Search for a command to run...
Linked Open Data (LOD) comprises of an unprecedented volume of structured data on the Web. However, these datasets are of varying quality ranging from extensively curated datasets to crowd-sourced or extracted data of often relatively low quality. We present a methodology for test-driven quality assessment of Linked Data, which is inspired by test-driven software development. We argue, that vocabularies, ontologies and knowledge bases should be accompanied by a number of test cases, which help to ensure a basic level of quality. We present a methodology for assessing the quality of linked data resources, based on a formalization of bad smells and data quality problems. Our formalization employs SPARQL query templates, which are instantiated into concrete quality test case queries. Based on an extensive survey, we compile a comprehensive library of data quality test case patterns. We perform automatic test case instantiation based on schema constraints or semi-automatically enriched schemata and allow the user to generate specific test case instantiations that are applicable to a schema or dataset. We provide an extensive evaluation of five LOD datasets, manual test case instantiation for five schemas and automatic test case instantiations for all available schemata registered with LOV. One of the main advantages of our approach is that domain specific semantics can be encoded in the data quality test cases, thus being able to discover data quality problems beyond conventional quality heuristics.