USER REPRESENTATION JOINT LEARNING FRAMEWORK BASED ON DIFFUSION SEQUENCE AND NETWORK STRUCTURE

In order to fully exploit the correlation between diffusion information and network structure, a user representation joint learning framework based on diffusion sequence and network structure is proposed. Two maximum likelihood estimation objectives were defined on the observation information diffusion sequence and social network structure respectively to learn user representation, and a multi task learning algorithm based on learning representation was designed for model optimization. In addition, a sampling-based algorithm was designed to optimize the joint model. Experiments on two social media datasets show that the model can achieve better prediction results in diffusion prediction and link prediction tasks, and has better robustness.