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
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: Satisfactory
Overall presentation: Average
Detailed Comments:
This review is in response to the authors' revision. I felt, on the whole, that the authors did not address my comments adequately. I also would recommend that the authors provide a point-by-point letter addressing comments - this would also have helped them adequately address my concerns (in my review, I listed numbered points that the authors did not reference in their response). I will use that same numbering system below for my justification for this rejection:
1. I asked for a unified example. While the authors did improve the figure, they did not integrate an running example in the text to make their ideas concrete. At each subsection of section 4, when they introduced a concept, they could have couched it in terms of a concrete example (e.g., based on Figure 2) - but they did not do this. This was my top concern for the authors, as this is not a typical paper for NAI (and does not have any neural component) - so it needed a strong use-case to both concretize and convince the reader of its utility. However, this is lacking.
2. The XUI components in section 3 were again absent from section 4 - which makes the paper disjoint. I did not see anything from the authors addressing this point.
3. The discussion on the selection of the XUI components was added, but it is very weak. The authors state it was based on a frequency of 1/2 to 1 yet some selected components were outside this range (2) and others not selected (at least 6) were in this range - so this was quite weak.
4. The authors did expand their discussion of user feedback in section 6.1.
5. The ML feedback mechanism was still lacking - and there is no use of details about ILP (though in sec 4.1 we are referred to ref 16, which appears to be a dissertation written in German - I think as NAI is an English-language journal, it would be important to include the relevant technical details, at least as an appendix to be self-contained). The lack of details on the ML component is another important reason to reject the paper from NAI - I had previously favored the paper despite not having a neural component on the assumption that the authors could provide more information on the ML interaction with the knowledge-graph based component and highly the UI aspects. The authors really needed to make a stronger case on the relationship between their specific use of ML, hybrid AI, and the UI aspects - which was not the case here.
6. There were minor comments dealing with typos.
On the whole, I think the authors could have an interesting contribution. I did have some reservations about topicality of NAI in my first review, but had hoped the feedback I gave the authors could have made it interesting enough. However, I do not believe that the authors have addressed enough key points to make this suitable for the NAI audience, and I would recommend that they move on to a different venue.
I would encourage the authors to address some of the above comments in a revision to a different journal, perhaps one that does not have the same focus as NAI. They could include a copy of this review in such a submission.
