In data integration and access to legacy data sources using end user-oriented
languages, the approach based on Virtual Knowledge Graphs (VKG) is gaining
importance [
As a consequence, management of VKG mappings throughout their entire
life-cycle is currently a labor-intensive, essentially manual e ort, which requires
highly-skilled professionals [
Commons License Attribution 4.0 International (CC BY 4.0).
In our ongoing work, we build on this key observation and provide the contributions described in the following. 2.1
We propose a catalog of mapping patterns that emerge when linking DBs to
ontologies. To do so, we build on well-established methodologies and patterns
studied in data management (such as W3C direct mappings { W3C-DM [
We organize patterns in two major groups: schema-driven patterns, shaped by the structure of the DB schema and its explicit constraints, and data-driven patterns, which in addition consider constraints emerging from speci c con gurations of the data in the DB. For each schema-driven pattern, we actually identify a data-driven version in which the constraints over the schema are now not explicitly speci ed, but hold in the data. But we provide also data-driven patterns that do not have a schema-driven counterpart. The two types of patterns can be used in combination with additional semantic information from the ontology, for instance on how the data values from the DB translate into RDF literals. These considerations lead us to introduce also pattern modi ers. Moreover, some of our patterns come with accessory views de ned over the DB-schema, which make explicit the presence of speci c structures over the DB schema that are revealed through the application of the pattern itself. Such views can be used themselves, together with the original DB schema, to identify the applicability of further patterns. 2.2
The proposed patterns can be employed in a variety of VKD design scenarios,
depending on which information artifacts are available, and which ones have to
be produced. Speci cally, we consider the following scenarios: (i) Debugging of a
VKG Speci cation, which arises when a full VKG speci cation is already in place
and must be debugged. (ii) Conceptual Schema Reverse Engineering, which aims
at inferring a conceptual schema of the DB that represents the domain of
interest by re ecting the content of a given full VKG speci cation. (iii) Mapping
Bootstrapping, where the DB and the ontology are given, but mappings relating
them are not, and patterns can be used to (semi-)automatically bootstrap an
initial set of mappings. These can then be further re ned and extended manually,
Title Suppressed Due to Excessive Length
possibly exploiting again the patterns. (iv) Ontology+Mapping Bootstrapping,
where neither the ontology nor the mappings are given as input, and have to be
synthesized. This scenario can be reduced to the previous one by rst inducing
a baseline ontology mirroring the structure of the DB schema. (v) VKG
Bootstrapping, where we just have a conceptual schema of the domain, and the goal
is to set up a VKG speci cation. The conceptual schema can be then
transformed into a normalized DB schema using well-established relational mapping
techniques (e.g., [
In our work, we have analyzed the concrete mapping strategies arising from a number of VKG use cases in order to understand how patterns occur in practice, and with which frequency. To this purpose, we have gathered 6 di erent scenarios, coming either from the literature on VKGs, or from actual real-world applications, covering a variety of di erent application domains. So far, we have manually classi ed a total of 1582 mapping assertions, falling in 367 pattern applications. We have studied the coverage of mappings appearing therein in terms of our patterns, as well as on how many times the same pattern recurs. Our investigation has shown that only 3% of pattern applications fall outside of our categorization, and it also gives an interesting indication on which patterns are more pervasively used in practice. 3
The work carried out so far is only a rst step, with respect to both categorization of patterns, and their actual use. Regarding the former, we are now exploring more in depth the interaction between patterns and pattern modi ers, such as value invention or identi er alignment. Regarding the latter, so far we have used patterns to investigate, and highlight, the speci c problems to address when setting-up a VKG scenario. We are now investigating solutions to these problems, by exploiting approaches from other elds, e.g., schema matching.
This research has been partially supported by: the EU H2020 project INODE; the Italian PRIN project HOPE; the European Regional Development Fund (ERDF) Investment for Growth and Jobs Programme 2014-2020 through the project IDEE (FESR1133); the Free University of Bozen-Bolzano through the projects KGID, GeoVKG, OntoGeo, and STyLoLa; the Wallenberg AI, Autonomous Systems and Software Program (WASP) funded by the Knut and Alice Wallenberg Foundation.