Taxonomy of Anomaly Types in Knowledge Graphs

Asara Senaratne, Peter Christen, Pouya Ghiasnezhad Omran, Graham J. Williams

Abstract: Detecting anomalies in Knowledge Graphs (KG) is a challenging task as the patterns of anomalies are unpredictable, unknown, diverse, likely rare, and often with no ground truth labels available. Hence, it is important to identify the types of such anomalies occurring in a KG, so domain experts can adopt measures to prevent anomalies occurring during KG construction, or remove anomalies from already constructed KGs, while also discovering knowledge. In such a process we can obtain a classification among these identified anomalies such that we know what anomalies are to be forwarded to domain experts for correction, and what can be corrected via automatic or semi-automatic techniques. However, to the best of our knowledge, there is no such pre-defined classification of possible common anomalies that could arise in a KG, which we could directly use to support anomaly classification. Hence, in this paper, we propose a taxonomy of possible anomaly types that can occur in a KG using the real-world KGs YAGO-1, DSKG, Wikidata and KBpedia.

External IDs:dblp:conf/esws/SenaratneCOW25