Visual Analysis of Complex Event Sequences

Event Sequences

Event Sequences

Event Sequences Data

We characterize an event sequence as a data type that is represented as a sequence of time stamps. Every time stamp indicates the occurrence of some measured phenomenon. In our case, event sequences don not have a type (such as therapy A, therapy B, etc.), it is just the signature of time stamps. Such event sequences are everywhere. Examples include:

  •     Rising stock prices
  •     Traffic accidents
  •     Earthquake aftershocks
  •     Tweets about some topic
  •     Happening of the Olympic Games (always every four years, isn’t it?!)
  •     Heartbeats
  •     Sleeping phases
  •     Etc. etc. etc.

 

Challenges

Event sequences may be long

  • Spanning across long time intervals, large number of events

Event sequence databases may be large

  • Thousands of sequences with a rich set of interesting patterns

Event sequences may be unknown

  • Making decisions without knowledge?!

 

Vision: Combining the strengths of humans and machines!

Algorithmic support is needed

  • Very good if data analysis problem can be solved automatically

 

Human supervision is needed

  • Very good if data analysis problem can NOT be solved automatically

 

Key Techniques

  • Metrics and Features: to characterize event sequences
  • Motif simplification: to substitute sequences by motifs
  • Grouping: to assign similar sequences to clusters
  • Ranking and Filtering: to enable changing data focus
  • Application: to apply techniques to real-world datasets