We demonstrate the concept of presenting a data schema of a knowledge graph, used to support visual schema presentation and visual queries over the data, as a knowledge graph itself. We demonstrate, how the visual schema analysis methods (visual schema presentation and visual query creation) can be applied to the meta-level schema, as well, including creation of queries involving both the data and its schema levels. Visual methods of data analysis, including visual data schema presentation and visual queries over the data hold a promise of enabling direct work with the data for a wider group of custom domain experts, as well as they can ease the work of data professionals by allowing to exploit their visual perception capabilities in their work.
visualization tools, as VOWL [
The tools for visual data exploration and queries involve OptiqueVQs [
Out of these tools, ViziQuer provides an option to combine the data schema presentation and visual
query facilities into a single visual environment [
The current solution enabling the shared visual data schema and visual query experience in the ViziQuer tool context foresees storing the data schema in a relational database (RDB). This solves the performance issues, even when the schemas of large data sets, such as DBpedia and Wikidata are stored. However, RDB storage is little helpful when it comes to explaining what data actually are stored in the database. The RDB format also makes it dificult to share the data schemas among the diferent users and to integrate their data into larger data ecosystems, typically ofered by the knowledge graphs.
In this work we outline the concept of creating a knowledge graph out of the visual data schemas themselves to enable them to be acted upon by various KG and semantic technology tools, including but not limited to the visual schema presentation and visual queries in the ViziQuer tool itself. We
CEUR
ceur-ws.org demonstrate the possibility of creating distributed visual queries in ViziQuer simultaneously over the original data and the data schema, as well.
In what follows, Section 2 reviews the visual schemas and queries in the ViziQuer environment context. Section 3 presents the data schemas as Knowledge graphs and Section 4 concludes the paper.
The material supporting the demonstration is available at https://github.com/LUMII-Syslab/ viziquer-tools-kg .
Figure 1 provides an example data set schema for a snapshot of the Nobel Prizes data set1. The schema lists the data classes as nodes and the properties available at each of them as attributes, or as links connecting the property source and target classes. There is frequency information for classes, attributes (in the context of their ascription class) and links (in the context of link source and target classes), as well as attribute and link cardinalities in their respective contexts. Note that the properties :hasPart and :isPartOf connect just the nobel:nobelPrize and nobel:LaureateAward classes, their applicability to the instances of the dbo:Award classes is covered by their ascription its subclasses (the schema provides the information about the “essential” ascription points of the properties). The D and R markers at the attributes and links depicting a property inform that the particular property visualization context corresponds to its domain (D) or range (R).
The data schema, as stored in the visual tool, allows creating visual queries for inspecting the data set contents, for instance, by asking for statistics of Nobel Prizes received by organizations across diferent prize categories, as shown in Figure 2.
A visual query is a rooted directed graph with data nodes corresponding to query variables (with an
option to assign class information, as well), the data links establishing their connections and attributes
building up the selection list. There are means for further query structuring as control nodes (unit or
union) and non-data edges (non-link edges and same-data edges). The reader may consult [
The data schemas (used for visual schema presentation and visual query support) in the ViziQuer environment2 are handled by a dedicated open-source Data Shape Server3 that can be adapted also to serve the data schema information to other clients (e.g., if the context-sensitive auto-completion of class and property names were to be implemented in a textual SPARQL editor). The Data shape server stores the data schemas in a relational PostgreSQL database, ensuring their technical availability, paying less attention to the ease of explainability of the schema structure.
For the data schema explanation and their integration into the knowledge graph landscape, the data schemas can be mapped into the Knowledge graph (KG) format. We propose an initial data schema conceptualization structure and manually create custom mappings for generating KG resources (together with their class assignment) from database records of diferent tables, as well as creating of the data triples in KG from the contents of these records. The mappings are then interpreted by a custom code that creates the actual data triples4.
Figure 3 contains the visualization of the meta-schema obtained from the analysis of the KG containing the Nobel Prizes data endpoint (the classes in the data schema would be shared for schemas of diferent endpoints; the entity frequency statistics would be diferent, though).
The central concepts of the data schema are :Classifier for presenting the data classes (and values in
2https://github.com/LUMII-Syslab/viziquer, see also https://viziquer.app
3https://github.com/LUMII-Syslab/data-shape-server
4Should a more standardized approach be required, a R2RML [
We note also the :importanceIndex attribute in the :ClassPropertyPair and :ClassProperty_Class_Rel classes. It informs (if the value is above 0) about when the class is important in the context of a property (or in the context of the property in the context of the “other end” class), so that the attribute or link holding the property name is to be ascribed in the schema diagram at the respective class.
Although the queries involving the meta-data level are available also on the data-level schemas and
the ViziQuer visual notation has means to support these (using, e.g., the explicit variables in the class
and property name positions in a schema diagram, cf. [
The ”most specific” class for an instance ?A is looked up in the named graph nobel_prizes_schema holding the schema information as an instance ?C of the class np:Classifier . The negation condition states that there is no other type ?T1 of the instance ?A such that it would be a subclass of the classifier instance ?C in the nobel_prizes_schema named graph5.
The availability of the data schema as a knowledge graph allows creating a wide range of metaanalysis queries over the data. The queries working over meta-schemas of diferent data sets are possible, as well.
This work has demonstrated the natural concept of presenting a knowledge graph data schema, as stored in a visual schema and query tool, itself as a data schema. Such a presentation has provided basis of the explanation of the data schema concepts, making it easier re-usable, as well as it allows using the knowledge graph analysis tools over the data schema representation.
The visual query environment allows building queries consuming data from diferent named graphs, corresponding to diferent schemas; we have shown an example of creating a query looking simultaneously both into the level of the data themselves, and the data schema. 5The global subquery construct is used in the visual notation to enforce the usage of the SPARQL MINUS construct.
A further work would be to expand the visual data schema visualization and visual query software to allow it to work directly with the schema stored as a knowledge graph, as well (without the need to store the schema in a relational database). Such an extension would ease re-using and sharing of schemas of data sets among diferent users (as the schema would be just a knowledge graph itself). This work has been partially supported by a Latvian Science Council Grant lzp-2021/1-0389 “Visual Queries in Distributed Knowledge Graphs”.