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.
The authors have revised the manuscript by improving the quality of figures and adding the new section of Discussion. I think the manuscript could be accepted in the current form.
[# PeerJ Staff Note - this decision was reviewed and approved by Jyotismita Chaki, a PeerJ Section Editor covering this Section #]
Dear authors,
Thanks for your revisions.
However, after my own evaluation of the revised version of the manuscript, I think the following revisions should be further made.
After that, I will be glad to accept the manuscript.
1) the improvement of the figures
In high-level journals, the quality of figures should be also very high.
Specially, the figures should be very clear, which are formatted or prepared in vector-based illustrations, such as in the formats of .eps, .emf. or .svg.
In the current form, the figures are not clear enough. I think they are formatted in .png or .emf. these are not suggested, please modify the figures and re-output the figures in vector-based formats, and then insert them into the main text.
You can use the software such as visio or Adobe Illustrator.
2) the section of discussion
In the current form, it lacks the section of discussion.
Here I need to clearly point out that, the section of Results and Analysis is different from the section of discussion.
Analysis is directly on the basis of the experimental results.
Discussion is much boarder, which should be focus on the advantages, shortcomings, and potential future work of the proposed method.
Please read other similar papers published in our journal to refer the preparation of Discussion.
**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.
1. please carefully check the reviewers’ comments, especially, please revise the figures to significantly improve the quality.
2. please provide more statistical tests and analysis to demonstrate that GeoDFNet is much better than baselines.
3. please enhance the contribution of the dual-fusion mechanism by explicitly comparing it to existing multimodal fusion approaches.
**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.
The manuscript presents a strong and innovative contribution to POI classification by integrating geospatial and textual features, with robust experimental validation across three datasets. The use of GAT and Transformer encoders, grounded in the Third Law of Geography, is a compelling approach that addresses a real problem in LBS applications. However, the manuscript would benefit from the following revisions:
(1) The abstract lacks a qualitative evaluation of the results, such as classification accuracy or other relevant performance metrics.
(2) The specific results of the model are very vague.
(3) The introduction could elaborate further on the specific knowledge gap being addressed. For instance, lines 75-79 mention misclassification issues (e.g., "Centennial Dragon Robe"), but a more detailed explanation of why textual methods fail in these cases would strengthen the justification (e.g., quantify the frequency of such ambiguities in real-world datasets).
(4) Figures like Fig. 6 (classification accuracy) and Fig. 7 (categorical F1) are clear, but the legends and axis labels could be more descriptive. For example, specify what "TextCNN" or "TextRNN" represent in the context of the comparison to avoid ambiguity for readers unfamiliar with these models.
(5) While the language is generally clear, minor grammatical errors exist. For example, in line 123, "These localized spatial representations are effectively integrated into model training" could be rephrased as "These localized spatial representations are effectively incorporated into the model training process" for clarity. I recommend a thorough review by a proficient English speaker or professional editing service to polish lines like 55-58, 75-79, and 103-104.
(6) The graph dataset construction (line 383-386) mentions connecting each node to its five nearest neighbors (k=5). The rationale for choosing k=5 is not provided. I suggest justifying this choice (e.g., based on spatial proximity thresholds or empirical testing) or conducting a sensitivity analysis to show how varying k impacts performance.
(7) While hyperparameters are provided (Tables 1-3), the manuscript could benefit from specifying the preprocessing steps for the textual and geospatial data (e.g., tokenization methods, normalization of coordinates). This would ensure full replicability.
(8) The results lack statistical significance tests to confirm that GeoDFNet’s improvements over baselines are not due to chance. I suggest adding a statistical test (e.g., paired t-test or Wilcoxon signed-rank test) to compare GeoDFNet’s performance against baselines like TextCNN or Transformer.
(9) Table 4 shows significant variation in category sizes (e.g., 61,050 Corporate Entities vs. 1,512 Scenic Spots). The manuscript does not discuss how class imbalance was handled (e.g., oversampling, weighted loss). Addressing this could strengthen the validity of the results, especially for underrepresented categories like Scenic Spots.
(10) The novelty of the dual-fusion mechanism (lines 124-129) could be better articulated by explicitly comparing it to existing multimodal fusion approaches (e.g., LIU Jiping et al., 2024, cited in line 169). A table summarizing differences in fusion strategies would help.
(11) Some figures (e.g., Fig. 5) are truncated in the text, and their full content is not described. Ensure all figures are fully explained in the text or captions. Additionally, Table 7 is incomplete in the provided document; ensure all metrics are fully reported.
(12) The methods section (lines 177-353) is highly technical, which is appropriate but may overwhelm readers unfamiliar with GNNs or Transformers. Simplifying some explanations (e.g., summarizing Eq. 1-6 in prose) or adding a high-level overview figure could improve accessibility.
The experimental design needs further improvement
The findings need to be further strengthened
Language needs to be improved, especially grammar
no comment
no comment
no comment
- in lines 80-83, you mentioned “The Third Law of Geography”. You should first introduce the laws of geography, then discuss why the third law is important here.
- in lines 94-95, you said "the synergistic integration of textual analysis and geospatial data remains insufficiently investigated in POI classification"; this needs more discussion. Please elaborate on why existing works are insufficient.
- The section about "Graph Neural Networks" needs more discussion. Since this is one of the central ideas of this paper, try to explain a little more about the appropriateness of the approach.
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.