Paper 190 (Research track)

Practical Implementation of Contextual Reasoning on the Semantic Web

Author(s): Sahar Aljalbout, Gilles Falquet

Full text: submitted version

Abstract: Annotating RDF data with contextual information such as provenance, space, temporal validity, etc. is becoming more and more crucial on the semantic web. However, defining an efficient representation and reasoning on such meta-knowledge remains challenging. This is due to the fact that the representation of statement annotations has sometimes been thought of as an RDF data problem without consideration of the logical formalism behind it. As a consequence, the underlying data structure lacks a clear formal semantics. In this paper, we study both levels: the logical formalism and the data model. First, we define a contextualized description logic. The formalism allows the representation of several contextual dimensions and considers contexts as first-class citizens, which offers an expressive formal language that supports the representation and reasoning of contextualized knowledge. Second, we present a contextual data model based on the N-ary pattern and we study the mapping of the contextual description logics constructors and axioms onto this model. Finally, we present a simple mechanism for reasoning with contextual statements. It involves the use of SWRL and SPARQL to implement inference rules that constitute a contextualized version of the OWL-2 profile rules and we show that this mechanism is technically straightforward to implement with the existing semantic web tools.

Keywords: Semantic Web; Contexts; Contextual Ontologies; Contextual reasoning; Contextual Description Logics; Contextual OWL

Decision: reject

Review 1 (by Zhisheng Huang)

(RELEVANCE TO ESWC) Contextual reasoning is one of important topics in the Semantic Web.
(NOVELTY OF THE PROPOSED SOLUTION) The paper proposes an approach of  contextual reasoning on the Semantic Web by defining a contextualized description logic, and presenting  a mechanism using SWRL and SPARQL.
(CORRECTNESS AND COMPLETENESS OF THE PROPOSED SOLUTION) The paper make a distinction  between the logical level (description logic) and the data level (RDF). The question is why the A-box in description logic cannot serve as the data level, and why RDFS cannot serve as the logic level. Conceptually it is quite confusing.
(EVALUATION OF THE STATE-OF-THE-ART) Relevant work have been discussed.
(DEMONSTRATION AND DISCUSSION OF THE PROPERTIES OF THE PROPOSED APPROACH) The paper does not report any evaluation on the proposed approach with some data tests.
The ideas are just illustrated by some toy examples.
(OVERALL SCORE) The paper proposes an approach of  contextual reasoning on the Semantic Web by defining a contextualized description logic, and presenting  a mechanism using SWRL and SPARQL.  The paper make a distinction  between the logical level (description logic) and the data level (RDF). The question is why the A-box in description logic cannot serve as the data level, and why RDFS cannot serve as the logic level. Conceptually it is quite confusing. 
The main problem of the paper is that it does not report any experiment with some data tests for an evaluation of the proposed approach. The ideas are just illustrated by some toy examples.

Review 2 (by anonymous reviewer)

(RELEVANCE TO ESWC) The paper considers the problem of context representation in Semantic Web languages: thus, it is clearly relevant to the topics of the conference.
(NOVELTY OF THE PROPOSED SOLUTION) The paper solution does not appear to be novel: the DL semantics is explicitly derived from the works of Klarman (e.g. [10]) while the RDF context representation is an application of known design patterns. Moreover, the motivating requirements and proposed SPARQL-based calculus for OWL-RL appear to be very similar to the works on the Contextualized Knowledge Repository (CKR) framework, see for example:
Bozzato, L., Ghidini, C., Serafini, L.: Comparing contextual and flat representations of
knowledge - a concrete case about football data. In: K-Cap 2013 (2013)
Bozzato, L., Serafini, L.: Materialization Calculus for Contexts in the Semantic Web. In Description Logics 2013: 552-572 (2013)
Serafini, L., Homola, M.: Contextualized knowledge repositories for the semantic web. J. of
Web Semantics 12, 64-87 (2012)
(CORRECTNESS AND COMPLETENESS OF THE PROPOSED SOLUTION) No evaluation for the approach is provided, either for the DL presentation or the RDF representation and SPARQL (prospective?) implementation. It is thus difficult to assess the correctness and applicability of the work.
(EVALUATION OF THE STATE-OF-THE-ART) While a short summary of the related works is provided, it is clearly very limited and does not provide sufficient comparison with the proposed approach: the logic based works for representation of context in Semantic Web languages (other that the already mentioned works on the CKR framework) have a wider literature and it might be reasonable to consider, for example some of the following works:
U. Straccia, N. Lopes, G. Lukácsy, A. Polleres, A general framework for representing and reasoning with annotated semantic web data, in: AAAI 2010 (2010).
L. Tanca, Context-based data tailoring for mobile users, in: BTW 2007, pp.282-295 (2007).
O. Udrea, D. Recupero, V.S. Subrahmanian, Annotated RDF, ACM Trans. Comput. Log. 11  1-41.(2010)
(DEMONSTRATION AND DISCUSSION OF THE PROPERTIES OF THE PROPOSED APPROACH) The paper does not provide sufficient justification for the claimed contributions.
In particular: 
- the relation between the DL based representation and the RDF representation based on the N-ary pattern is not formally established (are they equivalent?).
- the rule based inference system for OWL-RL is not proved to be correct w.r.t. the presented contextual semantics.
- while Section 2 presents a set of informal requirements, these are not proved to be fulfilled by the proposed solution.
(REPRODUCIBILITY AND GENERALITY OF THE EXPERIMENTAL STUDY) As commented above, no (experimental) study is provided on the properties of the presented contextual model.
The paper proposes a solution for a DL based representation of contextual knowledge in Semantic Web data.
After briefly discussing the intended requirements for context representation, the authors initially present a DL based formalization of the notion of contextualization (derived from [10]). In the following sections a RDF representation based on the N-ary pattern is provided. Finally the paper presents a rule-based inference system for "contextualized" OWL-RL and briefly discusses the possibilities for its implementation.
- The paper considers a relevant problem in Semantic Web data (i.e. the definition of a formally solid but practically applicable contextualization of data). On the other hand, it does not appear to consider the current efforts in tackling this problem and do not justify the contributions of the proposed approach in this direction.
- As noted above, the claims of the proposal are not supported by formal or experimental assesment (w.r.t. correctness of the DL vs. RDF representation, completeness of the OWL-RL inference system, practicality of the implementation with respect to other solutions, satisfaction of the requirements)
- The advantages of the DL and RDF representation are not sufficiently discussed in comparison with similar approaches.
- Actually, while the title mentions a "Practical Implementation for Contextual Reasoning", the implementation does not seem to be realized and the practicality of the solution is not evaluated.
- The preliminaries to DL languages are imprecise and do not cover all aspects used in the paper: in fact, while the DL formalization is provided for ALCO, then the rule-based calculus is presented for OWL-RL (corresponding to a larger fragment of sROIQ).
- Can you formally prove the correspondence between the DL formalization and the RDF based representation?
- Can you relate the presented requirements to the definition of contexts by McCarthy (as it seem implied in the conclusions)?
Added after rebuttal:
I acknowledge the authors' comments to reviews provided in their response letter: unfortunately, these does not allow to modify my evaluation.
While the motivations for the paper are clear, the claims of the work should be formally assessed and a stronger account of related works on contextualization of Semantic Web knowledge should be provided. Since these do not appear to be easily addable in a minimal revision of the paper, I still can not suggest the acceptance of the paper.

Review 3 (by anonymous reviewer)

(RELEVANCE TO ESWC) The paper discusses about some issues of implementing contextual description logics/OWL.
(NOVELTY OF THE PROPOSED SOLUTION) While the paper contains some interesting ideas, the novelty is a major concern of the work. Some related works are discussed but the motivation and formulation of the research questions to be tackled needed to be explained in a more convincing way.
(CORRECTNESS AND COMPLETENESS OF THE PROPOSED SOLUTION) The results looked sound but the work needs a stronger introduction and comparison to related work. Also, as it discusses about implementational issues, I'd like to see a prototype of implementation and some experimental results.
(EVALUATION OF THE STATE-OF-THE-ART) Neither implementation nor experimental results are provided.
(DEMONSTRATION AND DISCUSSION OF THE PROPERTIES OF THE PROPOSED APPROACH) It would be good if a comparison of the proposed approach to existing contextual description logics/OWL was conducted with depth.
(REPRODUCIBILITY AND GENERALITY OF THE EXPERIMENTAL STUDY) No experimental results are presented in the paper.
(OVERALL SCORE) The work is still premature in terms of its novelty, implementation and presentation.

Review 4 (by Antoine Zimmermann)

(RELEVANCE TO ESWC) The topic of contextual reasoning is obviously relevant to the Semantic Web.
(NOVELTY OF THE PROPOSED SOLUTION) The formalism defined for contextual DL is extremely similar to Klarman et al. [10]. Then, the concrete encoding in an RDF graph is reusing the well-known N-ary relation encoding from the W3C note [17].
(CORRECTNESS AND COMPLETENESS OF THE PROPOSED SOLUTION) The paper seems to be containing different, unrelated parts. For instance, an abstract syntax with a formal semantics is provided, and suddenly N-ary relations are used, not explaining how the abstract syntax is supposed to be interpreted in this form. Then entailment rules are given, without any proof of their correctness and completeness.
(EVALUATION OF THE STATE-OF-THE-ART) The state of the art is covering several aspects of contextual reasoning but misses others that are quite relevant. For instance, the phrases "Annotating RDF", "annotated RDF", "annotations of statements", "contextual annotations", etc. are used several times and yet, no reference to "Annotated RDF" by Udrea et al. (ESWC 2006, TOCL 2010) and "Annotated RDFS" by Straccia et al. (AAAI 2010), Zimmermann et al. (JWS 2012). The paper mentions "contextual OWL" but does not refer to C-OWL (Bouquet et al. 2003). Also relevant to the discussion in Sec.3.1, there is the combination of DDL and E-connection by Santipantakis & Vouros (Knowledge information systems 2015).
(REPRODUCIBILITY AND GENERALITY OF THE EXPERIMENTAL STUDY) There is no experimental study. What is provided is a rough description of how the entailment rules could be implemented using a combination of SWRL and SPARQL UPDATE.
(OVERALL SCORE) Comments after the authors' rebuttal:
I undestand the motivation and objectives of the paper, but at this stage there is still much work to be done to reach a contribution deserving publication at a conference like ESWC.
Concerning the proof of correctness and completeness of the rules: I do not agree that it is immediate. The fact that there is evidently a direct correspondence between how the semantics of the constructs is defined and the rules should be sufficient to establish correctness, but certainly not that easily establish completeness, especially since the rules are said to extend the OWL 2 RL/RDF rules which are *incomplete* (they only cover completely ABox reasoning).
Original review:
Summary of the Paper:
The paper discusses the problem of representing and reasoning on contextual information. The objective, it seems, is to be able to concretely represent contextual knowledge in an RDF, while committing to a formal semantic of a contextual logic. Based on the RDF representation, a rule set similar to the OWL 2 RL/RDF rules of the OWL 2 Profiles standard is provided and it is claimed that it is complete with respect to the chosen semantics. A short account of how the rule set could be implemented is previded.
Strong Points:
I think that a good point of the paper is that it does not limit itself to defining an abstract syntax for contextual reasoning, on which reasoning is defined (as many related approaches do), but it also present how the formalism is encoded concretely as an RDF graph.
Weak Points:
Many of the weak points have already been listed in response to the other evaluation points. There are problems of presentation, missing information, missing references, no evaluation, curious choices that are not well justified, and little to no discussion of the merrit and limitations.
Detailed comments:
One of the big problem is related to presentation. It is not clear what the contribution is at first as the paper jumps from one subject to another. As I am very familiar with the topics of the paper, I believe I managed to fill in the gaps, as there is a lot of unsaid information necessary to make the pieces fit together. The paper also gives the imprecion of an unfinished draft. For instance, the abstract finishes with three dots. In Section 5.2, the paragraph "Type assertions" finishes with " ....Check figure 3" unexpectedly. The following paragraph finishes with three dots. Other typos, etc. are detailed at the end of this review.
Another point is the lack of support for the choices made. Why is the formalism of Klarman et al. chosen? As mentioned by the authors, there exist many formalisms for contextual reasoning, some of which have concrete implementations. Then, the justification for the use of the N-ary relation model is poor. In the end, the author acknowledge that this is causing problems for representing rules in SWRL. Since the proposal is to use SPARQL UPDATE eventually, why not simply rely on named graphs?
One of the main result of the paper is the rule set in order to concretely reason with the contextual logic. However, the rules are provided without proof about their correctness and completeness. And of course, it is not sufficient to say "we can see that these rules are sufficient fto provide a complete reasoning system". In fact, it is very likely that the rules are not complete, because Theorem PR1 mentioned here stipulates that the rules are complete only for a subset of OWL 2 RL. In fact, they are incomplete for OWL 2 RL in general. BTW, the syntax defined in Section 4.2 is neither covering OWL 2 RL, nor convered by OWL 2 RL. So for sure, the rule set would not be correct and complete for the formalism defined in the paper.
Detailed and minor comments:
Abstract and keywords: remove the "..."
- "Anyone Can Make Statements About Any Resource" is not a slogan. It is just the title of a section in the RDF specification which highlights a fact of RDF. The capital letters are there because it is typical to capitalise the first letters of words in a section title
- "a significant number of researcher in AI that studies" -> "a significant number of researchers in AI that study"
Section 2:
- there is a missing full stop at the end of the first paragraph
- In Sec.2.1, "Some data models have been defined previously to solve this issue" -> if this is referring to reification, N-ary relation, singleton properties, etc., they are not data model. These models all rely on the same data model, RDF.
Section 3.1:
- there seems to be a confusion between bridge rules and domain relations. Domain relations relate instances, while bridge rules provides different types of relations between concepts, roles, individuals.
- "DDLs ... is very attractive" -> this is a matter of point of view. As far as I know, it is not used anywhere in practical applications. Why bridge rules would be more attractive than, say imports in package-based description logics (Bao et al. 2008), correspondences in Integrated Distributed Description Logics (Zimmermann 2007), or other mechanisms for combining and relating different ontologies?
- "mapping bridges" -> you mean "bridge rules"?
- "the domain of the ontologies to combine are completely disjoint" -> this is said in an informal part of the E-connection paper but formally, there is no constraint that makes the domains of interpretation disjoint. Having or not having a disjointness constraint does not change the semantics of the formalism anyway.
- "some authors [3] advocate also the representation of multiple contexts rather than the representation of one context" -> I don't understand this sentence. Isn't it the goal of all the approaches mentioned to be able to deal with multiple contexts?
Section 3.2:
- "The reification [4] is a general purpose technique for representing n-ary relations using a language such as RDF ..." -> this is quite true if you consider the notion of reification in general, but this is not so true for RDF reification. The purpose of RDF reification is to express statements about statements. A way to express statements about statements in RDF is to rely on n-ary relations, which have different ways of being encoded.
- "this is expressed as R(A, B, c)" -> in RDF, nothing is expressed like that
- "The cons arre three:(a)" -> missing space before "(a)"
- "the semantics are unclear" -> 1) "semantics" is uncountable and always singular 2) the semantics of the reification vocabulary is as clear as the rest of the semantics of RDF. To make it short and approximative, we can say that the reification vocabulary simply does not have any particular semantics
- "it is discouraged in Linked Open Data (LOD)" -> 1) this must be supported by a citation 2) the footnote that links to wikipedia is not very welcome in a research paper. In a research paper, you must rely as much as possible on primary sources.
- "using the fourth component to represent the provenance of a set of triples" -> 1) the fourth component of what? of a named graph? there is no fourth component in a named graph, since it is a pair (name,graph) 2) the name of the graph can be used to represent whatever we like. In fact, if you consider the semantics of named graphs given in [6], the name denotes the (name,graph) pair, and nothing else.
- "The singleton property usually lacks performance due to the number of predicates" -> this requires a supporting citation.
Section 4:
- "with a second dimensionof contexts " -> what was the first one?
- in the "Syntax" paragraph: "a concept expression (an element of Nkc)" -> elements of Nkc are concept names, not concept expressions
- The choice of adopting the "rigid designator hypothesis" is not motivated
- the interpretation of the contextual constructors have "D^{\mathcal{J}[y]}" but it should be "D^{\mathcal{I}[y]}"
- the interpretation of "\langle C\rangle r", with r a role name, is not defined
- in the definition of satisfaction, C and D are said to be in Nc, but they shoud be general concept expressions; also, there is \mathcal{I}(y) which should be \mathcal{I}[y]
- "if Kis not" -> missing space
Section 5:
- "e.g" -> missing dot
- in 5.1: "figure 2" -> Figure 2
- instead of footnote 4, use a proper bibliographic reference
- "Others formalism exists also and could" -> "Other formalisms could" + add examples to the sentence instead of putting a link to wikipedia as footnote
- the nary relation approach normally uses two properties, say worksFor1 and worksFor2, to connect the elements to the contextual relation. Using the same property is strange and can have unexpected consequences because of the domain and range of the property
- in 5.2: "Rdf:type" -> "rdf:type"
- the N-ary relation model is chosen, arguably because it is judged better in some ways, only to be left aside for rdf:type. It would make more sense to choose a model that works in all cases
- "what if the rdf:type is contextual ?" -> remove space before "?"
- "a class and an individual in the same time" -> at the same time
- the syntax "rdf:type(x,C)[d: v]" is not defined; moreover, v and d should be in math mode
Section 6:
- some variables or unknowns are not written in math mode
- the predicate Q is introduced as a quaternary predicate, and then it is said that it may not be quaternary after all.
- "Each contextual axiom of the form k:\phi [...] gives rise to a specific quadruple" -> it should give rise, in general, to several quadruples
- footnote 6 and 7 should be proper bibliographic references
- "satisfy the contextual OWL-RL syntactic restrictions" -> what is this?
- in Table 2, the terms "inSomeContextOf" and "inAllContextOf" have not been introduced
Section 7:
- in the INSERT query, there are spaces after each colons, which should not be there
- ref 1 has backslashes
- some references are missing capitalised letters (rdf, owl, etc.)

Metareview by Hsofia Pinto

This paper discusses the problem of representing and reasoning on contextual OWL/description logics. While an interesting topic the paper contribution is still too premature in terms of approach formulation presenting only toy examples and weak justifications. No full discussion, implementation or experimental results are included. Therefore, the paper was perceived as not yet meeting ESWC standards and the final recommendation is not to accept.
However, authors are strongly encouraged to work on the comments provided by the reviewers, improve from them and resubmit to an appropriate venue.

Share on

Leave a Reply

Your email address will not be published. Required fields are marked *