Convolutional Complex Knowledge Graph Embeddings

We investigate the problem of learning continuous vector representations knowledge graphs for predicting missing links. Recent results suggest that using a Hermitian inner product on complex-valued embeddings or convolutions real-valued can be effective means bring these insights together and propose ConEx—a multiplicative composition 2D convolution with embeddings. ConEx utilizes Hadamard to compose followed by an affine transformation in \(\mathbb {C}\). This combination endows capability (1) controlling impact embeddings, (2) degenerating into ComplEx if such degeneration is necessary further minimize incurred training loss. evaluated our approach five most commonly used benchmark datasets. Our experimental outperforms state-of-the-art models four datasets w.r.t. Hits@1 MRR even without extensive hyperparameter optimization. also indicate generalization performance increased applying ensemble learning. provide open-source implementation approach, including evaluation scripts as well pretrained (github.com/dice-group/Convolutional-Complex-Knowledge-Graph-Embeddings).