1. Visualizing User Activity History

2. Lots More Collaboration, Lots More History
Web services enabling socially constructed artifacts pose collaboration difficulties
People work on the same project/activity without ever knowing one another or communicating directly
“Long-term indirect collaboration” (Fischer 1992)
User activity can be recorded to use later
For backtracking/versioning and presentation
For improving indirect communication among designers
E.g., Reeve’s INDY (CSCW 1992)

3. Brent Reeves, Supporting collaborative design by embedding communication and history in design artifacts, Ph.D. Dissertation, 1993.

4. Visual Knowledge Builder (VKB)
Navigable History and History Viewer
2001

5. Visualizing History to Improve Users' Location and Comprehension of Collaborative Work
DoHyoung Kim and Frank M. Shipman III
Department of Computer Science and Engineering Texas A&M University

6. Issues with User History
User History is not User-Level History
Most systems record activity much like database transactions (for undo/redo)
This system-level record is at a different granularity than what users discuss
User history gets large and unwieldy for longer and/or larger collaborations
Difficult to find when a particular activity occurred
Takes time to identify what happened first

7. Approach: Infer and Present User-Level History
Two main goals
Identify meaningful aggregations of system-level history events
Use a combination of features of the events being grouped
Summarize and visualize aggregated history to support navigation/comprehension
Present information about the who, what, and when of the activity

8. Preliminary Designs from DoHyoung Kim’s Dissertation Work:
(Winter 2009)

10. Related Work History visualization History clustering
Time line visualizations (e.g. Plaisant’s LifeLines)
Edit or state visualizations (e.g. Viégas’ Wikipedia work)
Comic strip metaphor (e.g. Nakamura’s snapshots with callouts)
History clustering
SmartBack grouped Web navigation events
Time slicing for grouping edit events (Shurai)

11. CoActIVE
Collaborative Activity Interpretation and Visualization Engine (CoActIVE)
Developed as an application-independent component
Easy to integrate in systems with existing undo/redo functionality
Developers can customize or use as is
Novel aspects are in history interpretation and visualization

12. Visual Knowledge Builder (VKB)
Hierarchic workspace
Collect/author content
Expression via visual cues and spatial arrangement
Existing history mechanism

13. History Interpretation
Aggregation through hierarchic agglomerative clustering (HAC)
Allows for many levels of potential aggregation
Compares consecutive events for potential aggregation
Application-independent (e.g. time) or application-dependent distance measure
Dynamically determined thresholds determines depth of resulting tree
For integration in VKB a spatio-temporal distance measure is used
Inclusion of spatial distance matches common pattern of VKB use
Other applications could include content-based distance metric

14. Summary Generation & Keyword Extraction
Textual content
CoActIVE event object includes optional textual content field to indicate content associated with event
Potential keywords extracted using Java WordNet Library for each cluster
Nouns that are not stop words considered
TF/IDF used to identify significant keywords for clusters at each level of the tree

15. Visualizations Four visualizations of the interpreted history
History Session Viewer
Basic tree view – traditional view in VKB and many other applications
History Interpretation Viewer
Augmented tree view – includes time, user, and keyword information in overview
Filmstrip visualization with textual summaries
Augmented filmstrip that provides access to textual details
Filmstrip visualization with visual summaries
Augmented multilevel filmstrip

16. History Session Viewer

17. History Interpretation Viewer

18. Filmstrip Visualization (Top Level)

19. Textual Summary for Filmstrip Viewer

20. Visual Summary for Filmstrip Viewer

21. Evaluation Method Many features that could be evaluated
Quality of clustering
Interface to resulting clusters
Focused on alternative visualizations
24 participants used all four visualizations
Four similar collaboratively authored VKB spaces were chosen for tasks
Different tasks and visualizations were counterbalanced for ordering effects

22. Evaluation Activity
Participants answered 5 questions for each document
Locating specific activity by a specific user
“Find the event ID where Jill created the information object contained below.”
Recovering prior state of specific element
“What is the previous title of the “How to Start a Blog” collection?”
Comprehending the order of activities
“Which one of X and Y was created earlier in the document?”
Locating a particular state of the document
“Find the event ID where the collection, “CSCW without C” is displayed on a screen as shown below.”
Identifying all instances of a particular user working on a part of a document
“Find as many places as Jack worked on the “Current research” collection as possible in 3 minutes.”

23. Data Collected and Evaluation Metrics
First four questions
Time taken to answer question
Correctness of answer
Fifth question
Number of correct and incorrect answers in three minutes
Satisfaction with each interface probed just after use (7 point Likert scale)

24. Filmstrip + Text Summary Filmstrip + Visual Summary
Results – Overall Time
Time spent on first four questions by visualization
Difference between filmstrip visualizations and tree visualizations significant (p<.001)
Basic Tree
Augmented Tree
Filmstrip + Text Summary
Filmstrip + Visual Summary
Mean (minutes)
17.04
13.43
8.62
7.44
Standard deviation
5.99
4.01
2.71
2.81

25. Results – Timed questions
Time spent (and errors) by visualization
Difference for time between filmstrip visualizations and basic tree visualization is significant (p<.003)
Fewer errors (3 of 20 overall) in filmstrip views
Basic Tree
Augmented Tree
Filmstrip + Text Summary
Filmstrip + Visual Summary
Locating specific activity
4.16 (1)
3.70 (0)
2.55 (2)
2.02 (0)
Recovering prior state
3.56 (0)
1.62 (0)
1.49 (0)
1.25 (0)
Comprehending order
5.23 (4)
4.21 (1)
2.59 (0)
2.46 (0)
Locating particular state
4.09 (6)
3.90 (5)
1.98 (0)
1.70 (1)

26. Results – Identifying matching activities in three minutes
Locating periods where specific user worked on particular part of document
Difference between filmstrip + visual summary and other three visualizations significant
Basic Tree
Augmented Tree
Filmstrip + Text Summary
Filmstrip + Visual Summary
Mean (correct answers)
1.83
1.58
2.38
4.42
Standard deviation
1.37
1.47
0.92
1.25

27. Results – Summary Satisfaction data mirrored performance data Issues
Filmstrip with visual summaries was highest rated
Basic tree view was lowest rated
Issues
Navigation of filmstrip visualization was somewhat confusing
Current implementation shows final state in a history cluster meaning any movement beyond that state is in the next cluster

28. Future Work Lots of low-level improvements
Selection of state shown in filmstrip to represent segment of history
Keyword selection
Graphic design of visualizations
Evaluate alternative event clustering approaches
Does adding spatial distance or content distance improve over pure time-based distance metric?
How to make this work with synchronous collaborations (when subsequent events are by different users and not marking hand offs)?

29. Conclusions More long-term indirect collaborations than ever before
Records of user history can improve such collaborations
Size and nature of history records limit their usefulness
Event aggregation and segment visualization can help
Compared four visualizations based on a hierarchical aggregation of history
Filmstrip views outperformed tree views in all metrics
Filmstrip view with visual summary was most preferred
Future work can explore alternative segmenting techniques