A Knowledge Graph Embedding Model Based on Cyclic Consistency—Cyclic_CKGE
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsIn this paper, the authors introduce complete knowledge maps for medical diseases. A cyclic consistency-based knowledge graph embedding model is constructed and an intelligent medical system is developed using the proposed approach
Overall, this paper is interesting and nicely structured. I think that the contribution herein introduced may have several and useful practical implications. The paper appears to be solid as regards its basic ideas, as well as it is written by using an appropriate technical language. The notation is clearly defined and consistently used within the whole paper.
Nevertheless, there is some recommendation that can be taken into consideration to improve it.
- Authors said in the Abstract “… can predict not only the disease that the patient has, but also the complications that the disease may cause” How do you deduce complications? What method or inference mechanism do you use?
- Adding numerical results and some medical lessons learned to the abstract.
- In the Introduction section, be very detailed in the contribution of the proposed work compared to the work [26].
- This research should include a good motivation in the introduction that focused on disease-centered medical knowledge.
- How does the proposed solution-based knowledge graph fit into the intelligent medical test
- The medical data interoperability between different knowledge maps should be specified and justified. It can be also generic and interoperable.
- It would be helpful if authors provided some examples of queries in the section Experiment
- In the evaluation section, it will be good to compare the proposed approach with existing similar frameworks in the literature and based on accuracy and completeness.
Comments on the Quality of English LanguageMinor editing of English language required
Author Response
Dear reviewer:
Hello. Thank you very much for taking time out of your busy schedule to review my paper. We have revised this document with your professional comments. The modification results are as follows:
1. Principles and methods for inferring complications are added in sections 4.5 and 4.6 of this paper.
2. Section 4.5 details the important role of the rotation model in the construction of the medical knowledge map.
3. Finally, the development direction of medical knowledge map is prospected, and the shortcomings of this paper are analyzed.
4. In part of the experimental results, we respectively compared the existing models and analyzed the relevant results.
5. At the same time, we provide an example of a system in Section 5. The function of a test system is to verify the accuracy and completeness of the system.
Reviewer 2 Report
Comments and Suggestions for AuthorsThe paper presents a knowledge graph embedding model based on cyclic consistency.
1. Related work contains no comparison with the existing cyclic consistency model. Add a detailed comparison with existing cyclic models with your knowledge graph embedding model based on cyclic consistency.
2. There is no need for background and it consists of only one section
3. The model can be explained with the help of an example
4. Its better to give a name to the model instead of writing “ours” in Table 3
5. Add future work in conclusions
Comments on the Quality of English LanguageMinor English corrections are required
Author Response
Dear reviewer:
Hello. Thank you very much for taking time out of your busy schedule to review my paper. We have revised the paper through your professional review comments. The modification results are as follows:
1. The introduction of the existing circular consistency model is added in Section 3, and the new model based on the circular consistency principle is introduced in Section 4.
2. The test cases of the system are added in Section 5.3 to facilitate the intuitive perception of the system performance.
3. According to the expert review, we named the model Cyclic_CKGE.
Reviewer 3 Report
Comments and Suggestions for Authors
This paper discusses the construction of an intelligent medical system based on a medical knowledge graph. The system uses a knowledge graph embedding model based on cyclic consistency to predict missing links in the graph. It also includes a disease-centered medical knowledge map and can diagnose diseases and infer complications. The system standardizes text and uses the model to deduce relationships between disease entities. The document also presents the evaluation of the model's performance on different knowledge graph datasets and the functionality of the intelligent medical system. I found this paper interesting and the contribution of clear value. I would suggest revisions as detailed here below
1. How does the TPLinker model contribute to the construction of the intelligent medical system based on the medical knowledge graph?
2. How does the intelligent medical system handle the input of patients and their families?
3. Many recent researches used contrastive learning in medical AI. The authors should mention this in the introduction. Some references to build the related work
https://arxiv.org/abs/2310.07355
https://proceedings.mlr.press/v182/zhang22a.html
4. What are the potential applications of the constructed medical knowledge map and the intelligent medical system in the field of healthcare?
5. How does the intelligent medical system perform in terms of its functionality and usability?
Comments on the Quality of English Languageenglish is fine
Author Response
Dear reviewer:
Hello. Thank you very much for taking time out of your busy schedule to review my paper. We have revised the paper through your professional review comments. The modification results are as follows:
1. According to the review opinions of experts, the paper adds how Tplinker model helps to build an intelligent medical system based on medical knowledge graph in Section 4.5. At the same time, it analyzes how the intelligent medical system handles the input of patients and their families.
2. In recent years, many researches have adopted contrastive learning method to study medical intelligence. In section 7, we analyze the future trends of smart healthcare. What is the potential application of the constructed medical knowledge map and intelligent medical system in the medical field?
3. We have explained the function and usability of the intelligent medical system in Section 5.3 Functional test.
Reviewer 4 Report
Comments and Suggestions for Authors## Review summary
This work seems quite interesting, but I have some major concerns. First, the central section of the manuscript (i.e., Section 3) should be structured more clearly with appropriate references. Please also polish the notation! Second, the manuscript is somehow disproportionate: comprehensive introduction, related work, and background, whereas the results are presented very briefly and without proper discussion. The study uses a familiar methodology, but it lacks methodological soundness. There are significant gaps in the description and application of the methods that should be clarified and clearly explained (e.g., detailed description of a dataset, train/test split procedure). I recommend a thorough revision at this time, as some crucial methodological details have not been adequately addressed. I believe that further steps toward a more rigorous assessment would sufficiently support the authors' current findings and greatly improve the manuscript.
## Strong points
1. An intuitive and well-defined research question with clearly presented methods.
2. Comprehensive examination of different approaches to knowledge graph completion.
## Concerns/Weaknesses
1. The definition of knowledge graph should be extended according to the relevant literature (e.g., 1-3).
2. The Introduction section will benefit from a brief illustrative example of a knowledge graph.
3. I miss a discussion of the presented results. In addition, what are the limitations of your study and ideas for further work?
4. Some parts of the manuscript are confusing and do not allow the reader to replicate the results. The authors need to provide programming code. However, this work is currently not reproducible.
## Suggested references
[1] Paulheim H. Knowledge graph refinement: a survey of approaches and evaluation methods. Semant Web 2017; 8(3): 489-508. https://doi.org/10.3233/SW-160218
[2] Ehrlinger L, Wöß W. Towards a definition of knowledge graphs. In: Martin M, Cuquet M, Folmer E (eds.). SEMANTiCS (posters, demos, SuCCESS) 2016. Leipzig 2016: CEUR-WS.org; 4. https://ceur-ws.org/Vol-1695/paper4.pdf
[3] Ji S, Pan S, Cambria E, Marttinen P, Yu PS. A survey on knowledge graphs: Representation, acquisition, and applications. IEEE Trans Neural Netw Learn Syst 2022; 33(2): 494-514.
Comments on the Quality of English LanguageTypos!
Author Response
Dear reviewer:
Hello. Thank you very much for taking time out of your busy schedule to review my paper. We have revised the paper through your professional review comments. The modification results are as follows:
1. According to the expert review opinions, the paper provides test examples in section 5.3 of the system function test, and analyzes the results.
2. At the end of the paper, we analyze the shortcomings in the paper and explain the idea of further work.
Reviewer 5 Report
Comments and Suggestions for AuthorsThe work presented by the authors makes a contribution in knowledge graphs. The presented theory is generic, as are the evaluations/examples too. This part of the contribution could be sufficient for a publication.
The integration of this methodology in an application aimed to the medical domain, although treated as the focus of the work by the authors, at least in title, abstract and introduction, is not equally clear and convincing.
Detailed comments:
In page 5, start of section 3.1, we read "...Cycle-consistency (CC) was first proposed in the paper Cycle GAN (Cycle GAN)[40]...". But entry 40 in the bibliography is not related to this. In fact the paper the authors are referring to (Zhu, Jun-Yan, et al. "Unpaired image-to-image translation using cycle-consistent adversarial networks." Proceedings of the IEEE international conference on computer vision. 2017.) is not included in the bibliography.
Table 1 is confusing as it tries to include information from two tables (one on entities and one on relations) without visually separating the two. A vertical line between the second and third columns would probably not be enough, two separate tables seems a more reasonable and intuitive way to present this data.
In tables 2 and 3, it is not clear if there is a reason/meaning that everything is bold.
Table 3 (the second table 3), similarly to table 1, includes two sections of information without visually separating them, which can be confusing. A horizontal line before the second set of goals would help. A separate table might be even better.
The claim that the authors have presented a medical system is weak, mainly for the following reasons:
1. The methodology presented is generic, with no elements that are specific to the field of medicine.
2. All of the evaluation results that are presented are from generic datasets (eg wordnet) or specific to other domains (eg countries). There is no experimental application reported in the medical domain.
Since the authors seem to want to put emphasis on the medical application of their work, additional information is required on the actual medical data used, and evaluation results are needed that show how their approach produces better medical diagnostic results than other established methods.
The additional use of term *intelligent* medical application by the authors creates even higher expectations that are again not met. The feature(s) that form the intelligent aspect of the application need(s) to be presented and also thoroughly evaluated, in comparison with a similar system without the intelligent components.
Author Response
Dear reviewer:
Hello. Thank you very much for taking time out of your busy schedule to review my paper. We have revised the paper through your professional review comments. The modification results are as follows:
1. According to the expert review, "the authors' additional use of the term * smart * medical applications creates higher expectations that are not met." In contrast to similar systems without intelligent components, the characteristics that make up the intelligent aspects of the application need to be presented and thoroughly evaluated. "We added a test example of the system in the section 5.3 Medical System Functional Testing to improve the system functionality.
2. In view of the need for additional information about the actual use of medical data proposed by experts, we have also made some practical cases to supplement this, but due to the confidentiality of patient records, this paper also analyzes the shortcomings of the paper in Section 6.
Round 2
Reviewer 1 Report
Comments and Suggestions for AuthorsI found that the authors addressed most of my concerns satisfactorily, especially by offering arguments for the requested questions in the manuscript.
Author Response
Dear expert:
Thank you very much for taking time out of your busy schedule to read our paper and give comments. We are very happy for your encouragement. At the same time, we change some references and redefine the definition of knowledge graph in the introduction of this paper. Hope to make the article more clear. Finally, I wish you a happy life.
Reviewer 4 Report
Comments and Suggestions for AuthorsThe authors have put some additional effort into the manuscript, but I am not completely satisfied. The definition of the knowledge graph is still not appropriate. As I suggested in a previous round of revision, I encourage authors to study papers I recommend there. I highly recommend the consolidation of sections 6 and 7 into a singular section titled "Discussion."
Comments on the Quality of English LanguageSome level of polishing is strongly required.
Author Response
Dear expert,
Thank you very much for taking time out of your busy schedule to read our paper and give comments. We have given serious thought to your professional advice. The following modifications have been made to the paper:
(1) First of all, in response to the problem that the definition of knowledge graph is inappropriate, we have read through the three references you gave. We changed the definition of knowledge graph (marked with red notes) in the introduction of the first part of the paper by reading references. At the same time, the references were changed ([1]Paulheim, H., & Cimiano, P... (2017). Knowledge graph refinement: a survey of approaches and evaluation methods. Semantic Web, 8(3), 489-508.
[2]Ehrlinger L, Wßw. Towards a Definition of Knowledge Graphs[C]// Joint Proceedings of the Posters and Demos Track of, International Conference on Semantic Systems - Semantics2016 and, International Workshop on Semantic Change & Evolving Semantics. 2016.
[3]Ji, S. , Pan, S. , Cambria, E. , Marttinen, P. , & Yu, P. S. . (2021). A survey on knowledge graphs: representation, acquisition, and applications. IEEE Transactions on Neural Networks and Learning Systems, PP(99).)
2. In the paper, we combine the sixth and seventh sections into one section to make the structure of the paper reasonable.
Finally, thank you very much for your comments on our paper. Wish you a happy life.
Reviewer 5 Report
Comments and Suggestions for AuthorsThere is hardly any difference between the version of the manuscript I reviewed recently and the one I read today, with respects to the concerns I had.
As in the previous review I see the contribution in the theory of working with knowledge graphs and I find it sufficient in itself to merit journal publication. On the other hand, I do not see an adequate description of a medical system, thus the title, abstract, introduction and conclusions are not supported by the presented work.
Author Response
Dear expert,
Thank you very much for taking time out of your busy schedule to read our paper and give comments. We have given serious thought to your professional advice. The following modifications have been made to the paper:
(1) First of all, we thought about our paper and found your comments very valuable. Our paper really focused on the theoretical work of knowledge graph construction. Thus, the description of intelligent medical system is ignored. So we changed the title of the paper. The emphasis of this paper is to build the knowledge graph embedding model.
(2) We have changed the abstract of the paper. In the abstract, we no longer focus on smart healthcare. Instead, it focuses on describing the cyclic consistency knowledge graph embedding model. At the same time, after careful consideration, we think that if the paper only writes a knowledge graph embedding model, it is not enough to reflect its practical application value. Therefore, after completing the medical knowledge graph embedding model, we constructed the disease information database (detailed construction steps were described in sections 4.4, 4.5, and 4.6).
Finally, thank you very much for your comments on our paper. Wish you a happy life.
