NESS: Learning Node Embeddings from Static SubGraphs

We present a framework for learning Node Embeddings from Static Subgraphs (NESS) using a graph autoencoder (GAE) in a transductive setting. Moreover, we propose a novel approach for contrastive learning in the same setting. We demonstrate that using static subgraphs during training with a GAE improves node representation for link prediction tasks compared to current autoencoding methods using the entire graph or stochastic subgraphs. NESS consists of two steps: 1) Partitioning the training graph into subgraphs using random edge split (RES) during data pre-processing, and 2) Aggregating the node representations learned from each subgraph to obtain a joint representation of the graph at test time. Our experiments show that NESS improves the performance of a wide range of graph encoders and achieves state-of-the-art (SOTA) results for link prediction on multiple benchmark datasets.