Proximity Based Adaptation of Content to Groups of Viewers of Public Displays

International Journal of Ubiquitous Computing (IJU), Vol.9, No.1/2, April 2018
PROXIMITY-BASED ADAPTATION OF CONTENT TO
GROUPS OF VIEWERS OF PUBLIC DISPLAYS
Amir E. Sarabadani Tafreshi, Kim Marbach, and Gerhard Tröster
ETH Zürich University, CH-8092 Zürich, Switzerland
ABSTRACT
Responsive design adapts web content to different viewing contexts to deliver an optimal
viewing and interaction experience. Recent work proposed a model and framework for
proximity-based adaptation of web content as a new dimension for responsive web design.
While it was shown that the model improves the perception and user engagement for single
viewers, until now, the effect had not been investigated for multiple simultaneous viewers
who may be at different distances from the display. In this paper, we report on an initial
study that evaluated and compared the effects of using the average distance of viewers as
the basis for handling adaptation of content to multiple viewers with a classic one that
adapts content based only on display characteristics. Our results show that the adaptive
model provides a better view of the content and improves user engagement, but can be
confusing when serving multiple viewers.
KEYWORDS
Distance; large displays; Responsive design; multiple viewers.
1 Introduction
Despite the fact that viewers of public displays perceive the content of a display at dif-
ferent sizes according to their distance from the display, current responsive web designs
(RWD) only adapt web content based on the characteristics of the device and browser.
Therefore, in a recent work, Tafreshi et al. [1] proposed a model that extends current
responsive design techniques to take user proximity into account as part of the viewing
context for public displays. However, one of the open issues and major challenges is con-
sidering how to handle the fact that public displays typically have multiple simultaneous
viewers who may be at different distances from the display.
The proposed JavaScript framework, ResponDis, was designed to support experimen-
tation with variants of proximity-based adaptation model in both single and multi-viewer
contexts. However, until now only single-user scenarios had been investigated. In this
paper, we report on our initial investigations of multi-viewer contexts. Following one
of the suggested methods by Tafreshi et al. [1], we use the average viewer distance of
viewers to calculate the proximity of a group of users to a display. Our hypothesis is that
the average-based method that considers the proximity of all viewers would improve the
viewers’ perception and user engagement in general. A user study with 24 participants
was carried out to compare the resulting dynamic proximity-based adaptation of screen
content with an approach that performs a static adaptation of content based solely on the
DOI : 10.5121/iju.2018.9201 1

display characteristics. Our findings show that the proximity-based adaptive model im-
proved the user engagement and provided a better view of the content that avoided users
having to move close to the display. However, we observed that the adaptive approach
resulted in some confusion.
2 Related Work
HTML5 and CSS3 provide features that facilitate responsive design in which a website
adapts to the characteristics of a device or browser. Recent proposals for CSS4 media
queries take support for responsive design further by catering to other characteristics of the
viewing context, for example adjusting the design depending on the luminosity level [2].
Recent works [1,3] add another dimensions by also taking into account the distance and
walking-speed of the viewer as important factors for responsive web design, especially in
the case of public displays.
In all of these cases, each of the factors taken into account is associated with a single
input. For example, the size of the browser, the level of ambient light, or the distance of
the viewer to the display. However, since public displays can often have multiple view-
ers, there could be multiple inputs for viewer distance. Proximity-based responsive web
designs therefore need to somehow mediate in the case of multiple viewers. Strategies
previously proposed or envisaged in the literature to handle such situations can be classi-
fied into three approaches: (i) split-screen, (ii) selecting one target viewer out of a group,
and (iii) mixed.
A split-screen approach [4,5] splits the screen into multiple regions and allocates each
of them to a different viewer. Clearly, increasing the number of viewers leads to these
regions having smaller sizes which can negatively affect the perception of the viewers to
the content and also distract them. At the same time, it can waste screen space if regions
end up showing the same information.
The second approach has to decide on a strategy for selecting one of the viewers as a
target, thereby ignoring the others and adapting the content based on the distance of the
target viewer to the display [6]. One possibility would be to take the user first detected as
the target, while another is to select the viewer closest to the display.
The mixed approach [6] combines the distance of all the viewers and represents them
as a single parameter for design decisions. For example, the average distance of the
viewers could be used. However, it is not clear whether such an approach improves the
viewers’ experience, so we chose to investigate this.
3 Extension of the Model for Multiple Viewers
The proximity-based adaptation model [1] provides the optimal content resolution (CVR)
for a viewer who is in front of a display. The calculation of optimal content resolution is
not only based on the viewer’s distance to the display (VD) and the display’s diagonal size
(DS) in inches but also on the display’s native horizontal and vertical resolution in pixels,
denoted NHR and NVR, respectively. The equation that considers all of these factors to
calculate the CVR is equal to:
2

Figure 1: A sample use of the average proximity method to serve multiple viewers
CVR =
DS
q
((NHR
NVR )2 +1)×VD×tan( 1
60 × π
180)
(1)
Similar to the use of screen size in current responsive design methods, designers can
use the calculated CVR as a parameter for defining a set of media queries where each
media query will correspond to a distance range that effectively defines a zone in front of
the display. This means that the zones in which users are standing as well as their actual
distances from the display can be used to adapt content to multiple viewers.
4 User Study
We conducted a user study in a controlled lab setting to evaluate the proposed adaptation
model with groups of people. The goal of our experiment was two-fold. First, we wanted
to examine whether the proposed model improves the user engagement, usability and
viewer perception in the case of multiple viewers. Second, we aimed to evaluate how
the model would compare in a multi-viewer setting with current methods used to adapt
content based on display characteristics. Note that a key difference in these two methods
is that the former performs a dynamic adaptation as viewers move, while the latter is a
fixed adaptation. Figure 1 illustrates the sample use of the average proximity method to
adapt the content to groups of viewers. For our initial study, we decided to use the average
proximity method since it seems the fairest in terms of giving the same consideration to
all viewers.
4.1 Participants
We recruited 24 participants (8 females; 20–41 (median: 23) years) for the user study.
The participants were arranged in groups of three (8 groups). All of the participants had
normal or corrected-to-normal eyesight.
3

4.2 Method and Procedure
For each group of participants, we introduced the system and the goal of the study. We
also asked their consent to record the experiment using a video camera. The experiment
task was then summarised.
The task was to quickly find a specific character in a wimmelbook [7] picture, choos-
ing characters and pictures that are well known to many people. We carefully adjusted
and fitted the characters to be found so that they integrated well with other characters in
the picture. Our design decision on the size of the picture for the static UI was to provide
a fairly good view for both the furthest and closest viewers. Therefore, we adjusted the
size of the picture to 50% of the entire screen size characteristic. For the adaptive UI,
we designed the UI to resize the presentation of the picture for four zones, so that a user
standing in each zone had a close to optimal view of the picture.
Since our goal was to compare the adaptive and static approaches, we decided to keep
the type of adaptation as simple as possible so that the focus of the study was on the
approaches and not the adaptation strategy itself.
Each group performed the tasks using the two different interfaces. Each participant
in the group was equipped with a stopwatch. We showed a photo of a character and,
after counting from one to three, the participants started their stopwatches and could walk
around freely, while searching for the character in the picture. Once a user had found
the character, they had to press the stop button and stop moving until everyone had found
the character. This allowed us to identify and record the zones in which they found the
character as well as how long it took each of them to find it. This procedure was repeated
five times for each approach using different pictures (total: 10 conditions). The design of
the studies was counterbalanced in such a way that any group was equally likely to start
with one or other UI. Furthermore, the content orders in each case were randomised.
When participants had finished the tasks, they were allowed to freely move around and
test the two approaches. Then, each participant filled out a questionnaire and answered
some semi-structured questions about their experiences. The first part of the questionnaire
consisted of questions regarding demographic information and the visual acuity of the
participant. This was followed by Usability Scale (SUS) questions [8] and then a series of
questions assessing user engagement. At the end, the participants also provided an overall
rating on a 10-point Likert-scale for each approach.
10 questions, each on a 5 point Likert scale, were included in the SUS, allowing us to
calculate a single measure of usability within the 0–100 range. We considered the SUS
score as the main factor for evaluating the usability of both approaches.
To evaluate the user engagement, we used O’Brien and Toms’ [9] user engagement
scale (UES) which integrates a variety of user engagement factors and includes six dif-
ferent dimensions including Aesthetic Appeal, Endurability, Felt Involvement, Focused
Attention, Novelty, and Perceived Usability. Endurability evaluates the overall success of
the system, willingness of recommendation, and whether the viewer would use the sys-
tem again. Aesthetic Appeal measures the visual quality of the system. Felt Involvement
describes how involved the user felt with the system. Focused Attention describes to what
degree the system attains the user’s full attention. Novelty measures the viewer’s curiosity
and degree of interest. Perceived Usability reflects the user’s satisfaction with the system.
By considering different dimensions, we could check from different perspectives,
which approach enhances the user engagement more [9]. To evaluate the six-dimensions
4

of the UES, we had eleven questions that were also used to evaluate the model with single
viewers [1]. The questions were on a 5-point Likert-scale and we evaluated each dimen-
sion by summing the received scores of the corresponding questions.
To compare the attributes of the static and adaptive approaches, we used related-
samples Wilcoxon signed rank tests. Moreover, when the assumptions were met (i.e.
no normality violation as tested by Shapiro-Wilk test (p>0.05), etc. ), we used repeated
measure ANOVA. We considered p = 0.05 as the minimum significance level.
We conducted our experiments using a 27” LED display, configured in landscape
mode (see Fig. 2). The ResponDis framework was configured based on the display infor-
mation, i.e. DS=27, NHR = 1920, and NVR = 1080. For our experiment, we used the
default setting “averageProximity” to adapt the UI. We also used one Kinect which can
simultaneously track a maximum of six people.
4.3 Results
4.3.1 Viewer Perception
The average time measurements of the repeated experiments for five different pictures,
showed a trend (Z = 1.914, p = 0.056) toward less time required to find the charac-
ter using the adaptive approach (Median = 4.5870s) compared to the static approach
(Median = 6.1390s). This result suggests that the adaptive method improved the par-
ticipants’ perception of the content by 25.28%.
Moreover, the ending zones where the participants ended up finding the character
for both static (Median = 2.2) and adaptive (Median = 3) approaches, were statistically
significantly different (Z=-3.617,p<0.001). Therefore, using the adaptive approach par-
Figure 2: The study setup (picture published earlier in Tafreshi et al. [1]).
5

Factors of Engage-
ment
Median Z-value p-Value Conclusion
Aestethic Appeal
Adaptive: 4, Static: 3 Z = -1.624 p = 0.104 n.s
Endurability
Adaptive: 7, Static: 6 Z = -1.844 p = 0.065 n.s
Felt Involvement
Adaptive: 8, Static: 7 Z = -2.777 p = 0.005 3
Focused Attention
Adaptive: 7, Static: 8 Z = 0.253 p = 0.800 n.s
Novelty
Adaptive: 8, Static:
3.5
Z = -3.777 p < 0.001 3
Perceived Usability
Adaptive: 4.5, Static:
4
Z = -0.986 p = 0.324 n.s
Table 1: Comparison of different dimensions of user engagement for adaptive and static
approaches. 3: the adaptive approach performs better; n.s.: no difference was found.
ticipants needed to move less to adapt themselves to the picture on the display compared
to the static approach.
4.3.2 User Engagement
The results of the statistical analysis to evaluate the difference between the approaches on
different dimensions of the user engagement are presented in Table 1. The comparison
column indicates which approach had a statistically significantly higher value, or whether
there was no statistically significant difference. The sign 3 mark in the conclusion column
shows significant findings with p ≤ 0.05 if the adaptive approach performed better than
the static model.n.s. marks non-significant differences. There was no case where the static
approach was superior.
4.3.3 Usability and Overall rating
There was no statistically significant difference between the SUS score of the adaptive
(Median = 75) and static (Median = 77.5) approaches, Z = 1.333, p = 0.182. Further,
the overall rating on a 10-point Likert scale showed no statistically significant differ-
ence between the two approaches, Z = −0.185, p = 0.853. However, the adaptive ap-
proach achieved a relatively higher score (Median = 7) compared to the static approach
(Median = 6).
4.3.4 User Feedback
Participants considered the static UI as the “state of the art, nothing new”. (P 1.3). An-
other participant noted: “It was just a static picture. There is nothing special about that.”
(P 3.3). They also highlighted one of the issues of static UI design: “This is how we know
it from everywhere. I just have to go very close to see the small things on it.” (P 2.1). “The
pictures were just small, I do not see the point. ...” (P 3.2).
6

Using the adaptive design, some “... didnt understand what it adapts to” (P 6.1). One
participant wrote: “At first, I did not understand that the adaptive display was changing
the screen size in line with the distance. I was far from the screen. I thought it was
randomly re-sizing to make the task more difficult. This was probably because we were
multiple users, and the screen was trying to adapt to all of us at the same time. In the end
I found it confusing why it was re-sizing.” (P 8.3). Many would have preferred a smoother
transition between the zones: ”The state changes are too coarse. ...” (P 3.2) and many
suggested: “some kind of smoothed transition and/or a way to prevent too fast switching
between two states should be implemented for a calmer user experience”. (P 2.3).
Our adaptive approach was completely new to users and therefore a few of them were
confused, irritated and looking for a familiar feature: “it was frustrating to see the image
get smaller” (p 8.1). The abrupt transition “... distracts from whatever search pattern I
was on. I had to recollect myself for a moment and find out where I was again.” (P 8.2).
However, other participants found that “... the adaptive display is very helpful for people
who do not see very well.” (P 2.1). In addition, they found that the images “... were easier
to see because of the state changes ...” (P 3.2). Some future work was also suggested: “...
it would be interesting to see the difference it makes while trying to read something.” (P
2.3). Another participant noted that: “I think the time to find the character also depends
on the user’s familiarity with the character. I.e. because I am familiar with Pink Panther,
it was easy for me to spot it.” (P 6.2). Overall the adaptive approach was preferred: “... I
would really prefer the adaptive approach. It would help for example at the tram station.”
(P 2.1).
Choosing an appropriate method for groups of viewers was found challenging and re-
quires further investigation as one participant mentioned:“I feel that the adaptive display
would be better suited for an individual, not a group.”. (P 5.1). One participant also re-
ported one of the difficulties in decision making for serving multiple viewers: “The only
drawback of the adaptive approach in my opinion is the flickering due to multiple people
interacting and the fact that 2 people, one standing far away, the second one close result
in a state where one of those cannot see what he should.” (7.3).
5 Discussion
We observed a trend toward requiring less time to find the character using the adaptive
approach. While the insignificant improvement could be due to the approach, it might
also be a result of the observed behaviour of the participants: Some participants were
eager to find the character as fast as possible, and members of a group started competing
against each other. In contrast, other groups walked around slowly, trying to not disturb
one another.
The analysis of the zones where the participants ended up finding the character proved
that, using the adaptive approach, participants did not have to walk as close to the display.
This can be attributed to the fact that the display adapted itself to the viewers, rather
than the participants having to adapt their position to the display. Therefore, the adaptive
approach could be more effective in a public setting, as it requires less effort from the
viewers.
The issue of engaging the viewers of pervasive display systems is a well-known prob-
lem [10–12]. We observed a systematic difference in favour of the adaptive approach on
7

the Felt Involvement and Novelty factors of user engagement. The higher level of felt in-
volvement is likely due to the adaptation of the UI according to the position of the viewer.
The difference in the novelty can of course be accredited to the fact that current UI de-
signs use the static approach and participants had not previously encountered dynamic,
proximity-based adaptation.
Furthermore, viewers perceived using the adaptive approach to be more worthwhile.
However, it was observed that the adaptive approach could also be more confusing. There-
fore, the “average distance” strategy is probably not the best and other methods need to be
explored for serving multiple viewers. Reviewing the feedback and recorded videos, the
main confusion seemed to be due to the display UI flickering when a participant moved
forward and backward asymmetrically. Viewers also did not expect the abrupt change of
the UI. Some participants suggested that there should be a smoother transition between
the states. Also, it was clear that some participants did not realise how the UI actually
adapted to multiple viewers. Nonetheless, we found no systematic difference between the
approaches about the feeling of being annoyed.
The usability scores of both approaches were above average but not statistically sig-
nificantly different from each other. Therefore, using either of the approaches, there is no
risk of a low usability effect leading to low user engagement.
Although, the adaptive approach achieved a higher overall rating and provided viewers
with a better view, we observed no statistically significant difference between the overall
rating of the approaches.
6 Conclusion and Future Work
Unlike modern personal devices that have touch input [13], some larger displays are out
of reach or support no direct input on the display surface. We have presented a user
study that considers the average proximity of viewers as an input to handle proximity-
based adaptation of content to multiple viewers of public displays. In our study, we used
the average proximity of viewers as an input to a previously proposed proximity-based
adaptation model that integrates viewer distance proximity as an additional dimension for
responsive web design. Although the results of our study with multiple viewers showed
an enhancement in favour of the adaptive method, the improvement achieved was not
as significant as that found in the previous single-viewer user study [1]. Therefore, we
plan to investigate alternative methods and refinements for proximity-based adaptation in
multiple viewer settings in future work.
References
[1] Amir Esmaeil Sarabadani Tafreshi, Kim Marbach, and Moira C. Norrie. Proximity-
Based Adaptation of Web Content on Public Displays. In International Confer-
ence on Web Engineering (ICWE), pages 282–301. Springer International Publish-
ing, 2017. DOI: 10.1007/978-3-319-60131-116.
[2] Ben Frain. Responsive Web Design with HTML5 and CSS3. Packt Publishing Ltd,
2015.
8

[3] Amir E. Sarabadani Tafreshi, Adrian Wicki, and Gerhard Tröster. RDSpeed: De-
velopment Framework for Speed-Based Adaptation of Web Content on Public Dis-
plays. In 26th International Conference in Central Europe on Computer Graphics,
Visualization and Computer Vision (WSCG). WSCG, 2018.
[4] Till Ballendat, Nicolai Marquardt, and Saul Greenberg. Proxemic Interaction: De-
signing for a Proximity and Orientation-aware Environment. In ACM International
Conference on Interactive Tabletops and Surfaces (ITS), pages 121–130. ACM,
2010. DOI: 10.1145/1936652.1936676.
[5] Jakub Dostal, Uta Hinrichs, Per Ola Kristensson, and Aaron Quigley. SpiderEyes:
Designing Attention- and Proximity-aware Collaborative Interfaces for Wall-sized
Displays. In Proceedings of the 19th International Conference on Intelligent User
Interfaces, IUI ’14, pages 143–152, 2014. DOI: 10.1145/2557500.2557541.
[6] Miaosen Wang, Sebastian Boring, and Saul Greenberg. Proxemic peddler: A pub-
lic advertising display that captures and preserves the attention of a passerby. In
Proceedings of the 2012 International Symposium on Pervasive Displays (PerDis),
pages 3:1–3:6. ACM, 2012. DOI: 10.1145/2307798.2307801.
[7] Cornelia Rémi. Reading as playing. Emergent Literacy: Children’s books from 0 to
3, 13, 2011.
[8] Jeff Sauro. A Practical Guide to the System Usability Scale: Background, Bench-
marks & Best Practices. Measuring Usability LLC, 2011.
[9] Heather L OBrien and Elaine G Toms. Examining the Generalizability of the User
Engagement Scale (UES) in Exploratory Search. Information Processing & Man-
agement, 49(5):1092–1107, 2013.
[10] Amir E. Sarabadani Tafreshi and Moira C. Norrie. Screenpress: A powerful and
flexible platform for networked pervasive display systems. In Proceedings of the 6th
ACM International Symposium on Pervasive Displays (PerDis), pages 13:1–13:8.
ACM, 2017. DOI: 10.1145/3078810.3078813.
[11] Amir E. Sarabadani Tafreshi, Milan Bombsch, and Gerhard Tröster. Chained
Displays: Configuration of Multiple Co-Located Public Display. Interna-
tional Journal of Computer Networks & Communications (IJCNC), 10(3), 2018.
DOI: 10.5121/ijcnc.2018.10301.
[12] Amir E. Sarabadani Tafreshi, Andrea Soro, and Gerhard Tröster. Automatic, Gestu-
ral, Voice, Positional, or Cross-Device Interaction? Comparing Interaction Methods
to Indicate Topics of Interest to Public Displays. In Frontiers in ICT. Frontiers, 2018.
[13] Amir E. Sarabadani Tafreshi, Sara C. Sarabadani Tafreshi, and
Amirehsan Sarabadani Tafreshi. Tiltpass: Using device tilts as an authen-
tication method. In Proceedings of the 2017 ACM International Confer-
ence on Interactive Surfaces and Spaces (ISS), pages 378–383. ACM, 2017.
DOI: 10.1145/3132272.3134112.
9

Proximity Based Adaptation of Content to Groups of Viewers of Public Displays

More Related Content

Similar to Proximity Based Adaptation of Content to Groups of Viewers of Public Displays (20)

More from ijujournal (20)

Recently uploaded (20)

Proximity Based Adaptation of Content to Groups of Viewers of Public Displays