The capability to preserve, model, and predict information cascades over online social networks has many theoretical and practical implications, e.g., for marketing, recommendation filtering, and studying of societal behavior. Massive empirical data sets on users' online social activities are being collected, but they are often too big to analyze. The goal of this project is to design graph generative models and summarization techniques that can preserve pertinent information about online social interactions that lead to interesting events, e.g., viral diffusion of information or drastic change of user behavior. Towards this end, the project will develop generative models that can capture the dynamic evolution of user activity graphs (UAGs), which represent a sequence of inter-user communications/actions. At the microscopic level, the project will investigate user influence in the recruitment process. In addition, the project will design graph summarization techniques that can achieve good tradeoffs between data compression ratio, computational complexities, and accuracy in answering fundamental queries, such as identifying 'influential' sources of information cascades.
Broader Impact: If successful, this project will provide efficient methods for storing/archiving massive graph data to support longitudinal study on the dynamics of online social interactions, which has potential impact on multiple disciplines (e.g., economics, history, political sciences, and social science). This project will help train future researchers and practitioners in online social networking and network science through classroom curriculum development and online teaching. The PIs will continue ongoing diversity recruitment and outreach to K-12 students. Transfer of technology into commercial practice is made feasible through partnership with industry.
