Learning to Predict Social Influence in Complex Networks

In this project the following five major achievement are made. 1) Two kinds of information diffusion models incorporating asynchronous time delay and a method to select models that better explains the observation, 2) a method to learn and predict opinion share using a variant of voter model, 3) a method to detect changes in opinion share, 4) a method to detect changes in diffusion probability, and 5) a method to learn the strength of opinion. Each of them uses probabilistic models and machine learning techniques to learn the model parameters from the observation. These are important steps to construct basic methods for learning to predict social influence in complex networks. All of them have been published in international conferences and/or international journals. In total there are 17 publications and they are included in the final report.