Causal Knowledge Graphs for Counterfactual Claims Reasoning

AI approaches to misinformation detection.

Led by Björn Ross and Vaishak Belle with Xue Li (Postdoctoral Researcher)


Current AI approaches to misinformation detection often learn to recognise paraphrases of previously seen claims. Detecting new misinformation is much harder, and linguistic cues are not enough to distinguish fact from fiction. Our approach is grounded in knowledge graphs and the logic of causality. However, this approach has its own challenges. Much of the misinformation encountered is not limited to simple factual statements that can be tested against a structured representation of knowledge but it consists of more complex claims such as counterfactual statements (e.g. “this would never have happened if…”). To address this problem, we integrate approaches from different subfields of computer science, namely, computational logic, deep learning and natural language processing.