Abstract:
The existing attribute graph embedding algorithms cannot capture the influence of heterogeneous attribute information on topology structure flexibly and cannot deal with the heterogeneity of attribute well. A attributed graph embedding learning algorithm based on modal fusion is proposed. The algorithm could map different types of information smoothly into the same semantic space while preserving the topological structure. A modal fusion module was designed to utilize complementary information between different modes. Experimental results show that the proposed algorithm has better performance than other comparison methods on various tasks.