A Mobile Location-Aware Recommendation System
Semih Utku and Canan Eren Atay
Department of Computer Engineering, Dokuz Eylul University, Izmir, Turkey
Keywords: Context-Awareness, Location-Awareness, Mobile Technology, Data Mining.
Abstract: Improvements in mobile technology provide greater personal information accessibility, data incorporation,
and public resources accessibility, “anytime, anywhere”. Smartphones are not only devices that make phone
calls, but have also become a gateway to the Internet. Mobile devices offer the capabilities of usage
flexibility, mobility, fast wireless communication, and location-awareness. Location is determined by GPS
satellite tracking, position relative to GSM base stations, and the device's media access control. Similarly,
usage of social networks is increasing steadily. Widespread usage of social networks introduces new
requirements of Internet application. Users of such networks share their ideas and interests, as well as the
activities they plan to attend. In addition, they follow other users’ information and shape their planned
activities accordingly. In this study, an intelligent context-aware system is described. In this field, context-
awareness is a mobile paradigm in which applications can discover and take advantage of contextual
information, such as user location, nearby people and devices, and user activity. This system provides an
activity list that users plan to attend. Our recommender system creates results based on data mining
techniques, by using personal identification data and user activities. The recommender system brings novel
methodology to the activity-decision process by utilizing the right location and real-time information.
1 INTRODUCTION
Most people consider mobile technologies to be an
integral part of their lives. Recently released mobile
phones provide users not only with numerous
features (e.g., camera and video capture devices,
GPS localization, and Wi-Fi connectivity), but also
the capacity to program these mobile devices with
additional applications. Among the most popular
applications are location-based services (LBSs), in
which knowledge of the end user’s location is
utilized to deliver relevant, timely, and engaging
content (Rao and Minakakis, 2003).
The idea of context awareness is nearly as old as
the mobile devices themselves. Context awareness is
a mobile paradigm that uses context to provide
relevant information and/or services to the user,
where its relevancy depends on the user’s current
task (such as user location, nearby people and other
devices, and user activity). Context awareness in the
form of allowing designers to learn about what users
are doing with the app, where and when, is also
important to further improve that very same context
awareness of the applications (Fahy and Clarke,
2004). A mobile application can provide more
specific information that is suitable for the current
situation if it knows about its location and time,
which is particularly important since mobile devices
have certain limitations that affect the usability for
their users.
Social media too are becoming an important part
of everyday life. One goal of social navigation is to
utilize information about other people’s behavior for
our own navigational needs (Dieberger, 2003).
Social media encourage people to use resources
from these social networks and apply them to their
real lives, to avoid isolationism, and to understand
social trends by observing the users' contents on
their social networks. When Dourish and Chalmers
introduced the concept of social navigation in
(Dourish and Chalmers, 1994), they defined it as
“navigation towards a cluster of people or navigation
because other people have looked at something”. In
social navigation, people watch the activity of other
people to make choices about what is popular, to see
which paths to follow, and to find links to related
information (Hook et al., 2003). Social navigation
can potentially transform different spaces by
encouraging people to explore those spaces that
might otherwise be ignored or overlooked (Gay,
2009).
176
Utku S. and Eren Atay C..
A Mobile Location-Aware Recommendation System.
DOI: 10.5220/0005053001760183
In Proceedings of the International Conference on Knowledge Discovery and Information Retrieval (KDIR-2014), pages 176-183
ISBN: 978-989-758-048-2
Copyright
c
2014 SCITEPRESS (Science and Technology Publications, Lda.)
Data mining is the application of specific
algorithms for extracting patterns from data. Its main
goal is to extract information from a data set and
transform it into an understandable structure for
further use. Data mining is also used in location
management wherein the mobility patterns are
determined to predict the next location of the mobile
user in (Yavaş et al., 2005). In essence, activity
selection can be used in the development of a
collective spirit. Social media provides a rich source
of such information, for marketing professionals and
others interested in extracting opinions.
In this study, a mobile platform, LOAS
(LOcationAwareSystem) is presented, for
monitoring and collecting information about
people’s activities. We suggest not only a context-
aware mobile location-based service, but also one
that employs data mining techniques to relate them
to their hobbies and activities. In addition to places
and venues, collective activity is represented in real
time on a public map as a social-navigation
recommendation and guidance system. As far as we
know, there is currently no user study that
exclusively investigates people’s location-based
search motivations and the opportunities for context
aware data mining. In order to improve and optimize
location-based services, it is necessary to understand
people’s location-based information needs and the
context in which they occur. With this in mind, we
have implemented search recommendations based
on users' hotspots and social networks, with respect
to information privacy and security.
The rest of this paper is organized as follows.
Section 2 discusses some related work. In Section 3,
we describe the architectural, functional, and
implementation aspects of LOAS. The results of our
experimental evaluation are provided in Section 4.
In Section 5, we present our conclusions and plans
for future work.
2 RELATED WORK
Research topics related to our system include
location awareness, context-aware recommender
systems and campaigns in social network, mobile
information access and collaboration of these in data
mining.
Location used to be considered as important
context information (Baldauf et al., 2007). Location
awareness is a general term used for something that
can show that it is aware of your current location
services built on the location awareness capabilities
of mobile devices and networks (Lee et al., 2006).
Location detection techniques, including from
indoor and outdoor is provided in (Hightower and
Borriello, 2001). The major issue is that location
detection is hard to use in popular applications and
done in an ad-hoc way since it has no general
method for every device. (Fischmeister, 2003) uses
the concept of location awareness in which the
services can be accessed by the user with their user
agent. The infrastructure of the system provides
supports for proactive location-aware services.
Recommender systems (RS) are tools and
techniques used in various web-based applications to
provide users suggestions for items to be chosen
(Ricci et al., 2011). Recommender system consists
of three key components - users, items and user-item
matching algorithms. The utility of an item is
usually represented by a rating, measuring how
much a specific user is (or is predicted to be)
interested in a specific item. Depending on the
application, the ratings can either be specified by the
users, or computed by the application. RS techniques
are classified into six different classes of
recommendation approaches; context-based
collaborative filtering, demographic, community-
based and hybrid (Burke, 2007). Suggested system
uses hybrid recommendation approach because,
recommender is not only giving context-based
information, but also makes recommendations based
on personal interest. The issue of finding hidden
links between users and items based on the
similarity of the user preferences/interests and the
item content features is the essence of the already
presented hybrid recommender systems. Contextual
information is important to facilitate suggestions for
users. The context-aware RS will take into account
companion, time, atmospheres and weather that help
user to make better choice. In our system, we
integrated time, location and users’ social network
preferences into contextual information.
Context awareness has been the focus of many
research areas, especially in social network
structures. Most of the available applications focus
on human factors and physical environment. While
human factors can be categorized into information
on the user, the users’ social environment, the user’s
task; physical environments are structured into
location, infrastructure and physical conditions
(Bellavista et al., 2012). Context information is
shared by subscribing a platform designed using the
WASP Subscribing Language in (Costa et al., 2004).
Chen and Kotz introduced middleware, Solar, in
which a platform for context aware mobile
applications consisting of one star and several planet
nodes in (Chen and Kotz, 2002). Client applications
AMobileLocation-AwareRecommendationSystem
177
subscribe to context changes at the central star rather
than collect, aggregate or process context
themselves.
Our approach also related with mobile
information access that there are some remarkable
studies of user behavior of mobile devices. In
Church et al. (Church et al., 2008), the result of a
detailed analysis of mobile search behavior of over
2.6 million European mobile subscribers presented
that among those 11% executed at least one request.
This study was the first to analyze the click through
behavior of mobile searches. Result of this research
had interesting conclusions; mobile search is used by
only 8-10% of mobile internet users, mobile search
engines have widely adopted a traditional Web-
based approach to search, mobile searches queries
are short and users tend to focus on the first few
search results. Yi et al. investigated the search
characteristics from mobile devices (Yi et al., 2008).
The authors listed interesting points and trends from
this study; (i) there is the evidence of high variability
of mobile query patterns, (ii) statistics show
significant variations in the regional query patterns,
and (iii) the usage patterns are dynamic because
users are still figuring out how to take advantage of
new mobile devices and services. Another research
was the result of a Web-based diary study about
location based behavior search through a mobile
search engine illustrated in (Amin et al., 2006).
Authors were able capture users' explicit behavior
and implicit intention as well as spatial, temporal,
and social context of search. The results of this study
show that location-based searches are usually relied
on just-in-time information needs that are closely
related to social activity. Among the main findings
of this study, most location-based searches on
mobile devices are performed when users are along
with other people, such as relatives, friends, and
colleagues. The importance of taking into account
user information needs and context in search and
recommendation processes has inspired our
approach.
3 A FRAMEWORK FOR A
CONTEXT-AWARE SOCIAL
NETWORK
In this section, we will describe an intelligent mobile
location-based application which derives activity
data according to the users’ current locations and
their previously-attended activities, as well as
associated information from user-generated posts
online. With this approach, it helps users navigate
standard environments, as well as temporary events,
such as fairs, concerts, movies, and theaters in
particular. In general, the research platform
application is a social mobile service allowing users
to post their current location, and for suggestion
marks to be shown on a map, which can then be
utilized for social navigation. A typical use case
scenario is that a user wants to inform other users
about a positive or negative place, by inputting the
tag. Another scenario is that a user is curious and
wants to see where other users are located or what
are the best and the worst places to spend on a
specific day. Our application also has a built-in
calendar section, with which the user can arrange
his/her activities, as a smart calendar.
Figure 1: LOAS architecture.
LOAS is designed to support the following goals: i)
Suggest to users depending on their previous
choices. ii) Suggest to users depending on social
groups in which they are members. iii) Track all
status changes and share them with their social
groups. iv) Allow users to view the map. v) Allow
users to share activities and places, and to tag them.
vi) Allow users to organize daily dates and tasks on
a context-aware calendar. vii) Design an application
that should be understandable and manageable by
the typical user.
3.1 Recommender System
Recommender systems (RS) are tools and
techniques used in various applications to provide
users with suggestions for items to be chosen. Our
RS application has three main tasks. Firstly, it is
responsible for learning a user's friendships and
KDIR2014-InternationalConferenceonKnowledgeDiscoveryandInformationRetrieval
178
subsequently revealing hidden friendships about
previous activities and places in a given social
group. Secondly, it learns the user's location, and it
filters activities listed for a specific area. Thirdly, it
filters activities according to the data mining results
at any given time. We will now discuss these three
tasks and our solutions, in turn.
Data mining with similar user profiles: The
module includes the functionality of learning the
user, collecting data between all the users, and then
generating Association Rules based on the users’
data. The purpose of association rule mining is to
find frequent patterns, associations, correlations, or
casual structures of sets of items or objects in any
transaction database. The well-known Apriori
Algorithm was chosen for the association rule
mining; the implementation was written in C#. The
user reaches the suggestion list by calling a web
service and suggestion list function in WCF. This
service operates the data mining on the database of
LOAS.
Figure 2: Two-phased data mining of LOAS.
The processing of our algorithm has two phases of
data mining, which is illustrated in Figure 2. In the
first phase, the user-related data (interest, profiles,
etc.) are collected in the LOAS database. The
preprocessing was done on the server side, such that
it satisfies the user profile’s interest in the mobile
user and the cleaning of the data with any missing
values. This organizes the data for the next phase,
and it makes clear the related information before the
mining process. The second phase involves
processing the data using the association algorithm
and creating the association rules on the information
gathered. After the process, the results are returned
to the user as the basis of interest to mine the
relevant information.
Filter activities by Data Mining Results: The
LOAS system collects user preferences and exploits
this information to provide personalized
recommendations about points of interest in the
surroundings of the user’s current position.
Consider the set of user activities that will be
used in the data mining: Activity = {a
1
, a
2
, …, a
n
}.
After collecting the user activities, these features are
used to select the relevant attributes of User_i where
it is the raw data from the LOAS database. Let
SelectedTuples be the set of the selected tuples from
User_i.
User data are collected in the first phase of the
data mining by selecting the tuple(s) that satisfy the
Activity that will be included in SelectedTuples.
Then, these data will be processed by deleting the
missing values of the tuple(s). There are some
preprocessing techniques that fill up the missing
values, but this may consume a prohibitive amount
of time for processing. As a second phase in the
Data Mining, the association rule algorithm is
implemented on the data sets of the preprocessed
data. The association rules creates a rule set based
on the associated data sets. After the results are
acquired, the output of the data mining is sent to the
mobile side.
Filter activities by Location Information: Mobile
recommender systems make recommendations based
on a users’ interests and present location. Identifying
the position of the user is the first step to providing
location-aware services. Estimating location can be
done in several different ways. The user can be
located by the GPS (Global Positioning System)
module of the user’s device, which provides
accuracy of 1.5 – 20 meters. However, GPS data is
not always available indoors, and it may not work in
all urban areas. The second way is cell-based
solutions using a GSM (Global System for Mobile
Communications) base station in the network. The
position can be calculated based on the mobile
network cells. Other methods for locating the user
are device-based solutions that use the device's
media access control systems, such as WLAN
(Wireless Local Area Network, Bluetooth, or 3G).
There is a service working in the background of
LOAS detecting to the user’s current location every
5 minutes. Users are suggested activities according
to their profiles within a 25-km diameter area. All
activities are filtered using the Haversine Algorithm
(Sinnott, 1984). Furthermore, users are provided
with a map presenting a list of relevant activities and
places, including how to get to any selected activity
by car or by public transportation using the shortest
distance.
AMobileLocation-AwareRecommendationSystem
179
3.2 Server Side
Since mobile devices have less computational power
than desktops as well as relatively smaller screens,
in order to improve the response time of the system,
most of the heavy tasks have been put onto the
server side. The server has tables storing the users’
personal information, social networks the users are
members of, and lists of activities and places. These
data might also be used for other web-based
services.
3.3 Mobile Application
The LOAS mobile agent middleware provides an
advanced infrastructure that integrates the
application with core services and tools, to permit
communication to mobile elements. The mobile
agent core services layer consists of components for
position acquisition and filtering, background
services, a content provider, and the LOAS database.
The LOAS application layer has a social network
extension, a suggestion system, and a location-aware
component.
Figure 3: LOAS mobile agent middleware.
The general structure of our proposed model is
shown in Figure 3.
Figure 4: Structure of the LOAS model.
Activities: An activity has a name, an
explanation, a start date, a finish date, and its
location information. There are Hobbies in the
database of the system, which are added beforehand
and maintained statically. Thus we can give a hobby
type to each activity when any user adds a new
activity. An activity detail page has information
about an activity, a check-in button, a rating bar, the
user that created this activity, and top visitors of this
activity. Also, the user can see and add comments,
shown in the map, and invite friends to this activity
using the tabs on the bottom of the page.
MyCalendar: MyCalendar is a typical calendar
application, similar to many other calendar apps that
allow a user to organize daily dates and tasks. We
have included more functions to make it context-
aware. Typical calendar apps already provide the
concept of user-defined categories. For each date
entry, a user can define some planned activities. The
activities of the user are added to the calendar
according to their start date. When any activity
starts, the user is warned, via an alert. Any activity
that is added to the calendar is signed on the
interface. The color style changes according to the
selected activities. There are different ‘suggested’
and ‘selected’ color styles. Suggested activities that
are more relevant to the user’s activity list are
displayed more obtrusively (e.g., in bold), whereas
other activities are presented unobtrusively or even
hidden. Figure 5 shows an example MyCalendar
page containing some activities that are displayed in
purple (suggested by the RS system), while some
activities are in orange (those that have been
accepted). The recommendation system suggested a
Dance activity rather than the Art activity on June
10th because it takes more than 30 minutes to get to
where the Art activity is being held.
a)Listed activities b)Suggested
activities
c)Direction to
selected activity
Figure 5: Calendar page some activities are purple
(suggested by RS), some activities are orange (accepted).
Friends: With a Facebook extension, the friends’
activities and comments are collected and integrated
into our system. While a preferable activity list is
KDIR2014-InternationalConferenceonKnowledgeDiscoveryandInformationRetrieval
180
being prepared, the filtering is done according to the
user profile and the choices of similar profiles. Two
main processes are used for find matching profiles.
At first, the initial user interests can be collected
during the first registration by using the user's
personal identification data. Then, the transaction
data that is provided by users is added to the
database. The recommender system generates the
user’s possible activity list by using data mining
according to the user’s personal information,
location information, and activities.
Places: A place has a name, an explanation, some
location information describing where this place is
located, and a Place Type which gives information
about the type of this place. These Place Types are
kept in the server database statically. Users can add
a new place to this information. Users can list places
by a place type, within 1, 3, 5, 7, 10, and 15
kilometers distance.
4 MOBILE LOCATION-AWARE
MODEL
4.1 Data Mining
We proposed a framework for a location-based
application that uses data mining functionality to
provide users with suggestions depending on their
current locations and their previous activities. The
proposed model consists of two phases: the
generation of the activity data from the user profile,
and the activity suggestion based on the generated
rules. The LOAS database has a user attended
activities table storing that data for each user, as
shown in Table 1.
Table 1: User activities obtained from the LOAS database.
User ID User Activities
1 {cinema, dance, trip}
2 {tennis, fair,concert, theatre, art}
3 {concert, trip}
4 {fair, concert, art}
.. ..
We have applied the Apriori Algorithm in which the
main objective is to extract useful information from
large amounts of data. In our work, we collect the
activities of a user in the form of A = {((id
1
, a
1
), (id
1
,
a
2
)... (id
1
, a
n
)), ((id
2
, a
1
), (id
2
, a
2
),.., (id
2
, a
n
)), ....,(id
k
,
a
k
))} where id
1
denotes the ID number of the user
who engages in activities from a
1
to a
n
.
The execution of the Apriori mining algorithm
with supp
min
= 4 instances and using the activity
table, is illustrated in Tables 2 through Table 4. The
candidate patterns (C
1
), with a set of length-1 and
the large patterns (L
1
), are depicted in Table 2.
Table 2: Length-1 candidate patterns (C
1
) and length-1
large patterns (L
1
).
C
1
L
1
CANDIDATE SUPPLY PATTERN SUPPLY
{tennis} 12 {tennis} 12
{theatre} 11 {theatre} 11
{dance} 10 {dance} 10
{trip} 13 {trip} 13
….
Next, using the candidate generation algorithm
generates C2. In this algorithm initially the
candidates set is empty for each length-k large
pattern L determine all the cells which are neighbors
of Ik in G for each of these neighbor cells, v
generate a candidate by attaching v end of L. L1 is
used to generate C2 . Then, the supports of these
candidates are counted, and the patterns which have
a support value larger than suppmin are assigned to
the set L2. The sets C2 and L2 are presented in
Table 3.
CandidateGeneration()
// Generation of length-(k + 1) candidates
1. Candidates = ∅
2. foreach L = <I1, I2, ….., Ik>, L ∈ Lk
3. N+ = { v | there is an edge in G such as
Ik v }
4. foreach v ∈ N+(Lk)
5. {
6. C’ = < I1, I2, ….., Ik, v>
7. Candidates
Candidates C’
8. }
9. return Candidates;
Table 3: Length-2 candidate patterns (C
2
) and length-2
large patterns (L
2
).
C
2
L
2
CANDIDATE SUPP. PATTERN SUPP.
{tennis, theatre} 6 {tennis, cinema} 9
{tennis, dance} 2 {tennis, football} 5
{tennis, trip} 3 {tennis, concert} 6
{tennis, fair} 1 {tennis, theatre} 6
{tennis, concert} 6 {cinema, football} 6
{tennis, art} 1 {cinema, concert} 7
….
Having L2, C3 is generated using the
CandidateGeneration( ) function, and then the large
patterns in C3 are assigned to the set L3. These sets
are shown in Table 4.
All possible rules and their confidence values for
the activities are demonstrated in Table 5, as well as
AMobileLocation-AwareRecommendationSystem
181
Table 4: Length-3 candidate patterns (C
3
) and length-3
large patterns (L
3
).
C
3
CANDIDATE SUPPLY
{tennis, cinema, football} 4
{tennis, cinema, concert} 3
{tennis, cinema, theatre} 4
{tennis, cinema, dance} 2
{tennis, cinema, art} 0
….
L
3
PATTERN SUPPLY
{tennis, cinema, football} 4
{tennis, cinema, theatre} 4
{cinema, theatre, trip} 4
the set of length-4 candidate patterns, C
4
. If the
threshold confidence value, conf
min
is assumed to be
60, then the rules having a confidence bigger than or
equal to conf
min
will be the same as the rules in
Table 5, since all these rules have a confidence
bigger than conf
min
.
Table 5: All Possible Rules.
RULE CONFIDENCE
{theatre, trip} {cinema} 0.95
{dance} {cinema} 0.89
{theatre} {cinema} 0.82
{tennis, football} {cinema} 0.80
{tennis} {cinema} 0.75
{trip} {cinema} 0.67
{art} {concert} 0.67
{cinema, football} {tennis} 0.67
{tennis, theatre} {cinema} 0.67
4.2 Results of Data Mining
Our experiments are based on data obtained from 37
graduate and undergraduate students’ activities for 2
months. The results were evaluated utilizing
association rule mining techniques. Table 1 presents
the activity sequences entered by the users. Once
these values were analyzed, candidate (C1) item sets
with 10 different activity frequencies were
determined, as depicted in Table 2. Among these
groups, the movie with 23 instances has the
maximum value; the rally with 3 instances has the
minimum activities. When the value of suppmin is
chosen as 4, the activity rally with 3 instances is
removed from the length-1 large patterns (L1) set.
The rest of the cluster has created the large-1 item
set with 9 items.
There are 33 pairs in Candidate-2 item sets, in
which 14 of them have passed the threshold value,
supp
min
=4. In order to have a cluster of 3 pairs, 17 of
them were analyzed and 3 of them had passed the
minimum threshold level, and thereby formed
Large-3 pattern. A cluster with the candidate-4 item
set could not be found by using the Apriori
Algorithm. Table 4 depicts the last established item
set.
The rules obtained from these item sets are
shown in Table 5. The minimum confidence value is
taken as 60% when calculating the association rules.
By using this confidence value, a total of 14 rules
were obtained. According to the trends of these
users, it was observed that everyone who attends the
theatre and goes on a trip also goes to 100% of the
movies, and everyone who attends the theatre 82%
also goes to movies.
5 CONCLUSIONS AND FUTURE
WORK
Mobile social networks have been a challenging
research direction in recent times. Although the
number of users involved in social networks has
increasing tremendously in recent years, these
participants are lacking in coordination and
collaboration. It is hard to quickly and easily discern
where other friends in one's social network are
located, what other activities are available for a
given location and time, who else is participating in
those activities, and what are the shortest paths to go
there.
In this paper, we presented a mobile application
for monitoring and collecting information about
people’s activities in the social media space, to
facilitate user-centric collaboration and coordination.
Three promising contextual attributes - time,
location, and interest – have been taken into
consideration. We have designed a specialized
application for these social networks for use with
mobile devices, augmented with recommendation
systems and a data mining algorithm. Our proposed
model suggests to users various activities and places
depending on their social groups, their previous
choices, and their current location, by using data
mining techniques. The algorithm proposed is based
on mining the user profile, forming activity rules
from these patterns, and finally suggesting a mobile
user’s activities by using the mobility rules.
Suggested activities can be viewed on the map, and
directions can also be provided. Additionally, our
research shows that using the data mining function
enhances the context and location awareness of
KDIR2014-InternationalConferenceonKnowledgeDiscoveryandInformationRetrieval
182
mobile users. Another benefit of LOAS is that it will
enable users to choose more appropriate activities
and to share them with their friends, to socialize
more efficiently. The results of an evaluation
performed on real users show that the proposed
approach provides significant benefits in terms of
effectiveness compared with nonpersonalized
recommendation algorithms.
In this study, experimental tests have been
conducted to demonstrate the accuracy and
feasibility of the study. The next stages of the study,
people can be grouped according to the activities by
using k-means algorithm. Then, by using appriori
algorithm, more detailed guidance to these groups is
planned. The study was conducted to demonstrate
the use of contex-aware system by using collected
data in the mobile environment. Nowadays, it is
believed that the people can be consciously
directed with the participation of increasingly
widespread and social networks. In this preliminary
study, positive results have been taken in this
direction. In the next stages, evaluation of the results
is planned on more subjects by using clustering and
association rule mining. On the other hand, we plan
to combine the users’ demographic data and activity
trends, to be investigated in detail in future studies.
In addition, we intend to prepare a model for the
creation of specific groups of users and the
development of helpful guidance for new members
in these groups. Thus, we will be able to analyze and
make suggestions based on not only users but also
groups. Furthermore, augmented reality
implementations have become more widespread in
various mobile environments. In order to present
activity-related information more effectively, we
intend to include additional application into LOAS
to track the users' impacts.
REFERENCES
Rao, B., Minakakis, L. 2003. Evolution of mobile
location-based services, Communications of the ACM.,
46(12), pp.61–65.
Fahy, P., Clarke, S., 2004. CASS a middleware for
mobile context-aware applications. In Workshop on
Context Awareness, MobiSys.
Dieberger A., 2003. Social connotations of space in the
design for virtual communities and social navigation.
Computer Supported Cooperative Work, Springer
London, pp. 293–313.
Dourish, P., Chalmers, M., 1994. Running Out of Space :
Models of Information Navigation. HCI'94, Glasgow,
UK, pp.2–3.
Hook, K., Benyon, D., Munro, A., 2003. Designing
Information Spaces: The Social Navigation Approach.
Berlin: Springer-Verlag.
Gay, G., 2009. Context-Aware Mobile Computing:
Affordances of Space, Social Awareness, and Social
Influence. Synthesis Lectures on Human-Centered
Informatics, pp. 1–62.
Yavaş, G. et al., 2005. A data mining approach for
location prediction in mobile environments. Data &
Knowledge Engineering, 54(2), pp.121–146.
Baldauf, M., Dustdar, S., Rosenberg, F., 2007. A Survey
on Context Aware Systems. Int. J. Ad Hoc Ubiquitous
Comput., 2(4), pp.263–277.
Lee, J. et al., 2006. Location-Aware Agent Using Data
Mining for the Distributed Location-Based Services.
Computational Science and Its Applications – ICCSA.,
Springer Berlin Heidelberg, pp. 867–876.
Hightower, J., Borriello, G., 2001. A Survey and
Taxonomy of Location Systems for Ubiquitous
Computing. University of Washington, Seattle, WA.
Fischmeister, S., 2003. Mobile Software Agents for
Location-based Systems. In Proceedings of the NODe
Agent-related Conference on Agent Technologies,
Infrastructures, Tools, and Applications for E-
services. NODe’02. Berlin, Heidelberg: Springer-
Verlag, pp. 226–239.
Ricci, F., et al., 2011. Recommender Systems Handbook.
Springer.
Burke, R., 2007. The Adaptive Web – Hybrid Web
Recommender Systems. Springer-Verlag, pp. 377–
408.
Bellavista, P., et al., 2012. A Survey of Context Data
Distribution for Mobile Ubiquitous Systems. ACM
Computing Surveys., 44( 4):24.
Costa, P., et al., 2004. Towards a service platform for
mobile context-aware applications. ICUW, pp. 48–61.
Chen, G., Kotz, D., 2002. Solar: An open platform for
context-aware mobile applications. In Proceedings of
the First International Conference on Pervasive
Computing, pp. 41– 47.
Church K., et al., 2008. A Large Scale Study of European
Mobile Search Behaviour. MobileHCI’08, Amsterdam,
the Netherlands, pp.13–22.
Yi, J., Maghoul F., Pedersen, J., 2008. Deciphering
Mobile Search Patterns: A Study of Yahoo! Mobile
Search Queries. International World Wide Web
Conference Committee (IW3C2), Beijing, China,
pp.257–266.
Amin A., et al., 2006. Fancy a Drink in Canary Wharf?: A
User Study on Location-Based Mobile Search.
Proceedings of the 12th IFIP TC 13 International
Conference on Human-Computer Interaction, pp 736-
749.
Sinnott, R.W., 1984. Virtues of the Haversine. Sky and
Telescope, 68(2), p.159.
AMobileLocation-AwareRecommendationSystem
183
