Sequential Data Analysis with Temporal Graph and Skeletonization

Abstract: Sequential data analysis is a fundamental problem in data mining with applications in many science and business fields. Given the overwhelming scale and the dynamic nature of sequential data, new visions and strategies for sequential data analysis are needed to derive competitive advantages in real-world applications. In this talk, we introduce the “temporal skeletonization” framework, our approach to identifying the meaningful granularity for sequential pattern mining. We first show that a large number of symbolic features in sequential data can “dilute” useful patterns which may exist at a different level of granularity. This is so-called “curse of cardinality”, imposing significant challenges to the design of sequential analysis methods. To address this challenge, our key idea is to construct an undirected graph, and use the “skeleton” of the graph as a higher granularity on which hidden temporal patterns are more likely to be identified. Moreover, the embedding topology of the temporal graph allows us to translate the temporal content into a metric space which opens up new possibilities to explore, quantify, and visualize sequential data. In the talk, we will also review a series of extensions of temporal graph and skeletonization for emerging problems in marketing, finance, and location-based services.

Chuanren Liu is currently an Assistant Professor in the Department of Business Analytics and Statistics at the University of Tennessee, Knoxville, USA. His research interests include data mining and knowledge discovery, and their applications in business analytics. He has published papers in refereed journals and conference proceedings, such as IEEE Transactions on Data and Knowledge Engineering, INFORMS Journal on Computing, European Journal of Operational Research, IEEE Transactions on Cybernetics, Knowledge and Information Systems, Annals of Operations Research, and KDD, ICDM, SDM, AAAI, IJCAI, DSAA, etc.