Final Report: Local Structure and Evolution for Cascade Prediction

Information cascades in large social networks are complex phenomena governed by such diverse forces as the diffusion medium, user trends, and the information content itself. While these influences might be difficult to understand and model directly, the structure and evolution of the cascade can be used as a proxy for the sum effect. To this end, we study information cascades on Twitter and focus especially on the utility of local graph structure analysis for categorizing, understanding, and predicting cascade evolution. Specifically, after presenting basic statistics, we categorize cascades based on size and growth. We then count graphlet frequencies for different cascade categories in order to understand their structural differences. We also explore how these differences arise by counting graphlet frequencies at different points during the evolution process. Finally, we construct a machine learning framework for predicting cascade size.