Graph attention neural network method based on shortest path

The invention discloses a graph attention neural network method based on shortest path attention. The shortest path is the path with the minimum weight sum of each edge in the paths from the source node to the target node. The shortest path is used as an attention weight coefficient, and the obtained result has high interpretability. And meanwhile, the shortest path is directly calculated outside to serve as the attention coefficient weight, so that tedious calculation is not needed during training, and the training time is greatly shortened. And compared with some baseline methods on a cora data set, some more excellent results are obtained. 本发明公开了一种基于最短路径注意力的图注意力神经网络方法。最短路径即源节点到目标节点所经过路径中，各边权值和最小的一条路径。最短路径作为注意力权重系数，得到的结果具有较强的可解释性。同时因为在外部直接计算好了最短路径作为注意力系数权重，在训练时就不必再进行繁琐的计算，大大的降低了训练的时间。在cora数据集上与一些基线方法做了对比也得到了一些更优秀的结果。