ИСТИНА

Node2Vec is a widely used method for learning feature representations for nodes in graphs. However, obtaining embeddings may require quite a lot of time. In this paper, we propose a number of optimizations for the one of the most popular and high performance Node2Vec C++ implementations. Our optimizations accelerate Node2Vec by 2.5-5.1 times for various graphs. We demonstrate the preservation of the accuracy of the optimized algorithm by solving a node classification problem with multiple labels based on the learned embeddings. Keywords: node2vec, graph embeddings, graph representation learning, software optimization