Understanding Document Data Sources Using Ontologies with Referring Expressions

We show how JSON documents can be abstracted as concept descriptions in an appropriate Description Logic (DL). This representation allows the use of a DL ontology, which includes naming conventions (“referring expression types (RETs)”) for instances of certain primitive concepts, in order to locate (perhaps multiple) subdocuments of the original JSON document capturing information about some particular conceptual entity. Detecting such situations allows for normalizing the JSON document into several separate smaller documents that capture all information about each such conceptual entity. This transformation preserves all the original information present in the input document. The RET assignment enables more refined and normalized capture of documents, and lead to query answers that adhere better to user expectations. We also show how RETs allow checking for a document admissibility condition ensuring that each final subdocument describes a single conceptual entity.

1. 1.

   We assume that the reader is generally familiar with standard DL terminology, such as individuals, roles, TBox, and ABox, as well as JSON documents. For formal definitions of these please see Sect. 2.

2. 2.

   Some additional conditions must be imposed on PFDs and on conjunctions to guarantee PTIME bounds; see [7, 8] for details.

3. 3.

   A variety of equality generating dependencies, including keys, can be expressed with the use of a path functional dependency (PFD) concept description generated by the second production of this grammar.

4. 4.

   In Sect. 3 we also use the standard notion of a FunDL assertion box (ABox), a set of assertions of the form “C(a)”, “\(a = b\)”, and “\(f(a) = b\)” as defined in [7]. We elaborate on the relationship between knowledge bases that use a CBox and classical FunDL knowledge bases, i.e., with an ABox in that section.

5. 5.

   In [3] it was argued that determining ways to refer to individuals is an integral but distinct step of conceptual modelling/ontology design.

6. 6.

   This is a pattern language obtained by abstracting nominals in referring expressions, and by admitting a final production to express preference among referring expressions [2].

7. 7.

   Otherwise we can report a warning about an A entity that cannot be properly identified.

8. 8.

   Essentially, this entails ensuring that preference in \(\textsf{RTA}(A)\), for any primitive concept A, exhaustively accounts for any primitive concept B for which \((A \sqcap B)^{\mathcal{I}}\) is non-empty for some interpretation \(\mathcal{I}\).

Authors and Affiliations

1. Department of Computer Science, Rutgers University, New Brunswick, USA

   Alexander Borgida, Enrico Franconi, David Toman & Grant Weddell

2. KRDB Research Centre for Knowledge and Data, Free University of Bozen-Bolzano, Bolzano, Italy

   Alexander Borgida, Enrico Franconi & David Toman

3. Cheriton School of Computer Science, University of Waterloo, Waterloo, Canada

   Alexander Borgida, Enrico Franconi & David Toman

Corresponding author

Correspondence to David Toman .

Editors and Affiliations

1. University of New South Wales, Sydney, NSW, Australia

   Haris Aziz

2. University of Western Australia, Perth, WA, Australia

   Débora Corrêa

3. University of Western Australia, Perth, WA, Australia

   Tim French

Reprints and permissions

© 2022 The Author(s), under exclusive license to Springer Nature Switzerland AG

Policies and ethics