Attribute community detection based on attribute edges weights fusion and graph embedding factorization

In recent years, factorization combined with attribute information has played an important role in attribute community detection. However, previous studies focused more on connecting the original shallow topological and attribute data. They ignored the potential representation structure of the network. To solve the problem that shallow information cannot fully represent the network structure, this paper proposes an attribute node classification method based on Attribute Edges weights Fusion and graph Embedding Factorization, called AEFEF. First, AEFEF converts topological information and attribute information into corresponding matrices representing node associations. Then, AEFEF constructs a new adjacency matrix by increasing the weights of shared edges between the topology structure and attribute structure. This operation can strengthen the tightness between nodes. Second, to explore the potential community structure, feature embedding is obtained by factorizing the attribute similarity matrix. Meanwhile, the new adjacency matrix is designed as a weight matrix to make the feature embedding between related nodes more similar. Finally, semi Non-negative Matrix Factorization (NMF) is introduced to modify the feature embedding by converting the negative values into positive values. Then the embedding is factorized to generate the membership matrix. At the same time, the network with a rich structure is decomposed with topological data as the main component. Otherwise, attribute information is the main component of NMF used to increase the accuracy of node classification. AEFEF is compared with 10 state-of-the-art algorithms on 7 real network datasets. The results reveal that AEFEF can improve the precision of attribute community detection.