While knowledge graphs are increasingly adopted for supporting data analysis over linked data cubes, it is still challenging for end-users to personalize, preserve, and share cubes that are pertinent to their analytics objectives. Building on Personal Knowledge Graphs, this study introduces the notion of Personal Knowledge Graph Cubes (PKGCubes). PKGCubes serve as a mediator between the web of data cubes, and data analysis platforms. A demo of PKGCubes Manager is presented, enabling data analysts to create, publish, and reuse PKGCubes in standalone data analysis tools. This work contributes to ofering more personalized and self-service analytics tasks on the growing web of data.
Increased research eforts are aiming to leverage the expressive nature of knowledge graphs for
facilitating data analytics tasks [
While such eforts are providing greater data sharing and usability opportunities,
manipulating knowledge graphs for data analytics still poses some challenges to end-users [
Personal Knowledge Graphs (PKG) allow the representation of knowledge graph entities that
are relevant and of personal nature to a particular individual [
Figure 1 illustrates how we envision the PKGCube. It acts as a mediator between published
linked data cubes and data analytics tools. It serves as a personal and persistent snapshot view of
graph data, representing entities that connect to linked data cube sources, and feeds the personal
data of interest into analytics tools. Ontologically, we design a PKGCube as an extension of
the RDF Data Cube vocabulary [
Linked Data Cube
Datasets
Observation
Slice Create
Person
Link
Personal Knowledge
Graph Cube
Version PKGCube Description
Publish
CubeHash
Access SourceQuery
Data Analytics
Tools Reuse
We envisage a framework to create, publish, and reuse PKGCubes. The creation of the PKGCubes involves providing individuals access to navigate and select entities from linked data cube sources to include in their PKGCube based on their individual analysis objectives. The created PKGCubes are stored and published on a triplestore. To maximize reusability of the PKGCubes in a variety of data analytics tools, they can be further processed and transformed into more manipulable data formats such as tabular structure in the form of spreadsheets or Comma Separated Values (CSV).
We demonstrate the feasibility of PKGcubes through PKGCubes Manager, a Python Streamlit
online app1 that enables data analysts to create PKGCubes through a set of filters, publish the
1PKGCubes Manager is accessible at: https://linked.aub.edu.lb:8502/.
cubes to a triple store, and reuse them in analytics apps such as Microsoft Power BI [
The PKGCubes Publisher enables data analysts to specify a SPARQL endpoint that contains RDF data cubes. It ofers predefined endpoints in the drop-down menu, or new endpoints that can be provided by analysts. The selected endpoints need to store data cubes with explicit datasets, measures, and dimensions following the RDF Data Cube vocabulary. The publisher app scans the data available in the endpoint using SPARQL templates designed to detect the available datasets and their related entities. The SPARQL results are then used to populate the “Cube Filters” available in the app. Users can then filter the cubes based on the domains, datasets, and the available dataset measures and dimensions. After the selection, the tool builds a SPARQL query in the background based on the filters selection and presents the SPARQL results in tabular format. Users can then check the cube data loaded in the table, fine-tune the iflters if needed, and publish the cube.
To publish the cube, users need to provide their personal details including their name and email. Then the app executes (1) a Unique Resource Identifiers (URIs) and linkage generation step, (2) a versioning check, followed by (3) a data publication phase. In the first step, the PKGCubes URIs are generated based on the MD5 hash of the following combination: <email+dataset+measures+dimensions>. This configuration enables associating a unique oneway identification of the PKGCubes while preserving the users’ data privacy. It also helps with storing the configuration that the user followed to generate the PKGCube, and appropriate 2A video demonstration is available at: https://youtu.be/e9NPsrVSXXM linkages among cube versions. The publisher links the PKGCube URIs to the relevant entities (e.g., observations extracted from the endpoint, source query, and other elements mentioned in Figure 1) and the personal user URI generated based on the MD5 hash of their provided email. In the second step, a versioning functionality was implemented to keep track of the diferent versions of the same PKGCube. Versioning is valuable to have snapshots of the data saved at various points in time. The publisher app checks the version of the PKGCube at two levels. At the first level, the app checks whether the PKGCube URI was already published. If it’s a new PKGCube, the PKGCube entities and related files (i.e., CSV and RDF) are generated. If the cube exists, it checks the extracted content from the cube, compares it to the cube content hash of the latest version, and creates appropriate version linkages that users can explore. Finally, the generated PKGCube entities are published to a triplestore.
In the PKGCubes Explorer part, data analysts are able to browse the published PKGCubes details, their linked versions, and reuse the related data in external applications. Another feature of the tool is a “refresh” functionality that updates the PKGCube with the latest data available in the initial endpoint. This is useful to handling cases when the source RDF Data Cubes content changes, allowing analysts to update their PKGCubes with the latest data that can be seamlessly reflected in their external applications. To illustrate the reuse of data, Figure 2 showcases how a PKGCube’s linked CSV file was used in Microsoft Power BI to generate a dashboard on tourism index and guest houses around Lebanon3. This demonstrates the potential of PKGCubes to create personalized, uniquely referenced, and preserved data cubes that can be reused by analysts in their preferred data analysis environments.
We presented in this paper the notion of Personal Knowledge Graph Cubes, with a
demonstration of its application through the PKGCubes Manager online app. As part of future research,
this work can benefit from developing more robust management and access control features of
PKGCubes. This conforms with the Personal Knowledge Graph ecosystem laid out by
Skjaeveland et al. [
This work was partially supported by the Olayan School of Business (OSB) Research Initiative fund, and the American University of Beirut Research Board (URB). 3The PKGCube used to generate the Power BI visualizations is accessible at: http://linked.aub.edu.lb/pkgcube/ 551015b5649368dd2612f795c2a9c2d8