Social media data analysis with Graph Neural Networks (GNN)-1

Predicting Influence Probabilities using Graph Convolutional Networks

Citation: Jing Liu, Yudi Chen, Duanshun Li, Noseong Park, Kisung Lee, and Dongwon Lee, “Predicting Influence Probabilities using Graph Convolutional Networks,” In the 2019 IEEE International Conference on Big Data (IEEE Big Data), 2019.

Highlights

• Proposed a node-edge co-convolution graph neural network architecture
• Demonstrated its effectiveness in predicting information diffusion (i.e., influence probability) over social networks

The aim of this work is to predict the probability that one person influences another person over the social network so that it can more precisely to predicte information cascade over social network. Graph is used to model the social network, the influence probability to be predicted is one edge feature. We conjecture that the co-convolution of vertex features and edge features following the information flow direction could improve the prediction accuracy. The proposed node-edge co-convolution is illustrated as follows. The experiment result shows very high prediction accuracy with mean absolute percentage error less than 0.1.

Covolve Hidden vectors for node (left) and edge (right) following the information flow direction.

Overall Architecture