I590 Interactive Visual Analytics Week 11 | Nov 2, 2016 Task - PowerPoint PPT Presentation

I590 Interactive Visual Analytics Week 11 | Nov 2, 2016 Task abstraction Evaluation Khairi Reda | redak@iu.edu School of Informa5cs & Compu5ng, IUPUI

Administra/via… • Project 2 • 10 minutes presenta5on per team • 2 minutes for Q & A • Rehearse! • Please send me contribu5ons of individual team members • Project 3 • Proposals due next week in class (1 page) • Teams of 3 or less. Individual project is OK

Raw user data Visualization Visual Processed (multiple views structure data of visual things) pre- visual processing interaction encoding vis designer

Visualiza/on design process Study Design Evaluate Build

What is shown? Why is the user looking at it? How is it shown?

data abstrac/on

visual encoding

Task abstrac/on • Visualiza5on helps people carry out a task on a dataset • To design an effec5ve visualiza5on, we need to understand those tasks • Tasks come from the domain and the background of the users • Despite apparent differences, many tasks can be similar • “Contrast the prognosis of pa5ents who were admiXed to hospital to pa5ents receiving home care/rest” [epidemiologists studying flu] • “See if the results for 5ssue samples treated with LL-37 match up with the ones without pep5de” [biologists studying immune system response] • Both tasks are essen5ally about “comparing values between two groups” • Task abstrac5on allow us to iden5fy common visualiza5on designs despite apparent domain differences

Task abstrac/on {action, target} pairs discover distribu5on compare trends locate outliers browse topology Based on a slide by Miriah Meyer

Task abstrac/on { action , target} pairs Analyze Search Query

Discover vs Present Explore vs. Explain

Discover

Present

Enjoy

Task abstrac/on { action , target} pairs Analyze Search Query

Annotate

Mid level ac/ons { action , target} pairs Analyze Search Query

Mid level ac/ons { action , target} pairs Analyze Search Query

Example: Compare & Derive Via Alex Lex

Task abstrac/on {action, target } pairs

Target

Example Trends : How did the job market develop since the recession overall? Outliers : Looking at real estate related jobs Alex Lex

Real-world example of Domain task analysis and design

Inferring Grevy’s social interac5ons Mayank Lahiri

Domain Tasks • Find communi5es in zebra society, and influen5al individuals who play a role in shaping the social structure • Understand how the social structure of Grevy’s zebra evolve over-5me • Understand how Grevy’s zebra society responds to environmental variables

Domain Tasks • Find communi5es in zebra society, and influen5al individuals who play a role in shaping the social structure • Understand how the social structure of Ac/on: Explore (unknown target, unknown loca5on) Grevy’s zebra evolve over-5me Target: communi5es (groups of zebras that hang out together) • Understand how Grevy’s zebra society responds to environmental variables Rubenstein et al., 2015

Domain Tasks • Find communi5es in zebra society, and influen5al individuals who play a role in shaping the social structure • Understand how the social structure of Grevy’s zebra evolve over-5me • Understand how Grevy’s zebra society Ac/on: Explore & Compare responds to environmental variables Target: All communi5es over 5me

Time-changing groups A A A A C C Q Q X X C C Q Q X X A A C C Q Q X X B B B B R R Y Y R R Y Y B B R R Y Y T1 T1 T2 T2 T3 T3

Time-changing groups A A C Q X C Q X A C Q X B B R Y R Y B R Y T1 T2 T3 time Q R X Y group switch A B C

Time-changing groups individuals community

Domain Tasks • Find communi5es in zebra society, and influen5al individuals who play a role in shaping the social structure • Understand how the social structure of Grevy’s zebra evolve over-5me • Understand how Grevy’s zebra society responds to environmental variables Ac/on: Relate Target: Communi5es and environment

Social structure + geography Group movement Social + over space and 5me structure

Social structure + geography

Visualiza/on design process Study Design Evaluate Build

Why evaluate? • Evalua/on / valida/on is “about whether you have built the right product” • Does the visualiza5on serve its intended purpose? Does it enable users to perform their intended analysis tasks? • Is it “easy” to use? Are there any usability issues in the interface? • Does the visualiza5on enable accurate percep5on of values, distribu5ons, and/or trends in the data? • Does it provide new insights about the data? • Is the visualiza5on memorable and/or engaging? • How can we improve the visualiza5on?

What to evaluate? • Think about your contribu5on. That is, the new idea in your visualiza5on • The user interface? • The visual encoding? • The interac5on technique? • The abstract tasks you iden5fied from interviewing domain experts? • Or bits of the above?

Four nested levels of vis design Munzner, 2014

Four nested levels of vis design Study domain, interview users, iden/fy needs Munzner, 2014

Four nested levels of vis design Iden/fy tasks and data. Translate from domain-dependent to abstract tasks and data types Munzner, 2014

Four nested levels of vis design Sketch/design visual encoding and interac/on techniques Munzner, 2014

Four nested levels of vis design Implement visualiza/on using code Munzner, 2014

Threats to validity Munzner, 2014

Threats to validity Munzner, 2014

Evalua/on methods • Evaluate algorithm speed / memory usage • Controlled [lab] studies with any user • Qualita/ve studies • Insight-based evalua/on • Evalua/ng the data analysis process • Field deployment

Controlled [lab] studies - Usually done with any poten5al user - Goal is to control usage condi5on is much as possible - Allows for comparison between different techniques or representa5ons - Generally provides accurate conclusion, but results may not generalize beyond lab condi5ons or tested tasks - Typically quan/ta/ve in nature • Finely-scoped tasks • Measure accuracy and performance 5me

Qualita/ve studies - Usually open-ended usage scenarios - Smaller number of par5cipants compared to quan5ta5ve lab studies - More in-depth analysis - Analyze videos, audio, or comments from users - Can ask par5cipants to fill surveys, or provide subjec5ve feedback on the visualiza5on - Usually involves domain experts and target audience of the visualiza5on

Insight-based evalua/on The goal of visualiza5on is to generally generate new - insight Evalua5on should therefore include insight-genera/on - Think-aloud protocols: have the users say what they are - thinking Transcribe and code: - Observa5on - Hypothesis - Ques5on - Exploratory goal -

Evalua/ng the data analysis / explora/on process - The focus is on the visual analysis process of users, as opposed to the outcome of the analysis - Want to understand the analy5c dialogue between the user and the data - Want to capture and analyze mul5ple aspects: - Interac/ons with the visualiza/on - Interac5on logs, videos - Reasoning process - Think aloud protocol: have the par5cipant say what they are thinking - Eye gaze behavior

Filter Change view Hypothesis Change view Observe outliers Hypothesis Query Filter Decision making … … …

Evalua/ng visualiza/ons is s/ll tricky!