Marrying Uncertainty and Time in Knowledge Graphs

The management of uncertainty is crucial when harvest-ing structured content from unstructured and noisy sources.Knowledge Graphs (KGs) are a prominent example.KGsmaintain both numerical and non-numerical facts, with thesupport of an underlying schema. These facts are usually ac-companied by a confidence score that witnesses how likelyis for them to hold. Despite their popularity, most of exist-ingKGsfocus on static data thus impeding the availabilityof timewise knowledge. What is missing is a comprehensivesolution for the management of uncertain and temporal datainKGs. The goal of this paper is to fill this gap. We rely ontwo main ingredients. The first is a numerical extension ofMarkov Logic Networks (MLNs) that provide the necessaryunderpinning to formalize the syntax and semantics of un-certain temporalKGs. The second is a set of Datalog con-straints with inequalities that extend the underlying schemaof theKGsand help to detect inconsistencies. From a theoret-ical point of view, we discuss the complexity of two impor-tant classes of queries for uncertain temporalKGs:maximuma-posterioriandconditional probability inference. Due to thehardness of these problems and the fact that MLN solversdo not scale well, we also explore the usage of ProbabilisticSoft Logics (PSL) as a practical tool to support our reasoningtasks. We report on an experimental evaluation comparing theMLN and PSL approaches.