GraphPaper
目录
1. DDGK
Al-Rfou R, Perozzi B, Zelle D. Ddgk: Learning graph representations for deep divergence graph kernels[C]//The World Wide Web Conference. 2019: 37-48.
end-to-end supervised graph classification: learn a intermediate representation of an entire graph as precondition in order to solve the classification task;graph representation learning:
- feature engineering: graph’s
clustering coefficient,its motif distribution,its spectral decomposition, limited to composing only known graph- encode algorithmic heuristics from graph isomorphism literation
DDGK capture the attributes of graphs by usign them as features for several classificaiton problems, capturing the local similarity of graph pairs and the global similarity across families of graphs.
- Deep divergence graph kernel: learnable kernel does not depend on feature engineering or domain knowledge.
- Isomorphism Attention: cross-graph attention mechanism to probabilisticly align representations of nodes between graph pairs.
- Embedding based kernels
.1. Graph Encoding
an encoder is given asingle vertex and it is expected to predict its neighbors
.2. Cross-Graph Attention
an attention mechanism for aligning graphs basedon a set of encoded graph representations.
.3. DDGK
