• A
  • A
  • A
  • ABC
  • ABC
  • ABC
  • А
  • А
  • А
  • А
  • А
Regular version of the site

Article

A method for determining information diffusion cascades on social networks

Eastern-European Journal of Enterprise Technologies. 2018. Vol. 6. No. 2. P. 61-69.
Nguyen V. A., Duong N. S., Nguyen T. T., Sergey Kuznetsov, Nguyen T. Q.

Information diffusion on social networks has many potential real-world applications such as online marketing, e-government campaigns, and predicting large social events. Modeling information diffusion is therefore a crucial task in order both to understand its diffusion mechanism and to better control it. Our research aims at finding what factors might influence people in adopting a piece of information that is being shared on a social network. In this study, the traditional independent cascade model for information diffusion is extended with discrete time steps. The proposed model is capable of incorporating three different sources of diffusion influence: user-user influence, user-content preference, and external influence. Specifically, these sources of influence are quantified into real values of diffusion probability. To calculate user-user influence, we adopt and extend the disease transmission model according to the role of the user who diffuses the content. User-content preference, which measures the correlation between user preference and the adopted contents, is calculated based on a topic-based model. External influence is detected in a diffusion time step and is quantified and incorporated into our model for the next diffusion time step by applying and solving a logistic function. Moreover, the process of information diffusion is characterized by constructing a tree of information adoption and the diffusion scale is quantified by predicting the number of infected nodes. It is found that these sources of influence, especially external influence, play a significant role in information diffusion and eventually affect the shape and size of the diffusion cascade. The model is validated on both synthetic and real-world datasets. Experimental results confirm the advantage of our proposed method, which significantly improves over the previous models in terms of prediction accuracy.