Exploring and Mining Attributed Sequences of Interactions

We consider entities interacting over time: individuals meeting, customers buying products, etc., each entity being labeled with some information that may depend on time, and possibly extracted from the interaction nature. Capturing the dynamics as well as the structure of these interactions is of crucial importance for analysis. We are interested here in mining sequences of such interactions. For that purpose, we define core closed patterns in this context and introduce algorithms to enumerate them on a labeled stream graph. We run experiments on two real-world datasets, one representing interactions among students and the other representing citations between authors.