“SEISMIC: A Self-Exciting Point Process Model for Predicting Tweet Popularity.”
In this paper, author presents us a new social phenomenon that we call social information diffusion by a cascading behavior of re-sharing post of one post. Social network website like Facebook, twitter allow user to post, comment, and share post; and to understand such social phenomenon, it is important to analyze this in term of big information cascades and understanding such behaviors in forecasting information outbreak when one post got widely popular. In this paper, they have introduced us the theory of self-exciting point process to develop a statistical model that can predict the final number of sharing. Nowadays, the prediction of re-share size is based to type of approaching: feature based methods and point process based methods. Feature based methods are first extract an list of potential relevant feature then apply algorithms such as regression and probabilistic collaborative filtering etc. to estimate the size. This approaching is expensive and strictly important for quality of feature. The point process based methods is based on point process that directly models the information of an information cascades. This type of approaching is well developed for complementary problem of network inference. In their paper, they have develop a self-exciting model of information cascades for predicting the total number of re-shares of a given post. The different from SEISMIC to existing Hawkes process is that they introduced the process intensity depends on stochastic process. It is used to estimating the spreading rate of a given information cascade, determining the state of cascades, and predicting the final size of information cascades.
This is quite different from what we have learnt from class about a simple general cascade model. In chapter 16, the order of making decision is already determined, and it is not a real time cascades since we can’t go back to player one to change his decision when the later majority decision is opposite to his. Yet in real time, one can always change his mind; for instance, if I decide not to share a post today, but I can reconsider and share it tomorrow. Beside the above difference, in chapter 19, our model also exist a threshold used to determine whether one should follow the behavior of previous or neighbors. But in this case, it is a cycle which allow all player to change their mind along with time change by given initial number of people follow one behavior. This is likely to be the same approach to the research paper, but we are likely to be feature based cascades that we apply the probability along with our determination. Moreover, our model is all based on a drawing network not real time problem, therefore we will try to determine whether a complete cascades will achieve by initial adopters using relationship of clusters and cascades: for threshold of q, the remaining network contains a cluster of density greather than 1-q will cause the failure for initial adopters to cause a complete cascades. However, the real life problem will be impossible to cause everyone to follow the cascade, and it is nearly impossible even for a group of people.
Work cited page
1.Zhao, Qingyuan, et al. “SEISMIC: A Self-Exciting Point Process Model for Predicting Tweet Popularity.” Proceedings of the 21th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. ACM, 2015.
2.Easley, David, and Jon Kleinberg. Networks, Crowds, and Markets Reasoning about a Highly Connected World. New York: Cambridge UP, 2010. Print.