Review Comment:
The article presents a stream reasoning system that integrates a standard stream reasoner with answer set programming (ASP) and thus allows for more expressive reasoning (i.e., ASP inference rules are taken into account); additionally, it demonstrates that the approach is efficient. In a nutshell, the latter is achieved by reasoning in parallel, based on a new analysis of the relations between the input data in view of the rules. The article claims to extend a conference paper in that it provides new contributions:
1. a better characterization of the algorithm applied to analyze input dependencies
2. a process that uses this input dependency graph to construct a plan for partitioning the input
3. a proof of correctness of the approach
4. experiments using a full implementation of the approach
I opt for major revision but, currently, I have strong concerns that the paper’s shape can be changed in the way necessary. The reasons are detailed below. Due to the large number of actual issues, I only list some examples (marked by *) with the single reasons.
(1) First of all, there are too many language issues, which also make it hard to understand the intended meaning sometimes:
- AE and BE style are mixed
* Abstract: analyse
* p. 2, col 1: characterization
- there are mistakes such as wrongly chosen or separated words, and strange sentence constructions:
* p. 3, col 1: Accordingly with the database terminology,
he avails of an instance of the StreamRule system
* p. 4, col 1: Fig-ure 1
* p. 4, col 2: Note that this is a different approach than distributing the processing
- there are general style issues such as missing/too much spaces:
* p. 3, col 1: b ← nota, a ← notb.
* p. 4, col 1: in following sections . The logic program
- the style could be more professional:
* p. 4, col 2: here is when the problem begins
* p. 14, col 2: In terms of comparison with similar systems, we recently became aware of two engines for complex stream reasoning we didn’t know about at the time the paper was written and the experiments performed.
(2) The technical parts (the definitions and the proof) do not comply with the general standards:
- Some definitions use notation that occurs in other definitions, but is not explicitly introduced as notation; in contrast, other definitions explicitly mention that (e.g., p. 3, col 1, Given a rule r as above, …):
* p. 4, col 1: An input window (or window), W, …
* Def. 3 refers to N_P, E_P1 which are mentioned in Def. 1, but in a way that implies that the notation is arbitrary instead of fixed
- The paper is inconsistent in the entities it defines and does not define, in terms of granularity. It defines rather general notions, such as “connected component” (in a graph), but uses much more specific notions without explanation:
* p. 3, col 2, consistency is not defined (whereas satisfaction is defined some sentences before): … is logically closed if it is consistent or contains all literals.
- Some details in the definitions do not make sense to me
* Def. 3, (i): it is not clear to me why there is a sequence p_1, …, p_n of predicates; I would the think you just need four different predicates
* Def. 3, (i): why the unique existential quantifier and not a general one
- Some definitions are confusing words (esp. “predicate” and “atom”) or missing conditions, or are inconsistent otherwise:
* Def. 1: p_u, q_u, p_d, q_d are nodes and hence predicates, and hence cannot be literals
* Def. 1, (ii), (a): only fits to the rest (e.g., Fig. 2) if p_u\neq q_u
- Proofs of Prop. 1,2:
* “recursive” rule is important but not defined
* Why is that case with recursive rules special?
* For the multiple rules case, I think, you need induction
* pre(W_i) is a set of predicates and cannot be connected components
(3) Regarding the article in general and Contributions 1.-4.:
- Large parts of the article are taken from the conference paper; e.g., the text is only slightly extended sometimes and the examples are the same.
- Further, I actually cannot make out exactly what is meant by the first two contributions. With 2., I guess, the authors mean that the article does not only contain a textual description of the approach but a more formal presentation as algorithms (Algorithms 1-3). However, in my opinion, these three algorithms do not contain interesting details - the corresponding definitions and descriptions, which are also contained in the conference paper already, should explain the necessary processing sufficiently.
- Similarly, regarding Contribution 3: First of all, the proof is lacking formality and details (see the technical issues above). Second, it is not surprising, but should be (!) a simple induction proof along the definitions.
- I am no expert in the practical field, but the experiments seem to be sound, to me: they compare the system to similar, existing systems, use standard benchmarks, and consider important metrics.
(4) Related work: The paper generally describes related works sufficiently, although sometimes giving details I do not see the reason for. On the other hand, the two related approaches mentioned in the conclusions should be described in the related work section and, especially, the processing of the first, also ASP-based system, should be delimited from the approach presented in the article.
Altogether, I think, that the approach is interesting, and that the experiments alone would be sufficient to classify the article as “extension” of the conference paper. However, as mentioned above, in its current form, the submission does not comply with journal standards and needs major revision. Amongst others, I would suggest to the authors to let a native speaker proof read the paper.
|