Classification of complex network using bidirectional LSTM
María del Carmen Soto Camacho, Aldo Ramírez-Arellano · Proceedings of the World Congress on Electrical Engineering and Computer Systems and Science · 2024
Classifying complex networks has become highly significant in network analysis.Graph Neural Networks (GNNs) have successfully succeeded in this particular task.However, GNNs suffer from limited network representation as they rely solely on scalarbased features for node properties.GNNs' message-passing methods suffer from oversmoothing and lack global information on complex networks.Data augmentation is also impossible for GNN, so obtaining reasonable classifications in small datasets is an open issue.Deng's entropy of complex networks captures the network's topology and the valuable information generated by the nodes and edges to solve the problems arising from complex networks and unlock their full potential.Our proposed method utilizes Deng's entropy to calculate an entropy sequence incorporating local and global features at multiple scales.We then combine the entropy sequences for nodes and edges into a matrix fed into a bidirectional bLSTM network to perform complex network classification.