Visual notations for viewing RDF constraints with UnSHACLed

Abstract

The quality of knowledge graphs can be assessed by a validation against specified constraints, typically use-case specific and modeled by human users in a manual fashion. Visualizations can improve the modeling process as they are specifically designed for human information processing, possibly leading to more accurate constraints, and in turn higher quality knowledge graphs. However, it is currently unknown how such visualizations support users when viewing RDF constraints as no scientific evidence for the visualizations' effectiveness is provided. Furthermore, some of the existing tools are likely suboptimal, as they lack support for edit operations or common constraints types. To establish a baseline, we have defined visual notations to represent RDF constraints and implemented them in UnSHACLed, a tool that is independent of a concrete RDF constraint language. In this paper, we (i) present two visual notations that support all SHACL core constraints, built upon the commonly used visualizations VOWL and UML, (ii) analyze both notations based on cognitive effective design principles, (iii) perform a comparative user study between both visual notations, and (iv) present our open source tool UnSHACLed incorporating our efforts. Users were presented RDF constraints in both visual notations and had to answer questions based on visualization task taxonomies. Although no statistical significant difference in mean error rates was observed, all study participants preferred ShapeVOWL in a self assessment to answer RDF constraint-related questions. Furthermore, ShapeVOWL adheres to more cognitive effective design principles according to our performed comparison. Study participants argued that the increased visual features of ShapeVOWL made it easier to spot constraints, but a list of constraints - as in ShapeUML - is easier to read. However, also that more deviations from the strict UML specification and introduction of more visual features can improve ShapeUML. From these findings we conclude that ShapeVOWL has a higher potential to represent RDF constraints more effective compared to ShapeUML. But also that the clear and efficient text encoding of ShapeUML can be improved with visual features. A one-size-fits-all approach to RDF constraint visualization and editing will be insufficient. Therefore, to support different audiences and use cases, user interfaces of RDF constraint editors need to support different visual notations.