Graph comprehension model talk, Birkbeck and Toulouse Le Mirail, February 2012

University of Huddersfield
University of HuddersfieldReader in Cognitive Science at University of Huddersfield
A computational model of graph
comprehension
David Peebles
February 15, 2012
Outline of the talk
1. Introduction
• Empirical studies
• Previous models of
graph use
2. The ACT-R cognitive
architecture
3. An ACT-R model of graph
comprehension
Graphical displays of information
• Increasingly used in an ever-widening range of activities
• A current data explosion with new representations and software tools
being developed all the time
• “Open data” and greater public access to information visualisations:
• http://www.informationisbeautiful.net
• http://www.guardian.co.uk/news/datablog
• http://infosthetics.com/
• http://flowingdata.com/
• The emergence of visual analytics (Thomas and Cook, 2005)
• A growing need to teach graphicacy and understand the cognitive
and visual processes that underlie it
Costs and benefits of using graphical representations
Benefits
• Information can be grouped
by location, reducing search
• Allows perceptual reasoning
and rapid identification of
salient aspects of data
(Larkin and Simon, 1987)
Costs
• People interpret graphical
elements differently
(Zacks and Tversky, 1999)
• Interpretation can be
distorted by visual features
(Peebles, 2008)
Bar and line graphs
Bar graphs
• Interpreted as representing
categories
• People typically encode
height and separate values
• Better for comparing single
quantities
Line graphs
• Interpreted as representing
ordered/scaled data
• People typically encode
slope and continuous change
• Better for identifying trends
Line graphs
0
1
2
3
4
5
6
7
8
9
10
JanFebMar AprMayJun Jul AugSepOctNovDec
AmountProduced
Month
Amount of Silver and Gold Produced (in tonnes)
by ‘Precious Metals Plc.’ in 1996
Silver
Gold
Similar, informationally equivalent,
graphs not computationally
equivalent (Larkin and Simon,
1987)
0
1
2
3
4
5
6
7
8
9
10
0 1 2 3 4 5 6 7 8 9 10
Gold
Silver
Amount of Silver and Gold Produced (in tonnes)
by ‘Precious Metals Plc.’ in 1996
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Oct
Nov
Dec
“How much silver was produced
when 4 tonnes of gold were
produced? (Peebles and Cheng,
2003)
Police performance monitors (Peebles, 2008)
The case for simulation modelling
• There has been a great deal of graph comprehension research (e.g.,
Carswell and Wickens, 1990; Kosslyn, 1994; Pinker, 1981;
Shah and Carpenter, 1995)
• Still impossible to produce precise advice about optimal display for
particular communicative goal (Meyer, 2000; Kosslyn, 2006)
• Too many variables interacting in complex ways
• Definitive experiments difficult as too many variations in stimulus
conditions possible.
• Computer simulation models of graph reading and comprehension
• allow testing of graphical designs
• allow exploration of ideas and hypotheses about cognitive and
visual mechanisms (McClelland, 2009)
Previous process models of graph use
• GOMS-based (task-analytic) models of graph question answering
• UCIE (Lohse, 1993) MA-P (Gillan, 1994)
1. Identify perceptual and cognitive operators from task or
verbal protocol analyses
2. Construct sequences of operators (with execution times)
3. Predict scan paths and task completion times
• Computational models
• BOZ graph generation tool (Casner, 1991)
• CaMerA model of economics expert with multiple
representations (Tabachneck-Schijf et al., 1997)
• ACT-R graph reading model (Peebles and Cheng, 2003)
• None of these models are of graph comprehension
Cognitive architectures
• “Unified theories of cognition” (Newell, 1990)
• Theories of core cognitive components, representational formats,
and processes (e.g., memory, cognitive control)
• Main examples of symbolic cognitive architectures:
• Soar (Newell, 1990; Laird, 2012)
• Epic (Kieras and Meyer, 1997)
• ACT-R (Anderson, 2007)
• All three based on production system architecture
• All can interact with simulated task environments
• This allows models of interaction with external representations to be
developed
The ACT-R cognitive architecture
• Hybrid architecture with symbolic and subsymbolic components
• Production system model of procedural memory & cognitive control
• Semantic network model of declarative memory
• Activation-based learning, memory retrieval & forgetting mechanisms
• Simulated eyes & hands for interacting with computer-based tasks
Visual
Module
ACT−R Buffers
Environment
Pattern
Matching
Execution
Production
Module
Motor
Control
State
Problem
State
Memory
Procedural
Memory
Declarative
Interaction graphs
Low High
10
20
30
40
50
60
70
80
90
100
●
●
Percent Error as a function of Experience and Time of Day
PercentError
Experience
Time of Day
Day
Night
• Display data from 2 × 2
factorial research designs
• Novices often find
interpretation difficult as they
differ from more familiar bar
and line graphs
Low High
10
20
30
40
50
60
70
80
90
100
Percent Error as a function of Experience and Time of Day
PercentError
Experience
Time of Day
Day
Night
• Come in bar and line forms
(both equally likely)
• Primary task is to
comprehend relationships
between variables, not
identify specific values
How people use interaction graphs
Low High
10
20
30
40
50
60
70
80
90
100
●
●
Percent Error as a function of Experience and Time of Day
PercentError
Experience
Time of Day
Day
Night
• Novices’ interpretations of
line graphs significantly
worse than bar graphs
(Peebles and Ali, 2009)
• Drawn by salience of lines to
identify legend variable &
ignore x axis variable
Low High
10
20
30
40
50
60
70
80
90
100
Percent Error as a function of Experience and Time of Day
PercentError
Experience
Time of Day
Day
Night
• Expert users’ interpretations
also affected by graph type
• New line graph design
improved novice performance
to bar graph level (Ali and
Peebles, submitted)
Alternative designs
Low High
10
20
30
40
50
60
70
80
90
100
●
●
Percent Error as a function of Experience and Time of Day
PercentError
Experience
Time of Day
Day
Night
• “Combined” graph – no
significant improvement
10
20
30
40
50
60
70
80
90
100
●
●
Percent Error as a function of Experience and Time of Day
PercentError
Experience
Low High
Time of Day
Day
Night
• “Colour match” graph –
significant improvement to
bar graph level
An ACT-R model of interaction graph comprehension
Aims
• Simulate verbal protocols and eye movement scan paths of experts
and novices interpreting interaction graphs
• Provide a detailed characterisation of:
• the prior knowledge required to interpret interaction graphs
• the information extracted from the diagram
• the knowledge generated during the comprehension process
• the cognitive and perceptual operations involved in interpreting
interaction graphs
• the strategic processes that control comprehension
• what underlies the differences between expert and novice
performance
Stages of comprehension
• Verbal protocols reveal that experts produce qualitative descriptions
of the differences between conditions, not individual values.
• Comprehension is typically carried out in two main phases:
1. Variable identification. Labels categorised as dependent or
independent according to location, and associated with levels
and identifiers: left or right (x axis), or colour (legend)
2. Pattern recognition and description. Distances between plot
points compared and used to probe long-term declarative
memory for interpretive knowledge
• Success – retrieved knowledge is used to provide an
interpretation
• Failure – simply describe the identification or comparison
process being carried out
Prior declarative graph knowledge (1)
• Three types of knowledge required to interpret interaction graphs:
1. The typical allocation of DV and IVs to graph axes and legend
Low High
10
20
30
40
50
60
70
80
90
100
●
●
Percent Error as a function of Experience and Time of Day
PercentError
Experience
Time of Day
Day
Night
Prior declarative graph knowledge (2)
2. The principle that distance between graphical elements indicates
magnitude of relationship between conceptual entities the elements
represent
Low High
10
20
30
40
50
60
70
80
90
100
●
●
Percent Error as a function of Experience and Time of DayPercentError
Experience
Time of Day
Day
Night
Prior declarative graph knowledge (3)
3. Graphical/spatial indicators of three interpretive facts:
• Simple effects. Distance between two plot points
• Main effects. Differences in the y-axis location of the midpoints
between two pairs of plot points.
• Interactions. Differences in the inter-point distances between
levels, combined with information about their point ordering.
Low High
10
20
30
40
50
60
70
80
90
100
●
●
Percent Error as a function of Experience and Time of Day
PercentError
Experience
Time of Day
Day
Night
Representing and encoding information in the graph
• Model encodes four x-y
coordinate points and spatial
distances between them
• Encoded numerically then
translated to symbolic
descriptions (e.g., “small”,
“very large”). “Elementary
perceptual tasks”
(Cleveland and McGill, 1984)
• Encode spatial distance
between plot points
• Compare magnitude of
two distances and
produce a symbolic
description of difference.
Low High
10
20
30
40
50
60
70
80
90
100
●
●
Percent Error as a function of Experience and Time of Day
PercentError
Experience
Time of Day
Day
Night
Knowledge generated during comprehension
• Knowledge chunks
representing the variables
• Three knowledge structures
that accumulate and
associate items of graph and
interpretive information when
identifying:
• Simple effects
• Main effects
• Interactions
Stages of comprehension
• Comprehension proceeds in the following order:
1. Read title. Identify variable names and create memory chunks.
2. Seek variable labels, identify what they are by their location and
if required, associate with label levels
3. Associate variable levels with indicators (position or colour)
4. Look at plot region and attempt to interpret distances. If a highly
salient pattern exists (e.g., cross, large gap) process that first
5. Continue until no more patterns are recognised
• Individual production rule for each pattern
• No production rule then pattern not processed
Example model output
1 seek text at top of display. . .
2 [m-yield] = [var]
3 [as] [a] [function] [of] [p-density] = [var]
4 [and] [n-level] = [var]
5 seek text at far right of display. . .
6 [n-level] at [far right] = [independent] var
7 look to nearest text. . .
8 [low] = level of [n-level]
9 [high] = level of [n-level]
10 seek objects in plot region. . .
11 [blue] [line]
12 no memory for [blue] look to legend
13 [blue] [rect]. seek near text. . . [blue] = [low]
14 [green] [rect]. seek near text. . . [green] = [high]
15 seek text at far left of display. . .
16 [m-yield] at [far left] = [dependent] var
17 seek text at bottom of display. . .
18 [p-density] at [bottom] = [independent] var
19 look to nearest text. . .
20 [compact] = level of [p-density]. [compact] = [right]
21 [sparse] = level of [p-density]. [sparse] = [left]
Example model output
22 identify legend levels. . .
23 [0.0] diff [blue] = [no] [simple] effect [low] [n-level]
24 [0.5] diff [green] [modest] [simple] effect [high] [n-level]
25 compare [blue] & [green] levels. . .
26 [small] diff. [high] [n-level] > [low] [n-level]
27 [small] [main] effect [n-level]
28 identify x axis levels. . .
29 [0.0] diff [left] = [no] [simple] effect [sparse] [p-density]
30 [0.5] diff [right] = [modest] [simple] effect [compact] [p-density]
31 compare [left] & [right] levels. . .
32 [small] diff. [compact] [p-density] > [sparse] [p-density]
33 [small] [main] effect [p-density]
34 compare left and right patterns. . .
35 [0.5] diff in distance between points. [right] bigger
36 [modest] diff & [same] point order = [modest] [interaction]
37 for [sparse] [p-density] [low] [n-level] = [high] [n-level]
38 for [compact] [p-density] [high] [n-level] > [low] [n-level]
Limitations of the model
• Model currently focuses on graph knowledge and not the effects of
domain knowledge. Experts do display such domain-general graph
knowledge.
• Early visual processes are not specified. ACT-R models of visual and
spatial processing are being developed.
• Although based on human data, the model has not been rigorously
compared to eye movement and verbal protocol data yet.
• Model of novice users has not been developed yet
• Selectively remove production rules and interpretive declarative
knowledge.
Conclusions
• Currently the only computational model of graph comprehension
• Given a data set from a 2 × 2 factorial research design it will produce
an expert level description of the effects
• The model also describes the patterns used to produce the
interpretation
• The model associates variables and their levels, and levels with their
colour identifiers using mechanisms that are functionally equivalent
to the Gestalt laws of proximity and similarity respectively.
• The current model can be considered a first approximation to a more
detailed model that incorporates additional factors to broaden the
scope of behaviour accounted for.
• The model will form the basis of a more general graph
comprehension model that can interpret bar interaction graphs and
more general bar and line graphs.
References (1)
• Ali, N. and Peebles, D. (submitted). The effect of gestalt laws of perceptual
organisation on the comprehension of three-variable bar and line graphs
• Anderson, J. R. (2007). How can the human mind occur in the physical
universe? Oxford University Press, New York, NY
• Carswell, C. M. and Wickens, C. D. (1990). The perceptual interaction of
graphical attributes: Configurality, stimulus homogeneity, and object
interaction. Perception & Psychophysics, 47:157–168
• Casner, S. M. (1991). A task-analytic approach to the automated design of
graphic presentations. ACM Transactions on Graphics, 10:111–151
• Gillan, D. J. (1994). A componential model of human interaction with
graphs: 1. linear regression modelling. Human Factors, 36(3):419–440
References (2)
• Kieras, D. E. and Meyer, D. E. (1997). An overview of the EPIC architecture
for cognition and performance with application to human-computer
interaction. Human-Computer Interaction, 12:391–438
• Kosslyn, S. M. (1994). Elements of graph design. W. H. Freeman & Co.,
New York
• Kosslyn, S. M. (2006). Graph design for the eye and mind. Oxford
University Press, New York
• Laird, J. E. (2012). The Soar cognitive architecture. MIT Press, Cambridge,
Mass
• Lohse, G. L. (1993). A cognitive model for understanding graphical
perception. Human-Computer Interaction, 8:353–388
• Meyer, J. (2000). Performance with tables and graphs: effects of training
and a visual search model. Ergonomics, 43(11):1840–1865
References (3)
• McClelland, J. L. (2009). The place of modeling in cognitive science.
Topics in Cognitive Science, 1:11–38
• Newell, A. (1990). Unified theories of cognition. The William James
lectures. Harvard University Press
• Peebles, D. (2008). The effect of emergent features on judgments of
quantity in configural and seperable displays. Journal of Experimental
Psychology: Applied, 14(2):85–100
• Peebles, D. and Ali, N. (2009). Differences in comprehensibility between
three-variable bar and line graphs. In Proceedings of the Thirty-first Annual
Conference of the Cognitive Science Society, pages 2938–2943, Mahwah,
NJ. Lawrence Erlbaum Associates
• Peebles, D. and Cheng, P. C.-H. (2003). Modeling the effect of task and
graphical representation on response latency in a graph reading task.
Human Factors, 45:28–45
References (4)
• Pinker, S. (1981). A theory of graph comprehension. Occasional Paper 10,
Center for Cognitive Science, Massachusetts Institute of Technology
• Shah, P. and Carpenter, P. A. (1995). Conceptual limitations in
comprehending line graphs. Journal of Experimental Psychology: General,
124:43–62
• Tabachneck-Schijf, H. J. M., Leonardo, A. M., and Simon, H. A. (1997).
CaMeRa: A computational model of multiple representations. Cognitive
Science, 21:305–350
• Thomas, J. J. and Cook, K. A. (2005). Illuminating the path: The research
and development agenda for visual analytics. IEEE Press
• Zacks, J. and Tversky, B. (1999). Bars and lines: A study of graphic
communication. Memory and Cognition, 27(6):1073–1079
1 of 30

Recommended

Kaggle Days Paris - Alberto Danese - ML Interpretability by
Kaggle Days Paris - Alberto Danese - ML InterpretabilityKaggle Days Paris - Alberto Danese - ML Interpretability
Kaggle Days Paris - Alberto Danese - ML InterpretabilityAlberto Danese
340 views30 slides
Mobile-First SEO - The Marketers Edition #3XEDigital by
Mobile-First SEO - The Marketers Edition #3XEDigitalMobile-First SEO - The Marketers Edition #3XEDigital
Mobile-First SEO - The Marketers Edition #3XEDigitalAleyda Solís
471.4K views80 slides
What Makes Great Infographics by
What Makes Great InfographicsWhat Makes Great Infographics
What Makes Great InfographicsSlideShare
1.1M views1 slide
You Suck At PowerPoint! by
You Suck At PowerPoint!You Suck At PowerPoint!
You Suck At PowerPoint!Jesse Desjardins - @jessedee
1.4M views61 slides
STOP! VIEW THIS! 10-Step Checklist When Uploading to Slideshare by
STOP! VIEW THIS! 10-Step Checklist When Uploading to SlideshareSTOP! VIEW THIS! 10-Step Checklist When Uploading to Slideshare
STOP! VIEW THIS! 10-Step Checklist When Uploading to SlideshareEmpowered Presentations
932.2K views13 slides
How To Get More From SlideShare - Super-Simple Tips For Content Marketing by
How To Get More From SlideShare - Super-Simple Tips For Content MarketingHow To Get More From SlideShare - Super-Simple Tips For Content Marketing
How To Get More From SlideShare - Super-Simple Tips For Content MarketingContent Marketing Institute
1.1M views33 slides

More Related Content

Similar to Graph comprehension model talk, Birkbeck and Toulouse Le Mirail, February 2012

Csci101 lect00 introduction by
Csci101 lect00 introductionCsci101 lect00 introduction
Csci101 lect00 introductionElsayed Hemayed
12 views20 slides
Data visualization by
Data visualizationData visualization
Data visualizationMoushmi Dasgupta
350 views48 slides
The Machine Learning Workflow with Azure by
The Machine Learning Workflow with AzureThe Machine Learning Workflow with Azure
The Machine Learning Workflow with AzureIvo Andreev
6.4K views34 slides
The state of the art in integrating machine learning into visual analytics by
The state of the art in integrating machine learning into visual analyticsThe state of the art in integrating machine learning into visual analytics
The state of the art in integrating machine learning into visual analyticsCagatay Turkay
161 views43 slides
ExplainableAI.pptx by
ExplainableAI.pptxExplainableAI.pptx
ExplainableAI.pptxAndrea Morichetta
8 views16 slides
Activity Monitoring Using Wearable Sensors and Smart Phone by
Activity Monitoring Using Wearable Sensors and Smart PhoneActivity Monitoring Using Wearable Sensors and Smart Phone
Activity Monitoring Using Wearable Sensors and Smart PhoneDrAhmedZoha
340 views18 slides

Similar to Graph comprehension model talk, Birkbeck and Toulouse Le Mirail, February 2012(20)

The Machine Learning Workflow with Azure by Ivo Andreev
The Machine Learning Workflow with AzureThe Machine Learning Workflow with Azure
The Machine Learning Workflow with Azure
Ivo Andreev6.4K views
The state of the art in integrating machine learning into visual analytics by Cagatay Turkay
The state of the art in integrating machine learning into visual analyticsThe state of the art in integrating machine learning into visual analytics
The state of the art in integrating machine learning into visual analytics
Cagatay Turkay161 views
Activity Monitoring Using Wearable Sensors and Smart Phone by DrAhmedZoha
Activity Monitoring Using Wearable Sensors and Smart PhoneActivity Monitoring Using Wearable Sensors and Smart Phone
Activity Monitoring Using Wearable Sensors and Smart Phone
DrAhmedZoha340 views
Weave-D - 2nd Progress Evaluation Presentation by lasinducharith
Weave-D - 2nd Progress Evaluation PresentationWeave-D - 2nd Progress Evaluation Presentation
Weave-D - 2nd Progress Evaluation Presentation
lasinducharith1.6K views
Computational Thinking in the Workforce and Next Generation Science Standards... by Josh Sheldon
Computational Thinking in the Workforce and Next Generation Science Standards...Computational Thinking in the Workforce and Next Generation Science Standards...
Computational Thinking in the Workforce and Next Generation Science Standards...
Josh Sheldon2K views
Week 8 & 10 by Study Geek
Week 8 & 10Week 8 & 10
Week 8 & 10
Study Geek80 views
Lecture_2_Stats.pdf by paijitk
Lecture_2_Stats.pdfLecture_2_Stats.pdf
Lecture_2_Stats.pdf
paijitk2 views
Advantages and Limitations for Diagrams and Graphs by Hardik Bhaavani
Advantages and Limitations for Diagrams and GraphsAdvantages and Limitations for Diagrams and Graphs
Advantages and Limitations for Diagrams and Graphs
Hardik Bhaavani16.4K views
Graph Models for Deep Learning by Experfy
Graph Models for Deep LearningGraph Models for Deep Learning
Graph Models for Deep Learning
Experfy164 views
DataONE Education Module 09: Analysis and Workflows by DataONE
DataONE Education Module 09: Analysis and WorkflowsDataONE Education Module 09: Analysis and Workflows
DataONE Education Module 09: Analysis and Workflows
DataONE3.1K views
Only Time Will Tell: Modelling Information Diffusion in Code Review with Time... by Michael Dorner
Only Time Will Tell: Modelling Information Diffusion in Code Review with Time...Only Time Will Tell: Modelling Information Diffusion in Code Review with Time...
Only Time Will Tell: Modelling Information Diffusion in Code Review with Time...
Michael Dorner7 views
Big data visualization by Anurag Gupta
Big data visualizationBig data visualization
Big data visualization
Anurag Gupta1.3K views
Introduction to Data Visualization by Ana Jofre
Introduction to Data Visualization Introduction to Data Visualization
Introduction to Data Visualization
Ana Jofre2.5K views

More from University of Huddersfield

Multiple processes in graph-based reasoning by
Multiple processes in graph-based reasoningMultiple processes in graph-based reasoning
Multiple processes in graph-based reasoningUniversity of Huddersfield
106 views11 slides
Modelling alternative strategies for mental rotation by
Modelling alternative strategies for mental rotationModelling alternative strategies for mental rotation
Modelling alternative strategies for mental rotationUniversity of Huddersfield
103 views27 slides
Robert Gordon talk, 21 March 2018 by
Robert Gordon talk, 21 March 2018Robert Gordon talk, 21 March 2018
Robert Gordon talk, 21 March 2018University of Huddersfield
103 views26 slides
EWIC talk - 07 June, 2018 by
EWIC talk - 07 June, 2018EWIC talk - 07 June, 2018
EWIC talk - 07 June, 2018University of Huddersfield
96 views20 slides
Multiple representations and visual mental imagery in artificial cognitive sy... by
Multiple representations and visual mental imagery in artificial cognitive sy...Multiple representations and visual mental imagery in artificial cognitive sy...
Multiple representations and visual mental imagery in artificial cognitive sy...University of Huddersfield
157 views26 slides
Multiple representations talk, Middlesex University. February 23, 2018 by
Multiple representations talk, Middlesex University. February 23, 2018Multiple representations talk, Middlesex University. February 23, 2018
Multiple representations talk, Middlesex University. February 23, 2018University of Huddersfield
91 views27 slides

More from University of Huddersfield(12)

Multiple representations and visual mental imagery in artificial cognitive sy... by University of Huddersfield
Multiple representations and visual mental imagery in artificial cognitive sy...Multiple representations and visual mental imagery in artificial cognitive sy...
Multiple representations and visual mental imagery in artificial cognitive sy...
Machine Learning, Artificial General Intelligence, and Robots with Human Minds by University of Huddersfield
Machine Learning, Artificial General Intelligence, and Robots with Human MindsMachine Learning, Artificial General Intelligence, and Robots with Human Minds
Machine Learning, Artificial General Intelligence, and Robots with Human Minds
Diagrammatic Cognition: Discovery and Design workshop, Humboldt University, B... by University of Huddersfield
Diagrammatic Cognition: Discovery and Design workshop, Humboldt University, B...Diagrammatic Cognition: Discovery and Design workshop, Humboldt University, B...
Diagrammatic Cognition: Discovery and Design workshop, Humboldt University, B...
A cognitive architecture-based modelling approach to understanding biases in ... by University of Huddersfield
A cognitive architecture-based modelling approach to understanding biases in ...A cognitive architecture-based modelling approach to understanding biases in ...
A cognitive architecture-based modelling approach to understanding biases in ...

Recently uploaded

Distinct distributions of elliptical and disk galaxies across the Local Super... by
Distinct distributions of elliptical and disk galaxies across the Local Super...Distinct distributions of elliptical and disk galaxies across the Local Super...
Distinct distributions of elliptical and disk galaxies across the Local Super...Sérgio Sacani
33 views12 slides
Applications of Large Language Models in Materials Discovery and Design by
Applications of Large Language Models in Materials Discovery and DesignApplications of Large Language Models in Materials Discovery and Design
Applications of Large Language Models in Materials Discovery and DesignAnubhav Jain
13 views17 slides
Effect of Integrated Nutrient Management on Growth and Yield of Solanaceous F... by
Effect of Integrated Nutrient Management on Growth and Yield of Solanaceous F...Effect of Integrated Nutrient Management on Growth and Yield of Solanaceous F...
Effect of Integrated Nutrient Management on Growth and Yield of Solanaceous F...SwagatBehera9
5 views36 slides
1978 NASA News Release Log by
1978 NASA News Release Log1978 NASA News Release Log
1978 NASA News Release Logpurrterminator
11 views146 slides
Pollination By Nagapradheesh.M.pptx by
Pollination By Nagapradheesh.M.pptxPollination By Nagapradheesh.M.pptx
Pollination By Nagapradheesh.M.pptxMNAGAPRADHEESH
19 views9 slides
plasmids by
plasmidsplasmids
plasmidsscribddarkened352
13 views2 slides

Recently uploaded(20)

Distinct distributions of elliptical and disk galaxies across the Local Super... by Sérgio Sacani
Distinct distributions of elliptical and disk galaxies across the Local Super...Distinct distributions of elliptical and disk galaxies across the Local Super...
Distinct distributions of elliptical and disk galaxies across the Local Super...
Sérgio Sacani33 views
Applications of Large Language Models in Materials Discovery and Design by Anubhav Jain
Applications of Large Language Models in Materials Discovery and DesignApplications of Large Language Models in Materials Discovery and Design
Applications of Large Language Models in Materials Discovery and Design
Anubhav Jain13 views
Effect of Integrated Nutrient Management on Growth and Yield of Solanaceous F... by SwagatBehera9
Effect of Integrated Nutrient Management on Growth and Yield of Solanaceous F...Effect of Integrated Nutrient Management on Growth and Yield of Solanaceous F...
Effect of Integrated Nutrient Management on Growth and Yield of Solanaceous F...
SwagatBehera95 views
Pollination By Nagapradheesh.M.pptx by MNAGAPRADHEESH
Pollination By Nagapradheesh.M.pptxPollination By Nagapradheesh.M.pptx
Pollination By Nagapradheesh.M.pptx
MNAGAPRADHEESH19 views
A giant thin stellar stream in the Coma Galaxy Cluster by Sérgio Sacani
A giant thin stellar stream in the Coma Galaxy ClusterA giant thin stellar stream in the Coma Galaxy Cluster
A giant thin stellar stream in the Coma Galaxy Cluster
Sérgio Sacani17 views
Conventional and non-conventional methods for improvement of cucurbits.pptx by gandhi976
Conventional and non-conventional methods for improvement of cucurbits.pptxConventional and non-conventional methods for improvement of cucurbits.pptx
Conventional and non-conventional methods for improvement of cucurbits.pptx
gandhi97620 views
Structure of purines and pyrimidines - Jahnvi arora (11228108), mmdu ,mullana... by jahnviarora989
Structure of purines and pyrimidines - Jahnvi arora (11228108), mmdu ,mullana...Structure of purines and pyrimidines - Jahnvi arora (11228108), mmdu ,mullana...
Structure of purines and pyrimidines - Jahnvi arora (11228108), mmdu ,mullana...
jahnviarora9896 views
A Ready-to-Analyze High-Plex Spatial Signature Development Workflow for Cance... by InsideScientific
A Ready-to-Analyze High-Plex Spatial Signature Development Workflow for Cance...A Ready-to-Analyze High-Plex Spatial Signature Development Workflow for Cance...
A Ready-to-Analyze High-Plex Spatial Signature Development Workflow for Cance...
InsideScientific78 views
RemeOs science and clinical evidence by PetrusViitanen1
RemeOs science and clinical evidenceRemeOs science and clinical evidence
RemeOs science and clinical evidence
PetrusViitanen147 views
Light Pollution for LVIS students by CWBarthlmew
Light Pollution for LVIS studentsLight Pollution for LVIS students
Light Pollution for LVIS students
CWBarthlmew9 views
별헤는 사람들 2023년 12월호 전명원 교수 자료 by sciencepeople
별헤는 사람들 2023년 12월호 전명원 교수 자료별헤는 사람들 2023년 12월호 전명원 교수 자료
별헤는 사람들 2023년 12월호 전명원 교수 자료
sciencepeople58 views
Ellagic Acid and Its Metabolites as Potent and Selective Allosteric Inhibitor... by Trustlife
Ellagic Acid and Its Metabolites as Potent and Selective Allosteric Inhibitor...Ellagic Acid and Its Metabolites as Potent and Selective Allosteric Inhibitor...
Ellagic Acid and Its Metabolites as Potent and Selective Allosteric Inhibitor...
Trustlife100 views
Exploring the nature and synchronicity of early cluster formation in the Larg... by Sérgio Sacani
Exploring the nature and synchronicity of early cluster formation in the Larg...Exploring the nature and synchronicity of early cluster formation in the Larg...
Exploring the nature and synchronicity of early cluster formation in the Larg...
Sérgio Sacani910 views

Graph comprehension model talk, Birkbeck and Toulouse Le Mirail, February 2012

  • 1. A computational model of graph comprehension David Peebles February 15, 2012
  • 2. Outline of the talk 1. Introduction • Empirical studies • Previous models of graph use 2. The ACT-R cognitive architecture 3. An ACT-R model of graph comprehension
  • 3. Graphical displays of information • Increasingly used in an ever-widening range of activities • A current data explosion with new representations and software tools being developed all the time • “Open data” and greater public access to information visualisations: • http://www.informationisbeautiful.net • http://www.guardian.co.uk/news/datablog • http://infosthetics.com/ • http://flowingdata.com/ • The emergence of visual analytics (Thomas and Cook, 2005) • A growing need to teach graphicacy and understand the cognitive and visual processes that underlie it
  • 4. Costs and benefits of using graphical representations Benefits • Information can be grouped by location, reducing search • Allows perceptual reasoning and rapid identification of salient aspects of data (Larkin and Simon, 1987) Costs • People interpret graphical elements differently (Zacks and Tversky, 1999) • Interpretation can be distorted by visual features (Peebles, 2008)
  • 5. Bar and line graphs Bar graphs • Interpreted as representing categories • People typically encode height and separate values • Better for comparing single quantities Line graphs • Interpreted as representing ordered/scaled data • People typically encode slope and continuous change • Better for identifying trends
  • 6. Line graphs 0 1 2 3 4 5 6 7 8 9 10 JanFebMar AprMayJun Jul AugSepOctNovDec AmountProduced Month Amount of Silver and Gold Produced (in tonnes) by ‘Precious Metals Plc.’ in 1996 Silver Gold Similar, informationally equivalent, graphs not computationally equivalent (Larkin and Simon, 1987) 0 1 2 3 4 5 6 7 8 9 10 0 1 2 3 4 5 6 7 8 9 10 Gold Silver Amount of Silver and Gold Produced (in tonnes) by ‘Precious Metals Plc.’ in 1996 Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec “How much silver was produced when 4 tonnes of gold were produced? (Peebles and Cheng, 2003)
  • 7. Police performance monitors (Peebles, 2008)
  • 8. The case for simulation modelling • There has been a great deal of graph comprehension research (e.g., Carswell and Wickens, 1990; Kosslyn, 1994; Pinker, 1981; Shah and Carpenter, 1995) • Still impossible to produce precise advice about optimal display for particular communicative goal (Meyer, 2000; Kosslyn, 2006) • Too many variables interacting in complex ways • Definitive experiments difficult as too many variations in stimulus conditions possible. • Computer simulation models of graph reading and comprehension • allow testing of graphical designs • allow exploration of ideas and hypotheses about cognitive and visual mechanisms (McClelland, 2009)
  • 9. Previous process models of graph use • GOMS-based (task-analytic) models of graph question answering • UCIE (Lohse, 1993) MA-P (Gillan, 1994) 1. Identify perceptual and cognitive operators from task or verbal protocol analyses 2. Construct sequences of operators (with execution times) 3. Predict scan paths and task completion times • Computational models • BOZ graph generation tool (Casner, 1991) • CaMerA model of economics expert with multiple representations (Tabachneck-Schijf et al., 1997) • ACT-R graph reading model (Peebles and Cheng, 2003) • None of these models are of graph comprehension
  • 10. Cognitive architectures • “Unified theories of cognition” (Newell, 1990) • Theories of core cognitive components, representational formats, and processes (e.g., memory, cognitive control) • Main examples of symbolic cognitive architectures: • Soar (Newell, 1990; Laird, 2012) • Epic (Kieras and Meyer, 1997) • ACT-R (Anderson, 2007) • All three based on production system architecture • All can interact with simulated task environments • This allows models of interaction with external representations to be developed
  • 11. The ACT-R cognitive architecture • Hybrid architecture with symbolic and subsymbolic components • Production system model of procedural memory & cognitive control • Semantic network model of declarative memory • Activation-based learning, memory retrieval & forgetting mechanisms • Simulated eyes & hands for interacting with computer-based tasks Visual Module ACT−R Buffers Environment Pattern Matching Execution Production Module Motor Control State Problem State Memory Procedural Memory Declarative
  • 12. Interaction graphs Low High 10 20 30 40 50 60 70 80 90 100 ● ● Percent Error as a function of Experience and Time of Day PercentError Experience Time of Day Day Night • Display data from 2 × 2 factorial research designs • Novices often find interpretation difficult as they differ from more familiar bar and line graphs Low High 10 20 30 40 50 60 70 80 90 100 Percent Error as a function of Experience and Time of Day PercentError Experience Time of Day Day Night • Come in bar and line forms (both equally likely) • Primary task is to comprehend relationships between variables, not identify specific values
  • 13. How people use interaction graphs Low High 10 20 30 40 50 60 70 80 90 100 ● ● Percent Error as a function of Experience and Time of Day PercentError Experience Time of Day Day Night • Novices’ interpretations of line graphs significantly worse than bar graphs (Peebles and Ali, 2009) • Drawn by salience of lines to identify legend variable & ignore x axis variable Low High 10 20 30 40 50 60 70 80 90 100 Percent Error as a function of Experience and Time of Day PercentError Experience Time of Day Day Night • Expert users’ interpretations also affected by graph type • New line graph design improved novice performance to bar graph level (Ali and Peebles, submitted)
  • 14. Alternative designs Low High 10 20 30 40 50 60 70 80 90 100 ● ● Percent Error as a function of Experience and Time of Day PercentError Experience Time of Day Day Night • “Combined” graph – no significant improvement 10 20 30 40 50 60 70 80 90 100 ● ● Percent Error as a function of Experience and Time of Day PercentError Experience Low High Time of Day Day Night • “Colour match” graph – significant improvement to bar graph level
  • 15. An ACT-R model of interaction graph comprehension Aims • Simulate verbal protocols and eye movement scan paths of experts and novices interpreting interaction graphs • Provide a detailed characterisation of: • the prior knowledge required to interpret interaction graphs • the information extracted from the diagram • the knowledge generated during the comprehension process • the cognitive and perceptual operations involved in interpreting interaction graphs • the strategic processes that control comprehension • what underlies the differences between expert and novice performance
  • 16. Stages of comprehension • Verbal protocols reveal that experts produce qualitative descriptions of the differences between conditions, not individual values. • Comprehension is typically carried out in two main phases: 1. Variable identification. Labels categorised as dependent or independent according to location, and associated with levels and identifiers: left or right (x axis), or colour (legend) 2. Pattern recognition and description. Distances between plot points compared and used to probe long-term declarative memory for interpretive knowledge • Success – retrieved knowledge is used to provide an interpretation • Failure – simply describe the identification or comparison process being carried out
  • 17. Prior declarative graph knowledge (1) • Three types of knowledge required to interpret interaction graphs: 1. The typical allocation of DV and IVs to graph axes and legend Low High 10 20 30 40 50 60 70 80 90 100 ● ● Percent Error as a function of Experience and Time of Day PercentError Experience Time of Day Day Night
  • 18. Prior declarative graph knowledge (2) 2. The principle that distance between graphical elements indicates magnitude of relationship between conceptual entities the elements represent Low High 10 20 30 40 50 60 70 80 90 100 ● ● Percent Error as a function of Experience and Time of DayPercentError Experience Time of Day Day Night
  • 19. Prior declarative graph knowledge (3) 3. Graphical/spatial indicators of three interpretive facts: • Simple effects. Distance between two plot points • Main effects. Differences in the y-axis location of the midpoints between two pairs of plot points. • Interactions. Differences in the inter-point distances between levels, combined with information about their point ordering. Low High 10 20 30 40 50 60 70 80 90 100 ● ● Percent Error as a function of Experience and Time of Day PercentError Experience Time of Day Day Night
  • 20. Representing and encoding information in the graph • Model encodes four x-y coordinate points and spatial distances between them • Encoded numerically then translated to symbolic descriptions (e.g., “small”, “very large”). “Elementary perceptual tasks” (Cleveland and McGill, 1984) • Encode spatial distance between plot points • Compare magnitude of two distances and produce a symbolic description of difference. Low High 10 20 30 40 50 60 70 80 90 100 ● ● Percent Error as a function of Experience and Time of Day PercentError Experience Time of Day Day Night
  • 21. Knowledge generated during comprehension • Knowledge chunks representing the variables • Three knowledge structures that accumulate and associate items of graph and interpretive information when identifying: • Simple effects • Main effects • Interactions
  • 22. Stages of comprehension • Comprehension proceeds in the following order: 1. Read title. Identify variable names and create memory chunks. 2. Seek variable labels, identify what they are by their location and if required, associate with label levels 3. Associate variable levels with indicators (position or colour) 4. Look at plot region and attempt to interpret distances. If a highly salient pattern exists (e.g., cross, large gap) process that first 5. Continue until no more patterns are recognised • Individual production rule for each pattern • No production rule then pattern not processed
  • 23. Example model output 1 seek text at top of display. . . 2 [m-yield] = [var] 3 [as] [a] [function] [of] [p-density] = [var] 4 [and] [n-level] = [var] 5 seek text at far right of display. . . 6 [n-level] at [far right] = [independent] var 7 look to nearest text. . . 8 [low] = level of [n-level] 9 [high] = level of [n-level] 10 seek objects in plot region. . . 11 [blue] [line] 12 no memory for [blue] look to legend 13 [blue] [rect]. seek near text. . . [blue] = [low] 14 [green] [rect]. seek near text. . . [green] = [high] 15 seek text at far left of display. . . 16 [m-yield] at [far left] = [dependent] var 17 seek text at bottom of display. . . 18 [p-density] at [bottom] = [independent] var 19 look to nearest text. . . 20 [compact] = level of [p-density]. [compact] = [right] 21 [sparse] = level of [p-density]. [sparse] = [left]
  • 24. Example model output 22 identify legend levels. . . 23 [0.0] diff [blue] = [no] [simple] effect [low] [n-level] 24 [0.5] diff [green] [modest] [simple] effect [high] [n-level] 25 compare [blue] & [green] levels. . . 26 [small] diff. [high] [n-level] > [low] [n-level] 27 [small] [main] effect [n-level] 28 identify x axis levels. . . 29 [0.0] diff [left] = [no] [simple] effect [sparse] [p-density] 30 [0.5] diff [right] = [modest] [simple] effect [compact] [p-density] 31 compare [left] & [right] levels. . . 32 [small] diff. [compact] [p-density] > [sparse] [p-density] 33 [small] [main] effect [p-density] 34 compare left and right patterns. . . 35 [0.5] diff in distance between points. [right] bigger 36 [modest] diff & [same] point order = [modest] [interaction] 37 for [sparse] [p-density] [low] [n-level] = [high] [n-level] 38 for [compact] [p-density] [high] [n-level] > [low] [n-level]
  • 25. Limitations of the model • Model currently focuses on graph knowledge and not the effects of domain knowledge. Experts do display such domain-general graph knowledge. • Early visual processes are not specified. ACT-R models of visual and spatial processing are being developed. • Although based on human data, the model has not been rigorously compared to eye movement and verbal protocol data yet. • Model of novice users has not been developed yet • Selectively remove production rules and interpretive declarative knowledge.
  • 26. Conclusions • Currently the only computational model of graph comprehension • Given a data set from a 2 × 2 factorial research design it will produce an expert level description of the effects • The model also describes the patterns used to produce the interpretation • The model associates variables and their levels, and levels with their colour identifiers using mechanisms that are functionally equivalent to the Gestalt laws of proximity and similarity respectively. • The current model can be considered a first approximation to a more detailed model that incorporates additional factors to broaden the scope of behaviour accounted for. • The model will form the basis of a more general graph comprehension model that can interpret bar interaction graphs and more general bar and line graphs.
  • 27. References (1) • Ali, N. and Peebles, D. (submitted). The effect of gestalt laws of perceptual organisation on the comprehension of three-variable bar and line graphs • Anderson, J. R. (2007). How can the human mind occur in the physical universe? Oxford University Press, New York, NY • Carswell, C. M. and Wickens, C. D. (1990). The perceptual interaction of graphical attributes: Configurality, stimulus homogeneity, and object interaction. Perception & Psychophysics, 47:157–168 • Casner, S. M. (1991). A task-analytic approach to the automated design of graphic presentations. ACM Transactions on Graphics, 10:111–151 • Gillan, D. J. (1994). A componential model of human interaction with graphs: 1. linear regression modelling. Human Factors, 36(3):419–440
  • 28. References (2) • Kieras, D. E. and Meyer, D. E. (1997). An overview of the EPIC architecture for cognition and performance with application to human-computer interaction. Human-Computer Interaction, 12:391–438 • Kosslyn, S. M. (1994). Elements of graph design. W. H. Freeman & Co., New York • Kosslyn, S. M. (2006). Graph design for the eye and mind. Oxford University Press, New York • Laird, J. E. (2012). The Soar cognitive architecture. MIT Press, Cambridge, Mass • Lohse, G. L. (1993). A cognitive model for understanding graphical perception. Human-Computer Interaction, 8:353–388 • Meyer, J. (2000). Performance with tables and graphs: effects of training and a visual search model. Ergonomics, 43(11):1840–1865
  • 29. References (3) • McClelland, J. L. (2009). The place of modeling in cognitive science. Topics in Cognitive Science, 1:11–38 • Newell, A. (1990). Unified theories of cognition. The William James lectures. Harvard University Press • Peebles, D. (2008). The effect of emergent features on judgments of quantity in configural and seperable displays. Journal of Experimental Psychology: Applied, 14(2):85–100 • Peebles, D. and Ali, N. (2009). Differences in comprehensibility between three-variable bar and line graphs. In Proceedings of the Thirty-first Annual Conference of the Cognitive Science Society, pages 2938–2943, Mahwah, NJ. Lawrence Erlbaum Associates • Peebles, D. and Cheng, P. C.-H. (2003). Modeling the effect of task and graphical representation on response latency in a graph reading task. Human Factors, 45:28–45
  • 30. References (4) • Pinker, S. (1981). A theory of graph comprehension. Occasional Paper 10, Center for Cognitive Science, Massachusetts Institute of Technology • Shah, P. and Carpenter, P. A. (1995). Conceptual limitations in comprehending line graphs. Journal of Experimental Psychology: General, 124:43–62 • Tabachneck-Schijf, H. J. M., Leonardo, A. M., and Simon, H. A. (1997). CaMeRa: A computational model of multiple representations. Cognitive Science, 21:305–350 • Thomas, J. J. and Cook, K. A. (2005). Illuminating the path: The research and development agenda for visual analytics. IEEE Press • Zacks, J. and Tversky, B. (1999). Bars and lines: A study of graphic communication. Memory and Cognition, 27(6):1073–1079