基于重启随机游走的图自编码器

Adaptive graph auto-encoder based on restarted random walk

针对现有的图自编码器无法捕捉图中节点之间的上下文信息的问题,提出基于重启随机游走的图自编码器。首先,构造两层图卷积网络编码图的拓扑结构和特征,同时进行重启随机游走捕捉节点之间的上下文信息;其次,为了聚合重启随机游走和图卷积网络获得的表示,设计自适应学习策略,根据两种表示的重要性自适应地分配权重。为了证明该方法的有效性,将图最终的表示应用于节点聚类和链路预测任务。实验结果表明,与基线方法相比,提出的方法实现了更先进的性能。

Aiming at the problem that the existing graph auto-encoders can’t capture the context information between nodes of the graph,this paper proposed adaptive graph auto-encoder based on restarted random walk.It firstly constructed a two-layer graph convolutional network to encode the topology and features of the graph.At the same time,it carried out the restarted random walk to capture the context information between nodes.Next,it designed the adaptive learning strategy to aggregate the representations obtained by restarted random walk and graph convolutional network.It can adaptively assign weights according to the importance of the two representations.To prove the effectiveness of the proposed method,this paper applied the final representations of the graph to the task of node clustering and link prediction.The experimental results show that the proposed method achieves more advanced performance compared with the baseline methods.