Reviewer 2 has confusion concerning the unifying techniques section:
**what you refer to as "unifying techniques" is relational learning, I don't see why you refer to it in such an ambiguous way. Furthermore, grouping the techniques into three is so broad and doesn't correctly represent knowledge graphs methods. For example, matrix factorization and deep learning.
Matrix factorizations such as isomap, PCA, SVD and others are not knowledge graph representation techniques, but dimensionality reduction techniques. More importantly, you don't show how they can be applied or used in the context of knowledge graphs as claimed?
The "unifying applications" section shows how these techniques are being used on knowledge graphs, but it seems to me that the review was looking for examples within the unifying techniques category.
Fix here would be to add explanation text in the unifying techniques intro to solve the above problem or possibly change the title to be something like: "techniques for relational learning:.
FYI @agitter created a way to do spellcheck in manubot using appveyor. You may want to enable it and see how well it works in this project. I think the PR is open but it may be possible to easily port it here (or simply download and run it manually in the manubot environment): manubot/rootstock#333
Reviewer 2 comments that the knowledge graph example doesn't have arrows for some of the relationships:
**Figure 1 doesn't show the relationship direction, For example, "causes", "binds" and other relations don't clearly specify the source and destination nodes which can be confusing. Ideally, a knowledge graph should show that.
Suggested fix is to include an example of another knowledge graph that has arrows.
The reviewer pointed out that many of the graph examples in the first paragraph aren't knowledge graphs. I think that the best approach to respond to that point is to make the first paragraph be about analytical methods applied to graphs. Then update the lead in to the second paragraph, where you define knowledge graphs, to make clear that that is the point at which you're transitioning to knowledge graphs.
**word2vec is a shallow neural network (one layer for projection, no activation or non-linearity is used in this layer), therefore it is deep learning, nor any of the methods which use similar techniques.
Fix here is to change the deep learning section title to some form of the following:
"Artificial neural network methods"
"Shallow network approaches"
etc.
Reviewer 2 comments that references are for graphs not necessarily knowledge graphs:
**at least references 1, 2, 5,6 and 2 are concerned with methods for graphs not necessarily knowledge graphs, in which the edge label (i.e., relation is essential to its definition).
Based on this feedback, it is imperative to be explicit on the definition of a knowledge graph and adjust text to accommodate the change.