Effect of heuristic post-processing on knowledge graph profile patterns: cross-domain study

Sets of frequent schema-level patterns characterizing a given knowledge graph (KG) represent a central output of profiling tools such as ABSTAT, as they could provide a quick overview of the coverage of the KG and its adequacy for various tasks. However, the number of patterns may be huge, and the most frequent ones might not be the most useful ones for semantically characterizing the KG, since they might feature generic (OWL, SKOS, etc.) classes and even XML data types. We hypothesize that the pattern profile suitability for a ‘rapid skimming’ scenario might be improved by applying a stop-list of namespaces or individual schema IRIs by which the original pattern set is pruned. We experimented with post-processing the patterns returned by ABSTAT with regard to reducing the quantity of patterns and re-ranking the patterns appearing in the first positions of the frequency-ordered results. We processed the sets of KGs from two different domains – COVID-19 and linguistics/lexicography.