Learnable Models for Information Diffusion and its Associated User Behavior in Micro-blogosphere.

摘要

There are two major contributions: (1) a new efficient and effective method to detect a burst in information diffusion from the observed data, and (2) a new opinion formation model that takes people's past opinion into account. Both problems are formulated as a maximum likelihood problem in which the likelihood of observing the data from the model is maximized. For (1) the algorithm was tested against the real Twitter data of the 2011 Tohoku earthquake and tsunami, and for (2) the algorithm was tested against the real opinion diffusion data from a Japanese word-of-mouth communication site for cosmetics. Both confirmed that the algorithms are efficient and work as expected.