By Anonymous User
Review Details
Reviewer has chosen to be Anonymous
Overall Impression: Weak
Content:
Technical Quality of the paper: Weak
Originality of the paper: Yes, but limited
Adequacy of the bibliography: Yes
Presentation:
Adequacy of the abstract: Yes
Introduction: background and motivation: Good
Organization of the paper: Needs improvement
Level of English: Unsatisfactory
Overall presentation: Average
Detailed Comments:
I thank the authors for their hard work in updating the paper and answering my comments. While this paper has improved from its previous version I think there are a few problems that still remain.
*Methodology*
I still believe that the methodology has a problem (assuming I have understood the process). The authors have confirmed they run edge2vec on the entire graph before splitting. Then, to the best of my understanding they train a model and do link prediction. If that is so, I think that information is leaked during this process.
To respond to the authors, I do not think it matters that the task is unsupervised: if information from the edges is somehow propagated to the nodes to create the embeddings now any link prediction experiment on that graph is probably going to be biased. See for example: https://stellargraph.readthedocs.io/en/stable/demos/link-prediction/node... as an example of data splitting to make the experiment unbiased (in particular the section Construct splits of the input data)
I think the authors have correctly applied the transductive splitting, but they should have taken the embeddings part into account.
*Some additional writing issues*:
* Page 2) line 18: specially => especially?
* images still look bad
* I do not think the pygeometric version mentioned in the paper is the correct one (or at least is not the one reported in the requirements file in the code).
* there is no license attached to the code, so the authors cannot say "freely accessible with an open license"
* the bibliography again contains errors: I see a few "type: article." I see pairs of ".". Moreover, papers that have been published are reported as arxiv papers (node2vec).
In general, I believe the paper is still hard to read (I sometimes think some sentences could be better put together). I think the authors have worked hard on patching some of the issues but I believe there is more work to do to improve the paper.