<p>As a type of membrane computing model, the spiking neural P systems are inspired by neuronal communication achieved through spiking signals. In this study, the Controllable-Channel Hierarchical Spiking Neural P systems (CHSNP systems) are constructed. In CHSNP systems, astrocytes are used to control the synaptic connections of the neuronal network to achieve dynamic channel transmission between neurons. Most current heterogeneous graph neural networks are still deficient in capturing higher-order semantic information. Based on this, the controlled channel hierarchical spiking neural P systems based on heterogeneous graph node classification are developed in this study. Specifically, a new higher-order semantic fusion graph is constructed through multiple metapaths and learned on this new graph. In addition, we dynamically aggregate neighbor information by utilizing feature-similarity based attention. Finally, we demonstrate through extensive experiments that our model’s superiority over advanced models across three real-word datasets.</p>