A clustering method and system based on multi-view attribute missing graphs

The present invention discloses a clustering method for multi-view attributed graphs with missing attributes. First, a generative adversarial network structure comprising an encoder, a decoder, and a discriminator is adopted to impute the missing attribute information, obtaining a complete attribute matrix. Then, a structure consisting of one encoder and two decoders (a structure decoder and a feature decoder) is employed. After continuously training the encoder and decoders, the input data is fed into the encoder to obtain view-specific embeddings Z_m. Based on a self-attention mechanism, the obtained view-specific embeddings Z_m are fused according to view weights to produce the final graph embedding Z. Finally, clustering is performed on the graph embedding Z to partition the graph nodes into clusters, yielding the final clustering result. The present invention adopts a multi-view attributed graph, in which each view in the multi-view data is defined to include both attribute information X_m and structural information A_m. This addresses the technical issue of single-view perspectives in existing multi-view clustering methods, thereby enhancing the application value of the clustering method for multi-view attributed graphs with missing attributes.