Review Comment:
My opinion about the paper is that it has not received enough attention to be at its best.
It could be accepted given the potential of the topic.
However, it would be better to fully address the issues raised in the reviews before publishing it.
The paper has implemented various changes asked by the reviewers. It is clear from the changes that the author aimed at addressing more directly the relation with the semantic web.
However, the paper remains basically the same.
The 'Potential for Semantic Web' in which I asked for more material have rather been reduced, at least not increased; the paper is still not particularly concrete, at least not significantly more than in the previous version.
Table 1 has been added, and commented in Section 1.4. However this description does not tell the story: nothing explains how this works and how this take advantage of semantic web resources. There is more information in the table, but it is not very articulated. There is a lot of latitude to explain that mining the semantic web of recipes, it is possible to understand how some ingredient may be used in place of another and that this provides a differential effect, that doing it in another recipe, as long as no counter-indication is found in health data sources, may yield new surprising meals. Adding to that, add which techniques may be used for performing this: data mining/machine learning against web or semantic web recipes, querying remote resources or applying constraints to the search space, and finally blending or reasoning by analogy.
That would make an example (it can even be shorter) and tell how this could be seen.
Same thing about the robot scientist: we are told what it does, not how. That should be done with a few sentences, maybe extrapolated to take advantage of the semantic web (show the vision).
The design example starts better. However, turning to X-Y-Z within a few lines transforms the problem into a theoretical configuration problem. Apparently, the challenges are solved by 'so-called generative design' and since it is not a part of the semantic web, or the paper does not explain how it is related, this does not seem relevant.
Another place where an effort has been produced is the new 'Research direction' section, though it also use some materials that were present in the previous version.
However, it is unclear to me what this section addresses (see below).
Something which is unclear is the relationship between semantic web and artificial intelligence (I assume that this is the interpretation of the AI acronym which is in the title but not expanded) as well as computational creativity. The abstract mentions AI creativity, the title 'Creative AI', but it is also unclear if it is the same thing as computational creativity or if it is a subpart of it. At least, for me, the boundaries are blurred and it is difficult to understand, what is really specific to semantic web, as well as what is really new, e.g. 'scientific discovery' has been a topic of artificial intelligence since at least the 80's.
I certainly raised this point too shyly in the first version, but I actually did it with the 'we explored... creativity' comment.
This problem affects Table 1 in which all 'topics' above creativity types are supposed to be 'semantic web topics' which include multi-agent systems, constraint-based reasoning, for instance.
This also applies to the 'Research directions' section. It begins by a motivation by opposition to the previous section. The previous section was 'research questions regarding the potential of the semantic web wrt creativity', the new one is 'research directions wrt artificial intelligence'. It is unclear if the strong opposition is questions/directions or semantic web/artificial intelligence. Moreover, it is again surprising that it is about artificial intelligence and not semantic web. Actually, this section is also about semantic web, which make this introduction even more puzzling.
These issues are likely due to misunderstandings, but it would be better to dissipate those.
Details:
On several occasions the paper mentions 'and see what emerges'. This is a bit loose as a methodology.
Concerning recipes, there exists a case-based reasoning challenge which uses many recipes expressed in RDF: T{a}+ble (I can only find it in proceedings not the source itself).
p1, col2, l33: 'AI summer' of KR is a strange expression because KR is part of AI and the 'AI winter' expression means the winter of AI, hence this one would be the summer of AI of KR...
p1, col2, l48: by the users -> by users (or by their users)
p2, col1, l17-18: are starting to be addressed by AI: this has been quite some time.
p2, col1, l34-35: 'many researchers use the term 'concept' to refer to a range of things: ...' [reference needed] The point of this sentence is unclear.
p3, col1, l24: Best -> The best
p4, Table 1: It is difficult to read. Suggested improvement, write that there are three use cases in the caption, so that the reader immediately see that they are in column). Otherwise, since there are three category of information mentioned in the caption, they can only be in column. Use double \hline to separate these three categories.
p4, col2, l43-47: I would suppress the 'by authors' references are sufficient and the domain makes the point.
p6, col1, l11-13: 'Linked data ... coincide with the model of BisoNets as k-partite heterogeneous information networks': they do not really seem to coincide since the RDF model is not particularly 'k-partite' as far as I can tell.
p7, col1, l50-51: transferring and the use
p7, col2, l13-15: the analogy between CBR and design patterns is strange because design patterns are abstractions, though the CBR motto is to do instance analogy without abstraction.
|
