Abstract
The world of human-computer interaction (HCI)bases itself on three cyclic activ-
ities when designing interfaces: understanding users, designing, and evaluation.
As time has progressed, HCIhas grown to include other types of devices beyond
the desktop computers with a keyboard and mouse that are traditionally associ-
ated with it. Devices such as portable music players, digital cameras, and mobile
phones increasingly require methods and ideas that come from the world of HCI.
Methods for determining user needs usually are not dependent on the hardware
that will be used. However,many design metaphors that are standard on the desk-
top do not transfer to portable devices due to the special requirements of power and
input. An open question we try to answer is whether evaluation techniques that
were originally developed for desktop platforms could transfer to mobile devices?
If they can,what do they bring to mobile evaluation and are there any differences
that need to happen to these methods?We examine mobile phones specifically and
take a look at several evaluation methods.
We examine a method from the GOMSfamily of evaluation,the Keystroke-Level
Model. The Keystroke-Level Model gives the time it takes for an expert to do a
task error-free. It does this by using operators that represent keystrokes, mouse
movement and presses, the movement of hand between the two, and the the time
spent mentally preparing for an operator. While these operators are clearly linked
to the desktop there are analogs to these operators on mobile phones as well.
Creating a model for a task is straight-forward,but it has a potential for being au-
tomated. To help in this automation, we have developed a tool called KLM-Qt.
KLM-Qt is an open source tool that can examine events that are delivered to an
application and convert them into Keystroke-Level Model operators. This has the
advantage that all that needs to be done is demonstrate the application to get a
model, resulting in a savings of time. We include how the tool works along with
the details of its implementation. We also discuss changes that are made to make
it work better on mobile phones.
Besides describing the tool, we do several evaluations of some mobile phones us-
ing different methods. The methods include using KLM-Qt,usability testing,and
heuristic evaluation. All with the intent of discovering how well each method
works with mobile devices and to see what each approach can provide. We find
each method can provide insight into items that could be improved in an inter-
face. The KLM-Qt and the Keystroke-Level Model produce useful results but could
use some adjustments to help produce more accurate models,particularly when it
comes to handling input and calculating when certain operators are needed. Our
heuristic evaluation shows that the nature of the results is partially controlled by
the use selection of heuristics. Recommendations for the evaluated phones are also
reviewed.
We conclude with possible improvements that will help make the Keystroke-Level
Model method a better fit for mobile devices, and changes for KLM-Qt that will
make it a better tool. We also briefly discuss factors to keep in mind when doing
evaluation of mobile devices,regardless of the final methods that are used.