Multi Order Deep Learning Model for Community Detection

Mining the community structure inside the attribute network allows for analysing additional network nodes with much practical use. The structural information of the attribute network may be successfully included in the graph convolutional neural network, which also generates the node’s feature representation. This enables the community structure to be successfully implemented. However, most graph convolution algorithms now use fixed low-order graph convolution. This method ignores the variety of network architectures and only analyses neighbours in each node’s first or second order. It also does not make full use of the node connection. Furthermore, there is no way to avoid the initial network topology’s sparsity, which makes community discovery less effective. It is hypothesised that a K-order graph convolution community detection technique (KGCN) that incorporates attribute and structural information would overcome the issues above. This strategy, which exploits the nodes’ high-order structure to locate communities, successfully overcomes the original network’s sparsity. To lessen the sparsity of the actual network topology, the current network is redesigned first, utilising the node attribute information. The nodes are then encoded using the K-order graph convolutional encoder to generate their feature representation. Finally, spectral clustering is utilised to identify the network’s communities. The trial fimdings reveal that, when applied to four distinct real-world datasets, the KGCN technique outperforms other algorithms in community discovery.