Research Seminar by LI Yuchen

The prevalence of online social networks has prompted much attention on information diffusion, as a piece of information could quickly become pervasive through the “word-of-mouth” propagation among friends in the network. As a key algorithmic problem in information diffusion research, influence maximization (IM) has been extensively studied and has immense value in many applications such as viral marketing and political campaign. Nevertheless, existing techniques focus on homogenous and static influence but in reality social influences are inherently contextual, e.g., topic, ego and time. For example, messages with a different topic propagate to different group of audiences. Similarly, people have unique social influential characteristics over each other and the characteristics often vary with time. It is thus computationally infeasible to rerun the IM algorithm for each social influence network induced by a different contextual perspective.

In this talk, I will present my research on supporting scalable context-aware social influence processing. In particular, my approach is to construct compact indices to capture the commonalities among contextual perspectives, e.g., topics that are similar and time periods that are nearer. Furthermore, I will demonstrate how to utilize the indices to process contextual influence queries efficiently with theoretical guarantees. I will elaborate my approach using three distinct contextual influence applications: targeted viral marketing (topic), personalized influential selling point discovery (ego) and dynamic influential maximization (time). I will conclude the talk by discussing the challenges and opportunities in social influence researches.

 

About the Speaker