The core idea is to construct a smoothed classifier by randomly dropping edges and ag- gregating predictions over multiple randomized graph in- stances
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it