RDF Shape Induction using Knowledge Base Profiling

Knowledge Graphs (KG) are becoming the core of most artificialintelligent and cognitive applications. KGs describe the real-worldentities and their relationships. Popular knowledge graphs such asDBpedia, YAGO2, and Wikidata have chosen the RDF data modelto represent their data due to its capabilities for semantically richknowledge representation. Despite the advantages, there are chal-lenges in using RDF data, for example, data validation. Ontologies,the most common manner of specifying domain conceptualizationsin RDF data, are designed for entailments rather than validation.Most ontologies lack the granular information needed for validat-ing constraints. Recent work on RDF Shapes and standardization oflanguages such as SHACL and ShEX provide better mechanisms forrepresenting constraints for RDF data. However, manually creatingintegrity constraints for large KGs is still a tedious task. This bringsa clear need for methods and tools that could help to generate suchconstraints automatically or semi-automatically. In this paper, wepresent a data driven approach for inducing integrity constraintsfor RDF data using data profiling. Those constraints can be com-bined into RDF Shapes and can be used to validate RDF graphs.Our method is based on machine learning techniques to automati-cally generate RDF shapes using profiled RDF data as features. Inthe experimets, the proposed approach achieved 97% precision inderiving RDF Shapes with cardinality constraints for a subset ofDBpedia data.