单个较大非均匀超图聚类旨在将非均匀超图包含的节点划分为多个簇,使得同一簇内的节点更相似,而不同簇中的节点更不相似,具有广泛的应用场景.目前,最优的基于超图神经网络的非均匀超图聚类方法CIAH(co-cluster the interactions via attentive hypergraph neural network)虽然较好地学习了非均匀超图的关系信息,但仍存在两点不足:(1)对于局部关系信息的挖掘不足;(2)忽略了隐藏的高阶关系.因此,提出一种基于多尺度注意力和动态超图构建的非均匀超图聚类模型MADC(non-uniform hypergraph clustering combining multi-scale atten-tion and dynamic construction).一方面,使用多尺度注意力充分学习了超边中节点与节点之间的局部关系信息;另一方面,采用动态构建挖掘隐藏的高阶关系,进一步丰富了超图特征嵌入.真实数据集上的大量实验结果验证了MADC模型在非均匀超图聚类上的聚类准确率(accuracy,ACC)、标准互信息(normalized mutual information,NMI)和调整兰德指数(adjusted Rand index,ARI)均优于CIAH等所有Baseline方法.
Non-Uniform Hypergraph Clustering Model Combining Multi-Scale Attention and Dynamic Construction
A single large non-uniform hypergraph clustering is designed to divide the nodes contained in non-uniform hypergraphs into multiple clusters,so that the nodes in the same cluster are more similar,while the nodes in different clus-ters are less similar,which has a wide range of application scenarios.At present,the optimal hypergraph clustering method based on hypergraph neural network is CIAH(co-cluster the interactions via attentive hypergraph neural network).Although it can learn the relationship information of non-uniform hypergraphs better,there are still two deficiencies:(1)local relation information mining is insufficient;(2)the hidden higher-order relationship is ignored.Therefore,a non-uniform hypergraph clustering model based on multi-scale attention and dynamic hypergraph construction MADC(non-uniform hypergraph clustering combining multi-scale attention and dynamic construction)is proposed.On the one hand,multi-scale attention is used to fully learn the local relation information between nodes in the hyperedge.On the other hand,dynamic construction is used to mine hidden higher-order relationships,which further enriches hypergraph feature embedding.Extensive experiments on real datasets demonstrate that the clustering accuracy(ACC),normalized mutual information(NMI)and adjusted Rand index(ARI)of MADC model on non-uniform hypergraph clustering are better than all Baseline methods such as CIAH.
万方数据
维普
