A Double-layer Attention Heterogeneous Graph Neural Network Based on Coupled P System

Abstract

Heterogeneous graphs have various types of nodes and edges and its different types of nodes have features or attributes of different dimensions. Therefore, they can express rich and complex semantic information. This feature of heterogeneous graphs makes them widely used in researches of recent years. At the same time, the abundant semantic information also brings challenges to select heterogeneous neighbors of nodes for information processing process. It is difficult to aggregate nodes of different dimensions and types into the same feature space to compute node embedding. In addition, we also have difficulties in assigning different weights to various node types when aggregating heterogeneous neighbors. In this paper, we propose a double-layer attention heterogeneous graph neural network based on coupled P system (PDAHN) to compute the node embedding. The attention mechanism in this method is divided into two layers, node-layer attention and meta-path-layer attention. PDAHN can assign different weights to different nodes and meta-paths according to the different importance of them by weighted aggregation. The whole process is carried out in the coupled P system. Experiments on two datasets show that the method proposed in this paper outperforms the performance of state-of-the-art methods.