On Exploring Node-feature and Graph-structure Diversities for Node Drop Graph Pooling

A pooling operation is essential for effective graph-level representation learning, where the node drop pooling has become one mainstream graph pooling technology. However, current node drop pooling methods usually keep the top-k nodes according to their significance scores, which ignore the graph diversity in terms of the node features and the graph structures, thus resulting in suboptimal graph-level representations. To address the aforementioned issue, we propose a novel plug-and-play score scheme and refer to it as MID, which consists of a Multidimensional score space with two operations, i.e., flIpscore and Dropscore. Specifically, the multidimensional score space depicts the significance of nodes through multiple criteria; the flipscore encourages the maintenance of dissimilar node features; and the dropscore forces the model to notice diverse graph structures instead of being stuck in significant local structures. To evaluate the effectiveness of our proposed MID, we perform extensive experiments by applying it to a wide variety of recent node drop pooling methods, including TopKPool, SAGPool, GSAPool, and ASAP. Specifically, the proposed MID can efficiently and consistently achieve about 2.8\