Activity-Centered Domain Characterization Liz Marai Electronic Visualization Laboratory University of Illinois at Chicago
Domain characterization is difficult [Statistical and Applied Math Sciences Institute] [Munzner 2009: A Nested Model] [van Wijk 2006]
Vis models use Human-Centered Design Observation Ideation Testing Prototyping D. Norman, “The Design of Everyday Things” 2002 “Know your user”. You can’t design something for people without a deep, detailed knowledge of those people.
A few Vis HCD questions
Q1: Human-Centered Design and SwE Observation Ideation Testing Prototyping If writing functional specs is “like flossing” (everyone knows they should do it [Spolsky 2000] ) , where are the functional specs in Human- Centered Design vis models?
Q2: Value of Vis What is the relative value of a Cure Alzheimer’s visual computing project with only two domain expert users? “Cure Alzheimer’s” project “Explore pet-names” project # users
Q3: Workflows If visualization design models build on the Humans-Data-Tasks triad, where do user workflows fit in? Humans Data Tasks [shorpy.com: Efficiency Kitchen cca 1917]
“HCD Considered Harmful” “ HCD has become such a dominant theme in design that it is now accepted by interface and application designers automatically, without thought, let alone criticism. That’s a dangerous state --- when things are treated as accepted wisdom. ” (D. Norman 2005, “Human-Centered Design Considered Harmful”)
Activity-Centered Design (ACD) paradigm • ACD focuses on activities, not on the individual person (Activity Theory) • “ ...because people are quite willing to learn things that appear to be essential to the activity, activity should be allowed to define the product and its structure ” (D. Norman 2013, The Design of Everyday Things Revised and Expanded) • ACD is an enhancement of Human-Centered Design
An ACD model for domain characterization [Marai VIS’17 / TVCG Vol 24, Jan 2018 DOI: 10.1109/TVCG.2017.2744459 ] In a departure from existing vis models, this model: • Assigns value to vis based on activities • Ranks tasks > data • Incorporates workflows • Establishes the need for vis early • Extends the nested model with functional specs • Leads to a higher rate of project success
Functional specs as workflows • Allow the users to validate the designer’s understanding of the domain problem • A functional spec 1. Describes activity-features, not implementation • given as scenarios 2. Describes also what the system will NOT do 3. Lists nonfunctional requirements • Mandatorily reviewed and approved by users
Spolsky’s advice on functional specs • Templates considered harmful (Spolsky, “Joel on Software”, 2000) NOT: The user selects a biochemical model from the literature, and adds it as a new model entry with a single field “Model Name” into the visual analysis system. The system is web-based. BUT: Kermit the Frog, bored out of his mind, opens the latest issue of Nature Methods and spots a mouth-watering model of the fruit-fly response to allergens. Sticking his tongue out, Kermit runs to the browser, opens the visualization system, and types a new model entry with a single field called “Fruit-fly model”.
Does this ACD model matter?
Evaluation • Karl Popper (1902-1994), philosopher of science [http://wikipedia.org/] [https://i.redd.it/tlyoidfqe3gz.png] • “ A theory in the empirical sciences can never be proven, although it can be falsified ” (B. Gower, Scientific Method: An Historical and Philosophical Introduction)
Evaluation setup • Set A: 40 concrete short-term projects using prior models • generic agile model + nested model + pitfalls model [Sedlmair et al 2012] • Set B: 35 similar projects using the ACD model • same agile model + nested model + ACD model • All completed by young researchers training in interdisciplinary visualization • Required collaboration with experts in orthopaedics, biology, turbulent combustion, astronomy, machine translation,… • Project success defined as novelty and user adoption [F. Brooks 1996]
Evaluation: Supporting evidence • Set A: 25% success rate (10/40) • success: 2 by dual expertise researcher, 8 by weekly meetings with committed expert, 0 by non-weekly expert • failed: several data issues and communication issues, some despite weekly meetings with committed experts • Set B: 63% success rate (22/35) • success: 2 by dual expertise researcher, 10 by weekly meetings with committed expert, 10 by non-weekly expert • failed: 9 complete fails (improper Data Access, Func specs), 4 partial fails (Probes); several had weekly meetings with committed experts
Evaluation: Fit with existing reports and models • Agreement: • All vis models that include a Task axis [Springmeyer et al 1992] [Tory & Moller 2004] [Praetorius & Van Wijk 2009] … [Lloyd and Dykes 2011] !! • [Sedlmair et al 2012], [McKenna et al 2014] • • Agreement and partial disagreement • Nested model [Munzner 2009] : HCD validation X • Pure HCD models
ACD model Activity-centered frame and model for domain characterization that: • Assigns value to vis based on activities • Ranks tasks > data • Incorporates explicitly workflows • Establishes the need for vis early • Extends the nested model with functional specs • Leads to a higher rate of project success Designing for activity may improve the openness of users to novel, powerful visual encodings and interaction paradigms. Limitations and assumptions: see paper
Food for thought • There’s little in this model that makes it SciVis-restricted • There’s more than one model/theory for almost everything • does VIS pick one model and follow it for many years? (“human-centered- design”, “data-tasks-user”, ”overview first”, “five sheet design”…) • What counts as validation/evaluation of a model/theory? • if it provides a new interpretation of existing data/reports, why is that not enough?
Acknowledgments • NSF CAREER IIS-1541277, CBET-1250171, DMS-1557559, CNS-1625941 • NIH R01 CA214825, R01 CA225190, R01 LM012527 • The Feinberg Foundation • Collaborators and students • Electronic Visualization Laboratory faculty, staff and students • Anonymous reviewers, and T. Moller, A. Johnson, D. Laidlaw, T. Munzner
Food for thought • There’s little in this model that makes it SciVis-restricted • There’s more than one model/theory for almost everything • does VIS pick one model and follow it for many years? (“human-centered- design”, “data-tasks-user”, ”overview first”, “five sheet design”…) • What counts as validation/evaluation of a model/theory? • if it provides a new interpretation of existing data/reports, why is that not enough?
Questions? ACD model for domain characterization that: • Assigns value to vis based on activities • Ranks tasks > data • Incorporates workflows • Establishes the need for vis early • Extends the nested model with functional specs • Leads to a higher rate of project success
ACD Model Activity
Recommend
More recommend
