Inferbeddings is a link prediction framework that allows including First-Order background knowledge via adversarial training - the model is described in Adversarial Sets for Regularising Neural Link Predictors.