Modelling the Latent Semantics of Diffusion Sources in Information Cascade Prediction

Figure 4

The diagram of adversarial training framework. Firstly, the cascade sequences are divided into two categories according to the cascade sequences number of their diffusion sources. Secondly, the cascade sequences are encoded with the encoder-based cascade prediction model. Finally, the model is trained in an adversarial paradigm which minimizes the cascade prediction loss and meanwhile maximizes the discriminator loss.