010

Title:

Fast method for calculating links in spatial social networks

Discipline: Information Systems

Presenter:

Jiaxin Du

Abstract:

The weights of edges and the features of nodes within spatial (geolocated) social networks can be used to model and predict phenomenon such as information diffusion and disease spreading. Though the nodes features and information spreading outcomes in spatial social networks are visible, how these nodes are linked during the diffusion mechanism cannot be directly observed. Hence the structure of the diffusion network is unknown. Using traditional non-(semi-)parametric method to identify such links is time consuming and computationally intensive, which will make the method difficult to be methodologically extended and empirically applied. We propose a fast method of calculating the links in spatial social network utilizing only the features and status of nodes. It substantially reduces the calculation complexity compared to the non-(semi-)parametric method. We employ weighted dense networks and weighted sparse networks to validate the effectiveness of our method, because those networks are more representative of real world social networks than binary networks. The simulation results shows over $80\%$ accuracy even when we only have small amounts of observation time. This method was parallel computing in nature and can be extended to high-dimensional features.

Author(s):

Jiaxin Du

Funding Acknowledgements:

Anonymous