Much data with graph structures satisfy the principle of homophily, meaning
that connected nodes tend to be similar with respect to a specific attribute.
As such, ubiquitous datasets for graph machine learning tasks have generally
been highly homophilous, rewarding methods that leverag