Abstract: When there are different types of objects in the network,the relationships between objects would be various,and the structure of the network would become even more complex.To deal with the problem heterogeneousness of networks,this paper proposed a neural network based vectorization representation method of the heterogeneous networks.For the heterogeneous network with two types of images and text,multi-hierarchical convolution neural network was adopted to map the original image into a latent feature space,and fully-connected neural network was adopted to map the text object into the same latent feature space.In the latent feature space,the similarities between images,texts,and even image and text can be calculated by the same distance computing method.In the experiment,the proposed method is applied to test a variety of applications in heterogeneous networks,and the results show that the proposed method is effective.

Key words: Heterogeneous networks,Neural network,Vectorized representation,Embedded vector