Contextual patterns in mobile service...
Transcript of Contextual patterns in mobile service...
ORIGINAL ARTICLE
Contextual patterns in mobile service usage
Hannu Verkasalo
Received: 6 March 2007 / Accepted: 5 February 2008 / Published online: 4 March 2008
� Springer-Verlag London Limited 2008
Abstract Mobile services differ from other services
because of their temporal and spatial attributes. Mobile
services additionally differ from each other in their value-
added to the end-user. Some services—such as emailing
and voice—are more business oriented. On the other hand,
various free-time oriented services are provided in new
smartphones, such as imaging and music playback. The
present paper studies how mobile services are used in
different contexts. For this, the paper develops a special-
ized algorithm that can be used with handset-based usage
data acquired straight from end-users in an established
panel study process. Educated guesses can be drawn on the
user context based on the developed algorithm. In the
present exercise usage contexts were divided into home,
office and ‘‘on the move’’. The algorithm is used with
exemplary data from Finland and the UK covering 324
consumers in 2006. More than 70% of contextual use cases
are correctly classified based on raw data. According to
exemplary results particularly multimedia services are used
‘‘on the move’’, whereas legacy mobile services experience
more evenly distributed usage across all contexts. The
algorithm that identifies context based on raw data provides
a new angle to mobile end-user research. In the future, the
accuracy of the algorithm will be improved with the inte-
gration of seamless cell-id logging and GPS data.
Keywords Context-oriented services �Ubiquitous computing � Mobile services
1 Introduction
The mobile services industry is growing fast. According to
Nokia’s estimations, the number of cellular subscriptions is
likely to surpass three billion in 2008 (Nokia 2005). Many
people get actually their first Internet use experience with
mobile handsets. The portfolio of mobile services expands
also in other dimensions, for example, new mobile multi-
media and navigation services emerge on the market. The
most advanced pocket-sized mobile handsets (i.e. smart-
phones, converged devices) support many functions from
multimedia to personal information management. Person-
to-person services, however, still remain as the most
important service class. According to recent results from a
Finnish smartphone study in 2006 about 60% of smart-
phone usage time is still allocated on person-to-person
oriented services. This is down from 70% in 2005 (Ver-
kasalo 2007b).
As the evolution of the mobile services industry pro-
ceeds further, more value can be extracted from
understanding the usage patterns of individual customers.
New smartphone devices provide many different kinds of
value-added to the end-user. The understanding of new
services holds value because these new services might have
disruptive effects (Christensen 1997; Barsi 2002) on the
whole future of the mobile services industry (Verkasalo
2007a). The concept of ubiquitous computing in the mobile
domain is used while referring to today’s heterogeneous
mobile services. This concept refers to a trend in which
mobile services integrate with the environment, improving
the contextual value through, e.g., entertainment, infor-
mation or communication (see e.g. Ojala et al. 2003;
Abowd et al. 1997). To give an example, music playback
services provide lots of value-added when people are
commuting, but there is less value-added from such
H. Verkasalo (&)
Helsinki University of Technology, Helsinki, Finland
e-mail: [email protected]
123
Pers Ubiquit Comput (2009) 13:331–342
DOI 10.1007/s00779-008-0197-0
services at home because substitute devices (home stereo
sets) exist. Similarly the alarm clock service might be
practical at home, but there are no valuable use cases for
such service at office. In the future smartphones can assist
in various kinds of situations such as doing shopping
(payments, retrieval of product information), navigating to
a particular point of interest, retrieving weather forecasts,
confirming the airline check-in, etc. Context is much more
than the mere location; it is something that relates to all
situational factors around the user in a particular use case
(Kaasinen 2003).
This paper discusses the contextual dimension of mobile
services. Then a specialized context algorithm is developed
for a handset-based smartphone service research platform
(Verkasalo and Hammainen 2007). The algorithm analyses
usage-level data and divides individual customers’ usage
into three separate contexts (home, office, ‘‘on the move’’).
In the final part of the paper exemplary research results of
the algorithm are presented based on panel studies arranged
in the UK and Finland in fall 2006. This prototype algo-
rithm holds lots of future potential, particularly when
integrating it with seamless cell-id logging and utilization
of GPS data from handsets. No earlier approaches in
modeling the context of actual end-users exist in this scale.
Based on the outputs of the algorithm usage-level behavior
of smartphone users can be analyzed in great detail. In the
future, the algorithm will contribute to various disciplines
from end-user marketing to new service road tests and from
behavioral sciences to new business development.
2 Mobile services and context
Legacy mobile services include voice and text messaging.
These services have been embedded basically in all mobile
phones as we know them today (pocket-sized handheld
devices with digital cellular connectivity). Smartphones,
which combine person-to-person services with computer-
like applications such as document viewers, Internet
browsing and multimedia, have stretched the functionality
of these handheld devices. At the same time they have
made mobile phones even more all-around devices. Not
only do we see both hedonic and utilitarian mobile services
today, but also functionally these services provide many
kinds of value-added to the end-user.
A service is generally considered as a non-material
equivalent of a good. According to Kotler and Armstrong
(1996) services are ‘‘activities or benefits offered for sale
that are essentially intangible and do not result in the
ownership of anything’’. Electronic services are thereby
services, which involve electronic transactions. A mobile
service, further, is defined in this paper essentially as a
service that is consumed by the end-user with a mobile
handset. In this definition, a mobile handset means a
pocket-sized, handheld terminal which has at least a cel-
lular radio interface implemented.
In this paper, the focus is particularly on core mobile
services that provide value-added straight to the end-user.
Core services are built on top of enabler or supporting
services. Enabler and supporting services are needed in
implementing the core service. For example, a generic
packet data interface is needed in implementing video
streaming applications that utilize Internet protocols. In
this example, the packet data interface is a service enabler
whereas the video streaming application itself is the core
service. Different kinds of mobile services have different
characteristics, and therefore they should be managed and
deployed in different ways (Balasubramanian et al. 2002;
Heinonen and Andersson 2003). From a contextual point of
view, for example, multimedia services should be easy to
launch when walking on the street, mobile browsers should
be optimized for small displays and physical restrictions,
and future voice services should leverage, e.g. WiFi net-
works (with mobile VoIP) at home/office and cellular
network access (with legacy circuit-switched voice) when
outside of home/office (mainly because of more efficient
use of network resources).
Applications are here seen either as network applica-
tions implementing the service or then as stand-alone
applications in the handset (i.e. terminal applications).
Application is, therefore, a more technical term referring to
the solution itself, whereas service is better associated with
the value-added to the end-user. In this paper, all solutions
providing value-added to the end-user through mobile
handsets are collectively called as mobile services.
Mobile services have some special characteristics in
comparison to other types of services. The key differences
are related to the spatial and temporal components of ser-
vice usage. If you want to meet a bank teller, you have to
visit the bank (location) at a certain appointment (time).
These restrictions are sometimes present even when using
some electronic services, such as the Internet with fixed
access (i.e. wired access). Even though the fixed Internet
provides e.g. online banking services 24 h a day (thus
overcoming the problem of temporal availability), you still
need a computer with fixed access (which causes problems
in the spatial domain). Mobile services, used with handheld
mobile devices, overcome both spatial and temporal con-
straints (Heinonen and Pura 2006). One more interesting
dimension of mobile services is the potential for individual
personalization of service offerings (Clarke and Flaherty
2003). According to Kaasinen (2003) mobile services
should adapt in various dimensions (e.g. location and
context) to user preferences.
Many approaches to classify mobile services exist. To
mention some of the alternative approaches, the classification
332 Pers Ubiquit Comput (2009) 13:331–342
123
can originate from the end-user (subscribers, service users),
terminal (mobile phone, PDA, laptop computer, desktop
computer), network (operator, network technologies, serv-
ers) and academic research (diffusion/adoption research,
market studies, technical studies) points of view.
Verkasalo, Kivi and Smura (2007) identify three streams
of literature with regards to the mobile service classifica-
tion. First, the technical/business perspective exists as quite
a logical way to formulate a classification (see e.g. Ange-
hrn 1997; Basaure 2004; Cerqueira et al. 2005; ECOSYS
2004; Meuter et al. 2000; Mitchell and Whitmore 2003;
UMTS Forum 2001; Vesa 2005) framework. Technical and
business approaches are often tied together, and the per-
spective is external to the service user.
Secondly, the end-user perspective in classifying services
remains as the most common approach. This approach is
applied at least in Anckar and D’Incau (2002), Chande
(2005), Dabholkar (1996), Giaglis et al. (2003), Hyvonen
and Repo (2005), FMTC (2004), Jarvenpaa and Lang
(2005), Kaasinen (2005), Nysveen et al. (2005), O’Loughlin
(2000), Sullivan et al. (2005), Vanjoki (2003), Velez and
Correia (2002), Verkasalo and Hammainen (2006), Wil-
liamson (2003), Korhonen (2003), and Yoon et al. (2003).
The end-user perspective is much internal, looking at service
offerings from the end-user point of view.
Thirdly, an emerging stream of literature utilizes holis-
tically many softer research disciplines. This stream deals
with, e.g., context-specific use cases and special ways of
implementing particularly mobile services. Some of the
papers utilizing this approach include ARC Group (1999);
Balasubramanian et al. (2002); Heinonen and Pura (2006);
Heinonen (2004); Pura and Brush (2005); Van der Hejden
et al. (2005). This third approach increasingly combines
softer academic disciplines such as sociology or psychol-
ogy with older user-centric approaches of mobile service
literature. Heinonen and Pura (2006) claimed that the
spatial and temporal (space and time) dimensions can be
used in analyzing the characteristics of a mobile service
(see also Balasubramanian et al. 2002). The spatial and
temporal contexts can also be extended to include the
social setting (see e.g. Okazaki 2005). Furthermore, it is
suggested that the relationship between a customer and
service provider (e.g. the extent of personalization) should
be used in the service classification (Heinonen and Pura
2006). All these relate more or less to the sociological,
psychological or contextual service analysis. The context
perspective, that is the focus of this paper, is not far from
the end-user perspective in reality. This is because the
context is still perceived by the end-user (Kaasinen 2003),
and end-users should never be overlooked in mobile ser-
vice analyses (Holtel 2006).
The micro-level mobile service perspective (explained
above), focusing particularly on the contextual dimension
of service analysis, is now taken as the principal perspec-
tive of this paper. As mobile devices are carried basically
everywhere by end-users, they are therefore devices that
perfectly reflect the user’s context. Context is much more
than the location, including also the social setting and sit-
uational factors. Kaasinen (2003) argues that besides
location the other elements of context, however, are diffi-
cult to identify and measure. As the mobile services
industry and enhanced software platforms provide quite a
heterogenic portfolio of services today, it makes sense to
study these services not only from the user demographics
(Verkasalo 2005), but also from the contextual usage point
of view. The present research approach provides a signif-
icant contribution in understanding end-users by tackling
the contextual dimension with specific data mining tools.
Earlier studies that have discussed the modeling of con-
text and location with cell-ids include, e.g., Carmichael et al.
(2005), Trevisani and Vitaletti (2004) and Li et al. (2006).
The context algorithm introduced in this paper is used with a
recently developed handset-based mobile service research
platform (Verkasalo and Hammainen 2007). A handset-
based client extracts data on location (cell-ids) and time of
use. Based on this, educated guesses can be made on the
context of usage. In this paper, exemplary aggregated study
results are presented to demonstrate how the algorithm
could be used in large end-user panels with the associated
research platform. This research paper is pioneering in the
sense that the handset-based research platform has not been
earlier used in context research. Potential venues benefiting
from this research approach include
• Development of better UIs for different contextual
situations
• Adaptation of smartphone functionalities to different
contexts
• Potential improvements in presence services and
performance
• Extensions of mobile subscriber segmentation studies
• Analysis of social networks and micro-level behavior
of subscribers
• Preparation of techno-economic comparisons between,
e.g., competing radio access technologies (Smura 2006)
3 Identifying context from handset-based data
The solution to the identification of the context is explained
in this section. The presented algorithm is developed for a
specialized mobile service research platform. Handset-
based smartphone usage data is first acquired through a
novel service research platform. The platform consists of a
smartphone client running on subscribers’ Nokia S60
smartphones that acquires device actions and writes them
Pers Ubiquit Comput (2009) 13:331–342 333
123
to memory. Then this data is transmitted to centralized
servers for further analysis. For an overview of the process,
refer to Verkasalo and Hammainen (2007). The raw data
including all user actions with the smartphone over a panel
project of 2–3 months by several hundred subscribers is
used together with accurate hourly cell-if logs in deter-
mining the context of usage and then further in analyzing
usage-level differences between different contexts.
The approach is not unique, as Raento et al. (2005)
developed tools for identifying the context, too. Their cli-
ent automatically identified clusters of locations and
provided some artificial intelligence in making guesses on
users’ context and presence information. This client has
been used by Eagle (2005), too. The principal difference in
the approach of this paper is the data acquisition system. In
this paper, the data is centrally managed, and controlled
panel studies are analyzed in an aggregate fashion. In other
words, with the present approach much more data can be
collected and very accurate usage-level analyses between
contexts can be made. Whereas Raento et al.’s (2005) is a
more natural approach in building intelligence to the
handset, the approach of this paper better suits for large-
scale end-user research.
The input of the algorithm is a data file presenting each
user’s active cell-id (the unique cell-id identification code
to which the subscriber’s mobile phone is currently
hooked), logged once per hour for each panelist over the
whole panel period. Only this data file is required for the
proper functioning of the algorithm. Other data files rep-
resenting users’ smartphone actions (e.g. voice
communications, multimedia application usage, data ses-
sion details), equipped with time stamps, are later needed
in combining the contextual information with actual usage.
The stages of the context algorithm process are:
1. For each particular customer, the cumulative usage
(e.g. active hook-up hours) of each cell-id is calculated
over the whole panel period. Then the relative share of
usage for each of these cell-ids is calculated by
dividing the number of a particular cell’s active hours
by the total number of active hours spent in the panel.
2. For each cell-id achieving a relative share of usage
more than 10% (i.e. there is significant usage under a
particular cell) a hypothesis is made that the cell-id
must represent either home or office usage (both are
fixed locations with potential for a high number of
active usage hours).
3. The cells with at least a relative usage ratio of 10% are
analyzed in greater detail by weighting their hourly
and weekday-based distributions with corresponding
statistics (provided by Statistics Finland) indicating
people’s averaged location dynamics per weekday and
hour. Those cells that receive relatively more hits
during typical home context hours get more ‘‘home
points’’, and vice versa with office context.
4. ‘‘Home’’ and ‘‘office points’’ are summed up in order
to classify cells with relatively high cumulative usage
either into home or office contexts. If the cell received
more home points all usage taking place under that cell
is considered home usage, and vice versa for office.
5. All the cells receiving less than 10% of all cell-id hits
are considered as ‘‘on the move’’ context. The
assumption is thereby made that people do not spend
more than 10% of cumulative time in any of the ‘‘on
the move’’ cells.
6. All users for whom three separate contexts with
significant ([100 h) usage per context are identified
and included in the final dataset for usage-level
comparisons.
7. The outputs of the context algorithm are mapped
together with usage data in order to classify all other
user actions into one of the three contexts. Based on
this classification accurate usage-level comparisons
can then be made on the services under interest.
The context algorithm can be run on the whole aggre-
gated panel dataset. The algorithm is programmed into an
SPSS (2002) syntax file that is easy to apply for new panel
studies. The contribution of the algorithm is an automated
mapping of accurate logs on user location and timestamps
to statistics from Statistics Finland reflecting daily contex-
tual distributions of a high number of interviewed people.
The outputs of the algorithm are further easy to combine
with usage-level data. The major challenges include the
discontinuity of cell-id logging (only once per hour), net-
work conditions (change of the active cell-id because of
poor reception or network load even though the user does
not move) and overly mobile behavioral patterns (e.g.
people who do not have fixed daily office locations cannot
be easily analyzed). In the future, the context algorithm will
be improved by utilizing seamless cell-id logging (writing
logs every time the cell-id changes) and GPS chip data.
4 Empirical observations on contextual mobile service
usage
Based on the developed algorithm and available data,
demonstrative usage-level results of the context algorithm
are now presented. Though the panel studies (including
several hundred smartphone customers with the monitoring
software installed for ca. 2–3 months in a given market) are
geared towards early-adopter customers (see e.g. Rogers
1962), they nevertheless tell something about the most
advanced usage patterns that might evade to the mass-
market domain in the near future. This study sets out to
334 Pers Ubiquit Comput (2009) 13:331–342
123
provide some exemplary results on contextual usage-level
behavior. Future studies with the improved version of the
algorithm should be used in providing robustness studies
with better accuracy.
A dataset from Finland and the UK is utilized as a case
example of the utilization of the context algorithm. All in
all, the original dataset included 861 users. All of these
users had generated at least 3 weeks of active smartphone
usage. After running the context algorithm over the whole
dataset, 324 smartphone users were identified for whom the
context algorithm accurately identified three separate
contexts. For many users the context algorithm did not find
a proper office context. This indicates that those users were
perhaps not working regularly (e.g. unemployed people or
students who do not visit even school context too often),
they were exceptionally mobile workers, or they work from
multiple locations. Secondary handset usage (which makes
the dataset discontinuous) or poor cell-id reception (i.e. lots
of cell-id jumps in office context) complicate the context
identification process, too. The remaining 324 users for
whom the context algorithm worked well are included in
the results below. These 324 users generated an average of
67 usage days during the panel study. In terms of demo-
graphics they were mostly male and 20–39 years old
people, resembling demographic distributions observed in
other studies (see e.g. Verkasalo 2007b).
Figure 8 in Appendix A shows exemplary outputs of the
context algorithm for one particular user. As can be seen,
the algorithm divides all observed usage into three distinct
classes based on the heuristics as explained in the technical
part of this paper. In panel-wide studies, however, all data
is aggregated together and typically it makes more sense to
report, e.g., the share of all panelists who have explored a
particular service in a certain context, or the cumulative
share of all outbound voice calls that took place in home
context. Figure 8 nevertheless indicates the level of accu-
racy (hourly contextual information for every single
panelist) the context algorithm can reach.
During the panel study a possibility emerged to push
pop-up questions to users’ handsets through the handset-
based research platform. An interactive question related to
the user’s context was asked from more than 100 customers
after a random packet data session taking place during the
panel; 87 answers were received indicating the users’ own
perception of their context (home, office, ‘‘on the move’’).
This data was linked to the outputs of the context algo-
rithm; 70% of cases were correctly classified, i.e. the user’s
opinion and the guess of the context algorithm were
aligned. This is quite a satisfying success rate, as the
context algorithm in its current form still has the limitation
of only an hourly accuracy (therefore, the active context
might have changed after the hourly cell-id logging, and
thus the pop-up question’s answer is time-wise separate
from the output of the context algorithm). Additional error
is caused by people having mixed opinions on the type of
context (i.e. the context algorithm names typical weekday
daytime contexts into ‘‘office’’, though the user might
actually visit, e.g., alternative offices and consider this as
‘‘on the move’’ usage).
Figure 9 in Appendix A presents the distribution of
smartphone usage activity. The most active users have about
14 active smartphone usage hours per day, meaning that they
have taken actions with the smartphone on average during
14 out of 24 h of the day. On the other hand, users with very
low usage activity exist (e.g., only a couple of unique hours
of active smartphone usage per day). Those for whom a lot of
‘‘on the move’’ usage was identified, generated more
smartphone usage hours on average. Thus, more mobile
subscribers use smartphones more actively. Smartphones
are still meant in the first place for mobile usage.
Based on the aggregate data, panelists on average were
located 49.9% of weekday time in home context, whereas
office and ‘‘on the move’’ contexts represented 23.5 and
26.6% of the time, respectively. Home context acquires the
most time because people typically not only hang around a
lot at home, but they also sleep at home (thus generating a
lot of inactive presence at home).
Figure 1 presents averaged weekday context distribu-
tion. As can be seen, office context acquires the most usage
between 8 a.m. and 3 p.m., reflecting typical office hours.
Some ‘‘on the move’’ usage can be identified in the
nighttime, reflecting, e.g., foreign trips, holidays, visits to
friends, etc.
Figure 2 presents context distribution over the days of
the week. Some differences can be identified between
weekdays. For example, people on average spend 24.3% of
daily time (*5.8 h) at office on Monday and only 21.3%
(*5.1 h) on Fridays. Taking into account that the dataset is
quite extensive, these differences suggest that people spend
more time at work on Mondays, whereas an average
workday on Fridays is shorter. Very little office usage can
be identified on weekends.
Figure 10 in Appendix A illustrates aggregated time
allocation results per context. Multimedia usage experi-
ences more usage ‘‘on the move’’ than in office or home
contexts, suggesting that people utilize smartphone camera
and music playback functionalities relatively more some-
where else than at home or office. Voice call usage receives
relatively more usage at home than at office or ‘‘on the
move’’. Taking into account that voice is perhaps the most
context-independent service (most people have already
substituted mobile phones for fixed-line telephones), this
observation is plausible.
Figure 3 presents that on average panelists use the
smartphone the most at home. A typical panelist spends 69
min per day by doing something with the handset. It was
Pers Ubiquit Comput (2009) 13:331–342 335
123
earlier observed that 49.9% of all time is spent at home,
and Fig. 3 suggests that 51.8% of active smartphone usage
time (i.e. smartphone has been actively used) is spent in the
same context. The relative usage activity time (including
only active usage) follows the distribution of relative
context-specific time (including both inactive and active
usage). This is surprising, as one would assume that people
slept more at home than at office or ‘‘on the move’’. Per-
haps people still play around with smartphones quite a lot
at home, thus explaining the numbers of Fig. 3.
Figure 11 in Appendix A presents person-to-person
action distributions for weekdays. As can be seen, both
voice and SMS are all-around services. People utilize them
equally much in all contexts. SMS is a bit more home-
oriented service than voice, supporting earlier observations
suggesting that SMS is more free-time oriented. According
to earlier results SMS has caught relatively high usage in
the evening. Voice has been found more business-oriented
(Verkasalo 2005).
Figure 4 projects both mean and median durations for
inbound and outbound voice calls. In addition to normal
weekday contexts, also home and ‘‘on the move’’ weekend
contexts are included in this analysis. Voice calls tend to be
significantly longer at home, and this holds irrespective of
weekday or weekend. Perhaps in other contexts than home
there is less time available for casual chatting, whereas at
home people enjoy talking to friends for longer.
Figure 5 presents findings on more advanced services.
There are no significant differences between contexts when
comparing usage coverage numbers that reflect simply the
share of panelists who have at least once tried a particular
service in a given context. However, the fact that there is
less ‘‘explorative usage’’ in office context is clearly evident.
Figure 6 presents contextual usage patterns of particular
smartphone applications. Productivity applications such as
Weekday Hourly Location
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23
HourS
har
e o
f D
aily
Ho
urs
Sp
ent
Home Office On the move
Fig. 1 Weekday hourly
distribution of context
Weekly Location
10%
20%30%
40%
50%
60%
70%80%
90%
0%
100%
Tuesday Wednesday Thursday Friday SaturdayMonday Sunday
Day of Week
Sh
are
of
Wee
kly
Ho
urs
Sp
ent
Home Office On the move
Fig. 2 Daily distribution of
context
35.6
14.5
18.6
0
5
10
15
20
25
30
35
40
Home Office On the move
To
tal M
inu
tes
of
Usa
ge
/ Co
nte
xt /
Day
Fig. 3 Contextual smartphone usage minutes
336 Pers Ubiquit Comput (2009) 13:331–342
123
Adobe Reader catch quite a lot of usage ‘‘on the move’’. In
office and home contexts people perhaps have an access to
laptops or desktop computers and there are therefore fewer
needs to use productivity applications with mobile handsets.
File manager is a kind of configuration application (used when,
e.g., moving files and installing applications), and this kind of
usage mostly takes place at home. Finally, Clock application is
almost solely used for setting up an alarm. No surprise, almost
all usage of Clock application takes place at home.
Figure 7 reflects contextual usage patterns of multi-
media services. As can be seen, almost all but gaming
usage takes place ‘‘on the move’’. Multimedia applications
are—similarly to productivity applications—of such kind
that the value-added to the end-user is realized ‘‘on the
move’’. At home or office substitutes for these services
(music playback) exist or then there is simply no need for
the service (not that many people capture photos at home).
Interestingly, there is a lot of gaming usage at home. This
supports earlier speculations indicating that people might
play around with mobile handsets quite a lot at home.
Figures 12 and 13 in Appendix A tell about mobile
Internet usage. In general all three most important Internet
service categories (browsing, email and instant messaging)
experience the most usage ‘‘on the move’’. This can be
once again reasoned by the fact that there are simply fewer
needs for mobile Internet usage at home or office, because
substitute devices that are easier to use (desktop/laptop
computers) are available in these contexts. Browsing ses-
sions are longest ‘‘on the move’’, reflecting that people
have more hedonic (‘‘killing time’’) needs for browsing in
this context. At home or office there is probably a relatively
small need and, therefore, browsing sessions tend to be
shorter there.
All in all, these brief examples present the kinds of
usage-level analyses that can be pursued with the context
algorithm and the supporting handset-based mobile service
research platform. There are various usage patterns evident
proving that mobile services provide different kinds of
contextual value-added. Future usage-level studies could
pursue the following topics:
Fig. 4 Voice call durations in
different contexts
49%
9%
49%
19%
32%
46%
5%
42%
13%
19%
8%
49%
22%
32%
55%
0%
10%
20%
30%
40%
50%
60%
Browsing WLAN Access OfflineMultimedia
Playback
Radio VideoRecording
Sh
are
of
Pan
elis
ts T
ried
Home Office On the move
Fig. 5 Advanced service usage
in different contexts
Pers Ubiquit Comput (2009) 13:331–342 337
123
• Segmentation of users based on mobility patterns and
mobile service usage
• Visualization of mobile subscriber movement patterns
• Techno-economic calculations leveraging contextual
information (e.g., the potential of WiFi in substituting
for cellular access at home)
• Analyses of individual applications and their potential
for technical improvements (e.g., context adaptation)
The context algorithm together with the handset-based
end-user research platform provides unique data that is
difficult to obtain otherwise.
5 Conclusion
The present paper discussed first the special characteristics
of mobile services. Most importantly, mobile services can
be used from anywhere at any point of time. Mobile ser-
vices cover nowadays quite a portfolio of services from
hedonic to utilitarian ones. A need exists to empirically
model contextual mobile service usage.
A specialized context algorithm was developed to be
used together with a handset-based end-user research
platform. The new algorithm along with the associated
process of obtaining and analyzing huge amounts of usage-
level data automatically helps in assessing contextual
dimensions of mobile services. The new process was tested
with a case example including data from Finland and the
UK from fall 2006. According to the results, multimedia
and Internet services are used quite actively ‘‘on the
move’’, whereas legacy SMS and voice services experience
more evenly distributed usage among home, office and ‘‘on
the move’’ contexts. Different people have unique mobility
patterns, and therefore future end-user research should
78%
48%
29%
3%
22%
27%
19%
29%
44%
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
Clock applicationusage
File managerapplication usage
Adobe Readerapplication usage
Share of All Actions
Home Office On The Move
Fig. 6 Productivity application
and clock usage in different
contexts
31%
29%
24%
44%
37%
22%
21%
22%
28%
17%
47%
49%
53%
28%
46%
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
Offline multimedia player usage
Multimedia application usage:music/sounds
Radio application usage
Gaming usage
Camera usage
Share of All Actions
Home Office On The Move
Fig. 7 Multimedia usage in
different contexts
338 Pers Ubiquit Comput (2009) 13:331–342
123
focus on the micro-level of contextual service usage and
specialized segmentation studies. Other levers of the
developed algorithm are context-aware applications such
as presence solutions. The new algorithm and usage-level
data retrieval platform provide valuable information on the
behavior of mobile subscribers and service usage patterns.
This information might be useful among application
developers, strategic decision markets of mobile operators
and academic researchers.
Contextual analysis of end-user presence provides stra-
tegic insights. For example, handovers making seamless
transfer of voice calls from cellular to home WiFi networks
are nowadays possible with UMA technology. In order to
analyze network-level impacts of an introduction of WiFi/
cellular dual radio access, accurate data on consumer
contextual behavior is needed. Accurate analysis of mobile
services in different contexts also improves the under-
standing of service usage dynamics. This understanding is
needed in service commercialization and strategic posi-
tioning tasks.
The main venues for technical contribution are related to
an update of the algorithm with seamless cell-id logging
(continuous instead of hourly logging). In addition, future
handsets with embedded GPS chips could provide even
more accuracy to available location data. In particular, the
context could be identified with higher accuracy, and
moving buses can be separated from stable locations, e.g., a
cafeteria.
The heuristics presented in this paper did not identify the
context for everybody. The main source of error is too low a
frequency of cell-id logs and incapability to understand
geographical proximity (due to unavailability of GPS data
yet) and resulting inaccuracy in cumulating usage times per
context. With the planned improvements it is expected that
the presented heuristics can provide contribution to context-
oriented end-user research in the future. The exemplary
results reflecting differences in end-user behavior in dif-
ferent contexts will be further examined in the future. The
study platform will be improved to better account for both
explorative and confirmatory mobile end-user research.
The context algorithm presented in this paper is pio-
neering because it can be integrated into the end-user
research process than constantly includes thousands of
smartphone users all around the world. This provides both
a huge domain of end-users and at the same time a stan-
dardized process to implement panel studies.
6 Appendix: Other visualizations of the context
algorithm utilization
Figures 8, 9, 10, 11, 12, 13.
Fig. 8 Exemplary output of the
context algorithm
Pers Ubiquit Comput (2009) 13:331–342 339
123
Fig. 9 Distribution of active
usage hours
6%
5%
6%
27%
31%
25%
11%
12%
14%
12%
15%
12%
33%
26%
28%
13%
11%
14%
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
Home
Office
On the move
Percentage of Time
Browsing Messaging Multimedia PIM Voice Other
Fig. 10 Smartphone time
allocation in different contexts
31%
33%
36%
36%
25%
30%
31%
31%
44%
38%
33%
33%
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
Outbound voice
calling
Inbound voice
calling
Outbound sms
messaging
Inbound sms
messaging
Share of All Actions
Home Office On The Move
Fig. 11 Person-to-person
service usage in different
contexts
340 Pers Ubiquit Comput (2009) 13:331–342
123
References
Abowd GD, Atkeson CG, Hong J, Long S, Kooper R Pinkerton M
(1997) Cyberguide: a mobile context-aware tour guide. ACM
Wireless Netw 5:421–433
Anckar B, D’Incau D (2002) Value creation in mobile commerce:
findings from a consumer survey. J Inf Technol Theory Appl
4(1):43–64
Angehrn A (1997) Designing mature Internet business strategies: the
ICDT model. Eur Manage J 15(4):361–369
ARC Group (1999) Wireless industry showing us the future for the
Internet. In: Wireless Internet: applications, technology and
player strategies (1999–2004)
Balasubramanian S, Peterson RA, Jarvenpaa SL (2002) Exploring the
implications of M-Commerce for markets and marketing. J Acad
Market Sci 30(4):348–361
Barsi T (2002) Disruptive technology vs. disruptive applications.
Telephony Online. http://telephonyonline.com/news/telecom_
disruptive_technology_vs/index.html. Accessed on 6 June 2006
Basaure A (2004) Preliminary research on existing and planned
mobile data service solutions and value systems in leading
markets. LEAD project report. Networking Laboratory, Helsinki
University of Technology
Carmichael DJ, Kay J, Kummerfeld B (2005) Consistent modelling of
users, devices and sensors in a ubiquitous computing environment.
User Model User-Adapt Interact 15(3–4):197–234 (Aug. 2005)
Cerqueira, E, Neto A, Veloso L, Monteiro E, Mendes P (2005) Control
of multi-user services in mobile heterogeneous environments. In:
Proceedings of the third international working conference—
performance modelling and evaluation of heterogeneous net-
works, Het-Nets05, Ilkley, West Yorkshire, UK, July 2005
Chande S (2005) Mobile web services, Department of Computer
Science, University of Helsinki. http://www.cs.helsinki.fi/u/
chande/courses/cs/MWS. Accessed on 20 July 2006
Christensen CM (1997) The innovator’s dilemma. Harvard Business
School Press, Cambridge
Clarke I, Flaherty T (2003) Mobile portals: the development of M-
Commerce gateways. In: Mennecke BE, Strader TJ (eds) Book
mobile commerce: technology, theory, and applications. Idea
Group Publishing, Hershey, pp. 185–201
Dabholkar PA (1996) Consumer evaluations of new technology-based
self-service options: an investigation of alternative models of
service quality. Int J Res Market 13:29–51
Eagle, N (2005) Machine perception and learning of complex social
systems. Doctoral dissertation, Massachusetts Institute of
Technology
WEB Session Duration
5.15
4.024.06
3.48
6.02
4.43
4.96
4.19
5.17
4.13
0.0
1.0
2.0
3.0
4.0
5.0
6.0
7.0
Mean WEB session duration Median WEB session duration
WE
B S
essi
on
Du
rati
on
[M
inu
tes]
Home Office
On the move Home (weekend)
On the move (weekend)
Fig. 12 Average browsing
session durations in different
contexts
32%
25%
35%
27%
28%
21%
42%
48%
43%
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
Browsing packetdata usage
Email packet datausage
Internet servicespacket data usage
Share of All Actions
Home Office On The Move
Fig. 13 Internet service usage
in different contexts
Pers Ubiquit Comput (2009) 13:331–342 341
123
Ecosys (2004) First results on the economics of the new service
provisioning in 3G and beyond. The CELTIC initiative, techno-
economics of integrated communication systems and services
FMTC (2004) Mobile services market in Finland 2003 (in Finnish).
Finnish Ministry of Transport and Communications Reports
Giaglis GM, Kourouthanassis P, Tsamakos A (2003) Towards a
classification framework for mobile location services. In:
Mennecke BE, Strader TJ (eds) Book mobile commerce:
technology, theory, and applications, IGI Global, Hershey, pp
67–85, 22 Oct 2002
Heinonen K (2004) Time and location as customer perceived value
drivers. Doctoral Dissertation, Swedish School of Economics
and Business Administration, Helsinki, Finland
Heinonen K, Andersson P (2003) Swedish mobile market: consumer
perceptions of mobile services. Commun Strateg 49(1st quar-
ter):151–171
Heinonen K, Pura M (2006) Classifying mobile services. Presented at
Helsinki Mobility Roundtable, 1–2 June 2006, Helsinki
Holtel S (2006) A timeless way of service classification. Vodafone
R&D white papers. Understand the personality of future
communication services, Vodafone, 29 Mar 2006
Hyvonen K, Repo P (2005) Mobile services in the quotidienne of
Finns (Mobiilipalvelut suomalaisten arjessa), Vox consumpto-
ris—Kuluttajan aani, Kuluttajatutkimuskeskuksen vuosikirja
2005, Savion Kirjapaino Oy, Kerava
Jarvenpaa S, Lang K (2005) Managing the paradoxes of mobile
technology, information systems management. Fall 22(4):7–23
Kaasinen E (2003) User needs for location-aware mobile services.
Pers Ubiquitous Comput 7(1):70–79 (May 2003)
Kaasinen E (2005) User acceptance of mobile services—value, ease
of use, trust and ease of adoption. Doctoral dissertation, VTT
Publications 566, Finland
Korhonen P (2003) Technology leadership. Paper Presented at Nokia
Capital Market Days, New York, 24 November 2003
Kotler P, Armstrong G (1996) Principles of marketing. Pearson
Education, Inc., Upper Saddle River, New Jersey
Li W, Kilander F, Jansson CG (2006) Toward a person-centric
context aware system. In: Proceedings of 1st RSPSI International
Workshop, 7 May 2006, Dublin, Ireland
Meuter ML, Ostrom AL, Roundtree RI, Bitner MJ (2000) Self-service
technologies: understanding customer satisfaction with technol-
ogy-based service encounters. J Market 64(3):50–64
Mitchell K, Whitmore M (2003) In: Mennecke BE, Strader TJ (eds)
Location based services: locating the money. Idea Group
Publishing, Hershey, pp 51–66
Nysveen H, Pedersen PE, Thorbjørnsen H (2005) Intentions to use
mobile services: antecedents and cross-service comparisons.
J Acad Market Sci 33(3):330–346
O’Loughlin, MA (2000) Wireless Internet: opportunities and threats.
Ovum Group
Ojala T, Korhonen J, Aittola M, Ollila M, Koivumaki T, Tahtinen J,
Karjaluoto H (2003) SmartRotuaari—context-aware mobile
multimedia services. In: Proceedings of the 2nd international
conference on mobile and ubiquitous multimedia, Norrkoping,
Sweden
Okazaki S (2005) New perspectives on M-Commerce research.
J Electron Commer Res 6(3):160–164
Pura M, Brush G (2005) Hedonic and utilitarian motivations for
mobile service use. Paper presented at SERVSIG Research
Conference, Singapore, 2–4 June 2005
Raento M, Oulasvirta A, Petit R, Toivonen H (2005) ContextPhone—
a prototyping platform for context-aware mobile applications.
IEEE Pervasive Comput 4(2):51–59
Rogers EM (1962) Diffusion of innovation. Free Press, New York
Smura T (2006) Competition between emerging wireless network
technologies: case HSPA vs. WiMAX in Europe. In: 17th
European Regional ITS Conference, Amsterdam, Netherlands,
22–24 August 2006
SPSS (2002) SPSS 11.5 syntax reference guide, vol 1. SPSS, Chicago
Sullivan MG, Drennan J (2005) Marketing M-services: establishing a
usage benefit typology related to mobile user characteristics.
Database Market Customer Strateg Manage 12(4):327–341
Trevisani E, Vitaletti A (2004) Cell-ID location technique, limits and
benefits: an experimental study. Mobile computing systems and
applications, 2004. WMCSA 2004. Sixth IEEE Workshop on
Volume, Issue, pp 51–60, 2–3 December 2004
UMTS Forum (2001) The UMTS third generation market—phase II:
structuring the service revenue opportunities. Report No. 13.
UMTS Forum, April 2001
Van der Hejden H, Ogertschnig M, van der Gaast L (2005) Effects of
context relevance and perceived risk on user acceptance of
mobile information services. In: Proceedings of the Thirteenth
European conference on information systems (ECIS), Regens-
burg, Germany
Vanjoki A (2003) Multimedia—seed for growth. Nokia Capital
Market Days 2003. http://media.corporate-ir.net/media_files/
webcast/2003/nov03/Nokia/PPT/Drivingconsumermultimedia.pdf.
Accessed on 20 July 2006
Velez FJ, Correia LM (2002) Mobile broadband services: classifica-
tion, characterisation and deployment scenarios. IEEE Comm
Mag 40(4):142–150
Verkasalo H (2005) Handset-based monitoring of mobile customer
behavior. Master’s Thesis Series, Networking Laboratory,
Department of Electrical and Telecommunications Engineering,
Helsinki University of Technology, Espoo, Finland, September
2005
Verkasalo H (2007a) A cross-country comparison of mobile service
and handset usage. Licentiate’s Thesis, Helsinki University of
Technology, Networking Laboratory, February 2007
Verkasalo H (2007b) A snapshot on the evolution of Finnish mobile
service market. In: Conference on telecommunication techno-
economics (CTTE) 2007, Helsinki, Finland, 14–15 June 2007
(Submitted)
Verkasalo H, Hammainen H (2006) Handset-based monitoring of
mobile subscribers. Presented at Mobility Roundtable, Helsinki,
Finland, 1–2 June 2006
Verkasalo H, Hammainen H (2007) A handset-based platform for
measuring mobile service usage. INFO: J Policy Regul Strateg
9(1):80–96
Verkasalo H, Smura T, Kivi A (2007) A framework for analyzing the
usage of mobile services. Networking Laboratory, Helsinki
University of Technology (Submitted)
Vesa J (2005) Mobile services in the networked economy. IRM Press,
Hershey
Williamson J (2003) Content or brand? Paper presented at the 3GSM
World Conference, Cannes, France
Yoon JI, Suh HS, Leem CS (2003) A study on mobile business
customer satisfaction. Korea Institute for Industrial Engineering,
April 2003
342 Pers Ubiquit Comput (2009) 13:331–342
123