The paper introduces ConvE, a multi-layer convolutional network model for link prediction in knowledge graphs. ConvE uses 2D convolutions over embeddings to predict missing relationships between entities, aiming to improve expressiveness and efficiency compared to shallow models like DistMult. The model is designed to be parameter-efficient, achieving state-of-the-art results on several datasets while scaling to large knowledge graphs. It is particularly effective at modeling nodes with high indegree, which are common in complex knowledge graphs such as Freebase and YAGO3. The authors also address the issue of test set leakage in datasets like WN18 and FB15k, where inverse relations from the training set are present in the test set, and propose robust variants of these datasets to ensure fair evaluation. Experimental results show that ConvE outperforms previous models in terms of Mean Reciprocal Rank (MRR) across all datasets.The paper introduces ConvE, a multi-layer convolutional network model for link prediction in knowledge graphs. ConvE uses 2D convolutions over embeddings to predict missing relationships between entities, aiming to improve expressiveness and efficiency compared to shallow models like DistMult. The model is designed to be parameter-efficient, achieving state-of-the-art results on several datasets while scaling to large knowledge graphs. It is particularly effective at modeling nodes with high indegree, which are common in complex knowledge graphs such as Freebase and YAGO3. The authors also address the issue of test set leakage in datasets like WN18 and FB15k, where inverse relations from the training set are present in the test set, and propose robust variants of these datasets to ensure fair evaluation. Experimental results show that ConvE outperforms previous models in terms of Mean Reciprocal Rank (MRR) across all datasets.