Fig. 4
From: BioBLP: a modular framework for learning on multimodal biomedical knowledge graphs
Distribution of MRR (left) and H@10 (right) performance metrics for link prediction, across all values explored during hyperparameter optimization of the baseline models. TransE performs rather consistently, but poorly overall. ComplEx results on a broad range of outcomes, indicating that it is very sensitive to hyper-parameters. RotateE performs similar to ComplEx when comparing their medians, but has a smaller spread and with specific parameters it results in the best performance