Broad Learning Based on Subgraph Networks for Graph Classification

Abstract

Many real-world systems can be naturally represented by networks, such as biological networks, collaboration networks, software networks, social networks, etc., where subgraphs or motifs can be considered as network building blocks with particular functions to capture mesoscopic structures. Most existing studies ignored the interaction between these subgraphs, which could be of particular importance to represent the global structure at the subgraph level. In this chapter, the concept of subgraph network (SGN) is introduced and applied to network models, with algorithms designed for constructing the 1st-order and 2nd-order SGNs, which can be easily extended to build higher-order ones. Furthermore, these SGNs are used to expand the structural feature space of the underlying network, beneficial for network classification. The experiments demonstrate that the structural features of SGNs can complement that of the original network for better network classification. However, SGN model lacks diversity and is of high time-complexity, making it difficult to be widely applied in practice. Then, sampling strategies are introduced into SGNs and a novel sampling subgraph network (S2GN) model is designed, which is scale-controllable and of higher diversity. Further, a broad learning system (BLS) is introduced into graph classification, which fully utilizes the information provided by the S2GNs of different sampling strategies and thus can capture various aspects of the network structure more efficiently. Extensive experiments demonstrate that, by comparing with the SGN model, the S2GN model has much lower time-complexity, which together with BLS can enhance various graph classification methods.