ArcUno: A Topology-Sensing Tangible Toolkit for Lean Knowledge Graph Modeling
Knowledge graphs externalise entities and their relationships as a directed, labelled network, but constructing them is cognitively demanding and is typically supported by screen-, mouse-, and keyboard-based tools such as Protégé and yEd. Research on external cognition and tangible interaction suggests that a graspable, manipulable representation could support this activity.
This thesis presents ArcUno, a topology-sensing tangible toolkit for lean knowledge graph construction. Users assemble physical nodes and directed, articulated edges on a desk while an overhead camera detects ArUco fiducials, reconstructs the topology in real time, and exports the result as a knowledge graph in RDF/Turtle. Design requirements were derived from prior work on external cognition and tangible interaction, realised as a working prototype, and evaluated in a formative study with six domain experts combining think-aloud, observation, the exported graph, the System Usability Scale, and an adapted tangible-usability instrument.
The evaluation suggests that topology-sensing tangibles are well suited to exploratory knowledge graph construction when manipulation, revision, and inspection are central to the task. ArcUno preserved graph topology accurately across all six participant sessions, with every exported graph matching the final physical model, and was rated highly usable (SUS mean 84.2). However, participants’ difficulties concerned the semantics of the notation rather than the mechanics of interaction: relation domain and range were implicit, edge direction required careful attention, and physical properties such as colour were sometimes interpreted as meaningful. These findings indicate that future iterations should focus primarily on representational clarity instead of sensing reliability.
The thesis contributes a set of design constraints for tangible knowledge graph construction, an evaluated physical-digital prototype, and an evaluation approach that considers both the modelling process and the correctness of the resulting graph.
