Prof. Dr. Myra Spiliopoulou

Otto-von Guericke-Universität Magdeburg

Mining Streams of Temporal Objects

Zusammenfassung:

Many business applications require data mining upon structured objects that have temporal properties and exhibit stream behavior. Customer data are a typical example. First, they encompass demographics, transactions and product descriptions, all of which should be taken into account for customer segmentation or churn prediction. Furthermore, transactions constitute a stream, while customer preferences are of temporal nature and and may change due to external or endogenous factors (like ageing).

Model learning upon such data involves stream mining and model adaptation over interdependent streams of different speeds and data of temporal nature associated to them. The interdependent streams require solutions for stream synchronization and window-based monitoring. The temporal data call for methods that monitor how objects evolve and adapt the learned model accordingly. In this talk, we discuss methods for data management and model adaptation over streams of temporal objects. We report on experiments for model learning on bank customer data.