Shape Learning in Recursive SHACL: Foundations and Complexity

A recent publication on arXiv (ID 2607.27934) titled 'Shapes from Examples: Foundations of Shape Learning in Recursive SHACL' explores the automated learning of SHACL shapes using example data. The research emphasizes a fitting method, aiming to derive a shape expression that authenticates all positive examples while excluding negative ones from given sets of nodes in an input graph. It delves into shapes formulated within a core fragment of SHACL linked to the Description Logic ELI, analyzing well-founded, stable, and supported semantics for recursive shape catalogs. The authors present tight exponential-time upper limits for both fitting existence and the computation of the most specific fitting, along with polynomial bounds for particular cases. This study is crucial for knowledge graph applications, as SHACL shapes play a vital role in data graph validation, making automatic shape learning key to ensuring data quality. The paper was submitted to the Computer Science > Artificial Intelligence category and is accessible on arXiv.