A Comparative Analysis of Information Cascade Prediction Using Dynamic Heterogeneous and Homogeneous Graphs

Understanding information cascades in social networks is a critical research area with implications in various domains, such as viral marketing, opinion formation, and misinformation propagation. In information cascade prediction problem, one of the most important factors is the cascade structure of the social network, which can be described as a cascade graph, global graph, or an r-reachable graph. However, the majority of existing studies primarily focus on a singular type of relationship within the social network, relying on the homogeneous graph neural network. We introduce two novel approaches for heterogeneous social network cascading and analyze whether heterogeneous social networks have higher predictive accuracy than homogeneous networks, taking into account the potential differential effects of temporal sequences on the models. Further, our work highlights that the selection of edge types plays an important role in the accuracy of predicting information cascades within social networks.