Graph Neural Networks (GNNs) have become essential in interpreting relational data across various domains, yet, they often struggle to generalize to unseen graph data that differs markedly from training instances. In this paper, we introduce a novel framework called General Retrieval-Augmented Graph Learning (RAGraph), which brings external graph data into the general graph foundation model to improve model generalization on unseen scenarios. On the top of our framework is a toy graph vector library that we established, which captures key attributes, such as features and task-specific label information. During inference, the RAGraph adeptly retrieves similar toy graphs based on key similarities in downstream tasks, integrating the retrieved data to enrich the learning context via the message-passing prompting mechanism. Our extensive experimental evaluations demonstrate that RAGraph significantly outperforms state-of-the-art graph learning methods in multiple tasks such as node classification, link prediction, and graph classification across both dynamic and static datasets. Furthermore, extensive testing confirms that RAGraph consistently maintains high performance without the need for task-specific fine-tuning, highlighting its adaptability, robustness, and broad applicability.
Paper
Similar papers
Peer review
Summary
The paper propose a Retrieval-Augmented method to further assist graph in context learning. With the advancement of graph In-context learning and graph prompting, RAG is a natural technique that could be built upon them. The main contribution of the paper is to develop such a RAG pipeline on graph learning scenario by defining the graph database, graph retrieval pipeline, and training and inference of the model.
Strengths
The idea is novel in the field. And the implementation is clearly stated. Experiments show the effectiveness of the propose method.
Weaknesses
While the idea is novel. It is unclear why it works and how it is in par with RAG in NLP/CV. Specifically, it not sure how the “generation” in RAG works in the proposed framework. And the author did not explain why the proposed method would work — what pattern the model has learned from the retrieved graph pattern? Based on my limited knowledge, the key contributions of the paper would be how to construct the query data base and how to retrieve them. The methodology part doesn’t explain through why it’s design like so. Motivation of the adopted construction is missing.
Questions
1. Why do we need Toy Graphs Augmentation Strategy in 4.2 2. Why noise-based graph prompting is required? 3. If the model is trained to able to utilize retrieve data/embedding, is it still able to perform zero-short inference?
Rating
6
Confidence
3
Soundness
3
Presentation
2
Contribution
3
Limitations
None
Summary
The paper proposes RAGraph, a framework that enhances GNNs with RAG. RAG allows GNNs to utilize unseen data by retrieving relevant information. Extensive experimental results show the effectiveness of RAGraph.
Strengths
S1. A general and flexible framework. S2. Extensive experiments on various tasks (both node, edge, and graph level) S3. Structured and comprehensive discussion (e.g. details of experiments and comparison with PRODIGY).
Weaknesses
W1. The diversity of datasets needs to be improved. For link prediction, the paper is evaluated on mostly e-commerce data, where knowledge graph tasks should also be considered. For node classification, both homophilic datasets e.g. Cora, Arxiv; and heterophilic datasets should be evaluated. W2. Lack of large scale experiments, e.g. those OGB node/link/graph datasets. W3. The model seems complicated with a big hyperparameter search space, e.g. the weights of time, structure, environment, and semantic similarities in Eq. 1.
Questions
NA
Rating
5
Confidence
3
Soundness
3
Presentation
3
Contribution
3
Limitations
NA
Response to authors
I've carefully read the rebuttal and appreciate the authors' efforts in addressing the weaknesses I identified. Regarding W1, while I acknowledge the additional experiments, since academic graphs like MAG are known to be homophilic, the experiments still predominantly reflect a homophilic setting. For W2, I appreciate the inclusion of additional experiments, and the results are indeed convincing. However, for W3, the authors did not directly address my concern. Given that the major concerns are still valid and my scores are already positive. I'm keeping my original score.
Kindly Requesting for Review of Our Rebuttal and Reconsider Our Rating
Dear Reviewer sivC, We would like to express our gratitude for taking the time to review our paper and provide us with valuable comments. We understand that you have busy schedules and apologize for any inconvenience caused by our urging letter. **With only 7 hours left in the discussion period, we hope to receive feedback from the reviewers: did our response address your concerns, and what can we do to further improve our score?** Thank you for your consideration, and we look forward to receiving your feedback soon. Best regards, Authors of the paper 8566
Response to W1: The experiments still predominantly reflect a homophilic setting.
Dear Reviewer sivC, Thank you for your careful reading of our rebuttal and for providing constructive feedback. We greatly appreciate the time and effort you have invested in evaluating our manuscript. Regarding Weakness 1, it appears there may have been a misunderstanding. We adhere to the definition of a heterophilic graph as outlined in the HAN [1] in Introduction **<Heterogeneity of graph>**, which describes it as "**a heterogeneous graph is a special kind of information network containing either multiple types of entities or multiple types of links**". In our rebuttal experiments, the Microsoft Academic Graph (MAG) [2] contains **four types of entities**: papers, authors, institutions, and fields of study, as well as **four types of relationships** that connect these entities: an author is "affiliated with" an institution, an author "writes" a paper, a paper "cites" another paper, and a paper "has a topic of" a field of study. **Thus, the MAG dataset fulfills the criteria for heterophilic settings**. Additionally, in our experiments, we utilized dynamic bipartite graphs such as TAOBAO, KOUBEI, and AMAZON, which also contain **two types of entities** and are classified as **heterogeneous graphs**. Furthermore, we also conducted experiments on a more heterogeneous knowledge graph, ICEWS, as part of our rebuttal. Moreover, we included additional experiments using datasets with more widely recognized **heterogeneous graphs**, as employed in HAN [1], specifically IMDB (three types of entities and two types of relationships) and ACM (three types of entities and two types of relationships). Specifically, we strictly follow the experiment configuration of HAN and leverage HAN as backbone model: | **Methods** | **ACM Micro F1** | **ACM Macro F1** | **IMDB Micro F1** | **IMDB Macro F1** | |-----------------|------------------|------------------|-------------------|-------------------| | **Backbone (HAN)** | 87.48 | 87.50 | 57.93 | 53.82 | | **PRODIGY/NF** | 86.73 | 86.69 | 57.24 | 52.36 | | **PRODIGY/FT** | 87.61 | 87.63 | 58.20 | 53.97 | | **RAGraph/NF** | 87.58 | 87.54 | 58.16 | 53.95 | | **RAGraph/FT** | 87.77 | 87.80 | 58.49 | 54.30 | | **RAGraph/NFT** | **88.14** | **88.16** | **58.55** | **54.42** | The results from these datasets also **demonstrate the effectiveness of our approach**. We will add these experiment results to the final version. [1] Wang X, Ji H, Shi C, Wang B, Ye Y, Cui P, Yu PS. Heterogeneous graph attention network. In WWW 2019. [2] Kuansan Wang, Zhihong Shen, Chiyuan Huang, Chieh-Han Wu, Yuxiao Dong, and Anshul Kanakia. Microsoft academic graph: When experts are not enough. In Quantitative Science Studies 2020. Best regards, Authors of the paper 8566
Response to W3: The model seems complicated with a big hyperparameter search space.
Dear Reviewer sivC, Thank you for your careful reading of our rebuttal and for providing constructive feedback. We greatly appreciate the time and effort you have invested in evaluating our manuscript. For Weakness 3, we apologize for not providing a direct response in our initial rebuttal. We have now prepared a more detailed analysis. Firstly, the extensive hyperparameter search space in RAGraph is not merely a complexity for its own sake; it also **facilitates the integration of more dimensions of information**, such as $w_1\sim w_4$ balancing the importance of different similarities and $\gamma$ weighing the significance of task-specific output vectors versus hidden embeddings. This, in turn, **enhances RAGraph's performance** across a range of tasks, including node-level, edge-level, and graph-level scenarios. By accommodating a comprehensive search space, our model can effectively capture the intricate relationships between nodes and edges, leading to superior performance across diverse tasks. Secondly, aside from those sensitive parameters $k, topK$ which we have already supplemented in sensitivity experiments in Figure 3, and apart from the augmentation number $K=50$, we also attempt to remove $\alpha, \lambda, \gamma, w_1\sim w_4$, and conduct additional experiments to assess the impact of these hyperparameters, denoted as "w/o hyperparameters": | Methods | PROTEINS (5-shots) | PROTEINS (5-shots) | ENZYMES (5-shots) | ENZYMES (5-shots) | TAOBAO | TAOBAO | AMAZON | AMAZON | | -------------------------------- | ------------------ | ------------------ | ----------------- | ----------------- | ------ | ------ | ------ | ------ | | | Node Level | Graph Level | Node Level | Graph Level | Recall | nDCG | Recall | nDCG | | RAGraph/NF | 40.27 | 55.16 | 48.41 | 25.83 | 22.13 | 21.40 | 18.11 | 05.94 | | RAGraph/NF + w/o hyperparameters | 39.86 | 54.82 | 47.39 | 25.50 | 22.08 | 21.32 | 17.95 | 05.88 | The results demonstrate that even **without these hyperparameters, RAGraph still exhibits strong performance**, with only a slight decrease compared to the tuned settings. This robustness underscores the strength of our model’s design, highlighting its ability to maintain high performance across varying conditions. Consequently, **these hyperparameters perform well under a broad range of search spaces**. This also further validates that **when employing optimization algorithms like Bayesian optimization, RAGraph does not require as extensive a search space** compared to traditional method like grid search. Thirdly, the challenge of dealing with a large search space is inherent and formidable in most deep-learning models. In order to expedite the process of finding the most suitable hyperparameters while reducing search space, we utilized Optuna [2], an advanced hyperparameter optimization framework. Optuna employs an efficient sampling algorithm that **significantly reduces the size of the search space** by intelligently selecting candidate hyperparameters based on their potential impact on model performance. Unlike grid search or random search, Optuna dynamically prunes unpromising trials, focusing on the most promising regions of the search space, thus accelerating the overall training and improving the efficiency of our experiments. Lastly, to empirically validate the effectiveness of Optuna, we conducted experiments using ACM dataset [1] and set hyperparameter $\alpha, \lambda, \gamma, w_1\sim w_4$ between range [0,1]. We compared the results of Grid Search and Optuna under the same settings. For Grid Search, with each parameter having a search step size of 2, we explored a total of 2^7 combinations. With the HAN backbone taking nearly 1.2 seconds per epoch and running for 10 epochs, the total time for Grid Search was **[27 minutes and 39 seconds]**. In contrast, using Optuna with 100 trials under the same settings, the total time was reduced to **[4 minutes and 28 seconds]**. **This empirical analysis demonstrates that Optuna significantly reduces the search space of RAGraph and accelerates the hyperparameter optimization process**. We hope these additional points will further clarify your questions. If you have any further questions, please do not hesitate to contact us. We are committed to improving the manuscript and ensuring that all concerns are adequately addressed. Thank you again for your valuable suggestions and for maintaining a positive assessment of our work. [1] Wang X, Ji H, Shi C, Wang B, Ye Y, Cui P, Yu PS. Heterogeneous graph attention network. In WWW 2019. [2] Takuya Akiba, Shotaro Sano, Toshihiko Yanase, Takeru Ohta, and Masanori Koyama. 2019. Optuna: A Next-generation Hyperparameter Optimization Framework. In KDD. Best regards, Authors of the paper 8566
Summary
This paper aims to leverage the retrieval-augmented generation method to improve the generalisation capability of pretrained graph neural networks (GNNs) to unseen data. To this end, a framework named RAGRAPH is proposed. RAGRAPH first constructs a toy graph vector library (key-value pairs) by chunking from resource graphs, where the keys store some features of the master nodes of the toy graphs and the values are the corresponding node representations and task-specific outputs. When making predictions for unseen nodes/graphs, RAGRAPH retrieves top-$K$ most similar toy graphs from the vector library and augment the nodes/graphs with the retrieved toy graphs. Moreover, to mitigate the challenge of retrieving related but irrelevant graphs, the paper also proposes a prompt tuning method to finetune the pretrained GNNs. The main idea of this prompt tuning method is to explicitly inject some noises into the retrieved toy graphs during the finetuning stage.
Strengths
1. The paper propose a novel framework RAGRAPH, which leverages RAG methods to enhance the generalisation capability of pretrained GNNs. 2. The paper is well-structured and easy to follow. 3. Experimental results on three graph learning tasks, i.e., node classification, graph classification and link prediction, demonstrate the effectiveness of the proposed RAGRAPH framework. 4. The code is provided for reproducibility.
Weaknesses
1. The RAGRAPH framework is quite complicated, consisting of several steps in the pipeline. However, some design choices within the pipeline are not properly justified. For example: (1) When constructing the toy graph vector library (section 4.1), RAGRAPH uses an inverse importance sampling strategy and some augmentation strategies to sample toy graphs to be stored in the library. It is currently unclear if we actually need all these steps to achieve satisfactory performance and how different choices would affect the final performance. (2) In the toy graph retrieval process (section 4.2), RAGRAPH uses four different sources of similarities, i.e., time, structure, environment and semantic similarities, to compute the similarities between the centre node in the query graph and the master nodes in the toy graphs. It is also unclear if all these four similarities all contribute to the performance improvement. Moreover, it is unclear how to set the different weights for these four similarities. (3) In the knowledge fusion layer (section 4.3.2), RAGRAPH fuses the Decoder output with the aggregated task-specific output vector, which is obtained from the retrieved toy graphs, to make final prediction. However, it is unclear if this fusion method can outperform using the Decoder output only or the aggregated task-specific output vector only. Therefore, I would recommend conducting some ablation studies to justify the specific choices in the RAGRAPH framework. 2. There are lots of hyperparameters in the RAGRAPH framework, such as the balance weight $\alpha$, the scaling constant $K$ in toy graph augmentation, the hyperparameters in different augmentation methods, the reweighting hyperparameter $\gamma$, ect. However, it is unclear how to set these hyperparameters to obtain the best performance. Additionally, the specific settings of these hyperparameters used in the experiments are also absent from the paper. 3. In line 332-333, the paper states that “RAGRAPH outperforms all the baselines across the three graph tasks”. However, this is inaccurate given that RAGRAPH does not always obtain the best performance in the link prediction task (see Table 2). 4. I am confused about the analyses in line 348-353. The paper states that “PRODIGY/NF and RAGRAPH/NF are inferior to Vanilla/NF, indicating that …”. However, the experimental results in Table 1 and Table 2 actually indicate that both PRODIGY/NF and RAGRAPH/NF outperform Vanilla/NF in almost all the cases.
Questions
1. What are the justifications for the specific design choices in the proposed RAGRAPH framework? Can they all contribute to the performance improvement? 2. How to set the hyperparameters in RAGRAPH? It is unclear whether there is a general setting that can achieve decent performance across different datasets, or if the hyperparameters need to be adapted for each specific dataset. 3. Can you provide some qualitative analyses to provide some insights regarding why retrieving toy graphs can help improve the performance?
Rating
6
Confidence
4
Soundness
2
Presentation
3
Contribution
3
Limitations
The authors have discussed limitations and broader impacts in their paper.
Kindly Requesting for Review of Our Rebuttal
Dear Reviewer 3qQ8, We would like to express our gratitude for taking the time to review our paper and provide us with valuable comments. We understand that you have busy schedules and apologize for any inconvenience caused by our urging letter. With only 1 day left in the discussion period, we hope to receive feedback from the reviewers: did our response address your concerns, and what can we do to further improve our score? Thank you for your consideration, and we look forward to receiving your feedback soon. Best regards, Authors of the paper 8566
Thanks the authors for providing the responses, which address my major concerns regarding the effectiveness of each components. I have updated my score accordingly.
Thank you!
Dear Reviewer 3qQ8, We are delighted to hear that your concerns have been satisfactorily answered! Thank you once again for recognizing the contributions of our work. Best regards, Authors of the paper 8566
Summary
### Summary: The paper presents RAGRAPH, a pioneering framework that integrates Retrieval-Augmented Generation (RAG) with pretrained Graph Neural Networks (GNN) to bolster their generalizability on unseen graph data. The authors construct a toy graph vector library capturing key attributes, which aids in the retrieval of analogous graphs to enrich the learning context during inference. RAGRAPH demonstrates superior performance over existing methods in tasks such as node classification, link prediction, and graph classification, showcasing its adaptability and robustness across diverse datasets without the need for fine-tuning. ### Contributions: 1. The introduction of RAGRAPH, the first of its kind to merge RAG techniques with pre-trained GNNs, offering a significant leap in model generalization. 2. The work creates a novel library that stores key graph attributes, facilitating the retrieval of similar graphs to enhance learning. 3. The proposed RAGRAPH outperforms state-of-the-art methods across multiple graph learning tasks, highlighting its effectiveness. Besides, the framework maintains high performance across different tasks and datasets, emphasizing its robustness.
Strengths
1. RAGRAPH's strength lies in its innovative use of retrieval mechanisms to enhance graph learning tasks. By retrieving and integrating external graph data, it effectively broadens the context for learning, leading to improved performance and generalization. 2. The whole process is reasonable and solid, with clear presenation. 3. The ability to maintain high performance across various datasets without task-specific fine-tuning is a significant strength. This adaptability makes RAGRAPH a robust choice for diverse real-world applications. Moreover, RAGRAPH's design as a plug-and-play module allows for seamless integration with pre-trained GNNs.
Weaknesses
I have only one concern about the cons of this work: The major weakness is the difficulty to construct and maintain high-quality graph vector base for different tasks, since according to my experiences, the performance is highly dependent on the quality of the external knowledge. This may influence the application of the proposed method in more diverse real-world cases.
Questions
See the concern above.
Rating
6
Confidence
4
Soundness
3
Presentation
3
Contribution
3
Limitations
NA
Kindly Requesting for Review of Our Rebuttal
Dear Reviewer va3W, As the discussion phase is ending tomorrow, we extend our gratitude once more for your valuable and insightful comments! We have provided careful and detailed responses to all your questions. It would be greatly appreciated if you could kindly let us know whether we have answered all your questions. Please also kindly let us know if you have any further questions, and we would like to try our best to resolve them before the deadline. Best regards, Authors of the paper 8566
Dear authors, Thanks for your rebuttal. I think your replies shows the efforts **needed** to address the challenge I proposed in my review. It does make sense to me. I believe more efforts will be needed when applying to problems beyond academic graphs. Therefore, I would like to keep my rating unchanged. Best, Reviewer
Response: The experiments are still limited in academic graphs.
Dear Reviewer va3W, Thank you for your careful reading of our rebuttal and for providing constructive feedback. We are delighted to know that **your main concerns have been resolved**! We greatly appreciate the time and effort you have invested in evaluating our manuscript. Regarding your remaining concern, we believe there may have been a misunderstanding. In our experiments, we utilized a diverse set of datasets, including academic datasets such as PROTEINS, BZR, Cora, IMDB, ACM and large-scale datasets like MAG, supplemented by Rebuttal. In addition to these, we employed **several industrial graph datasets**, including TAOBAO (from the largest online shopping platform in China, https://ali-home.alibaba.com/en-US/about-alibaba), KOUBEI (from the large consumption platform, https://www.koubei.com/), and AMAZON (from the largest online shopping platform in USA, https://www.amazon.com/). On large-scale dynamic recommendation graph data—specifically TAOBAO, KOUBEI, and AMAZON—our performance significantly **outperforms that of GraphPro** [1]. **GraphPro has already been successfully deployed on a large-scale online platform for dynamic streaming data (as referenced in Section 4.5 of GraphPro)**, where it has achieved notable improvements in CTR prediction performance. In contrast to GraphPro, our algorithm **not only delivers superior performance but also offers faster inference speed**. To prove this point, we conducted training and inference on 204,168 nodes in the dynamic TAOBAO dataset, and calculated their time consumed under the same settings: | **Aspect** | **GraphPro** | **RAGraph** | **Efficiency Improvement** | |--------------|-----------------|-------------|----------------------------| | **Training** | 21.87s | **19.14s** | **1.143 $\times$** | | **Inference**| 12.11s | **10.08s** | **1.201 $\times$** | As illustrated in the Table, this advantage stems largely from **RAGraph's k-hop subgraphs—query graphs—for inference**, whereas GraphPro requires inference over the entire graph. Furthermore, RAGraph's capability to directly retrieve historical evaluation data **without the need for model fine-tuning** enables it to achieve superior results, as demonstrated by the comparison between GraphPro/NF and Vanilla/NF in Table 2 in the main Paper. **Therefore, in terms of performance and efficiency, our model outperforms GraphPro which has been deployed online on these three large-scale dynamic industrial graph datasets** In addition, our **toy graph base can also be cached to accelerate deployment in the industrial area**. Given the proven success of GraphPro in online deployment, we are optimistic about the future application of RAGraph in the industrial GNN field, i.e., in the recommendation system. We also anticipate that RAGraph could potentially achieve notable results in RAG within NLP, leveraging the potential success of large graph models. At last, please also kindly let us know if you have any further questions or what can we do to further improve our score? And we would like to try our best to resolve them before the deadline. [1] GraphPro: Graph Pre-training and Prompt Learning for Recommendation. In WWW 2024. Best regards, Authors of the paper 8566
Dear authors, Sorry for the misunderstanding in the provided real-world datasets beyond academic graphs. Then the increased efforts may not be more than the proposed efforts in your rebuttal. I would like keep my rating unchanged. Best, Reviewer
Kindly Requesting for Reconsider Our Contributions
Dear Reviewer va3W, Thank you for your thoughtful feedback and for taking the time to review our rebuttal. We are delighted to know that your main concerns and the misunderstanding have been resolved! We greatly appreciate the time and effort you have invested in evaluating our manuscript. We understand that you have decided to keep your rating unchanged, and we respect your decision. However, we would like to kindly ask you to reconsider our contributions and the efforts we have put into addressing your concerns and improving our work. In our revised manuscript, we have not only clarified the use of a diverse set of datasets, including large-scale, real-world datasets from industrial applications, but we have also demonstrated significant performance improvements over existing methods. Specifically, our approach shows clear advantages in terms of efficiency and scalability, particularly in handling large industrial graphs—a crucial aspect in real-world scenarios that we believe align closely with the goals of our field as suggested by you. Furthermore, we have incorporated additional experiments and analyses to provide a more comprehensive evaluation of our method’s effectiveness. These efforts were made to ensure that our work contributes meaningfully to both the academic community and industry applications. **We understand the importance of maintaining rigorous standards, and we are committed to further refining our work based on your guidance. We would be grateful if you could reconsider your rating in light of these additional efforts and the potential impact of our contributions.** Thank you once again for your time and consideration. Best regards, Authors of the paper 8566
Dear authors, I appreciate your efforts in this work, however, according to my experience in this domain, RAGraph is not that **big and novel** idea in terms of research contributions. The major contribution of this work, in my opinion, come from the technical and experimental parts, for which I gave a weak accept. I would like keep my rating unchanged. Best, Reviewer
Thank you
Dear Reviewer va3W, Thank you for your thoughtful feedback and for taking the time to review our rebuttal. We greatly appreciate the time and effort you have invested in evaluating our manuscript. Regarding novelty, we believe that the introduction of RAGraph represents a meaningful step forward in the GNN field by innovatively incorporating external graph data to enhance generalization in unseen scenarios. While the concept might not seem groundbreaking in isolation, the framework's ability to dynamically retrieve and integrate relevant data during inference, without the need for fine-tuning, highlights its adaptability and potential for wide-ranging applications. We are particularly optimistic about its future impact, especially in areas like recommendation systems and potentially in RAG within NLP as large graph models continue to develop. **We sincerely thank you again for recognizing the strengths of our technical and experimental work, and we are grateful for your evaluation of our manuscript.** Best regards, Authors of the paper 8566
Kindly Requesting for Review of Our Rebuttal and Reconsider Our Rating
Dear Reviewer v5NH, We would like to express our gratitude for taking the time to review our paper and provide us with valuable comments. We understand that you have busy schedules and apologize for any inconvenience caused by our urging letter. **With only seven hours left in the discussion period, we hope to receive feedback from the reviewers: did our response address your concerns, and what can we do to further improve our score?** Thank you for your consideration, and we look forward to receiving your feedback soon. Best regards, Authors of the paper 8566
Gratitude for the Review Process and Feedback on Paper #8566
Dear AC, Senior AC, NeurIPS PCs, and Reviewers, I hope this message finds you well. As the review process for our paper, [RAGraph: A General Retrieval-Augmented Graph Learning Framework] comes to a close, **I wanted to take a moment to express my sincere gratitude to you and the reviewers for your time, effort, and valuable suggestions**. First and foremost, **we are very grateful to the AC for your timely reminders and for ensuring the smooth and effective progression of the entire review process**. I would also like to extend my deepest thanks to each of the reviewers. We sincerely appreciate your recognition of our work, particularly your acknowledgment of **RAGraph's novelty, as well as potential and promising applications**. We are also delighted to know that all of the reviewers' concerns have been satisfactorily addressed! Your feedback has played a crucial role in refining our research, and we are grateful for your thoughtful and constructive suggestions. In the future, we will incorporate the dataset experiments, ablation studies, and other improvements discussed during the rebuttal process into the main text. Finally, we deeply appreciate the time and effort each of you has invested in this process. Your contributions have been invaluable, and we are grateful for the opportunity to engage with such a dedicated group of professionals. Best regards, Authors of the paper 8566
Decision
Accept (poster)