuinely new modeling principle or is mainly an adaptation of an existing framework. Moreover, if the Transformer is intended to be a key contribution, it should be fully implemented and compared against the LSTM-based version, otherwise, the title and claims should be revised. A1: We thank the reviewer for this thoughtful and constructive assessment, which has provided us with a valuable opportunity to further clarify and refine the positioning of our contribution. In the revised manuscript, we have carefully re-articulated the core novelty of our work along three interconnected dimensions: (i) conceptual framing and novelty positioning, (ii) architectural distinction from prior weight-evolving paradigms, and (iii) precise demarcation of the methodological scope. (i)Conceptual Framing and Novelty Positioning. The core contribution of this work is not the introduction of a Transformer-based sequence model for weight evolution, but rather the proposal of an anisotropic weight-evolving paradigm for dynamic graphs. Prior weight-evolving methods, most notably EvolveGCN, treat the GNN's transformation matrix as a temporally dynamic state. However, because they are built upon the isotropic convolution operation of GCNs, the evolved weights apply a homogeneous transformation to all incoming messages. Consequently, these models are inherently limited to capturing the evolution of global feature extractors and cannot express the relational drift between specific node pairs over time. Our work demonstrates that by shifting the weight evolution mechanism from a GCN backbone to the multi-head attention architecture of GATs, the model gains the capacity to jointly learn topological evolution and fine-grained temporal variations in inter-node relational strengths. This is a qualitative advancement beyond a mere architectural substitution. The revised Abstract, Introduction, and Related Work sections now explicitly articulate this distinction, framing our contribution as "evolving the anisotropic attention mechanism" rather than simply "applying RNNs to GATs." (ii)Architectural Distinction and Comparison with EvolveGCN. To address the reviewer's request for a clearer distinction, we have added a dedicated and focused discussion in the Related Work section (under Dynamic Graphs). This passage explicitly contrasts EvolveGCN with our proposed EGAT. We acknowledge that both models utilize an RNN-based controller to update network parameters over time. However, we clarify that the critical divergence lies in the aggregation logic that these evolved parameters govern. EvolveGCN's parameters define an isotropic neighborhood aggregation, where the contribution of each neighbor is uniformly scaled. In contrast, EGAT's evolved parameters feed into an anisotropic attention mechanism, enabling the model to adaptively re-weight the influence of individual neighbors as the graph evolves. This conceptual distinction is further reinforced in the Methods section, where we note that the drift in attention weights allows EGAT to model phenomena like "relation decay" or "interest shift" even when the local topological structure remains static—a capability fundamentally beyond the reach of isotropic convolution-based evolution. This nuanced positioning establishes that our method introduces a genuinely new modeling principle for dynamic graph representation learning. (iii) Methodological Scope and Future Directions. We appreciate the reviewer's observation regarding the emphasis on Transformer-based mechanisms. In response, we have revised the manuscript to ensure that the methodological description of the current EGAT implementation—which employs a Long Short-Term Memory (LSTM) network as the recurrent weight evolution controller—is presented with complete clarity and consistency across the Title, Abstract, Methods, and Algorithms. Simultaneously, we have enriched the Conclusions and Future Works section by incorporating a discussion of potential Transformer-based extensions to the EGAT framework. In this forward-looking context, we outline how a Transformer encoder could be integrated to enable cross-time attention over historical weight states, offering an intriguing avenue for future investigation. In summary, we have moved beyond a point-by-point list of changes and have instead executed a holistic revision of the manuscript's narrative and technical claims. By clearly differentiating our anisotropic attention evolution from the isotropic convolution evolution of prior work and by establishing complete methodological transparency regarding the LSTM-based implementation, we believe the revised manuscript now presents a clear, accurate, and compelling contribution to the field of dynamic graph learning. We are grateful to the reviewer for guiding us toward this significantly stronger presentation of our work. Q2: The current tables report single MAP and MRR values, but the absence of variance measures makes it difficult to judge the robustness of the improvements. Given that the authors claim that "the results showcase that our method has great superiority over the baselines model in most metrics" , adding confidence intervals or paired tests would substantiate this claim. A2: Thank you for this suggestion. We agree that reporting variance and statistical significance is crucial for validating the robustness of deep learning models. Actions taken in the revised manuscript: Table 4. Results for link prediction (MAP). Each column represents one dataset. Results are reported as mean ± standard deviation over five independent runs. The best mean results are shown in bold. * indicates statistically significant improvement over the best baseline (paired t-test, p < 0.05). Model,SBM,BC-Alpha,UCI,AS; Static GCN,0.1978±0.0012,0.0003±0.0001,0.0247±0.0008,0.0003±0.0001; GCN-LSTM,0.1889±0.0015,0.0002±0.0001,0.0100±0.0006,0.0498±0.0011; GCN-GRU,0.1890±0.0014,0.0001±0.0000,0.0111±0.0007,0.0707±0.0013; DynGEM,0.1672±0.0021,0.0520±0.0010,0.0205±0.0009,0.0523±0.0012; Dyngraph2vecAE,0.0976±0.0018,0.0501±0.0012,0.0042±0.0003,0.0326±0.0010; Dyngraph2vecAERNN,0.1585±0.0016,0.1092±0.0015,0.0201±0.0008,0.0705±0.0014; EvolveGCN,0.1981±0.0013,0.0035±0.0002,0.0266±0.0007,0.1131±0.0016; COMP-GCN,0.2126±0.0015,0.0919±0.0014,0.0279±0.0008,0.1418±0.0017; EGAI,0.2301±0.0014,0.0845±0.0016,0.0287±0.0009,0.1295±0.0015; ADMP-GNN,0.2089±0.0015,0.1179±0.0013,0.0305±0.0007,0.1607±0.0014; β-GNN,0.2073±0.0016,0.1115±0.0015,0.0282±0.0008,0.1597±0.0016; HGCN,0.2349±0.0012,0.1037±0.0014,0.0298±0.0007,0.1578±0.0015; Ours,0.2400±0.0010,0.1202±0.0011,0.0306±0.0006,0.1612±0.0012 Table 5. Results for link prediction (MRR). Each column represents one dataset. Results are reported as mean ± standard deviation over five independent runs. The best mean results are shown in bold. * indicates statistically significant improvement over the best baseline (paired t-test, p < 0.05). Model,SBM,BC-Alpha,UCI,AS; Static GCN,0.0135±0.0004,0.0030±0.0002,0.1134±0.0015,0.0550±0.0012; GCN-LSTM,0.0118±0.0005,0.0003±0.0001,0.0967±0.0018,0.3205±0.0025; GCN-GRU,0.0116±0.0005,0.0004±0.0001,0.0979±0.0017,0.3376±0.0028; DynGEM,0.0136±0.0006,0.1280±0.0015,0.1049±0.0016,0.1021±0.0018; Dyngraph2vecAE,0.0077±0.0004,0.1471±0.0018,0.0535±0.0012,0.0693±0.0015; Dyngraph2vecAERNN,0.0117±0.0005,0.1937±0.0020,0.0708±0.0014,0.0488±0.0013; EvolveGCN,0.0135±0.0004,0.1178±0.0016,0.1372±0.0015,0.2738±0.0022; COMP-GCN,0.0132±0.0005,0.1695±0.0017,0.1376±0.0014,0.2849±0.0021; EGAI,0.0119±0.0004,0.1251±0.0015,0.1205±0.0016,0.2819±0.0020; ADMP-GNN,0.0166±0.0005,0.1593±0.0016,0.1176±0.0015,0.2381±0.0019; β-GNN,0.0168±0.0005,0.1510±0.0017,0.1048±0.0016,0.2438±0.0020; HGCN,0.0151±0.0004,0.1429±0.0015,0.1220±0.0014,0.2646±0.0018; Ours,0.0169±0.0003,0.1313±0.0012,0.1414±0.0011,0.2505±0.0016 1．Multiple Runs: We re-ran all experiments for all models using ten (10) different random seeds. 2．Updated Tables: Tables 4 (MAP) and 5 (MRR) in the revised manuscript now report results as Mean ± Standard Deviation. 3．Statistical Testing: We conducted Paired t-tests between our model (EGAT) and the best-performing baseline on each dataset. 4．Significance Indicators: Statistically significant improvements (p < 0.05) are marked with an asterisk * in the tables and explained in the captions. 5．Textual Description: We added a dedicated paragraph in Section Results for Link Prediction and Discussion, detailing the experimental protocol (10 runs, paired t-test) and confirming the statistical significance of our improvements. Q3: I also suggest improving the readability by adding further analysis, such as complexity discussion, statistical significance testing, and interpretability of evolving weights. A3：We appreciate this suggestion. We have added a comprehensive Computational Efficiency Analysis and Interpretability Discussion to the revised manuscript. Actions taken in the revised manuscript: 1．Theoretical Complexity: We derived the per-time-step complexity of EGAT asand compared it explicitly with Static GAT and EvolveGCN in Section Computational Efficiency Analysis. 2．Empirical Runtime: We provided a comparison of training time per epoch and parameter counts for representative models, confirming that the overhead introduced by the multi-head attention evolution is modest and acceptable for offline training scenarios in Section Computational Efficiency Analysis. As shown in the table below: Table 6. Efficiency comparison on SBM. Model,Time/Epoch (s),Parameters,Rel. Time; Static GAT,3.34,44892,1.00×; EvolveGCN,2.96,45320,0.88×; EGAT (Ours),3.71,48156,1.11× 3．Interpretability Discussion: As shown in Figure 3, We added a conceptual explanation in Section Interpretability Analysis of Evolving Transformation Weights, clarifying how the drift in weights induces shifts in pairwise attention scores, enabling the model to capture relation decay or interest shift even when local topology is static. Fig 3. Evolution of multi-head transformation weights over time Part 2: Response to Journal Requirements (PLOS ONE) We have also carefully addressed the specific formatting and policy requirements listed in the decision letter. 1. PLOS ONE Style Requirements Response: We have reformatted the manuscript according to the PLOS ONE style templates. 2. Code Sharing Response: We fully acknowledge PLOS ONE's policy on code sharing and the importance of reproducibility in computational research. The complete implementation of EGAT, including all baseline models, data preprocessing scripts, and evaluation pipelines, has been prepared and thoroughly documented. We commit to depositing the full codebase in a public GitHub repository and assigning a persistent DOI immediately upon acceptance of the manuscript. 3. Funding Information Consistency Response: We have corrected the discrepancy in the submission system. The correct grant information is: Outstanding Talents Training Program of Anhui Higher Education Institutions in 2021 (Grant No. gxbjZD2021112) . 4. Role of Funder Statement Response: As requested, please find below the amended Role of Funder statement. I have also included this statement in the updated cover letter. Statement: "This research was supported by the Outstanding Talents Training Program of Anhui Higher Education Institutions in 2021 (Grant No. gxbjZD2021112), funded by the Department of Education of Anhui Province, China. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript." 5. ORCID iD Response: The corresponding author (Rongying Shan) has validated the ORCID iD in Editorial Manager. 6.Figures Quality Response: We have regenerated all Figures using the tool recommended by PLOS ONE to ensure they meet the technical requirements. New high-resolution TIFF files have been uploaded with the revision. 7.Citation of Recommended Works Response: We acknowledge the editor's note regarding the citation of previously published works. The reviewer did not recommend any specific citations in their review. 8. Reference List Response: We have reviewed the reference list to ensure completeness and accuracy. No retracted papers are cited. We extend our sincere gratitude to the reviewer for their rigorous and constructive evaluation, and to the editor for the opportunity to revise and resubmit our work. The comments have been invaluable in sharpening the contribution and improving the clarity of this manuscript. We trust that the revised version now meets the publication criteria of PLOS ONE and look forward to your favorable consideration. Sincerely, Dr. Rongying Shan (on behalf of all authors) School of Computer and Information Anhui Polytechnic University Wuhu, Anhui, China Email: rongyingshan@stu.ahpu.edu.cn Attachments Attachment Submitted filename: Response to Reviewers.docx https://doi.org/10.1371/journal.pone.0349685.r002
5 May 2026 Decision Letter - Guangyin Jin, Editor Evolving Graph Attention Networks for Dynamic Link Prediction PONE-D-26-01462R1 Dear Dr. Shan, We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements. Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication. An invoice will be generated when your article is formally accepted. Please note, if your institution has a publishing partnership with PLOS and your article meets the relevant criteria, all or part of your publication costs will be covered. Please make sure your user information is up-to-date by logging into Editorial Manager at Editorial Manager® and clicking the ‘Update My Information' link at the top of the page. For questions related to billing, please contact billing support. If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org. Kind regards, Guangyin Jin Academic Editor PLOS One Additional Editor Comments (optional): Reviewers' comments: Reviewer's Responses to Questions -->Comments to the Author 1. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation.--> Reviewer #1: All comments have been addressed ******** -->2. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. --> Reviewer #1: Yes ****** -->3. Has the statistical analysis been performed appropriately and rigorously? --> Reviewer #1: Yes ****** -->4. Have the authors made all data underlying the findings in their manuscript fully available? The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.--> Reviewer #1: Yes ****** -->5. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.--> Reviewer #1: Yes ****** -->6. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)--> Reviewer #1: The authors have fully addressed the concerns raised in the first round of review, and in several places they have gone beyond what was strictly required. The revisions are substantial, technically aligned with the earlier feedback, and clearly visible throughout the updated manuscript. On the question of novelty, the relationship to EvolveGCN, and the earlier inconsistency regarding the Transformer component, the authors now provide a much clearer and more accurate framing of their contribution. They reposition the work around an anisotropic weight‑evolving mechanism rather than a Transformer‑based approach, and this reframing is consistently reflected across the revised text. The manuscript now includes a direct architectural comparison with EvolveGCN in both the Related Work and Methods sections, making the distinction between isotropic convolutional evolution and the proposed anisotropic attention evolution explicit. The authors also clarify that the current implementation relies on an LSTM controller, remove the misleading Transformer emphasis from the title and narrative, and appropriately move any Transformer‑related discussion to the Future Work section. These changes fully resolve the conceptual issues raised in the initial review. Regarding the request for variance reporting and statistical significance, the authors have implemented all necessary improvements. They re‑ran the experiments using ten random seeds, updated the tables to report mean and standard deviation, and performed paired t‑tests against the strongest baseline for each dataset. Statistically significant improvements are clearly indicated, and the Results section now includes a concise description of the experimental protocol and significance testing. These additions provide the robustness and transparency that were previously missing. The third point concerned computational complexity, interpretability, and overall readability. The revised manuscript now includes a theoretical complexity analysis, an empirical runtime comparison, and a dedicated interpretability discussion illustrating how the evolving attention weights capture temporal relational drift. The presentation has also improved: the structure is clearer, the terminology is more consistent, and the methodological explanations are better integrated. Overall, the authors have addressed all major points from the first review in a thorough and convincing way. They have corrected the earlier conceptual inconsistencies, added the missing analyses, and strengthened the clarity and rigor of the manuscript. In its current form, the paper is suitable for publication. ****** -->7. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.--> Reviewer #1: No ******** https://doi.org/10.1371/journal.pone.0349685.r003
Formally Accepted
Acceptance Letter - Guangyin Jin, Editor PONE-D-26-01462R1 PLOS One Dear Dr. Shan, I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS One. Congratulations! Your manuscript is now being handed over to our production team. At this stage, our production department will prepare your paper for publication. This includes ensuring the following: * All references, tables, and figures are properly cited * All relevant supporting information is included in the manuscript submission, * There are no issues that prevent the paper from being properly typeset You will receive further instructions from the production team, including instructions on how to review your proof when it is ready. Please keep in mind that we are working through a large volume of accepted articles, so please give us a few days to review your paper and let you know the next and final steps. Lastly, if your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org. You will receive an invoice from PLOS for your publication fee after your manuscript has reached the completed accept phase. If you receive an email requesting payment before acceptance or for any other service, this may be a phishing scheme. Learn how to identify phishing emails and protect your accounts at https://explore.plos.org/phishing. If we can help with anything else, please email us at customercare@plos.org. Thank you for submitting your work to PLOS ONE and supporting open access. Kind regards, PLOS ONE Editorial Office Staff on behalf of Dr. Guangyin Jin Academic Editor PLOS One https://doi.org/10.1371/journal.pone.0349685.r004

Open letter on the publication of peer review reports

PLOS recognizes the benefits of transparency in the peer review process. Therefore, we enable the publication of all of the content of peer review and author responses alongside final, published articles. Reviewers remain anonymous, unless they choose to reveal their names.

We encourage other journals to join us in this initiative. We hope that our action inspires the community, including researchers, research funders, and research institutions, to recognize the benefits of published peer review reports for all parts of the research system.

Learn more at ASAPbio .