Masterarbeit
Quality Assessment for Federated University Knowledge Graphs Using Machine-Learning-Based Anomaly Detection: A Case Study on the Across Alliance
Completion
2026/07
Research Area
Students
Farid Mammadov
Advisers
Sandra Schaftner M.Sc.
Description
Federated university Knowledge Graphs (KGs) are increasingly used to integrate heterogeneous institutional data such as research outputs, teaching materials, and organizational information across distributed environments. While these KGs enable interoperability and data reuse, ensuring high Data Quality remains a major challenge due to inconsistencies, semantic heterogeneity, and the lack of centralized control in federated settings. Existing Data Quality approaches often face challenges in scaling, incorporating contextual information, and identifying complex or unexpected errors, which negatively affects the reliability and usefulness of such KGs for downstream applications.
This thesis proposes a semantic quality management framework for federated university KGs grounded in FAIR and Linked Data principles. The proposed solution combines established Data Quality models and metrics with automated verification and validation techniques based on constraint languages such as SHACL and SPARQL-based testing. In addition, machine- learning-based methods, particularly Graph Neural Networks (GNNs), are employed to support anomaly detection across distributed KG sources. Specifically, relational and multi-relational GNN architectures such as Relational Graph Convolutional Networks (R-GCN) and Composition-based Graph Convolutional Networks (CompGCN) are explored to capture structural and semantic patterns and identify inconsistencies and misalignments. Furthermore, the thesis experimentally compares these GNN-based approaches with KG embedding models such as TransE and DistMult regarding their suitability for Data Quality assessment and anomaly detection. The framework supports quality assessment in a federated data environment.The objective of this master thesis is to design, implement, and evaluate a comprehensive quality management approach for federated university KGs. This includes analyzing the state of the art in Data Quality assessment, FAIR principles, Linked Data Quality frameworks, and machine-learning-based anomaly detection techniques. In addition, a prototype of the framework will be implemented. The evaluation focuses on detecting and mitigating Data Quality issues, maintaining semantic consistency and interoperability, and demonstrating practical applicability and scalability in a federated academic environment.
