Global and local evaluation of link prediction tasks with neural\n embeddings

We focus our attention on the link prediction problem for knowledge graphs,\nwhich is treated herein as a binary classification task on neural embeddings of\nthe entities. By comparing, combining and extending different methodologies for\nlink prediction on graph-based data coming from different domains, we formalize\na unified methodology for the quality evaluation benchmark of neural embeddings\nfor knowledge graphs. This benchmark is then used to empirically investigate\nthe potential of training neural embeddings globally for the entire graph, as\nopposed to the usual way of training embeddings locally for a specific\nrelation. This new way of testing the quality of the embeddings evaluates the\nperformance of binary classifiers for scalable link prediction with limited\ndata. Our evaluation pipeline is made open source, and with this we aim to draw\nmore attention of the community towards an important issue of transparency and\nreproducibility of the neural embeddings evaluations.\n