Learning Node Representations

Representation learning is one of the foundations of Deep Learning and allowed big improvements on several Machine Learning fields, such as Neural Machine Translation, Question Answering and Speech Recognition. Recent works have proposed new methods for learning representations for nodes and edges in graphs. In this work, we propose a new unsupervised and efficient method, called here Neighborhood Based Node Embeddings (NBNE), capable of generating node embeddings for very large graphs. This method is based on SkipGram and uses nodes’ neighborhoods as contexts to generate representations. NBNE achieves results comparable or better than state-of-the-art feature learning algorithms in three different datasets and, differently from our main baseline (Node2Vec), which needs to have its parameters tuned in a validation set, is completely unsupervised.