18/11/2018
1
PV182
Human Computer Interaction
Lecture 10
Cognitive Models
Fotis Liarokapis
liarokap@fi.muni.cz
19th November 2018
Cognitive Models
Cognitive Models (Low Level)
• Sources:
– Marti Hearst (SIMS, UC Berkeley)
– Robert Stevens (www.cs.man.ac.uk)
– Susan E. Brennan
(www.psychology.stonybrook.edu)
– Rebecca W. Boren (Arizona State University)
Cognitive Modeling Based Evaluation
• Fitts’ Law
– Used to predict time needed to select a target
• Keystroke-Level Model
– Low-level description of what users must do to
perform a task
• GOMS
– Structured, multi-level description of what users
must do to perform a task
Model of Human Processing The Model Human Processor
• Perceptual system
– Sensors
• Cognitive system
– Processors
• Motor system
– Effectors
(Card, Moran, & Newell, 1983)
18/11/2018
2
Important Parameters
• Memory capacity
• Decay
• Representation
• Processing cycle time
Sample Times
• Eye-movement = 230 [70~700] ms
– Typical time = 230 ms
– “Fastman” = 70 ms
– “Slowman” = 700 ms
• Perceptual processor: 100 [50~200]
• Cognitive processor: 70 [25~170]
• Motor processor: 70 [30~100]
Model of Simple RT Problem
• Task: Press button
– When symbol appears
Model of Simple RT Problem .
• Task: Press button
– When symbol appears
• 1. Perceptual processor
captures it in the visual
image store & represents
it in working memory
– 100 [50~200]
Model of Simple RT Problem ..
• Task: Press button
– When symbol appears
• 2. Cognitive processor
recognizes the
presence of a symbol
– 70 [25~170]
Model of Simple RT Problem ...
• Task: Press button
– When symbol appears
• 3. Motor processor
pushes the button
– 70 [30~100]
18/11/2018
3
Model of Simple RT Problem ....
• Task: Press button when symbol appears
• 1. The perceptual processor captures it in the visual
image store and represents it in working memory
– 100 [50~200]
• 2. The cognitive processor recognizes the presence
of a symbol
– 70 [25~170]
• 3. The motor processor pushes the button
– 70 [30~100]
• Total time?
Model of Simple RT Problem .....
• Each of these action primitives takes some
small amount of time (in msec.)
• The Model Human Processor provides a range
of parameters you can use to predict precisely
how long something will take, or to compare
the time needed for alternative actions
Hick’s Principle of Uncertainty
• Predicts how long a response will take in a
given situation, based on how likely (or
uncertain) the different possibilities are
Decision Complexity
• The speed with which an action can be selected
is strongly influenced by the number of possible
alternative actions that could be selected
• Hick-Hyman Law of reaction time shows a
logarithmic increase in reaction time (RT) as the
number of possible stimulus-response
alternatives (N) increases
– Humans process information at a constant rate
RT = a + bLog2N
Hick-Hyman Law Hick’s Principle of Uncertainty
• RT = a + b log2N
– RT = reaction time
– a, b = constants
– N = number of possible responses,
– assuming all are equally probable
• +1 is due to uncertainty whether to respond
18/11/2018
4
Decision Making Process
• The most efficient way to deliver a given
amount of information is by a smaller number
of complex decisions rather than a large
number of simple decisions
• An example is this decision making process:
– Would you like to have a big long-hair dog or a
small nervous dog or a black cat or a small no-hair
cat ?
• vs.
• Dog or a cat ? … dog
• Big or small ? … small
• Quiet or nervous ? … quiet
Power Law of Practice
• When something is done again and again,
performance follows a power law:
– You keep improving with practice, but as you
become an expert, you improve less and less
Power Law of Practice Note:
• The power law of practice describes
quantitative changes in skilled behavior (both
cognitive and motor), but not qualitative
changes (changes in strategies)
Fitt’s Law
• Moving a mouse to a target:
• What can vary?
• how long it takes
• how far you have to move
• how big the target is
Models movement time for selection tasks
The movement time for a well-rehearsed selection task
• increases as the distance to the target
increases
• decreases as the size of the target
increases
Fitt’s Law
18/11/2018
5
Time (in msec) = a + b log2(D/S+1)
where
a, b = constants (empirically derived)
D = distance
S = size
ID is Index of Difficulty = log2(D/S+1)
Fitt’s Law
Same ID → Same Difficulty
Target 1 Target 2
Time = a + b log2(D/S+1)
Fitt’s Law
Smaller ID → Easier
Target 2Target 1
Time = a + b log2(D/S+1)
Fitt’s Law
Larger ID → Harder
Target 2Target 1
Time = a + b log2(D/S+1)
Fitt’s Law
Keystroke-Level Model
• Another “discount” usability method
• Main idea:
– Walk through the interface, counting how many
operations it would take an expert user to perform
– Look for ways to optimize
– Look for potential sources of error
• KLM is very low-level (tiny operations)
Keystroke-Level Model
• How to make a KLM
– List specific actions user does to perform task
• Keystrokes and Button presses
• Mouse movements / Pointing
• Hand movements between keyboard & mouse / Homing
• Drawing
• System Response time (if it makes user wait)
– Add Mental operators
– Assign execution times to steps
– Sum execution times
• Only provides execution time and operator sequence
18/11/2018
6
KLM times
• operator remarks time(s)
• K Press key
• good typist (90wpm) 0.12
• poor typist (40wpm) 0.28
• non-typist 1.20
• B Mouse button press
• down or up 0.10
• click 0.20
• P Point with mouse
• Fitt’s law 0.1 log2(D/S+0.5)
• average movement 1.10
• H home hands to/from kbd 0.40
• D drawing / domain dependent •
M mentally prepare 1.35
• R response from system – measure KLM
Example
• Replace all instances of a 4-letter word.
– (example from Hochstein)
GOMS model of a system usage
• A family of user interface modeling techniques
• Goals, Operators, Methods, and Selection rules
– Higher-level than KLM
– Input: detailed description of UI and task(s)
– Output: various qualitative and quantitative measures
GOMS (Card, Moran, & Newell)
• Goal - what the user wants to achieve
• Operator - elementary perceptual, motor, or cognitive act
• Method - a series of operators that forms a procedure for
doing something
• Selection rule - how the user decides between methods
(if...then...).
GOMS (continued)
• Examples:
• Goal - editing a paper (high level)
• cutting and pasting text (low level)
• Operator - typing a keystroke
• Method - set of operators for cutting
• Selection rule - how the user chooses a method
Applications of GOMS analysis
• Compare UI designs
• Profiling
• Building a help system
– GOMS modeling makes user tasks and goals explicit
– Can suggest questions users will ask and the answers
18/11/2018
7
What GOMS can model
• Task must be goal-directed
– Some activities are more goal-directed than others
– Even creative activities contain goal-directed tasks
• Task must a routine cognitive skill
• Task can include serial and parallel tasks
GOMS Output
• Functionalitycoverage and consistency
– Does UI contain needed functions?
– Are similar tasks performed similarly?
• Operator sequence
– In what order are individual operations done?
– Abstraction of operations may vary among models
How to do GOMS Analysis
• Generate task description
– Pick high-level user Goal
– Write Method for accomplishing Goal
(may invoke subgoals)
– Write Methods for subgoals
• This is recursive
• Stops when Operators are reached
• Evaluate description of task
• Apply results to UI
• Iterate
Operators vs. Methods
• Operator: the most primitive action
• Method: requires several Operators or subgoal invocations
to accomplish
• Level of detail determined by
– KLM level – key press, mouse press
– Higher level - select-Close-from-File-menu
– Different parts of model can be at different levels of detail
GOMS Example
• Move text in a word processor
– (example from Hochstein)
GOMS Example
• Move text in a word processor
– (example from Hochstein)
18/11/2018
8
Advantages of GOMS
• Very general purpose
• Allows for individual differences
• Much predictive power about timing
• Good at predicting "ideal" performance
• Gives several qualitative and quantitative
measures
• Model explains why the results are what they are
• Less work than user study
• Easy to modify when interface is revised
• Research ongoing for tools to aid modeling
process
Disadvantages of GOMS
• Not so good at predicting errors
• Takes a long time to conduct analysis
• Whole may not be the sum of the parts
• Ignores the nature of internal symbolic
representations - focus is very low-level
• Not as easy as heuristic analysis, guidelines,
or cognitive walkthrough
Disadvantages of GOMS
• Only works for goal-directed tasks
• Assumes tasks are performed by expert users
• Evaluator must pick users’ tasks/goals
• Does not address several important UI issues,
such as
– readability of text
– memorability of icons, commands
• Does not address social or organizational
impact
Summary
• We can use Cognitive Modeling to make
predictions about interface usability
• Complementary to Usability Studies
• In practice:
– GOMS not used often
– Fitt’s law often used for determining best case for
new kinds of input methods
Questions Acknowledgements
• Prof. Ing. Jiří Sochor