Refining Transitive and Pseudo-Transitive Relations at Web Scale

The publication of knowledge graphs on the Web in form RDF datasets, and subsequent integration such are both essential to idea Linked Open Data. Combining can result undesirable graph structures even logical inconsistencies. Refinement methods that detect repair therefore crucial importance. Existing refinement for often domain-specific, limited single relations (e.g. owl:sameAs), or scale. We present a challenge consisting number datasets transitive pseudo-transitive hand-labeled gold standards, as well baselines. introduce an efficient web-scale algorithm works relations. Our analyses structure, allows use weighting schemes heuristically determine which possibly erroneous edges should be removed make cycle free. When compared against general-purpose algorithms perform same task, our removes least amount cycle-free while maintaining better precision identifying measured human gold-standard.