Journal of Data Mining and Knowledge Engineering ISSN: 3139-5341 (Online)

Abstract

Knowledge graphs (KGs) represent entities and relations in structured form, enabling intelligent reasoning and semantic search. However, large-scale KGs often contain anomalous or inconsistent relational patterns caused by noise, incompleteness, or malicious insertion. Relational anomaly detection aims to identify such irregular structures considering both entities and relationships. This paper reviews the foundations, methods, datasets, and challenges of relational anomaly detection in knowledge graphs. We categorize approaches into statistical, embedding-based, graph neural network, rule-based, and hybrid techniques. Comparative analysis highlights strengths and limitations across scalability, interpretability, and detection accuracy. Applications in fraud detection, cybersecurity, biomedical discovery, and data quality management are also discussed. Finally, open challenges such as evolving graphs, explainability, and privacy-aware anomaly detection are outlined. The review provides a consolidated understanding for researchers and practitioners working in graph mining and knowledge engineering.

Keywords: Knowledge graph, relational anomaly detection, graph mining, link prediction, graph neural networks, outlier detection, semantic data quality