vast 2008 challenge - university of maryland€¦ · 26/10/2008 · on day 7, this group dropped...
TRANSCRIPT
VAST 2008 ChallengeThe VAST 2008 Challenge Co-chairs
Georges Grinstein, University of Massachusetts LowellCatherine Plaisant, University of MarylandJean Scholtz, Pacific Northwest National Laboratory
The VAST 2008 Challenge CommitteeSharon Laskowski, National Institute of Standards and TechnologyTheresa O’Connell, National Institute of Standards and TechnologyMark Whiting, Pacific Northwest National Laboratory
The Challenge Committee wishes to thank
The VAST and VisWeek organizers All participantsAll students who supported the project
Loura Costello, Heather Byrne, Adem Albayrak
The analysts and judgesOur sponsors
Thanks to our Sponsors
NVAC
NSF
NIST
Battelle Memorial Institute
IARPA A-SpaceX
VAST 2008 ChallengeNew format
4 Mini-ChallengesEach has own dataset and tasksTeams may enter one or more
Grand ChallengeAnalyze all 4 datasets – Integrate to answer
No “1st place” – but instead many awards for outstanding aspects of the entriesBoth visualization and analysis awards
The VAST ChallengesDemonstrate the visual analytics capabilities
of your tools against an invented scenario and synthetic datasets.
Challenge themes included geospatial, activity and behavior, text processing, and social network analyses.
Strong InterestOver 400 downloads of 2008 data set
Compared to 150 to date for 2007 data set73 submissions
Cell phone (social network) 22Boats landings (geo spatial) 13Evacuation Traces (geo spatial) 20Wiki pages (finding factions) 12Grand Challenge 6
28 organizations12 student teams13 countries
This morningNow
Overview of datasets, questions, judgingAwards and summary of submissions to each of the 4 mini challenges
After the breakChallenge Panel
Data and interactive sessionPresentation by 3 Grand Challenge teamsQuestions to teamsOpen Discussion
Challenge Scenario: the dataInformation
A web page about the Paraiso movement covering the history and beliefs of the group
Data1. wiki edits2. cell phone call logs3. Coast Guard interdiction records4. records of an evacuation of a Department of
Health Building
Challenge Scenario: the tasksParticipants in each Mini Challenge were required to analyze a single data setsParticipants could enter more than oneParticipants in the Grand Challenge were required to analyze all four data sets
Grand Challenge tasksAssess the beliefs of the movement and the activities they’re involved inDetermine if the movement advocates violence
JudgingJudging was done by a committee of 15
7 members were professional analysts8 members have usability and visualization expertise
Criteria usedUtility in arriving at the answerAccuracy of the answerQuality of the visualizationsFunctionality of the analytic environment
AwardsVisualization awards
System awards
Analysis awards
Award policy – only one award per teamAwards were specific to the various mini challenges and to the quality of the submissions this year. We will not necessarily be giving the same awards next year.
Intuitive visualizationsInnovative visualizationsOutstanding functionalityLevel of integrationAccuracyHigh quality report
AwardsVisualization awards
System awards
Analysis awards
Award policy – only one award per teamAwards were specific to the various mini challenges and to the quality of the submissions this year. We will not necessarily be giving the same awards next year.
Intuitive visualizationsInnovative visualizationsOutstanding functionalityLevel of integrationAccuracyHigh quality report
AwardsVisualization awards
System awards
Analysis awards
Award policy – only one award per teamAwards were specific to the various mini challenges and to the quality of the submissions this year. We will not necessarily be giving the same awards next year.
Intuitive visualizationsInnovative visualizationsOutstanding functionalityLevel of integrationAccuracyHigh quality report
Cell Phone Challenge
Cell Phone LogsDataset
10 days of cell phone call logsCaller id / callee id / cell tower(=location) / date / length of call
Analytic SituationA social faction (supporters of the “Paraiso”movement) was a close knit group of individuals who maintained cell call contact. On day 7, this group dropped their phones and picked up another set.
Cell Phone QuestionsWhat is the Catalano/Vidro social network, as reflected in the cell phone call data, at the end of the time period? Characterize the changes in the Catalano/Vidro social structure over the ten day period?
GeoTemporalNet
MobiVis
Oculus Info Inc.
Almost all teams visualized the social network with a node/link diagram
INRIA Bordeaux - VizWhiz
Other Representations
Calls from 400 number to 398 number (123), R2=10%, XY projection
Nyenrode Business Univesiteit University College Dublin -SocialDynamicsVis
Other Representations
CORE
Palantir
Prajna Project
Visualization/Analysis of the geospatial components of the data
ICAVEAttributeLayoutViz
Visualization/Analysis of the geospatial components of the data
Visualization/Analysis of the geospatial components of the data
SocialDynamicsVis
PALANTIR
SPADAC
Handling TimeUsing Histograms
University of Maryland
Handling TimeUsing Stacked Filled Curves
VizWhiz
NEVAC
ComVis
SocialDynamicsVis
Coordinated Displays
Oculus Info Inc.
Coordinated Displays
University of IllinoisNational Center for Supercomputing
Coordinated Displays
Cell Phone Awards
Beijing University of Posts and TelecommunicationsSocial Network Accuracy
The submitted social network was analyzed using a computed average of node/link precision and recall. Highest accuracy was 55.9%
University of California, DavisIntuitive Social Network Graphs
University College DublinNode-Link Animation
University of Maryland – SocialActionTime Visualizations of Cell Phone Activity
Vision Systems & Technology, Inc.Effective Toolkit Integration
Boats Challenge
Boat LogsDataset
Logs of landings or interception by Coast GuardLaunch point / landing point / type of vessel / date / passengers / whether or not intercepted
Analytic SituationBoat migrants (Paraiso supporters) were leaving the Isla del Sueno for Florida in massive numbers. The US Coast Guard were picking them up with increasing success. Their landing strategy moved westward until successful landing began on the Yucatan Peninsula, Mexico.
Boat QuestionsCharacterize the choice of landing sites and their evolution over the three year
Characterize the geographical patterns of interdiction over the three years
What is the successful landing rate over the time period
Map or no map?
Grid with lat/long coordinates Beijing
Many used Google EarthSome: own GISSome: no map, or optional.
CMU team
U of Singapore
U of Illinois
Grouping and clustering
SPADAC
ParvacU of Washington
Aggregation (and dealing with overlap)
Parvac – U of WashingtonAggregation by regionX axis is region, e.g. L5 = Mexico
OculusIrregular areaAggregation on the map
Representing TimeAnimation
Those who relied solely on animation missed the large number of Mexico landings (they overlapped in a very small area)
Timelines
VSTI Prajna
3D
Hybrid Projection
Representing Time
Oculus
Tacc
Symbol used on the map
Dots - CORE
Lines - Tacc
Centroid - Tacc
SPADAC - Arrows
Coordinated displays
GET COMVIS EXAMPLE
Zagreb/VrVis/VTech
Palantir
Analysis SupportMany merely reported numbersSome made hypothesesA few provided some analysis or reporting support
Oculus
Analysis thoroughnessSome teams provided the minimum answersOther looked at the data very thoroughlye.g. looking at every angles of the data, exploring seasonal patterns, finding suspects’ names amongst passengers
Recommendation
TaccX is time, Y latitudeBlack lines are coast guard patrol traces
A few gave suggestionse.g. new areas for the Coast Guard to patrol
Boat Awards
SPADAC Inc.Analysis Summary
Summary pointed out that Mexico landings could be seen as a land bridge to Florida and noted that these were successful because US Coast Guard would have no jurisdiction in Mexico.
VRVis – U. of Zagreb – Virginia TechSimple and Effective Integrated Display
Traces Challenge
Traces LogsDataset
RFID logs during a building evacuationX, Y coordinates over time period per ID
Analytic SituationA small improvised explosive device (IED) was set off in a at a (fictitious) Miami, Florida Department of Health (DOH) Building, resulting in casualties and moderate damage. RFID was used to track all employees & visitors. A Paraiso support set off the device, but there were witnesses.
Trace Challenge QuestionsWhere was the device set off? Identify potential suspects or witnessesIdentify any suspects and/or witnesses who managed to escape the building. Identify any casualtiesDescribe the evacuation
Sorting it all out Staining by Southern Illinois
Who did it? Where did they do it? Wisdom of the crowds (users) finger Ramon Katalanow
UBari – IVU Lab
Who saw it?
Anton Knapp (trace visible) turns towards Ramon Katalanow (circled in red) during the evacuation.
Katalanow then moves away from Knapp to other exit.
VSTI-Prajna
Looking at the Exits
UCD Protovis Timeline
Who Escaped?
Fraunhöfer Institute
Results Overview
VRVis-ComVis
Unique Submission Delivery
Victoria University of Wellington
Evacuation Traces Awards
Fraunhöfer InstituteTool Integration
Southern Illinois University EdwardsvilleInnovative Trace Visualization
University of Bari User Testing to Obtain Consensus
Users were asked to view a visualization and “place” a bomb in the right place
The majority of the users (53.84%) marked a point inside the room where suspects 18 & 50 were.
Users speculated that the main suspect either “threw”the device inside that room or activated it upon entering that room.
Wiki Challenge
Wiki logsDataset
Wiki edit comment data Who / what type of edit performed
Analytic SituationThe Paraiso Wikipedia entry became a focal point for conflict between the followers and the general population. The wiki edits pages showed some of the social relationships and dynamics involved.
Wiki QuestionsWhat are the factions represented in the edit pages and who are its members?Is the Paraiso movement involved in violent activities?
Visualizing Edit ActivityPatterns of Edits Over Time
Used a database table and applied various filters to find frequent editors and also color coded entries based on identification of who edits who.
Pacific Northwest National Laboratory
Visualizing Edit ActivityPatterns of Edits Over Time
National Center for Supercomputing Applications at Urbana-Champaign
Palantir
Visualizing Edit ActivityPatterns of Vandalism
U. Washington - PARVAC
Visualizing Edit ActivityPatterns Based on File Changes
File changes over 20,000 bytes stand out showing a deletion and restoration of the entire record. This pattern suggested instances of vandalism where corrections are made by bots
A close view of File Changes under 20,000 bytes reveals patterns indicating editing wars; subsequent offsetting changes in file size by alternating editors.
Oculus Info, Inc.
Visualizing Edit ActivityPatterns of Usage
Carnegie Mellon UniversityPalantir Technologies
Visualizing Individual Supporters/Opponents Revert and Undo Patterns
Vision Systems & Technology, Inc
NEVAC – Penn State University and Drexel University
Visualizing Individual Supporters /Opponents Similarity of Contributors
Oculus Info, Inc.
Palantir Technologies
U. Washington PARVAC
Visualizing FactionsSocial Networks
Inria
Beijing University of Posts and Telecommunications
Visualizing Factions
Vision Systems & Technology, Inc.Oculus Info, Inc.
Wiki Edits Awards
There were none made as outstanding performance in the wiki edits were subsumed in the Grand Challenge awards
Grand ChallengeBased on ALL the data available (i.e. using the data from all 4 Mini Challenges) what is the social network of the Paraiso movement at the end of the time period?What name or names can be associated with individual activities?What is the geographical range of the Paraiso Movement and how does it change over time?How do the major beliefs of the Paraiso movement affect their activities?
Data IntegrationDynamic ontology allows disparate data to be combined and searched
Can build structured data from unstructured text
Metadata is stored with entries (can be edited and shared)
Palantir Technologies
History mechanism available. Using the metadata and the history mechanism, analyst can generate a Powerpoint presentation
Data Integration
Association Graph
NEVAC
Integration of Analysis
Analysts do separate analysis of minis
Share folders which contain sandbox analysis
Another analyst can pick this up and combine in another sandbox
Oculus Info, Inc
Analysis Support
Hypotheses can be added as nodes and links into graphs
NEVAC
Grand Challenge Awards
Oculus Info Inc.Support for Diverse Analytic Techniques
Palantir TechnologiesInteractive Visual Analytic Environment
Pennsylvania State University-NEVACData Integration
AFTER THE BREAK
PanelInteractive session overview3 Grand Challenge teams presentQuestions to teamsOpen discussionNext year plans
BREAK
VAST 2008 ChallengeThe VAST 2008 Challenge Co-chairs
Georges Grinstein, University of Massachusetts LowellCatherine Plaisant, University of MarylandJean Scholtz, Pacific Northwest National Laboratory
The VAST 2008 Challenge CommitteeSharon Laskowski, National Institute of Standards and TechnologyTheresa O’Connell, National Institute of Standards and TechnologyMark Whiting, Pacific Northwest National Laboratory
Summary of overall event
See slides 2 to 13
Challenge EmphasisThe process and visualizations used to locate the “answers” are AS IMPORTANT as the answers
This is the concept that distinguishes the VAST Challenge from other contestsTherefore we are judging not just the answer but the utility of the visual environment in arriving at that answer.
Judging the EntriesJudges are professional analysts and the VAST Challenge committee Entries are reviewed and both qualitative and quantitative measures are collected
Accuracy scores are also given As the problems and data become more realistic the “answers” are not black and white
Answers need to be supported by evidence
SUNDAY eveningParticipant Session
Opened to Challenge participants only
For each mini-challengeSummarized entriesDiscuss solutions and problemsSuggestions
Monday EveningInteractive Session
3 teams participated 30 minutes to load and preprocess datasets similar to the ones used in the VAST Challenge30 minutes to train analyst, explain system concept and functionalities2 hours for team + analyst to analyze dataObservers
Took notes to provide each team with feedbackGeneral discussion of analysis session
PanelistsDon Pellegrino, Drexel University (NEVAC)Lynn Chien, Oculus Info, Inc.Brian Lee, Palantir
10 minutes each to explain their systemSeveral minutes for questions to each panelistPanelists respond to committee questionsOpen question time
NEVAC– Penn State
Oculus Info, Inc.
Palantir Technologies
Questions- Grand ChallengeWhat was your motivation for entering?What was the most difficult aspect of the Grand Challenge?What was the most rewarding aspect?Can you share with us lessons learned during your work on the Grand Challenge?Would you say that entering the Grand Challenge was worth the effort?
Questions – Interactive SessionAbove and beyond the lessons learned in the Grand Challenge were there additional lessons learned in the interactive session?What was the most difficult aspect of the interactive session?What was the most enjoyable aspect?If a team contacts you for advice on entering the VAST 2009 Challenge what would you tell them?
Audience Questions
Data Set Availability?2008 VAST data sets/questions
Available to all (still on website)If you submit your answers, we will give you the solution
2008 Interactive session data sets/questionsA smaller problemAvailable only to instructors (for training, class projects)2006 and 2007 datasets also availableSolutions and videos given to instructors
VAST 2009 ChallengeVAST 2009 Challenge - Similar formatwww.cs.umd.edu/hcil/VASTchallenge09Practice using previous datasets
Deadlines for submission (June/July)Target dataset delivery in Spring (March?)
Contact any of us if you have suggestions or [email protected]