If someone learns Tableau and is interested, Prof. Schneiderman would appreciate a 10-15 min presentation about it.
Goals for the application project:
- Deep experience with a particular tool
- Learning thought patterns of an analyst
- 3 "juicy" headlines and 500-1000 words
It's ok to use more than one tool (compare and contrast them).
Some groups already on team website.
Help was solicited for various projects:
- Undergraduate class scheduling
- Versioning machine
- Geospatial accident visualization w/ filtering
- Temporal visualization of medical or weblog data
- Initial project proposal is due 26 Feb.
- Next deliverable will be an annotated bibliography.
There was also a brief discussion of the midterm exam. There will be at least one question of the form "design a visualization for this data set."
Observation: it's difficult to detect patterns when all the data are categorical.
Question: what is the definition of "interesting" in a temporal data set?
Lin, J., Keogh, E., Lonardi, S., Visualizing and discovering non-trivial patterns in large time series databases, Information Visualization 2005, Vol 4, No 2 http://www.palgrave-journals.com/ivs/journal/v4/n2/pdf/9500089a.pdf
Two papers using other methods for time series visualization are:
- JJ van Wijk, ER van Selow, Cluster and Calendar based Visualization of Time Series Data, Proceedings of IEEE Symposium on Information Visualization, 1999.
- Marc Weber, Marc Alexa, Wolfgang Müller, Visualizing Time Series on Spirals, Proceedings of the IEEE Symposium on Information Visualization 2001 (INFOVIS'01)
What are we looking for?
- Motif - frequently occurring pattern
- Anomaly - infrequently occurring pattern
- Semantic representation - SAX (Semantic Aggregate approXimation):
- Turns a time series into a series of letters
- Number of letters is manually chosen
- All letters are equally probable
- Each period categorized as a letter
- Sequence trees
- Each item in a sequence is encoded as a path in the tree
- Easy to see motifs and anomalies
- Numerosity reduction
- Non-overlapping windows
- Recording only different patterns
- Series comparison metrics
Demo of Viztree.
- Tree structure visualization
- Mapping using color and thickness
- Automatic pattern identification (speed and generality)
- Simultaneous view of subsequence matches
- Level/section zooming
- Dynamic parameter response
- Automatic suggestion for parameter
- Individual subsequence selection in time-zoom panel
- HCIL-style selectors for tree focus
- Calendar clusters
- Automatic periodicity detection
- Better motif recognition with smaller parameters (ie. more data compression)
- Conversion to character strings results in great increases in processing speed
State of the Union
Looking for term usage spikes with Spotfire.
Changes in control panel.
- drops: "budget" and "weapons"
- highly variable: "world" and "social"
- spike: "security" and "sadaam"
- low variance: "cheney" and "bless"
Future work on this topic is possible.