This is a repository for the code and the dataset of our work: Approximate Pattern-Based Reasoning (APR)*.

Most existing knowledge graph refinement approaches either focus on adding missing knowledge to the graph, i.e., completion, or on identifying wrong information in the graph, i.e. error detection.

For the former case, current approaches often assume that there is no schema for knowledge graphs or such schema can be ignored. In this work, we propose a novel approach to develop an approximate reasoning to validate the triples that are generated from several knowledge graph completion algorithms, including those claimed to be correct by the triple classification service.

Our somehow surprising results suggest that the above schema-less assumption might need to be reconsidered.

• APR now is being under review as demo paper on the CIKM 2018 (http://www.cikm2018.units.it/).