|
|
| |
|
Web
Page Viewing Behavior
This
research explores the determinants of eye movement behavior on
a single web page. We sought to understand whether viewing behaviors
are determined by individual differences between subjects (including
demographic variables and familiarity with different websites),
different types of websites, the order of pages bieing viewed,
or tasks at hand. Our results indicate that the gender of web
viewers, web page viewing order, and the interaction between page
order and site type influences online ocular behavior. Task instruction
and background knowledge in terms of familiarity with the websites
do not significantly affect web viewing behavior.
Correlates of Visual Disconnect
This paper analyzes three characteristics inherent to
webpage elements -- location, size, and information density --
to determine the likelihood that each of these qualities will
influence a viewer's pattern of attention. We conducted an analysis
of eye movements across three commerical home pages, and found
that location was most influential in directing attention-based
eye movements, highlighting the importance of selectively allocating
screen real-estate. Furthermore, we present the concept of 'visual
disconnect' to reveal specific instances in which the aforementioned
properties fail to consistently influence user attention.
AdTrak
This study uses web-based online advertising as a medium through
which to understand eye movements and the process of selective
attention. We explore how a number of different components, including
size, location, and ad type affect and individuals likelihood
of recognizing, responding to, and remembering an online advertisement.
Recommendations and suggestions for the future of online marketing
and the structure of online advertisements are proposed.
Online Decision-Making
This study uses an online search interface as a medium
through which to explore how ocular behavior can offer insight
into detecting the decision-making process involved in document
selection. Existing research in decision-making shows that there
are three stages to this process -- orientation, evaluation, and
verification. We seek to verify whether this process also holds
constant in the decision-making of online document selection.
|
| |
Search
Behavior, Trust, and Decision Making
We
have conducted studies with users engaged in different kinds of
search tasks using Google as their search engine. Looking specifically
at how searchers make decisions regarding the relevancy of the
returned search results we have been able to assess global ocular
behavior involved in how users make their ultimate decisions.
Results thus far indicate that 96% of the time users make a decision
based on the first page of returned results. Furthermore, users
in general look only at the first two ranked abstract, but chose
the first abstract most of the time. After the second abstract,
fixations drop significantly, and very few users ever look at
abstracts beyond the page break. If the first two abstracts are
not deemed relevant, users typically will skip and jump to other
abstracts and then regress to review the first two abstract again.
These
results have since been replicated and extended. Specifically,
we have looked at differences in ocular behavior as a function
of gender and task, as well as what happens when we manipulate
the perceived ranking of the abstracts returned. With regard to
the former, significant differences did emerge between men and
women, particularly with respect to the number of abstracts viewed,
and the linearity of the scan patterns prior to decision making.
Men looked at more abstracts and pages of abstracts than women,
and displayed fewer instances of regressing back to previously
viewed abstracts. Not surprisingly, for conceptually more difficult
tasks, all ocular indices were significantly greater.
Unbeknownst to some users, we also manipulated the relevancy of
the returned results to some users. These searchers received the
top ten abstracts in reverse order, such that Google’s number
1 ranked abstract was displayed as the 10th ranked abstract, the
second ranked abstract by Google was displayed as the 9th ranked
abstract, and so on. The results of this study indicated that
users in this condition still chose the top two displayed abstracts
as often as those users who received Google’s actual ranked
results. Furthermore, and perhaps more interesting, their ocular
behavior indicated that on some implicit level, users who viewed
reverse-ordered abstracts knew that something was awry even though
their decision making was not different from controls, nor could
they explicitly articulate any knowledge of the rouse after the
experiment was over. Particularly, these users spent more time
on each abstract page, had more regressions to earlier abstracts,
spent more time per page, and looked at more abstracts. In addition,
users in the reversed condition had more fixations on the 9th
and 10th abstract (the actual top two abstracts returned by Google),
and when pupil dilation was compared within this group for the
1st and 2nd displayed abstract and the 9th and 10th displayed
abstract, there were no significant differences. Thus, users demonstrate
dissociation between their ocular behavior and their ultimate
decision, indicated a greater trust in Google’s ranked display
than their own evaluation.
Current
studies are extending the generalizability of these results by
including different search engines, more tasks, and great diversity
amongst the users. |
Average Pupil
Dilation Across Abstracts for Reversed Condition |
| |
The
Average Scan Pattern and What It Might Tell Us
Ocular
data provides numerous indices that have been used to make inferences
regarding attention, preference, arousal, response to novelty,
task difficulty etc. While fixation data and saccadic indices
have been employed as individual measures, less frequent analysis
has been attempted on the sequence or pattern data that these
two indices together create. In the current work we explore a
method for deriving an “average” scan pattern aggregated
from many users viewing the same visual stimulus (web site). To
do this we us a multiple sequence alignment algorithm to extract
similarities among multiple patterns. The end result is a single
scan pattern that can represent the behavior of an entire group.
This method has been assessed visually and numerically, and has
fared well on in both cases. We are developing experiments currently
to test the validity of this method more systematically. |
| |
We envision a measure of an average scan pattern to be fruitful
in both theoretical and applied domains. For example, it seems
reasonable that once such this method has been validated we
could use it to discriminate between different users under different
circumstances, extending to possibilities that the scan pattern
might but useful as an input parameter to search engine retrieval
algorithms. Similarly, it seems within the realm of possibility
that scan patterns of those particularly proficient at a task
(e.g. information search, medical diagnosis, search and rescue)
could be used to train novices on the same task.
At
a theoretical level we are interested in pursuing this pattern
data from a network perspective, specifically from a network
motif angle in trying to discover their underlying structural
properties. Network motifs are “…recurring, significant
patterns of interconnections” [Milo, et al. 2002]. We
imagine that pattern motifs within the average scan pattern
could be used as blueprints for discriminating between different
kinds of cognitive processing.
|
| |
|
|
