基于自我监督学习策略的层智能图卷积网络

为了解决当前图卷积网络需要依赖大型数据集,从而导致时间和空间复杂度上升问题,提出了基于自我监督学习策略的层智能图卷积网络(RRLFS-L-GCN)。首先,通过在层智能图卷积网络(layer-wise graph convolutional network, L-GCN)中添加多任务机制以提高算法的泛化能力;然后,设计一种随机删除固定步长边(aandomly remove links with a fixed step, RRLFS)的自我监督学习策略,从而提出基于自我监督学习策略的层智能图卷积网络算法;最后,通过边预测验证RRLFS-L-GCN的性能。实验结果表明,该算法的识别率最高可达97.13%。对于Cora测试集,该算法所得识别准确率比未改进的层智能图卷积网络算法提高了6.73%。对于PubMed测试集,该算法所得识别准确率比未改进的层智能图卷积网络算法提高了8.13%。与图卷积网络相比,在Citeseer数据集上,识别准确率提高了18.43%。

Layer-wise graph convolutional network based on self-supervised learning strategy

To solve the problem that the current graph convolutional network needs to rely on large datasets, which leads to increased time and space complexity, this research proposed a layer-wise graph convolutional network based on self-supervised learning strategy(RRLFS-L-GCN). Firstly, it added an multi-task mechanism into the layer-wise graph convolutional network(layer-wise graph convolutional network, L-GCN) to improve the generalization ability of the algorithm. Then, it designed a self-supervised learning strategy that randomly removed fixed-step links(randomly remove links with a fixed step, RRLFS). Therefore, it proposed a layer-wise graph convolutional network algorithm based on a self-supervised learning strategy. Finally, it used link prediction which was to verify the performance of RRLFS-L-GCN. Experimental results show that this algorithm has the highest recognition rate, up to 97.13%. For the Cora testset, this algorithm obtains 6.73% accuracy higher than that of the unimproved layer-wise graph convolutional network algorithm. For the PubMed testset, this algorithm obtainss 8.13% accuracy higher than that of the unimproved layer-wise graph convolutional network algorithm. Compared with the graph convolutional network, it improves the recognition accuracy rate on the Citeseer dataset, which is 18.43%.