Fig. 10From: Understanding the limitations of network online learningCumulative reward results using node2vec features. Improvement using the embedding as features is not straightforward. In the BA and DBLP cases, performance is the same or worse. In the BTER case, performance improves towards the beginning, but does not decisively outperform the default features in the end. Only in Cora does it appear that NOL-HTR with node2vec features decisively outperforms all of the other methods, but NOL with node2vec only outperforms the other methods towards the end of the experimentBack to article page