Graph pooling plays an important role in graph-level tasks, which cannot be ignored. Most of the current hierarchical graph pooling methods rely on top-k for node selection. However, this top-k node selection is data sensitive and usually requires experience to select the best k value. In addition, the scoring of nodes takes into account graph information from a single view. In this article, we introduce multiview learning and deep reinforcement learning (RL) into graph pooling and propose a new graph pooling HMTPool that incorporates multiview information and adaptively selects the best k value when using top-k for node selection. The effectiveness and flexibility of our proposed HMTPool have been validated on several graph classification datasets. To the best of our knowledge, we are the first to apply deep RL to graph pooling, which opens up the possibility of combining deep RL and graph pooling.

Graph neural networks (GNNs) are extensions of deep neural networks to graph-structured data. It has already attracted widespread attention for various tasks such as node classification and link prediction. Existing research focuses more on graph convolutional neural networks (GCNs). However, it is usually overlooked that graph pooling can obtain graph representations by summarizing and down-sampling node information. Meanwhile, existing graph pooling methods mainly use top-k for node selection, but most of them consider only single view information when scoring nodes, and the k values in top-k are usually selected empirically. This work proposes the hierarchical multiview top-k pooling (HMTPool) with deep-Q-networks, which scores nodes taking into account multiview information (considering graph structure and features) and does not rely on the empirical adaptive selection of the best k value. HMTPool is a two-stage process. It first uses a variant GCN and multilayer perceptron to scor...