Network structures are prevalent in many real-world datasets, such as social networks among people, geospatial networks among city regions, and biological networks among genes or brain regions. The underlying structures not only affect the global and local properties of these networks but also guide the evolution of different network processes. For example, a gene behaving abnormally affects other genes working together with it, leading to a disease such as cancer. Similarly, rumors are often spread via the friendship among users in an online social network. Understanding and predicting these network processes are thus important but challenging tasks that find applications in many different real-world scenarios.
In this talk, I will first present a coupled tensor decomposition for predicting group-level popularity of online content. Here, the user groups in online social networks are found by clustering a network-constrained activity graph of users. Second, I will discuss the problem of summarizing network processes with a set of compact and representative patterns that give insights into how these processes happen and evolve. To deal with NP-hardness and the exponential subgraph search space, I devise a greedy algorithm with the help of sampling.