A causal-based symbolic reasoning framework for uncertain knowledge graphs

Recently, reasoning methodologies for uncertain knowledge graphs have been extensively proposed. However, symbolic reasoning for uncertain knowledge graphs has rarely been studied. There are multiple paths between the subject and object entities, which makes it a challenge to deduce the confidence of triples base on symbolic reasoning. In this paper, we develop a causal-based symbolic reasoning framework UKGCSR, which aims to infer object entity and triple confidence through multi-hop reasoning and causal inference. The multi-hop reasoning module establishes the reasoning process as a Markov decision process, excavates paths and reliability between entities through pathfinding. Then, the causal inference module constructs a causal diagram and generates counterfactuals. It evaluates each path's contribution to the triple, so as to calculate the confidence of prediction facts. Our model provides the interpretability in reasoning process and shows relative high-performance in experimental results.