Social interactions among a group of friends will typically have a certain recurring rhythm. Most people interact with their own circle of friends at a range of different rates, and through a range of different modalities (by phone, instant messaging, face-to-face meetings etc.). When these naturally recurring interactions are maintained effectively, people feel at ease with the quality and stability of their social network. Conversely, when a person has not interacted with one of their friends for a longer period than they usually do, a problem can be identified in that relationship which may require action to fix. We present Stay-in-touch, an ambient information system which provides peripheral cues to the user which serve as occasional recommendations of which of their friends they should contact soon in order to keep their social network in a healthy state.
Author Keywords Ambient displays, reminder systems, information visualisation, social networks analysis.
INTRODUCTION
When modelling the social interactions among a group of
friends, a certain recurring rhythm is identified, which will
differ between social groups. Within this group, a single
person may have a range of different rhythms with each of their
friends, due to the similarity of their schedules, the differing
strengths of those friendships, and a range of other social
factors. When these rhythms are maintained well — that is,
the person interacts with that friend at the regularity that they
normally do — the health of that friendship will feel
natural. If on the other hand the friendship falls out of rhythm,
through neglect or unfortunate circumstance, and the two
people do not see each other or otherwise interact, this gap
will be felt, though perhaps not always understood.
A person’s “social rhythm” describes the rate and regularity
with which they interact with the various people they know.
In this regard it is a fuzzy metric; if asked how often you
interact with a certain friend of yours, you will probably reply
with “about twice a week” or “most days”, not something
like “once every 37 hours”. Stay-in-touch is an application we
have developed to notify a user when it detects a lull in one
of their social connections. It does so by analysing their
social interaction history through data gathered from a multitude
of sources, and attempting to quantify their social rhythm.
A person’s ability to regulate their own social rhythm relies
on their perception of time running like clockwork, but the
human mind’s perception of the passage of time is capricious
at best [
Stay-in-touch provides the observer with suggestions of
actions they can take to maintain the stability of their social
network through a visual interface. This encourages users to
contact their friends regularly, but also helps them to
identify problems with certain friends early, so that they can take
steps to correct a deviation before it becomes more
pronounced. Thus, if a user tends their network well, they will have
stronger ties with a wider and more diverse set of friends.
We are particularly interested in studying the effectiveness
of these social reminders in the context of calm computing,
as defined by Weiser [
In the next section we will describe some social network analysis research that is relevant to this project, followed by a discussion of the the design of the system we have built. Next we discuss the applicability of this type of data to ambient information systems and describe how we could evaluate this sort of system.
SOCIAL NETWORK ANALYSIS
Previous studies have analysed social rhythms in
sociotechnical systems, although the focus of these studies was
on the general trends of social rhythms apparent on a
large scale. Golder et al. studied interactions between
college students over the social networking site Facebook, and
found that students’ social calendars were heavily influenced
by their school schedule [
Online social networking sites are generally used to
maintain social connections which were originally forged
offline [
That said, a person’s social network cannot be described by data from any one source. Though the majority of a user’s friends may indeed be present in an online social networking website, they will also have friends that they interact with purely offline, or mostly by phone or email. These ongoing social interactions are equally valid in characterising a user’s circle of friends.
VISUALISING SOCIAL INTERACTIONS
There have been many visualisations generated of social
networks, particularly since the rise of social networking
websites and the rich data sets they present. Many visualisations
use a familiar node-link diagram of a graph [
Data Sources Ambient systems can leverage the vast amounts of data available from the physical and virtual worlds. We now leave digital traces of most of our social interactions: all of our email is archived on a server somewhere, our instant messages are logged locally and remotely, posts to social network profiles are publicly visible, and so on. Even co-location data can be recorded if the users both wear a tracking device of some kind, allowing the identification of events like two people conversing in a research lab.
Though all of this data is attractive, for this first version of the Stay-in-touch system, we decided to focus on records of mobile phone interactions, which we are able to access from our telecommunication provider’s website. The software has been built to be agnostic to the nature of the interactions, so adding support for emails in future, for example, is a matter of writing a small client to parse the user’s inbox and find mails that they have sent or received from their friends. These, along with other discrete interactions, can then be entered into the system.
The visual display
Our visualisation is built using Processing [
Stay-in-touch presents a time series plot, visible in Figure 1. Each row represents a person’s social interactions with one of their friends, showing mobile phone interactions; blue dots indicate phone calls, with the size of the dot reflecting the length of the call, while red dots indicate text messages and are uniformly-sized. Weeks are subtly delineated by differing background colours to provide users with an indication of their longer-term habits at a glance.
The current day is highlighted, and the next week is visible on the right of the display. Cues for future interactions are displayed in this area in the form of hollow circles. Their colour and size indicate the type of interaction suggested, based on a prediction algorithm that we have written for this purpose. Predicted social interactions are drawn on the day that our algorithm has calculated to be most likely for them to occur, but the user can see them a week in advance. This gives the user several opportunities to act on the information being presented to them at an appropriate time.
If the user does not interact with their friend in any way
before the suggested interaction , an “X” is marked at this
position and this is counted as a “miss”. The prominent marking
of these events (or non-events, if you will) serve to draw the
user’s attention to these more critical cues.
AT ODDS WITH AMBIENCE
Neely et al. have previously explored their hypothesis that
some context sources are more applicable to being
presented in an ambient manner than others [
We add to these three properties of the system’s primary
output which we think make it a good candidate for application
in an ambient display — the reminders are passive, temporal
and simple. Passive means that changes in the information
do not always require immediate attention; users can take
note of reminders but choose not to act on them until later.
Temporal means that the data changes over time; if the
display remains at the periphery of a person’s attention, they
can monitor for changes while concentrating on other
activities. Simple means that the information can be digested
easily; at a basic level, a reminder simply consists of the name
of a friend who they should contact soon. Other information
may be present, such as a suggested contact time or medium,
but this only serves to augment the primary information.
These three properties correspond well to the
interaction, reaction and comprehension model proposed by
McCrickard et al. [
This example highlights the fact that only certain classes
of notifications suit application as an ambient display. This
leads us to our key question: to what degree can reminders
and notifications be incorporated into ambient systems,
before the notion of calm computing is rendered meaningless?
FUTURE APPLICATIONS
The implementation described above could be used as both
an interactive display, where a user filters the information
processed by the system manually to achieve insights into
their social trends, or as a information display, which allows
a user to passively get a feel for the general health of their
social environment in an instant. However, the application of
the Stay-in-touch model to a more inherently ambient
solution could take a radically different form. An example of such
an ambient display in the area of social information is the
Whereabouts Clock developed at Microsoft Research [
Since the critical information output for the user — reminders indicating when a friendship is stagnating — is atomic and relatively simple, it could be used in conjunction with a number of lo-fi data delivery methods. The user could subscribe to receive suggestions as text messages on their mobile phone, or through email or twitter tweets, informing them of the person they need to catch up with. There is also the possibility of using small personal devices, such as the Chumby or iPhone, to allow more convenient access to reminders and provide a simplified version of the main visualisation. FUTURE EVALUATION We see an ideal evaluation of the system as a diverse userbased study; one would record the suggested social interaction time and type and along with the actual time that the person next contacted the friend in question. This data would be collected over a sustained period, and two parallel sets of social interaction data would emerge; the smaller the discrepancy between these two data sets, the more helpful the system is. A control group would be made up of a subset of the testers who would have their social interactions predicted as normal, but not shown to them. If the average difference between prediction and interaction is much lower for the informed group, then this would suggest that not only are the reminders influencing those users’ social habits, but that they are promoting a more regular and sustained social rhythm. In addition, if the study was administered to several groups of people, each with different levels of notification, we could get a good idea of when adding more frequent notifications or more vivid animations stopped being helpful and starting being distracting.
CONCLUSIONS We have presented Stay-in-touch, which presents ambient information to a user based on the interactions that they have with members of their social network, and suggest how this kind of information can help a user to keep their social network in a healthy state. Given sufficiently careful treatment, infrequent notifications can become a useful addition to an ambient display. We have postulated that certain traits are desirable in an ambient reminder system; these are a long possible response time, variance in the timing and meaning of reminders, and simple, easy to interpret reminder information. Only extensive user testing can confirm that these qualities give an indication of the suitability of a type of reminder to ambient systems.
Acknowledgements: This work is supported by Science Foundation Ireland through an Undergraduate Research Experience and Knowledge grant (UREKA), and under grant number 03/CE2/I303-1, “LERO: the Irish Software Engineering Research Centre.”