Node Embeddings via Neighbor Embeddings
About
Node embeddings are a paradigm in non-parametric graph representation learning, where graph nodes are embedded into a given vector space to enable downstream processing. State-of-the-art node-embedding algorithms, such as DeepWalk and node2vec, are based on random-walk notions of node similarity and on contrastive learning. In this work, we introduce the graph neighbor-embedding (graph NE) framework that directly pulls together embedding vectors of adjacent nodes without relying on any random walks. We show that graph NE strongly outperforms state-of-the-art node-embedding algorithms in terms of local structure preservation. Furthermore, we apply graph NE to the 2D node-embedding problem, obtaining graph t-SNE layouts that also outperform existing graph-layout algorithms.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Node Classification | Photo (test) | Mean Accuracy94.3 | 241 | |
| Link Prediction | Citeseer | AUC100 | 174 | |
| Link Prediction | Cora | AUC (Cora)100 | 94 | |
| Node Classification | Cora (random) | Accuracy84.3 | 79 | |
| Link Prediction | Photo | AUC-ROC99.7 | 52 | |
| Link Prediction | Computers | AUC-ROC99.4 | 50 | |
| Link Prediction | arXiv | AUC100 | 40 | |
| Link Prediction | Pubmed | AUC-ROC100 | 29 | |
| 2-hop neighbor recall | Citeseer | Top-10 2-hop Neighbor Recall44.5 | 12 | |
| 2-hop neighbor recall | Cora | Top-10 2-hop Neighbor Recall46.3 | 12 |