ions, impact evaluation) but also on the ecosystem concerns (provisioning costs, business models) that often prove critical to developing commercially-viable solutions.
INTRODUCTION
Ambient information systems help users stay connected to
relevant but non-critical information in a non-intrusive way.
Mobile handsets make ideal personal ambient information
appliances due to both their penetration among the global
consumer population and their rich contextual knowledge.
The always-on, always-connected nature of handsets could
conceivably enable “what’s happening” style interfaces [
The latter can be segmented further into distributed display
architectures (e.g., Eye-Q [
DESIGN PRINCIPLES
We begin by defining the dimensions of an effective mobile
interstitial display. We borrow heavily from Pousman and
Stasko’s taxonomy of ambient information systems [
Information Capacity This denotes the nature and amount of information that can be effectively depicted on an ambient display. The passive nature of these systems and the low interruption-tolerance of mobile users suggest that interstitial information systems are best used to convey non-critical, delay-tolerant content for casual consumption. Deployment experience and user studies indicate that environmental information (e.g., local weather), general news (e.g., sports, entertainment) and updates from the user’s primary social group (e.g., music and motion presence) are viewed as being appropriate for interstitial consumption. From the mobile perspective, the amount of information conveyed can be characterized by factors like the channel bandwidth (number of concurrent information channels supported as ambient interfaces), the information density (ratio of information value to message size) and hysteresis (rate of decay in perceived value of item over time).
In our experience, effective mobile experiences may be able to support channel bandwidths of 3-5 channels*, with content characterized by a high information density coupled with a low hysteresis (1-5 hours). However, future work may discover ways of conveying more information in a glanceable manner. High information density (e.g., a sports score, or a stock quote that conveys high-value information in just a few bytes) is especially important for mobile devices, given the relatively high cost of cellular bandwidth and the limited cognitive bandwidth display real-estate available to users. Emphasis on ‘hysteresis’ is also a function of the high cost of data delivery to the handset. To conserve valuable bandwidth and battery, mobile information appliances often resort to ‘cache-and-render’ models that leverage periodic bulk transfers to the mobile device (in lieu of expensive incremental or continual realtime information updates).
Notification Level
Notification levels reflect the degree of interruption that is
acceptable to users. It is typically dependent on the user’s
interest in his current task – which varies from inattention to
divided attention to more focused attention. Accordingly,
researchers [
A second aspect of representational fidelity (from a mobile perspective) focuses on nesting – i.e., allowing the user to “snack” superficially on a number of content channels but then enabling him to drill down further to obtain additional details on items of specific interest.
Aesthetic Emphasis
Aesthetics is essentially viewed as a subjective discussion
[
Clouds – can be rendered within a carousel or as a
widget. These abstractions adapt to real estate available
but focus on displaying aggregate data rather than
details [
Based on our experiences with SCREEN3’s carousel mode, we see value to a comparative study of such ‘ticker-style’ interfaces that use wipe-based transitions for ambient information visualization. Different classes of tickers can be envisioned – continuous scrolling (steady rate), discrete scrolling (employing scroll pause scroll cycles) and serial presentation (no scrolling, just item replacement). Our initial thoughts favor the use of discrete scrolling since this allows users to consume sufficient information without additional effort on their part – also giving them sufficient time to react to information by clicking-through for details. Fairness We augment Pousman and Stasko’s taxonomy with the dimension of fairness, as a way for passive interfaces to support a larger information capacity without active user navigation. It’s quite common in channel-oriented mobile systems (e.g. SCREEN3 below) for the number of channels to exceed the display capacity† of the device, and for the number of items per channel to exceed what can be shown to the user at one time. The definition of fairness for ambient information systems parallels its usage in distributed systems – namely that every ‘channelized’ information item will have fair access to “face-time” with the user – even if the user does not actively navigate to it directly. The carousel model adopted by SCREEN3 is a good example of a fair ambient information system with a non-weighted, round-robin selection scheme. However, one can imagine a number of alternative ambient interfaces that support fairness. We note that fairness becomes particularly important in commercial systems where many third-party providers of such ambient content have a vested interest in having their content seen by the user at some point. This is different from the case where users elected to receive specific content of interest – i.e., the user knows the content exists and can navigate to it if desired, while in the earlier case users remained unaware of its existence unless they ‘stumbled’ upon it by accident.
Privacy Concerns
The mobile phone is a deeply personal device, thus
naturally privacy concerns arise when talking about a
medium on which to display contextualized and possibly
personalized information that may be viewable by others.
As much as ambient visualization is a central means for
user awareness, the fact that in certain situations, it may be
publicly viewable raises concerns about how the
information needs to be presented. A lot of this concern is
with protecting the user from undesirable situations [
The SCREEN3 model’s emphasis has been on providing users with a zero-click, lean-back experience for lazy content consumption – with the ability to transition to a more lean-forward, interactive experience as desired. These different degrees of user engagement are supported by a “bite–snack–meal” approach to information delivery as shown in the figure. The “bite” contains headline-quality information for content items, enabling users to consume it at a glance (passive awareness). For items of interest, users can easily transition to a lean-forward (more interactive) experience by clicking through to receive a “snack” – typically a cached extended summary for the headline article. For more information, users can then click through the snack to request a complete “meal” – typically a link to a network repository containing the complete article with richer media attachments that the user can browse online or download for consumption.
SCREEN3 currently adopts a carousel model for displaying information bites – the carousel can be navigated manually (to enable scrolling through the channels, or through items within a channel) but is typically animated, automatically scrolling through channels and items in sequence for a true zero-click consumption experience. The server caches client state in order to decide what updates need to be delivered to the mobile device. However, the content being displayed remains still fairly static in terms of both its applicability to the user (i.e., it may be a channel that the user subscribed to in the past, but hasn’t actively in recent times) and its visualization on the device.
EFFICACY MEASURES FOR INTERSTITIAL SYSTEMS User interactions convert “ambience” into “intent”. This tipping point is of value from a business perspective, as it bridges content and commerce. Where passive viewing a news item represents casual interest in something, actively interacting with it might indicate sufficient interest in the topic or item to merchandise related goods and services. Identifying user interests can also provide personalization of ambient information. This improves not only system efficiency (ability to prioritize fetching and caching of content that is likely to be of more interest to this user) but also the hit-rate (user click-through) for interstitial consumption of related content.
The latter is particularly important on mobile devices.
Display constraints limit the information capacity, allowing
only a few items to become visible in any limited time
window. Further, interstitial consumption patterns imply
that the face-time afforded to ambient information systems
is usually limited to short ‘windows of opportunity’ in
between other user tasks. Thus, items must now compete
for the user’s attention within a given opportunity. Item
selection becomes key to either holding the users’ interest
(thus providing opportunities for other items to be shown)
or losing it (thus ensuring that he or she remains unaware of
the existence of items that are relevant and interesting).
So how can we measure the impact of ambient information
systems? We propose various measures that reflect different
degrees of user engagement and for different rendering
abstractions (e.g., carousel-based, widget-based)
• Attention Measures. Identifies the minimal level of
engagement with ambient information. E.g., time spent by
user in scrolling through channels, or time for which an
item was in focus with user present at device. While the
notion of an attention measure could apply equally well
to PCs and TVs, the fact that the phone is a personal
device on which content consumption is a deliberate
decision, is likely to provide “clean”, high quality
attention data. As shared devices, both the PC and TV
suffer from the “who’s watching” drawback (according to
an industry statistic, over 50% of the time, the TV is on
with nobody watching). Simple context enablers can
disambiguate mobile-in-the-pocket and
mobile-to-one’sear situations, and allow more accurate measurements of
content viewing on the handset.
• Action Measures. Identifies a higher level of user
engagement related to a specific item or channel,
particularly since the user is potentially aware of possible
delays in fulfillment such as for network downloads. E.g.,
user click-through (bite snack meal) or hide reveal
transitions for specific widgets [
Transaction measures are more interesting from a business perspective since they translate more directly to commerce. However, they are also a more difficult measure to evaluate since the correlation between point-of-viewing (on ambient interface) may be temporally or spatially distant from the point-of-purchase (e.g., at a later time, on a potentially different application). Correlations may be simplified in cases where the transaction is driven directly off the ambient interface (e.g., via menu actions).
More complex solutions can involve correlating short-term activity history to long-term monitored user behaviors. For instance, short-term history can link an ambient display item to the user (e.g., click-through captured showing user viewed extended information about concert on the ambient display). The long-term observations (across devices and domains over a longer period of time) can be analyzed to infer that a subsequent user activity was influenced by this recently viewed item – e.g., a weather item indicating rain in the forecast was viewed some time before the purchase of an umbrella was recorded on the user’s credit card. In general, we view handsets as ideal devices for gathering the raw data required for deriving such metrics. It inherently provides a source for fresh, personalized user data. QUESTION FOR DISCUSSION This discussion was motivated by a single but multi-faceted question: What differentiates the design of a mobile ambient information system that emphasizes interstitial consumption of content? We believe these systems are useful, are viable (given existing technologies) and are of commercial interest (both for differentiating devices with an enhanced user experience, and in creating opportunities that convert ambience to action and ultimately to commerce). With this paper, our goal is to initiate deeper discussions on design issues and key applications for such systems. ACKNOWLEDGMENTS We thank the various members of Motorola Labs and the SCREEN3 team who helped us gain a better understanding of the operation and utility of this interface on the handset.