SHACLens: A Visualization Workflow for SHACL Violation Exploration in Knowledge Graphs

Christian Alexander Steinparz^1*Andreas Hinterreiter¹Labinot Bajraktari²Vitaly Sedlyarov²Markus J. Bauer¹Thomas Zichner²Marc Streit²

• ¹Department of Computer Science, Faculty of Science and Natural Sciences, Johannes Kepler University of Linz, Linz, Austria
• ²Boehringer Ingelheim RCV GmbH & Co KG, Vienna, Austria

Validating large knowledge graphs with the Shapes Constraint Language (SHACL) often yields violation reports too large to interpret and trace to root causes, especially in industry-scale datasets such as pharmaceutical omics pipelines. We present SHACLens, an interactive visualization workflow—developed with a major pharmaceutical partner—that links ontology, instance data, and violation reports across multiple coordinated views. We contribute a practitioner-informed workflow co-designed with pharmaceutical data-analysis experts, along with design lessons distilled: We first analyzed their existing practice, and optimized it into a multiple coordinated-view system. A Node-Link View combines ontology and groups of equivalent violations, a projection view reveals clusters of nodes with similar errors, a LineUp table combines instance data with violation information, a Class Tree offers a class-hierarchy overview, and an integrated LLM assistant provides contextual explanations and can operate the system via natural-language commands. Within this workflow, selections and filters propagate across views, exposing co-occurring errors and their likely upstream causes. Analysts iteratively identify violation clusters, inspect correlations, and trace the detailed cause of errors. We evaluated SHACLens through an iterative expert-in-the-loop design process with the partner team and a qualitative study on a transcriptomics dataset containing 5,203 violating nodes with the same experts. In this study, SHACLens efficiently surfaced repeated sets of errors due to missing objects and schema inconsistencies, supporting goal-oriented analysis and serendipitous findings.