Learning Spread Probability Based on User’s Influence and Flavour

Abstract

In this paper, we propose a new topology free model of social network information dissemination, IPM for short, based on the representation learning method. We construct two latent spaces: user influence space and user interest space. Each user and each propagation item are embedded to feature vectors in latent space. When the model predicts the probability of a user receiving a propagation item, it considers not only the influence from other users but also the user’s flavour for the propagation item. We calculate propagation probability according to the distance between vectors. Experimental results on real data show that the model can simulate diffusion and predict more accurately. It is superior to the state-of-the-art model in many metrics.