DecisionFlow: Visual Analytics for High-Dimensional Temporal Event Sequence Data

Temporal event sequence data is increasingly commonplace, with applications ranging from electronic medical records to financial transactions to social media activity. Previously developed techniques have focused on low-dimensional datasets (e.g., with less than 20 distinct event types). Real-world datasets are often far more complex. This paper describes DecisionFlow, a visual analysis technique designed to support the analysis of high-dimensional temporal event sequence data (e.g., thousands of event types). DecisionFlow combines a scalable and dynamic temporal event data structure with interactive multi-view visualizations and ad hoc statistical analytics. We provide a detailed review of our methods, and present the results from a 12-person user study. The study results demonstrate that DecisionFlow enables the quick and accurate completion of a range of sequence analysis tasks for datasets containing thousands of event types and millions of individual events.

1 Introduction

Temporal event data is nearly ubiquitous in this era of mobile devices, electronic communication, and sensor networks. It can be found in everything from social network activity, to financial transactions, to electronic health records. More than ever before, large collections of this sort of data are being recorded that capture (a) what type of event is happening, (b) when it happens, and (c) the entities (e.g., customer, bank account, or patient) involved.