This is a state-of-the-art method for producing graph embeddings that can be used for node classification. ARCTE stands for Approximate Regularized Commute-Time Embedding.

It produces the graph embeddings through an improved implementation of personalized PageRank. Additionally, a supervised feature weighting step is applied in order to further boost the importance of the framework.

The method and its associated study has been published on PLOS One and its implementation is publicly available on GitHub. An overview of its experimental results is available on this page