TRIC: A Triples Corrupter for Knowledge Graphs.

We study the problem of corrupting triples in Knowledge Graphs (KG) for the purpose of assisting anomaly detection and error detection techniques developed for KG quality enhancement. Our goal is to provide users with the highest possible level of control over the triples corruption process, and simultaneously develop a solution that scales to large KGs. Hence, we introduce TRIC, an approach for corrupting triples considering both semantic and type information to generate errors in a KG. In this paper, we discuss how the problem of triples corruption is challenging, and different from existing negative sampling techniques used in link prediction. To the best of our knowledge, there is no approach in the literature dedicated for generating abnormal triples in KGs to support anomaly detection and error detection tasks.