Property Assertion Constraints for an Informed, Error-Preventing Expansion of Knowledge Graphs

Abstract

The expansion of knowledge graphs (KGs) by new triples is an elementary process, which is needed for enriching and extending the represented knowledge. In order to function correctly and reliably, it is of utmost importance for many knowledge-driven applications that the expansion only adds semantically correct statements to the KG. Existing validation methods however, including description logic reasoning, SHACL and ShEx, can only detect wrong statements after they have materialized in the KG. They are of no or limited value for preventing errors when expanding KGs. To solve that problem, Property Assertion Constraints (PAC) are introduced as main contribution of this paper. For the context of a given instance and property, a PAC can identify all valid instances, which result in semantically correct property value assertions. By only offering them to users as options to choose from, the creation of semantically wrong statements is prevented and users can greatly benefit from this informed preselection. The main principle of PAC consists in the restriction of a property’s range definition by additional logic, which needs to be fulfilled additionally to the range. Similar to SHACL, PAC utilize SPARQL for defining the constraints, which can comprise almost arbitrarily complex conditions or business logic. The fundamental difference to SHACL and other integrity constraint approaches however is that PAC quasi negate the principle of formalizing constraints from constraints that detect erroneous (already materialized) facts into constraints that do the complete opposite, namely finding all semantically correct assertions (before materializing them).