Research Methods for HCI:

Cognitive Modelling

BCS HCI Tutorial

1st September, 2008
What is a cognitive model?
Specification of mental representations,
operations and problem-solving
strategies that occur during execution of
computer-based tasks.
Can be:
General description of steps required to
complete a task
Sophisticated computer simulation of users
performing tasks with computers
Anna Cox, UCL Interaction C 2
Why build cognitive models?
Help us develop a precise
understanding of HCI
Sometimes enable us to make
predictions about performance
Can be used early on in design as well
as to evaluate existing designs

Anna Cox, UCL Interaction C 3

Engineering models
Used to predict aspects of human
performance before users are
introduced to a system
Evaluate speed and number of
operations required to perform different
tasks

Anna Cox, UCL Interaction C 4

Engineering models: GOMS
Goals
Operators
Methods
Selection rules

Anna Cox, UCL Interaction C 5

Engineering models: KLM
Keystroke-level model
Restricted version of GOMS (no G or S!)
Predicts execution time to perform a task
All operators have specific execution time
Time to complete the task = sum of
execution times of all the operators
required to complete the task

Anna Cox, UCL Interaction C 6

Cognitive Architectures
Sophisticated and complex programming
environment within which psychological
theory is represented
Can build models of a user completing a
particular task within the architecture
The architecture constrains the way in which
this can be done, thus ensuring that the model
complies with existing psychological theory

Anna Cox, UCL Interaction C 7

Cognitive Architectures: ACT-R
Theory of human cognition
Represents aspects of human memory,
speech, vision, audition & motor
systems
Built on top of LISP (so it helps if youre
familiar with that programming
language)

Anna Cox, UCL Interaction C 8

 Peebles & Cheng (2003)
an experiment, eye movement study and
cognitive modelling analysis to investigate the
cognitive, perceptual and motor processes
involved in a common graph reading task
using two different types of Cartesian graph
Aim: to investigate how graph users ability to
retrieve information can be affected by
presenting the same information in slightly
different types of the same class of diagram

Anna Cox, UCL Interaction C 9

 Experiment
participants presented with the value of a
given variable and required use the graph
to find the corresponding value of a
target variable,
when the value of oil is 2, what is the value of
gas?.

Anna Cox, UCL Interaction C 10

Anna Cox, UCL Interaction C 11
Results
The RT data showed that the graph
used and the type of question asked
both had a significant effect on the time
it took for participants to retrieve the
answer.

Anna Cox, UCL Interaction C 12

Anna Cox, UCL Interaction C 13
 Eyetracking data
In 62.7% of all trials (irrespective of the graph
used or question type being attempted), after
having read the question at the start of a trial,
participants redirected their visual attention to
elements of the question at least once during the
process of problem solving with the graph.

Anna Cox, UCL Interaction C 14

Hypotheses about what could be
going on
a) Ps initially encode the three question elements
but are unable to retain all of them in working
memory or retrieve them when required due to
the cognitive load involved in solving the
problem,
b) to reduce the load on working memory,
participants break the problem into two sections,
the first allowing them to reach the given
location and the second to then proceed to the
target location corresponding to the solution.

Anna Cox, UCL Interaction C 15

 What was going on?
During the process of model development it was found
that on a significant proportion of trials the model was
not able to retrieve question elements at the later stages
of the trial because their activation had fallen below
the retrieval threshold. (i.e. option a)
As a consequence new productions had to be added to
allow the model to redirect attention to the question in
order to re-encode the element and then return to
solving the problem.
This was precisely the behaviour observed in the eye
movement study.
Anna Cox, UCL Interaction C 16
The numbered circles on the model screen shot indicate the sequence of fixations
produced by the model

Anna Cox, UCL Interaction C 17