Detecting and Explaining Anomalous Influence Cascades in Social Networks

While Ghawi et al. (2021) observe that small influence cascades are becoming more prevalent over time among intellectuals, current analyses typically focus on aggregate statistics rather than the surprising, outlier cases. This project proposes building machine learning tools that flag anomalous cascades—instances where influence spreads much more (or less) than expected, considering network structure and historical diffusion. By integrating explainable AI, the system wouldn’t just detect these outliers, but also generate interpretable explanations (e.g., "this cascade was unusually broad given the low initial influencer centrality"). This approach extends the work of Ghawi et al. by shifting the focus from broad temporal trends to the investigation of individual, unexpected diffusion events, potentially uncovering hidden mechanisms, emergent behaviors, or external triggers. The significance lies in enhancing our ability to spot and react to rare, high-impact diffusion phenomena—crucial for applications from viral marketing to misinformation control.

References:

1. Diffusion dynamics of influence in a social network of intellectuals. Raji Ghawi, Cindarella Petz, J. Pfeffer (2021). Social Network Analysis and Mining.
2. Diffusion dynamics of influence in a social network of intellectuals. Raji Ghawi, Cindarella Petz, Jürgen Pfeffer (2021). Social Network Analysis and Mining.