Knowledge Graph Quality Revisited: Recent Methods, Emerging Trends, and Open Challenges

Knowledge graphs are now a mature method for representing knowledge in a machine-readable form. In practice, knowledge is often fragmented and stored in different formats and levels of granularity. Consequently, knowledge graphs must integrate heterogeneous data from multiple sources with schemas of varying complexity. Knowledge graphs must also be able to manage evolving knowledge and resolve conflicts between sources. These requirements make them difficult to create and maintain and lead to various quality issues in terms of both data and structure. Poor-quality knowledge graphs cause a variety of problems in applications such as decision-making and artificial intelligence results, leading to financial losses. At the same time, knowledge graphs are increasingly used to make artificial intelligence systems practical for business. They can ground large language models in verified organizational information, thereby improving the accuracy and reliability of their responses and reducing hallucinations. However, these benefits depend directly on the quality of the underlying knowledge graph: incomplete, outdated, or incorrect knowledge can be propagated into downstream results. Until now, research on quality aspects has traditionally focused on relational databases. We therefore conducted a systematic literature review of current methods for assessing the quality of knowledge graphs and compared them with traditional approaches in the field of relational data. In addition to analyzing the state of the art, we discuss which methods from the relational world can be adopted, which specific differences need to be considered in the case of knowledge graphs, and where we see open questions and future developments.