Information Diffusion Algorithms Based On Latent Space Vector Models

With the rapid growth of social networks, information diffusion has drawn widely attention in recent years. Information diffusion is defined as the process by which information is reached through interactions between users. Information diffusion can be divided into four research categories,such as burst topic, community detection, information source detection,and link prediction, and diffusion prediction is a branch of link prediction.Diffusion prediction aims to predict which users will involve in the diffusion process. At present, the traditional methods of information diffusion can be divided into similarity model and probability model, but these models have limitations: 1) For similarity model, it needs the underlying structure network and other user profile as the input feature,nevertheless, the data has problem in acquisition; 2) For probability model,it focuses on the global information structure, with low efficiency,therefore it is unable to deal with large-scale datasets.On the basis of the above limitations, this paper proposes an information diffusion algorithm based on the latent space vector algorithm.With using triplet, the model transforms the listwise problem into pairwise,and it enhances the efficiency a lot, and therefore it reduces the time complexity. The model was tested on three real datasets with three other baseline models, and the latent space vector model performs the best.Finally, with topic attributes taken into consideration, the paper proposes an optimization model.