Increasingly rich temporal databases enable researchers to conduct detailed studies of temporal event sequences. When researchers have hypotheses of which events lead to other events, current tools assist them to find those events, even in complex data sets. Generating hypotheses of temporal associations among temporal events are challenging because of the large number of possible complex patterns and the knowledge discovery task in an exploratory context.
PairFinder creates visualizations of pairs of events from the temporal event sequences. As a first step toward generating hypotheses of causal relations, the goal of the project is efficient discovery of hypotheses of the associations between pairs of events.
An alignment framwork is used to align the records by one focal event. The distribution of the related events in the relative time frame is summarized as a histogram. See Figure 1. for an example.
With interestingness measures, users can sort the event pairs by interestingness values and start examining from the top-ranked pairs.
The records may have additional attributes, with which PairFinder can filter the records or compare histograms created from different group of records.
We thank the National Institutes of Health (Grant RC1CA147489-02) for partial support of this research.
Hsueh-Chien Cheng, Catherine Plaisant, and Ben Shneiderman.
Identifying and Measuring Associations of Temporal Events. HCIL Tech Report HCIL-2012-05.