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