Discovering Graph Patterns for Fact Checking in Knowledge Graphs

Abstract

Given a knowledge graph and a fact (a triple statement), fact checking is to decide whether the fact belongs to the missing part of the graph. This paper proposes a new fact checking method based on supervised graph pattern mining. Our method discovers discriminant graph patterns associated with the training facts. These patterns can then be used to construct classifiers based on either rules or latent features. (1) We propose a class of graph fact checking rules (\(\mathsf {GFCs}\)). A \(\mathsf {GFC}\) incorporates graph patterns that best distinguish true and false facts of generalized fact statements. We provide quality measures to characterize useful patterns that are both discriminant and diversified. (2) We show that it is feasible to discover \(\mathsf {GFCs}\) in large graphs, by developing a supervised pattern discovery algorithm. To find useful \(\mathsf {GFCs}\) as early as possible, it generates graph patterns relevant to training facts, and dynamically selects patterns from a pattern stream with small update cost per pattern. We further construct two \(\mathsf {GFC}\)-based models, which make use of ordered \(\mathsf {GFCs}\) as predictive rules and latent features from the pattern matches of \(\mathsf {GFCs}\), respectively. Using real-world knowledge bases, we experimentally verify the efficiency and the effectiveness of \(\mathsf {GFC}\)-based techniques for fact checking.