P AIR F INDER : I DENTIFYING AND M EASURING T EMPORAL A SSOCIATIONS FROM T EMPORAL E VENT S EQUENCES Hsueh-Chien Cheng, Catherine Plaisant, Ben Shneiderman Department of Computer Science Human-Computer Interaction Lab University of Maryland 5/22/2012
T EMPORAL E VENTS ICU 19:35 Oct. 1 Admission 19:28 Oct. 1 Floor 03:19 Oct. 4 23:06 Oct. 7 Discharge
T EMPORAL A SSOCIATIONS Consider both order and relative time difference “Discharge” occurred 92 hours after “Floor” “ICU” occurred 52 hours before “Floor” AdmissionICUFloorDischarge What if we have a large number of records?
A LIGNMENT F RAMEWORK Align the records by one event and move the others accordingly in the relative time frame
A LIGN Records Aligned records 1 Align 2 Aggregate 3 Summarize Focal event, FloorRelated event, ICU
A GGREGATE Aligned records Aggregation 1 Align 2 Aggregate 3 Summarize Focal event, FloorRelated event, ICU
S UMMARIZE Aligned records Aggregation 1 Align 2 Aggregate Histogram 1 day 3 Summarize Focal event, FloorRelated event, ICU
O RGANIZING H ISTOGRAMS Even with a small number of event types, there are many event pairs. We need a better way to organize the histograms 10 event types 10 * 9 = 90 event pairs
P AIR F INDER 1. Shows the histograms summarizing the associations between all pairs of events 2. Applies interestingness measures to locate interesting histograms easily
G RADUATE S TUDENT D ATASET A synthetic dataset with 1000 records and 7 event types Event Type Class Signup Paper Submission Software Release Masters Degree Proposal Defense Job Interview
H ISTOGRAMS Histogram helps understand the associations between a pair of events o Students had defenses 13 months after their proposals o No defense occurred before proposal
I NTERESTINGNESS M EASURES Interestingness measure helps arrange a large number of histograms in a meaningful order Less interesting More interesting
I NTERESTINGNESS M EASURES Not Interesting o Which related event occurred mostly after the focal event? o Which related event occurred periodically after alignment? Not Interesting
C ASE S TUDY 5 case studies were done to demonstrate the potential of PairFinder.
C ONCLUSION PairFinder uses histograms to summarize the association between all pairs of events. applies Interestingness measures to order the histograms by their interestingness. Work with us --
T AKE A WAY M ESSAGES Visualizations help summarize complicated relations and enable further interpretation of the data. Organizing a large number of visualizations facilitates knowledge discovery. We thank the National Institutes of Health (Grant RC1CA ) for partial support of this research. Project page at