Modular Framework for Uncertainty Reasoning in Knowledge Graphs

A recent thesis presents a modular approach designed for scalable uncertainty reasoning within knowledge graphs, tackling three specific types of uncertainty: vague attribute values, probabilistic existence of triples, and incomplete schema knowledge. This framework incorporates probabilistic literals along with a query algebra tailored for continuous attributes, a method that converts SPARQL provenance into manageable probabilistic circuits, and topology-aware geometric embeddings for statistical schema reasoning. The research seeks to address the absence of inherent uncertainty support in existing Semantic Web standards, which often results in computational challenges. The primary assertion is that distinct reasoning methods—algebraic, logical, and geometric—can effectively manage each level of uncertainty. The thesis can be found on arXiv under ID 2605.16568.