Ensemble-Based Fact Classification with Knowledge Graph Embeddings

Numerous prior works have shown how we can use Knowledge Graph Embeddings (KGEs) for ranking unseen facts that are likely to be true. Much less attention has been given on KGEs fact classification, i.e., mark either as true or false. In this paper, tackle problem with a new technique exploits ensemble learning and weak supervision, following the principle multiple classifiers make strong one. Our method is implemented in system called $$\mathsf {DuEL}$$ . post-processes ranked lists produced by embedding models classifiers, which include supervised like LSTMs, MLPs, CNNs unsupervised ones consider subgraphs reachability graph. The output of these aggregated using weakly does not need ground truths, would expensive obtain. experiments show produces more accurate classification than other existing methods, improvements up 72% terms $$F_1$$ score. This suggests promising perform KGEs.