PreprintPDF Available

Collaborating with Users in Proximity for Decentralized Mobile Recommender Systems

Preprints and early-stage research may not have been peer reviewed yet.
Preprint

Collaborating with Users in Proximity for Decentralized Mobile Recommender Systems

Abstract and Figures

Typically, recommender systems from any domain, be it movies, music, restaurants, etc., are organized in a centralized fashion. The service provider holds all the data, biases in the recommender algorithms are not transparent to the user, and the service providers often create lock-in effects making it inconvenient for the user to switch providers. In this paper, we argue that the user's smartphone already holds a lot of the data that feeds into typical recommender systems for movies, music, or POIs. With the ubiquity of the smartphone and other users in proximity in public places or public transportation, data can be exchanged directly between users in a device-to-device manner. This way, each smartphone can build its own database and calculate its own recommendations. One of the benefits of such a system is that it is not restricted to recommendations for just one user - ad-hoc group recommendations are also possible. While the infrastructure for such a platform already exists - the smartphones already in the palms of the users - there are challenges both with respect to the mobile recommender system platform as well as to its recommender algorithms. In this paper, we present a mobile architecture for the described system - consisting of data collection, data exchange, and recommender system - and highlight its challenges and opportunities.
Content may be subject to copyright.
Collaborating with Users in Proximity for
Decentralized Mobile Recommender Systems
Felix Beierle
Service-centric Networking
Telekom Innovation Laboratories
Technische Universit¨
at Berlin
Berlin, Germany
beierle@tu-berlin.de
Tobias Eichinger
Service-centric Networking
Telekom Innovation Laboratories
Technische Universit¨
at Berlin
Berlin, Germany
tobias.eichinger@tu-berlin.de
Abstract—Typically, recommender systems from any domain,
be it movies, music, restaurants, etc., are organized in a cen-
tralized fashion. The service provider holds all the data, biases
in the recommender algorithms are not transparent to the user,
and the service providers often create lock-in effects making
it inconvenient for the user to switch providers. In this paper,
we argue that the user’s smartphone already holds a lot of the
data that feeds into typical recommender systems for movies,
music, or POIs. With the ubiquity of the smartphone and other
users in proximity in public places or public transportation, data
can be exchanged directly between users in a device-to-device
manner. This way, each smartphone can build its own database
and calculate its own recommendations. One of the benefits of
such a system is that it is not restricted to recommendations for
just one user – ad-hoc group recommendations are also possible.
While the infrastructure for such a platform already exists –
the smartphones already in the palms of the users – there are
challenges both with respect to the mobile recommender system
platform as well as to its recommender algorithms. In this paper,
we present a mobile architecture for the described system –
consisting of data collection, data exchange, and recommender
system – and highlight its challenges and opportunities.
Index Terms—Social Networking Services; Ubiquitous Com-
puting; Mobile Computing; Smartphones; Context Data; Device-
to-Device Communication; Recommender Systems
I. INTRODUCTION
Recommender systems are ubiquitous in several different
domains of everyday life. They recommend restaurants to
go to (e.g., Yelp, Google Maps), music to listen to (e.g.,
Spotify, Deezer), or movies to watch (e.g., IMDb, Netflix). The
recommender algorithms usually are limited to recommending
items that the platform offers (Spotify, Netflix) or at least items
that are indexed with the provider (Yelp, Google Maps). There
might be certain biases, for example, towards recommending
products that create the biggest margin for the service provider.
Furthermore, as the service provider is interested in retaining
its users, there are certain lock-in effects that try to make
the user stay with the current service provider instead of
switching to another one. In that sense, from an organizational
perspective, the examples given above are centralized – a
single service provider has all the data and decides how
recommendations are calculated.
On top of that, users have privacy concerns regarding the
use of the data they share [1]. Recently, Google was fined 50
million euros due to a violation of the new European privacy
laws (GDPR)1.
The infrastructure that could be a solution for both chal-
lenges of lock-in effects and privacy concerns at the same time,
is already in the palms of its users. The smartphone can store
lots of information about its user and his/her interests, e.g.,
regarding preferred restaurants, music, or movies. Equipped
with capabilities for device-to-device communication, users
can exchange data between each other. When considering
recommender systems based on content-based filtering or
collaborative filtering, data about similar items and similar
users is needed. Data about the properties of items can be
retrieved through public APIs (e.g., Google Places, Spotify,
Open Movie Database (OMDb)). Finding similar users might
be simple with smartphones: spending time at the same
location might imply similarity – at least to a certain degree.
Additionally, from our previous research, other methods of
determining similarity between users based on smartphone
data are available [2], [3]. Thus, exchanging data between
smartphones in proximity in a device-to-device fashion allows
to create local databases that allow to filter for similar users.
This data can be used for on-device recommender systems that
are independent of external service providers.
Typically, each of the mentioned existing recommender
systems only offers recommendations for a single user. The
ubiquitous system we propose in this paper offers the possi-
bility of spontaneous ad-hoc recommendations for groups of
users in proximity.
Combining and expanding approaches from device-to-
device computing [4], [5] and decentralized recommender
systems [6]–[8], in this paper, we propose a modular archi-
tecture for recommender systems for virtually any domain,
building on the existing infrastructure of smartphones. The
architecture consists of collaborative data collection paired
with data exchange via device-to-device communication and
local recommender systems running on each device, supported
by third-party service providers where appropriate. There are
some challenges to overcome when developing such a plat-
1https://www.cnil.fr/en/cnils-restricted-committee- imposes-financial-
penalty-50- million-euros- against-google- llc
form. Some data is already readily available on smartphones,
for example the most frequently visited locations. Other
user preferences/ratings that cannot be assessed automatically
might have to be entered manually or retrieved from external
service providers, e.g., music listened to or favorite movies.
While some short-distance wireless technologies, like NFC,
Bluetooth, or WiFi Direct, are available on most modern
smartphones, data exchange between users remains a challenge
to be implemented for multiplatform apps.
The main contributions of this paper are:
proposing a general modular architecture for a service-
provider-independent mobile platform for recommender
systems
developing a comprehensive approach for collecting data
and exchanging data in a device-to-device fashion for
multiplatform apps (Android and iOS) to enable domain-
independent recommender systems
The remainder of this paper is structured as follows: In
Section II, we introduce a general architecture for the proposed
mobile recommender system platform. In Section III and IV,
we illustrate our approach for collecting and exchanging data.
In Section V, we highlight what challenges and opportunities
the recommender systems in our proposed platform will face,
before pointing out related work in Section VI.
II. PRO PO SE D ARCHITECTURE
In order for the described architecture to be feasible, there
are two main requirements:
R1 The system has to be multiplatform.
R2 The system has to be designed in a modular way.
Android and iOS are the remaining relevant mobile operating
systems. They have a combined market share of about 100%2.
The main challenge this brings is related to data exchange, as
we describe in Section IV. The system should be developed
in a modular way (R2) in order to be able to exchange
components easily. Consider short-distance wireless interfaces:
In the past, infrared was used. Technological advances now
offer larger transmission ranges, shorter connection times, and
higher bandwidths via, for example, Bluetooth or WiFi Direct.
Similarly, advances in recommender systems and machine
learning might offer better recommendations, creating the need
to replace the module or offload certain tasks to components
available from external service providers.
In Figure 1, we illustrate our proposed general modular
architecture. The three main components of the system are
Data Collection,Data Exchange, and Recommender System.
Data Collection is responsible for getting data about the
user. Data Exchange is responsible for getting data from
other users. The Recommender Systems utilizes all available
data for recommending items to the user. The mobile OS
provides components for sensors (for example for tracking the
user’s location for inferring his/her favorite POIs) and wireless
interfaces (for exchanging data).
2https://www.statista.com/statistics/266136/global-market-share-held-by-
smartphone-operating- systems/
As we further analyze in Section III and IV, external
service providers might be needed (or be useful) in order
to retrieve metadata about items, utilize existing systems, or
offload data or computational tasks. Figure 1 shows dashed
lines for optional connections to third party service providers.
Data Collection might use this to retrieve data about the user
or to enrich already available data, e.g., find out the genre of
the songs the user listened to. More details are given in Section
III. The Recommender System can optionally be relayed to an
external service provider.
Mobile Component
Application
Mobile OS
Local DB
Data Collection
Data Exchange
Recommender System
Sensor APIs
Wireless Interfaces
App Data
Other Devices
Fig. 1: Architecture components of the proposed system.
III. COLLECTING DATA
We identify three different possibilities to retrieve user data:
a) Tracking data automatically: Smartphones contain a
multitude of sensors that are often used for context-aware
applications. Frameworks like the Google Awareness API3
yield the location, weather, etc. for the user. Such data can
easily be tracked and so preferences or implicit ratings for,
e.g., locations or POIs can be inferred. In previous work,
we demonstrated an Android application that tracks a large
variety of context data sources [9]–[11]. The data that can be
tracked automatically on iOS might differ. In order to create
a multiplatform system and ensure that the same data points
are available on all systems, additional ways of retrieving the
user’s ratings are necessary. Figure 2a shows the sequence
diagram of automatic context data tracking.
b) Retrieving data from existing service providers: In
order to minimize necessary user effort, the second method we
suggest is retrieving data from existing service providers. For
example, Spotify’s API enables application developers to fetch
recently played tracks4. Regularly doing this yields a complete
music listening history indicating implicit user ratings. Figure
2b shows the sequence diagram for the collection of data from
a third-party service provider.
3https://developers.google.com/awareness/
4https://developer.spotify.com/documentation/web-api/reference/player/get-
recently-played/
Alice
Smartphone
loop
collect data
(a) Automatic context
data tracking.
Alice
Smartphone
Service Provider
authorize Service Provider
loop
[in predened frequency]
request data
data
(b) Retrieving data from third-party service
providers.
Alice
Smartphone
Catalog Service Provider
get items to show to user
return items
show items
rate items
(c) Letting the user manually rate items.
Fig. 2: Data collection mechanisms.
c) Letting the user manually rate items: For data that is
neither automatically trackable nor available via third parties,
the user should be able to enter it manually. By defining
an ontology for categories and terms that can be exchanged
between users, compatibility between the data from different
collection methods can be ensured. Pre-defined categories can
be movies, music, or restaurants, where recommender system
are often used, but any other category would be possible too.
Service providers like The Open Movie Database API5, for
example, can be used to help employ globally valid identifiers
for each item, in this case, for each movie. Figure 2c shows the
sequence diagram for manual data collection. Catalog Service
Provider denotes a service provider that offers structured
information about a specific category, like the mentioned Open
Movie Database.
IV. EXC HA NG IN G DATA
The idea behind the exchange of data is that when users
pass each other, their preference data, i.e., their ratings, are ex-
changed automatically in a device-to-device manner, building
up each user’s local database with more data. In order for such
a system to work unobtrusively and without user interaction,
the exchange of data should be done in the background
(broadcasting), without establishing explicit connections be-
tween smartphones. While device-to-device communication
has gained some attention in research, practical application
is still lacking, especially when considering broadcasting, and
especially when considering R1 (multiplatform app for both
Android and iOS). In this section, we give an overview of
related work, related applications, and propose a technical
solution to be used in our proposed architecture.
During the advent of mobile phones, researchers suggested
using device-to-device peer discovery and communication
in order to stimulate social interactions. There are several
papers between 2005 and 2010 describing exchanging data
via Bluetooth, sometimes combining the direct data exchange
with retrieving data from a central server [12]–[15]. In both
[16] and [17], the authors suggest using WiFi SSIDs and
the Bluetooth discovery protocol in order to exchange data
between devices. Only very small amounts of data can be
5http://www.omdbapi.com/
transferred that way but the benefit is that no proper connection
has to be established between two devices, thus allowing for
broadcasting data to devices in proximity.
With the advent of smartphones, wireless interfaces im-
proved and computing power increased, thus making it worth
looking into more recent publications. A lot of work related
to device-to-device communication focuses on challenges like
offloading, content dissemination, or energy efficiency rather
than the application layer [4]. Yet, there are some projects
related to applications that describe actual implementations
or suggestions for implementations. In [18], the authors use
the WiFi ad-hoc mode on an Android device. This mode is
not available by default, an extension had to be compiled
into the Linux kernel. There is still a lack of support of
WiFi ad-hoc since publication of that paper (2016), which
shows that only a very limited set of devices would be
able to run such an application. In [19], the authors use
Bluetooth Low Energy (BLE) in IoT scenarios for ad-hoc
communication. The framework they develop allows devices to
communicate without predefined roles like client/server. Other
recent works suggest using WiFi Direct for exchanging data
between smartphones [20], [21], which is not readily supported
on iOS devices.
Besides the academic work in the field, it is also worth
looking into what related applications are actually available.
In the following, we thus look into existing applications
and technologies for device-to-device data transfer. Overall,
such technologies seem to only operate with devices from
the same manufacturer or the same mobile OS. Hand-held
gaming devices from Nintendo and Sony are offering data
exchange with nearby players in proximity6. This is a feature
specific to each gaming system and does not work across
devices from Nintendo and Sony. Both Apple and Google
offer frameworks that enable software developers to interact
with nearby devices. Apple’s framework is called Multipeer
Connectivity7and only works with other Apple devices.
Google introduced its framework Google Nearby8with two
different APIs: Nearby Connections and Nearby Messages.
6https://www.nintendo.com/3ds/built-in-software/streetpass/how-it- works
and http://us.playstation.com/psvita/apps/psvita-app- near.html
7https://developer.apple.com/documentation/multipeerconnectivity
8https://developers.google.com/nearby/
Nearby Connections allows for device-to-device data transfer,
but only between Android devices. Nearby Messages is only
available when the devices are connected to the internet and
allows only small payloads, but is available for both Android
and iOS. There are multiplatform apps offering device-to-
device data transfer solutions, e.g., SHAREit9. Such apps often
let one user open a WiFi hotspot that is then joined by a second
user. Data exchange via Bluetooth between Android and iOS
devices is not readily available and typically requires explicit
user interaction.
Reviewing related work in academia, software development
frameworks, and apps, we summarize that the issue we are
facing stems from the interaction of the following three factors:
(a) broadcasting: In an optimal solution, exchanging pref-
erence data between passing users in proximity happens
in the background without user interaction.
(b) multiplatform app: In order to be able to provide the
proposed system for virtually all smartphone users, the
system has to be available on both Android and iOS.
(c) large message size: In order to exchange user pref-
erences data needed for a recommender system, larger
messages have to be exchanged (in the range of several
KB or MB).
Solutions for combinations of two of those aspects are
available: (a) + (b): Using Google Nearby Messages, WiFi
SSIDs, or Bluetooth Discovery Protocol messages, broadcast-
ing between Android and iOS devices is possible, but only
with small payloads. (b) + (c): Apps like SHAREit allow
the explicit connection establishment between two devices in
order to transfer large amounts of data. Typically, local WiFi
hotspots are used. It might be possible to implement a combi-
nation of (a) + (c) with OS specific solutions (iOS: Multipeer
Connectivity, Android: Google Nearby Connections).
For enabling a combination of all three aspects tough, a
workaround is necessary. Building on the existing approaches,
we present one workaround to facilitate a multiplatform ap-
proach with broadcasting that does not require user interaction
and alleviates the issue of size limitations, see Figure 3.
First, Alice authorizes the system to access her account at
some Cloud Storage Provider (CSP) like Dropbox, Google
Drive, etc. Alternatively, she could use her own cloud storage.
In some predefined frequency, Alice’s data is then uploaded
to the CSP and shared via a public URL. This URL is
then broadcasted via Google Nearby Messages or some other
technique that allows multiplatform broadcasting. As only the
URL is shared, which can be further shortened via a URL
shortener service, the small payload restrictions of multiplat-
form broadcasting technologies should suffice. Another user,
Bob in Figure 3, receives the broadcast with the URL and
can download Alice’s publicly shared data. Optimizations like
waiting for a WiFi connection can easily be implemented.
Note that the only required user interaction by Alice or Bob
is the authorization of the Cloud Storage Provider. Deeper
investigations have to address the limitations posed by such
9https://www.ushareit.com/
a solution, for example considering potential attack vectors
created by such an approach.
Alice
Smartphone A
Cloud Storage Provider
Smartphone B
Bob
authorize
Cloud Storage Provider
loop
[in predened frequency]
upload data
URL
loop
[in predened frequency]
broadcast URL
loop
[in predened frequency and when
download conditions are met, e.g.,
WiFi available]
request data
data
Fig. 3: Data exchange via a third-party cloud storage provider.
Even though technologies like BLE are in widespread use,
a deeper look reveals that actually exchanging data between
Android and iOS without user interaction is not a trivial task.
Even if future developments might reveal that the implemen-
tation of multiplatform broadcasting gets even harder, our
proposed architecture could still enable ad-hoc group recom-
mendations with explicit connection establishment. A group of
users in proximity could explicitly enable the data exchange
among the group and locally calculate recommendations or
query a third-party service provider for recommendations.
V. RECOMMENDING NEW ITEMS
Decentralized recommender systems traditionally use peer-
to-peer networks [6], [7], [22]. Gossip protocols leverage
commonly encountered small world properties of overlay
topologies in file-sharing networks and allow to find and gather
peers with similar preferences quickly and reliably. Once
established, communities of interest exchange item ratings
among each other. In contrast to such an approach, our
ad-hoc fashion of connection and data exchange between
smartphones is a network that is essentially fully disconnected.
Furthermore, most of the given approaches of decentralized
recommender systems deal with personal computers, while we
focus on ubiquitous scenarios with mobile devices.
Other related fields are those of ubiquitous recommender
systems and context-aware recommender systems (CARS).
They consider items in proximity or consider the user’s current
context while recommending items, respectively [23]. In con-
trast to these approaches, our proposed system is general in the
sense that any type of item can be recommended, independent
of the item’s physical proximity or the user’s context.
Another field, that has gained less attention in industry
and academia, is that of group recommender systems [24],
[25]. With its ad-hoc nature and immediate preference data
exchange, our proposed system is ideally suited to be used for
pervasive group recommendation scenarios. Exchanging data
between several users in a group setting, a local recommender
system can calculate recommendations based on the given
data, considering the preferences of each user. When utilizing
an external service provider for a recommendation, most likely,
before contacting it, the preferences of each group member
have to be combined into one group profile as most providers
will only recommend items for a single user.
When employing a local recommender system on the smart-
phone, additional data is needed. For content-based filtering,
the properties of items have to be known. Third-party service
providers can help with retrieving such needed metadata about
items. For user-based collaborative filtering, information about
the similarity of users is utilized. While services like Spotify
or Netflix have very large databases with millions of users,
the local databases in our proposed architecture will be much
smaller and thus there is a lower likelihood of finding similar
users.
We see two possible solutions for this problem. First, we
could let each user disseminate more than just his/her own
item preferences/ratings and let him/her also send data from
previous encounters – this would also address the cold start
problem new users will face. Another approach is to calculate
the similarity of users in a different way, independent of the
users’ ratings. In psychology, the propinquity effect is the
well-studied effect that physical proximity is a good predictor
of forming interpersonal bonds [26], [27]. Having unique
identifiers for each user and counting the number of times
and/or the duration of being in proximity would then likely
predict a higher bond. Additional methods are available for
determining similarity in proximity-based applications. In [2],
we developed and evaluated a method for estimating similarity
based on the exchange of the users’ context data using
probabilistic data structures in device-to-device scenarios. In
[3], we developed a privacy-preserving method for determining
the similarity of two users based on their text messaging data.
Both of those methods can be implemented in our proposed
architecture to find similar users, without having the need
to have users that rated the same items. Future work will
have to show to what extend those similarity metrics and
the propinquity effect yield valuable similarity indications
for user-based collaborative filtering. Future work could also
investigate the feasibility of approaches like federated learning,
effectively exchanging trained models or updates to models for
recommendations [28].
VI. RE LATE D WORK
Most related work is from the field of social networking
services. Often, the related work does not specifically focus
on recommender systems but on other aspects like privacy,
utilizing opportunistic networks, or encouraging real-life user
interactions.
In our own previous work, we outlined a general concept
for a social networking service utilizing context data from
smartphones, focusing on how connections between users
can be established [29]. In [30], we proposed to utilize the
relationship between smartphone data and personality traits
for friend recommendations in device-to-device scenarios.
In [12], the authors let users sent their identifiers of a social
networking services (via Bluetooth) to enable receivers to visit
their publicly available profile. The idea is to encourage social
interaction. The authors of [31] developed a framework for
ephemeral social networks. The goal is content dissemination
in opportunistic networks in scenarios with large crowds like
sports events. Westerkamp et al. propose a decentralized social
networking service in [32]. The goal is avoiding censorship by
utilizing the Ethereum Name Service for identity management
and by using self-hosted data storages – or trusted third-
party service providers – for each user. With such a solution,
the challenges of lock-in effects and lack of privacy are
tackled and the user is put in control of his/her data. In our
paper, we followed a similar approach, focusing on a different
application domain (recommender systems) and on pervasive
scenarios.
[5] is a short survey paper that highlights the challenges
of device-to-device computing. Regarding the wireless net-
work interfaces, the authors consider cellular networks, WiFi,
Bluetooth, and NFC. In our paper, we gave details about
the challenges when implementing data exchange between
devices.
Regarding decentralized recommender systems, [6] and [8]
follows similar approaches compared to our proposed system.
In [6], decentrally stored data from the web is used for a
recommender system running on the user’s personal computer.
Since that paper’s publication (2005), the development of mo-
bile devices enable mobile and ubiquitous scenarios depicted
in this paper. In [8], the authors propose that smartphones
exchange data in a device-to-device fashion and calculate their
own recommendation via collaborative filtering. The focus of
that paper is on the recommender algorithm that is evaluated
with a music data set. For device-to-device communication,
WiFi Direct is proposed.
VII. CONCLUSION AND FUTURE WORK
Current recommender systems often exhibit a lock-in effect
for the user and are connected to privacy concerns. Their rec-
ommendations are typically for single users rather than groups
and might be biased according to the interests of the providing
platform. We proposed a decentralized mobile architecture for
recommender systems that leverages the preferences/ratings
from users that are or have been in proximity. The introduced
system runs on the users’ smartphones and utilizes existing
external third-party service providers.
The system consists of three main parts, data collection,
data exchange, and recommender system. We developed three
ways to collect data: tracking context data automatically,
retrieving data from third-party service providers, and letting
the user manually rate items. We highlighted that while short-
range wireless transmission technologies are implemented
on all modern smartphones, exchanging larger amounts of
data without user interaction on a system available for both
Android and iOS remains a challenging task. We proposed a
workaround by broadcasting URLs of cloud storage providers
from which the receivers can then download the sender’s data.
Regarding the recommendation process, when considering
user-based collaborative filtering, we looked into ways for
determining the similarity between users, including methods
independent of the users’ ratings.
Future work will be to implement and test the proposed
device-to-device data exchange in order to evaluate its reli-
ability, transmission times in real user scenarios, and battery
consumption. Regarding the recommender system, future work
includes the implementation of a mobile recommender engine.
A simulation with a real data set can help evaluate the quality
of the recommendations that such a system can provide.
Furthermore, potential attack vectors and the overall security
of the system should be investigated.
ACKNOWLEDGMENT
We are grateful for the support provided by Robert Staake,
Jan Pokorski, Simone Egger, and Yong Wu.
REFERENCES
[1] M. Falch, A. Henten, R. Tadayoni, and I. Windekilde, “Business models
in social networking,” in CMI Int. Conf. on Social Networking and
Communities, 2009.
[2] F. Beierle, “Do You Like What I Like? Similarity Estimation in
Proximity-Based Mobile Social Networks,” in Proc. 2018 17th IEEE
International Conference On Trust, Security And Privacy In Computing
And Communications (TrustCom). IEEE, 2018, pp. 1040–1047.
[3] T. Eichinger, F. Beierle, S. U. Khan, R. Middelanis, V. Sekar, and
S. Tabibzadeh, “Affinity: A System for Latent User Similarity Com-
parison on Texting Data,” in Proc. 2019 IEEE International Conference
on Communications (ICC). IEEE, 2019 (in press).
[4] M. Ahmed, Y. Li, M. Waqas, M. Sheraz, D. Jin, and Z. Han, “A
Survey on Socially Aware Device-to-Device Communications,” IEEE
Communications Surveys Tutorials, vol. 20, no. 3, pp. 2169–2197, 2018.
[5] W. Zhang, H. Flores, and P. Hui, “Towards collaborative multi-device
computing,” in 2018 IEEE International Conference on Pervasive Com-
puting and Communications Workshops (PerCom Workshops). IEEE,
2018, pp. 22–27.
[6] C.-N. Ziegler, “Semantic Web Recommender Systems,” in Current
Trends in Database Technology - EDBT 2004 Workshops, ser. Lecture
Notes in Computer Science, W. Lindner, M. Mesiti, C. T ¨
urker, Y. Tz-
itzikas, and A. I. Vakali, Eds. Springer, 2005, pp. 78–89.
[7] R. Baraglia, P. Dazzi, M. Mordacchini, and L. Ricci, “A peer-to-peer
recommender system for self-emerging user communities based on
gossip overlays,Journal of Computer and System Sciences, vol. 79,
no. 2, pp. 291–308, Mar. 2013.
[8] L. N. Barbosa, J. Gemmell, M. Horvath, and T. Heimfarth, “Distributed
User-Based Collaborative Filtering on an Opportunistic Network,” in
2018 IEEE 32nd International Conference on Advanced Information
Networking and Applications (AINA). IEEE, 2018, pp. 266–273.
[9] F. Beierle, V. T. Tran, M. Allemand, P. Neff, W. Schlee, T. Probst,
R. Pryss, and J. Zimmermann, “TYDR – Track Your Daily Routine.
Android App for Tracking Smartphone Sensor and Usage Data,” in
2018 IEEE/ACM 5th International Conference on Mobile Software
Engineering and Systems (MOBILESoft). ACM, 2018, pp. 72–75.
[10] ——, “Context Data Categories and Privacy Model for Mobile Data
Collection Apps,” Procedia Computer Science, vol. 134, pp. 18–25,
2018.
[11] F. Beierle, V. T. Tran, M. Allemand, P. Neff, W. Schlee, T. Probst,
J. Zimmermann, and R. Pryss, “What data are smartphone users willing
to share with researchers? Designing and evaluating a privacy model
for mobile data collection apps,” Journal of Ambient Intelligence and
Humanized Computing, 2019 (in press).
[12] A. Beach, M. Gartrell, S. Akkala, J. Elston, J. Kelley, K. Nishimoto,
B. Ray, S. Razgulin, K. Sundaresan, B. Surendar, M. Terada, and R. Han,
“WhozThat? Evolving an Ecosystem for Context-Aware Mobile Social
Networks,” IEEE Network, vol. 22, no. 4, pp. 50–55, 2008.
[13] N. Eagle and A. Pentland, “Social Serendipity: Mobilizing social soft-
ware,” Pervasive Computing, IEEE, vol. 4, no. 2, pp. 28–34, 2005.
[14] A.-K. Pietil ¨
ainen, E. Oliver, J. LeBrun, G. Varghese, and C. Diot,
“MobiClique: Middleware for Mobile Social Networking,” in Proc. 2nd
ACM Workshop on Online Social Networks (WOSN). ACM, 2009, pp.
49–54.
[15] Z. Yang, B. Zhang, J. Dai, A. Champion, D. Xuan, and D. Li, “E-
SmallTalker: A Distributed Mobile System for Social Networking in
Physical Proximity,” in 2010 IEEE 30th International Conference on
Distributed Computing Systems (ICDCS). IEEE, 2010, pp. 468–477.
[16] J. Manweiler, R. Scudellari, and L. P. Cox, “SMILE: Encounter-based
trust for mobile social services,” in Proc. 16th ACM Conference on
Computer and Communications Security. ACM, 2009, pp. 246–255.
[17] J. Teng, B. Zhang, X. Li, X. Bai, and D. Xuan, “E-Shadow: Lubri-
cating Social Interaction Using Mobile Phones,” IEEE Transactions on
Computers, vol. 63, no. 6, pp. 1422–1433, Jun. 2014.
[18] Z. Lu, G. Cao, and T. L. Porta, “Networking smartphones for disaster
recovery,” in 2016 IEEE International Conference on Pervasive Com-
puting and Communications (PerCom). IEEE, 2016, pp. 1–9.
[19] J. Yang, C. Poellabauer, P. Mitra, J. Rao, and C. Neubecker,
“BlueNet: BLE-based ad-hoc communications without predefined roles,”
in 2017 IEEE SmartWorld, Ubiquitous Intelligence Computing, Ad-
vanced Trusted Computed, Scalable Computing Communications, Cloud
Big Data Computing, Internet of People and Smart City Innovation
(SmartWorld/SCALCOM/UIC/ATC/CBDCom/IOP/SCI). IEEE, 2017,
pp. 1–8.
[20] Z. Mao, J. Ma, Y. Jiang, and B. Yao, “Performance Evaluation of WiFi
Direct for Data Dissemination in Mobile Social Networks,” in 2017
IEEE Symposium on Computers and Communications (ISCC). IEEE,
2017, pp. 1213–1218.
[21] J. Shu, S. Kosta, R. Zheng, and P. Hui, “Talk2Me: A Framework for
Device-to-Device Augmented Reality Social Network,” in 2018 IEEE
International Conference on Pervasive Computing and Communications
(PerCom). IEEE, 2018, pp. 1–10.
[22] G. Ruffo and R. Schifanella, “A Peer-to-Peer Recommender System
Based on Spontaneous Affinities,” ACM Transactions on Internet Tech-
nology, vol. 9, no. 1, pp. 1–34, Feb. 2009.
[23] C. Mettouris and G. A. Papadopoulos, “Ubiquitous recommender sys-
tems,” Computing, vol. 96, no. 3, pp. 223–257, Mar. 2014.
[24] A. Crossen, J. Budzik, and K. J. Hammond, “Flytrap: Intelligent group
music recommendation,” in Proc. 7th International Conference on
Intelligent User Interfaces. ACM, 2002, pp. 184–185.
[25] F. Beierle, K. Grunert, S. G¨
ond¨
or, and A. K¨
upper, “Privacy-aware
Social Music Playlist Generation,” in Proc. 2016 IEEE International
Conference on Communications (ICC). IEEE, 2016, pp. 5650–5656.
[26] D. M. Marvin, “Occupational Propinquity as a Factor in Marriage Se-
lection,” Quarterly Publications of the American Statistical Association,
vol. 16, no. 123, pp. 131–150, 1918.
[27] L. Festinger, S. Schachter, and K. Back, “The spatial ecology of group
formation,” Social pressures in informal groups, pp. 141–161, 1950.
[28] B. McMahan, E. Moore, D. Ramage, S. Hampson, and B. A. y Arcas,
“Communication-Efficient Learning of Deep Networks from Decentral-
ized Data,” in Proceedings of the 20th International Conference on
Artificial Intelligence and Statistics. PMLR, 2017, pp. 1273–1282.
[29] F. Beierle, S. G¨
ond¨
or, and A. K¨
upper, “Towards a Three-tiered Social
Graph in Decentralized Online Social Networks,” in Proc. 7th Interna-
tional Workshop on Hot Topics in Planet-Scale mObile Computing and
Online Social neTworking (HOTPOST). ACM, 2015, pp. 1–6.
[30] F. Beierle, K. Grunert, S. G¨
ond¨
or, and V. Schl ¨
uter, “Towards
Psychometrics-based Friend Recommendations in Social Networking
Services,” in 2017 IEEE International Conference on AI & Mobile
Services (AIMS). IEEE, 2017, pp. 105–108.
[31] A. Boutet, S. Frenot, F. Laforest, P. Launay, N. Le Sommer, Y. Maheo,
and D. Reimert, “C3PO: A Network and Application Framework for
Spontaneous and Ephemeral Social Networks,” in Web Information
Systems Engineering – WISE 2015, ser. Lecture Notes in Computer
Science, J. Wang, W. Cellary, D. Wang, H. Wang, S.-C. Chen, T. Li,
and Y. Zhang, Eds. Springer, 2015, pp. 348–358.
[32] M. Westerkamp, S. G¨
ond¨
or, and A. K¨
upper, “Tawki: Towards Self-
Sovereign Social Communication,” in Proc. IEEE International Confer-
ence on Decentralized Applications and Infrastructures (DAPPCON).
IEEE, 2019 (in press).
ResearchGate has not been able to resolve any citations for this publication.
Article
Full-text available
Context-aware applications stemming from diverse fields like mobile health, recommender systems, and mobile commerce potentially benefit from knowing aspects of the user’s personality. As filling out personality questionnaires is tedious, we propose the prediction of the user’s personality from smartphone sensor and usage data. In order to collect data for researching the relationship between smartphone data and personality, we developed the Android app track your daily routine (TYDR), which tracks and records smartphone data and utilizes psychometric personality questionnaires. With TYDR, we track a larger variety of smartphone data than many other existing apps, including metadata on notifications, photos taken, and music played back by the user. Based on the development of TYDR, we introduce a general context data model consisting of four categories that focus on the user’s different types of interactions with the smartphone: physical conditions and activity, device status and usage, core functions usage, and app usage. On top of this, we developed the Privacy Model for Mobile Data Collection Applications (PM-MoDaC) specifically tailored for apps that are related to the collection of mobile data, consisting of nine proposed privacy measures. We present the implementation of all of those measures in TYDR. Our experimental evaluation is based on data collected with TYDR during a two-month period. We find evidence that our users accept our proposed privacy model. Based on data about granting TYDR all or no Android system permissions, we find evidence that younger users tend to be less willing to share their data (average age of 30 years compared to 35 years). We also observe that female users tend to be less willing to share data compared to male users. We did not find any evidence that education or personality traits are a factor related to data sharing. TYDR users score higher on the personality trait openness to experience than the average of the population, which we assume to be evidence that the type of app influences the user base it attracts in terms of average personality traits.
Conference Paper
Full-text available
As of today, web-based social communication platforms , such as WhatsApp, Twitter, or Facebook, are almost exclusively realized via centralized platforms, based on proprietary interfaces, protocols, and data formats. In consequence, even though social communication being a decentralized, peer-to-peer phenomenon, web-based communication today is implemented via closed, proprietary data silos, which not only lock-in users into their service platforms, but also control and own exchange information and data, including personal and sensitive data such as photos, messages, or contact information. In this paper we present Tawki, a decentralized service architecture for social communication. Using Tawki, users remain in full control of their personal data, which is stored and managed by personal data storages. Each data storage is accessible via a unified Tawki API, which allows users to send and request data to and from other users' personal data storages. Following this approach, social communication is again peer-to-peer without involving a third party controlling and monitoring the process. Tawki uses the Ethereum Name Service (ENS) for both the management of user identities and resolving identifiers to the respective user's personal storage location. Leveraging the immutability of the Ethereum Blockchain, identity management and discovery of personal data storages is secured against censorship and control through any third party.
Conference Paper
Full-text available
While existing social networking services tend to connect people who know each other, people show a desire to also connect to yet unknown people in physical proximity. Existing research shows that people tend to connect to similar people. Utilizing technology in order to stimulate human interaction between strangers, we consider the scenario of two strangers meeting. On the example of similarity in musical taste, we develop a solution for the problem of similarity estimation in proximity-based mobile social networks. We show that a single exchange of a probabilistic data structure between two devices can closely estimate the similarity of two users - without the need to contact a third-party server.We introduce metrics for fast and space-efficient approximation of the Dice coefficient of two multisets - based on the comparison of two Counting Bloom Filters or two Count-Min Sketches. Our analysis shows that utilizing a single hash function minimizes the error when comparing these probabilistic data structures. The size that should be chosen for the data structure depends on the expected average number of unique input elements. Using real user data, we show that a Counting Bloom Filter with a single hash function and a length of 128 is sufficient to accurately estimate the similarity between two multisets representing the musical tastes of two users. Our approach is generalizable for any other similarity estimation of frequencies represented as multisets.
Article
Device-to-device (D2D) communication is continuously evolving and engulfing a wide diversity of proximate services and applications, which is considered as one of the key technologies in the 5G ecosystem. It is argued that direct communications paradigm advantages may not be entirely captured without embracing the concepts of the social-domain. Following the explosion of social networks and smart mobile devices, social-interactions among mobile users enable the promotion of D2D co-operations. In this article, we first present a systematic overview of the socially-unaware technical challenges and the prospective applications to consolidate a background on D2D communications. Next, all the social features utilized in tackling and improving the technical issues and the applications are briefed, respectively. Then, we provide a comprehensive survey on socially-aware D2D communications, where various features of social-domain are leveraged such as ties, community, trust, and selfishness to tackle the D2D technical issues and improve the performance of prospective D2D applications. We categorize state of the art socially-aware D2D communications into two main sections. The first section considers technical issues, while the second section pertains to their applications to provide insightful information and motivate more in-depth studies in this area. For each aforementioned section, we overview technical aspects and classify the core contributions based on approaches used and the social features utilized. In the discussion section, we summarize the socially-aware D2D communications and present lessons learned. Finally, we highlight the open research challenges in this field and predict future research trends.