Relgraph: A Multi-Relational Graph Neural Network Framework for Knowledge Graph Reasoning Based on Relation Graph

Round 1

Reviewer 1 Report

The paper is very well written and contributes to the KG reasoning problems,

There are, however, two elements I miss in this paper.

1. Computational complexity of your approach. How does it compare to other approaches In terms of time and/or other resources needed?

2. In section 5.1 you describe on which datasets your model performs better, but why is it? What are the characteristics of the given dataset that influence the performance of your method? In other words for what type of data would it be most beneficial to apply this model? 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Relgraph: A Multi-relational Graph Neural Network Framework for Knowledge Graph Reasoning based on relation graph

The paper introduces a novel framework called Relgraph for knowledge graph reasoning tasks, aiming to address the limitations of existing multi-relational graph neural network (GNN) models in capturing interactions between relations. The authors propose incorporating a relation graph, which explicitly models the interactions between different relations by treating relations as nodes and entities as relations. This approach aims to enhance the representation learning and reasoning capabilities of multi-relational GNNs on knowledge graphs (KGs).

Pros:

1. Novelty: The introduction of the relation graph concept is a fresh and innovative approach to modeling predicate interactions in KGs. By explicitly representing the relationships between relations, the Relgraph framework addresses a significant limitation of existing multi-relational GNNs, which often assume independent impacts of relations on entities.

2. Versatility: The Relgraph framework is designed to be versatile and can seamlessly integrate with various traditional representation learning algorithms, such as TransE and RotatE. This flexibility allows the framework to be applied to a wide range of KG reasoning tasks and models.

3. Empirical evaluation: The authors have conducted rigorous experiments on benchmark datasets (WN18RR, FB15K-237, and UMLS) and a real-world drug repurposing task. The results demonstrate the effectiveness of the Relgraph framework in improving the performance of knowledge graph reasoning models, particularly on datasets with numerous predicates.

4. Analytical experiments: The paper includes a thorough analysis of various hyperparameters and settings, providing valuable insights into the model's behavior and performance under different conditions.

Cons:

1. You should take care of punctuation marks for equations. There are absent commas and periods at the end of equations through the whole document.

2. At the beginning of lines 236 and 238 we can see commas, but they should be placed at the end of equations above these lines.

3. It can be seen that abbreviation GNNs starts at line 27, but it starts again for “Graph Neural Network” at line 80. It seems that one time at line 27 is enough.

4. In line 212, the "where" after the equation does not begin with a paragraph, but, for example, in line 193 or 174, the sentences come from a paragraph. Similarly, the sentences in lines 240, 222, 205 do not start with a paragraph. It is necessary to revise the text again and harmonize these aspects.

5. Sometime authors write bold titles and after it text from the new paragraph. Sometimes they are not. As an example, we can compare lines 305-306, 311, 317, 328, etc. It is important to make these moments consistent.

Clarity and Organization:

1. The paper is generally well-written and organized, with a clear introduction, related work, methodology, results, and discussion sections.

2. The use of illustrative examples and diagrams (such as Figure 2) helps in understanding the proposed framework, but additional visual aids or step-by-step walkthroughs could further enhance the clarity of the methodology.

3. The authors could discuss potential extensions or variations of the Relgraph framework, such as incorporating additional types of relationships or exploring alternative attention mechanisms.

Overall, the Relgraph framework proposed in this paper presents a promising approach to enhancing knowledge graph reasoning by explicitly modeling interactions between relations. The authors have demonstrated the framework's effectiveness through comprehensive experiments and provided valuable insights into its behavior and performance. I can recommend this article after taking into account the corresponding comments. 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Please see the attachment.

Comments for author File: Comments.pdf

I think the English language used in this manuscript is fit for a research journal, as only a couple of mistakes have been found throughout the paper

Author Response

Please see the attachment.

Author Response File: Author Response.pdf