Structural Entropy Guided Graph Hierarchical Pooling
This work addresses limitations in graph pooling for machine learning tasks like graph and node classification, offering an incremental improvement over existing hierarchical methods.
The paper tackles the problems of local structure damage and suboptimal performance in hierarchical graph pooling by proposing SEP, a structural entropy-guided pooling method that globally optimizes cluster assignments without layer-specific compression quotas. The results show that SEP outperforms state-of-the-art methods on graph classification benchmarks and achieves superior performance in node classification tasks.
Following the success of convolution on non-Euclidean space, the corresponding pooling approaches have also been validated on various tasks regarding graphs. However, because of the fixed compression quota and stepwise pooling design, these hierarchical pooling methods still suffer from local structure damage and suboptimal problem. In this work, inspired by structural entropy, we propose a hierarchical pooling approach, SEP, to tackle the two issues. Specifically, without assigning the layer-specific compression quota, a global optimization algorithm is designed to generate the cluster assignment matrices for pooling at once. Then, we present an illustration of the local structure damage from previous methods in the reconstruction of ring and grid synthetic graphs. In addition to SEP, we further design two classification models, SEP-G and SEP-N for graph classification and node classification, respectively. The results show that SEP outperforms state-of-the-art graph pooling methods on graph classification benchmarks and obtains superior performance on node classifications.