Knowledge Graph Completion Based on Entity Descriptions in Hyperbolic Space

Round 1

Reviewer 1 Report

Authors present a knowledge graph completion method using entity text and entity descriptions. The idea of hyperbolic space is used to represent more compact and enrich hierarchical text representation.  The algorithmic approach is not clearly described and sometime reader has to guess about the possible processing. The motivation of transforming representation space is not justified; it also need to discuss why the improvement is there with this representation.  There are several recent papers on KG completion missing from citations.  The evaluation reported in the experimental studies also not discussed and justified clearly. I have following technical questions for the authors:

1.  Why should not the author discuss the time and space analysis for Learning DMuPR?

2.  What unbalance energy represents in this problem domain?

3. How the balance factor is learned? What are the relationship with the balance factor to the graph completion task?

4. There should be a study that justified HIT@10 selection? Did the author also evaluate different level of HITs?

5. What are the motivation of ablation studies? What and how this evaluate the completion task?

6. The role of fusion is not discussed. The factor of fusion should be study separately? The base line is not clearly set? 

 

There are some important works that the author missed. Here are a few pointers:

1. Zamini, M.; Reza, H.; Rabiei, M. A Review of Knowledge Graph Completion. Information 2022, 13, 396. https://doi.org/10.3390/info13080396

2. Ji, Shaoxiong, et al. "A survey on knowledge graphs: Representation, acquisition, and applications." IEEE Transactions on Neural Networks and Learning Systems 33.2 (2021): 494-514.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

This paper proposes Knowledge Graph Completion Based on Entity Descriptions in Hyperbolic Space. It is claimed that this method overcomes the obstacles in the last published papers in this field. The paper is well-organized and well-structured and it also lacks some details for readers to understand. I would suggest the authors address the following concerns in order to meet the publication requirement:

1. The introduction section lacks fluidity and readability. 

2. The experimental study is very trivial. 

3. Include & cite recent publications from the current year 

4. Authors are suggested to include more discussion on the results and also have some explanation regarding the justification to support why the proposed method is better in comparison to other methods. 

5. Authors are suggested to highlight their exact best results in comparison to other methods to justify the advantages of their proposed method. 

6. Explain why the current method was selected for the study, and its importance and compare it with traditional methods. The following papers are good examples:

https://doi.org/10.1016/j.eswa.2021.115406

https://doi.org/10.1038/s41598-021-90428-8

https://doi.org/10.1155/2021/5597222

https://doi.org/10.1155/2022/4703682

https://doi.org/10.1007/s10479-022-04755-8

7. About the literature and method, each paper should clearly specify the proposed methodology, novelty, and results. 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

The manuscript is consistent, the purpose and objectives of the paper are clearly presented, the research logic is clear and the results are presented in detail and compared with the results of other similar studies.

The related work section may be elaborated, with critically identifying the pros and cons of similar studies. The motivation of the proposed work has to be arrived.

Does the proposed algorithm address heterogeneous and unstructured data?

Can the algorithm be adapted to dynamic graphs?

How do you compare your approach with supervised learning approaches?

Why only two datasets are considered. Addition two more datasets could help in a better evaluation.

 

 

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

All my concerned are very well addressed. 

Reviewer 2 Report

I carefully read the revised version of this manuscript. As can be understood, my questions are clarified, and previous issues are resolved. This manuscript is suitable for acceptance.

 

Reviewer 3 Report

I appreciate the efforts made to revise the manuscript. All the points are addressed.