Prediction Model for Non-topological Event Propagation in Social Networks

The spread of events happens all the time in social networks. The prediction of event propagation has received extensive attention in data mining community. In prior studies, topologies in social networks are usually exploited to predict the scope of event propagation. User’s action logs can be obtained in reality, but it is difficult to get topologies in social networks. In this paper, NT-GP, a prediction model for non-topological event propagation, is proposed. Firstly a time decay sampling method was used to extract the walk paths from user’s action log, and then deep learning method was applied to learn the sampling paths and predict the future propagation range of the target event. Extensive experiments demonstrate effectiveness of NT-GP.