=Paper=
{{Paper
|id=None
|storemode=property
|title=Selective Propagation of Social Data in Decentralized Online Social Network
|pdfUrl=https://ceur-ws.org/Vol-730/paper2.pdf
|volume=Vol-730
}}
==Selective Propagation of Social Data in Decentralized Online Social Network==
https://ceur-ws.org/Vol-730/paper2.pdf
Selective Propagation of Social Data in Decentralized
Online Social Network
Udeep Tandukar and Julita Vassileva
Department of Computer Science, University of Saskatchewan, Saskatoon, Canada
udeep.tandukar@usask.ca, jiv@cs.usask.ca
Abstract. In Online Social Networks (OSNs) users are overwhelmed with huge
amount of social data, most of which are irrelevant to their interest. Due to the
fact that most current OSNs are centralized, people are forced to share their data
with the site, in order to be able to share it with their friends, and thus they lose
control over it. Decentralized OSNs provide an alternative which allows users
to maintain control over their data. This paper discusses an approach for
propagation of social data in a decentralized OSN so as to reduce irrelevant data
among users. The approach uses interaction between users to construct
relationship model of interest. This relationship model acts as a filter later while
propagating social data of the same interest group. This paper also presents a
plan of a simulation to analyze our approach.
Keywords: Online Social Network, Decentralization, Peer-to-peer system,
Information propagation, Relationship modeling
1 Introduction
Online Social Networks (OSNs) have become a common ground where people are
generating and consuming huge amount of information. This information varies from
personal thoughts (like status updates) to global news (such as wars, world cup, etc.).
There have been growth in the number of providers of such OSNs and user’s data are
scattered over these different providers. In OSNs there is a huge flow of information
of which only a fraction is relevant to the users. Since decentralized OSNs also inherit
most of the issues from OSN, they have to deal with the question of how to provide
relevant information to users and filter out the irrelevant information.
The currently popular OSNs are centralized which means they store all the
information that people are generating or consuming. This information is mostly
private and people voluntarily share it with the site, in order to be able to share it with
their friends. However, in this way, they have less control over their own data and
their data is scattered over the internet in different OSN providers which usually do
not support data interoperability (apart from trivial user profile information).
This issue of control and privacy in centralized OSNs has recently motivated
research into decentralized OSNs where the data of users are kept in their own clients
(nodes). There have been several attempts to build decentralized OSNs and these
2 Udeep Tandukar and Julita Vassileva
projects are still going on [3], [7], [15]. Among several approaches to build
decentralized OSN, a peer-to-peer (P2P) infrastructure has also been proposed. P2P
has been popular among file-sharing applications, but not in OSNs. The inherent
nature of how people connect with each other in a social network makes peer-to-peer
architectures suitable for building decentralized OSNs.
This paper proposes an approach to deal with both problems mentioned above –
information overload and privacy – using a P2P infrastructure and relationship models
according to interest groups among users to filter out irrelevant information from
flowing out of the source. These relationship models are updated depending on
feedback resulting from the interaction between users and what they do with the
information they receive. The relationship models are used later to route the
information that a user sends to her friends appropriately according to its semantic
meaning.
The rest of the paper is structured as follows. A review of related works is
presented in Section 2, followed by definition of our research problem in Section 3.
Section 4 discusses our approach on filtering out irrelevant social data using
relationship models. The proposed plan of simulation and evaluation of the discussed
approach is presented in Section 5. Finally, Section 6 concludes the paper.
2 Background and Motivation
The section begins with an overview of online social networks and decentralized
online social networks as an alternative to centralized ones. Then there is discussion
about using peer-to-peer architecture for decentralized online social networks. Then
we mention about information dissemination in social networks, and user modeling
which is related to relationship modeling in our work. And the problem statement of
our research is covered at the end of this section.
2.1 Online Social Networks
An Online Social Network (OSN) is defined as web platform in which a person can
create a profile, connect to other people, view and traverse network of connections
within the system, share resources and information within the system, and use social
applications with which people within the system can interact and collaborate with
each other [5], [9]. With the growth of internet usage, OSNs first came into existence
in the form of SixDegree.com (in 1997) which had basic OSN features [5]. In the
following years many more OSN service providers (like Friendster, MySpace,
Last.FM, Hi5, Twitter, Facebook, etc.) came into existence, some of which have
grown to have millions of active users. All these OSN followed client-server
architecture in which the service provider is centralized. This architecture supports
high accessibility since users can access the service from any web-browser wherever
and whenever they desire. But due to this centralized nature, these OSN have inherent
issues like single point of failure, central administration that can control activities of
users, privacy issues due to central data storage, and requirement of larger servers and
Selective Propagation of Social Data in Decentralized Online Social Network 3
bandwidth to accommodate the growing number of users. With these issues in
centralized OSN we saw discontinuation of some popular OSN services like
SixDegree.com and Friendster [5], and the users lost all their social contacts and data
when they lost access to those services. Some centralized OSN services like Twitter,
due to growth in their popularity were having many performance scalability issues
and still face some frequently slow response or even unresponsiveness [8]. In addition
to these technical issues, there are also social issues arising with the rapidly growing
popularity of social networking. People became more conscious about the information
that they share in their social networks. Services, like Facebook which have millions
of users, frequently have to deal with privacy issues. For example, Facebook’s
Beacon online ad system was tracking activities of the users in third party websites
even when users were logged off from Facebook and had declined to broadcast their
activities [14]. The system caused an outrage among Facebook users, and Facebook
quickly discontinued Beacon. However, present centralized OSN still have full
control over user data once shared, the user losses control over it and cannot remove it
or export it into another OSN. Although there are various OSN available in web, there
is no easy interoperability across them. Users of OSN (e.g. MySpace) cannot interact
with users of another OSN (e.g. Facebook). This has lead OSNs being viewed as
“information silos” [23]. As alternative to this centralized OSN, we can consider
decentralized OSN in which users have control over their data.
2.2 Decentralizing Online Social Networks
Decentralized online social networks have distributed computing structure with
trusted network of servers or peer-to-peer network. In [23], the authors suggest that
decentralized OSN will give back to users the control of their data with respect to
privacy, data ownership and information dissemination.
Users hosting their own social data. According to Yeung et.al [23], in decentralized
OSN the user is not required to be a part of social networking services like Facebook,
Twitter, etc. to maintain his/her online social presence. The user can host a FOAF [6]
(Friend-Of-A-Friend) file, an activity log, photos/videos, and social client in a trusted
server. They will have full control over whom and what to share out of his/her social
data. The authors describe how the functionality offered by popular social
applications, such as “Personal Wall”, “Photos”, and “News Feed” can be
implemented in decentralized OSN. In the proposed system, the user shares and
communicates social data with other users by using WebDAV [22] or SPARQL
Update [18] protocols. As a prototype they have developed “Tabulator” [3] which is a
generic data browser and editor of linked RDF (Resource Description Framework)
data [4]. These types of decentralized OSN encourage users to store their social data
on the web in standard format such as RDF and it should be accessible through URI
(Universal Resource Indicator). Therefore, the user does not have to rely on only one
social application, and can use any social application that support these open
technologies.
4 Udeep Tandukar and Julita Vassileva
Using P2P infrastructure. Decentralized OSN can also be implemented with the use
of Peer-to-Peer (P2P) network. A P2P network is a distributed network in which
nodes are connected with each other to participate in processing, memory, and
bandwidth intensive tasks. These networks scale better than centralized server
architectures without the need of costly centralized resources. P2P networks have
been popular mostly as file sharing networks (such as KaZaA, BitTorrent, etc.) and
sometimes as collaborative sharing networks (such as Skype), but have not been used
as a medium for online social networking. The inherent nature of peer-to-peer
connection between users in a social network makes OSN suitable for peer-to-peer
architecture [9]. In file sharing P2P systems like Gnutella, most of the users are free
riders [1]. In contrast, P2P applications like Skype, where users tend to stay connected
to the network to receive calls from their friends, shows the potential that P2P holds
as implementation infrastructure for OSN.
To accommodate the familiar functionality of centralized OSN (like status updates,
photo uploads, commenting, rating) in decentralized OSN, there are various
challenges. Since the social data is stored at the peers, the availability of the social
data depends on the online behaviour of peers. The storing of data on the peers allows
encryption of these data which can ensure privacy while transmitting data from peer
to another. The propagation of social data or updates among users in the OSN should
be managed so that there is less duplication and no latency. These and other
challenges have been discussed in detail in [9].
Some P2P systems have exploited properties of social networks (like trust,
collaboration) in other to improve the performance of P2P networks. System like
Tribler [15], has adopted social networking on BitTorrent based P2P file-sharing
network in order to recommend, search, and download contents. The authors have
used “Buddycast Algorithm” that exchanges preferences of peers in the implicitly
defined social network to generate recommendation list and search contents. By using
a collaborative download protocol called “2Fast”, in which users collaborate to
contribute their bandwidth, download performance also improved. Some also used the
graph topology of real social networks in order to form a P2P overlay topology and
used it to improve lookups and scalability [2].
PeerSoN [7] is an effort to build decentralized OSN over theP2P architecture. It
has implemented encryption of user data to protect privacy and ensure direct
exchange of data between devices for delay-tolerance and opportunistic networking. It
uses OpenDHT [17], a Distributed Hash Table (DHT) service, for look-up service to
find other peers in the P2P network and also to store data like, IP address, file
information, and notifications for peers. In DHT [16], {key, value} pairs are stored in
a distributed nodes and node can retrieve the value associated with any key
efficiently. The prototype of PeerSoN provides functionality like social links
(becoming friends), storage to maintain profile and post by their friends,
asynchronous messaging, and live chatting.
Selective Propagation of Social Data in Decentralized Online Social Network 5
2.3 Information dissemination
Nowadays, OSN produce huge amount of information and propagation of this
information to its destination has to be well coordinated so as to reduce information
overload, duplication, latency and to ensure quality.
In a file-sharing P2P network, the location of a resource is a very important piece
of information. Generally, users send queries about resources of interest and system
returns lists of their locations (i.e. peers that store these resources). In contrast to this
process of searching, [13] discuss “selective information push” where user posts her
profile to “super-peers” and receives notifications about resources that match her
interests as these resources become available. Since the user is both consumer and
producer of the resources shared in the network, she can also post advertisements of
her resources and the super-peer will push notifications about these resources to
relevant peers (peers with matching interests). This mechanism depends on the
preferences of the user querying super-peers and if user has a more general interest
then she might get lots of notifications. Here, the super-peer can be taken as a
recommender that is pushing information according to the peer’s interest.
The Push-poll recommender algorithm [20] propagates information through
implicit social network, formed by peers with similar interests, using “word of
mouth” mechanism. This system also takes in account feedback from the recipient to
determine the future influence of sender on recipient. KeepUp recommender system
[21] is based on the Push-poll algorithm. It allows user to interactively adjust the
amount of influence that her neighbours have on the recommendations she receives.
This gives power to user to decide indirectly what and how much information is
propagated to her.
GoDisco [10] focuses on dissemination of social data according to the context of
the information. The nodes gossip about their interests and strength of these interests
with their neighbours in a regular interval. They also keep track of the behaviour of
their neighbours (like activeness, forwarding behaviour). This knowledge of each
other is used in the dissemination phase where the messages are assumed to have
some semantic value that can be mapped to the interests of the nodes. Our work will
consider the degree of relationship between these nodes and influence of this on
dissemination. We will be creating relationship model of neighbouring nodes to
determine degree of relationship.
2.4 User Modeling
Each user has her own characteristics, e.g. interests, preferences, etc., and we can
utilize it to provide her with relevant social data in social network. These
characteristics comprise the user model for the system. In our work, we will be
building user models of the neighbours of a user within particular domain of interest
in order to evaluate which information to forward to them or not.
Since the user will be using her OSN data on different devices and possibly across
different applications, interoperability of the user models along with other social data
is very important. That is why it is desirable for the representation of user data in the
6 Udeep Tandukar and Julita Vassileva
user model to follow some ontology so that it could be understood and interpreted
outside of the context of the application in which the model was created. Using RDF-
based user model like UserML [12] can enable distribution of the user model among
different devices. UserML divides user model dimension in three parts: auxiliary,
predicate and range. If we want to express the interest of a user in UserML then
auxiliary will be “hasInterest”, predicate will be “reading” to indicate her interest and
range can be “low-medium-high”.
To stimulate cooperation while sharing resources in P2P system, Sun et.al [19] has
applied user modeling and modeling of relationships between users. With the help of
user models of interest, they were able to route information to other users with similar
interests. Using relationship modeling between users, they were able to determine
typical time patterns of neighbour’s behaviours to ensure better quality service. The
authors created an overlay topology over the P2P network, where a relationship
between users is created when a user successfully downloads a file from another user
and the strength of the relationship grows with the number of successful interactions
between these users.
With the increasing interest of users in social networking on web, there has been
significant growth in research related to OSN. Users are becoming more and more
sensitive to their data and decentralized OSN holds a key for these users to use web
with full control over their data. As discussed earlier for decentralized OSN users can
either choose secure server to host their data or use P2P infrastructure. As a new
domain for P2P infrastructure, OSN holds lots of challenges to the researchers. As
OSN is based on sharing of information, well-coordinated propagation of information
is very important to handle information overload, latency, and repetitions. In our
work, we will be focusing on using models of interest of neighbouring users for
proper propagation of information.
2.5 Problem Statement
Online social network (OSN) has provided a medium for people to communicate
and share information (social data). People share their thoughts, photos, videos, links
to web pages, etc. in this network. The network in OSN constitute of members that
are interconnected with each other through some relationship like friendship, common
preferences, etc. When shared to the network, the shared social data propagate to each
and every member of the network whether it is relevant to them or not. From the
viewpoint of a sender, she is sharing only one social data at a moment. But from the
viewpoint of a receiver, there can be more than one sender. According to statistics
from Facebook [11], on average users have 130 friends. If all of the friends share a
social data then a user will get 130 different social data. In this way, the user’s stream
is flooded with huge amount of social data, most of which are irrelevant to the user’s
interest.
Most of the available OSNs are based on client-server architecture in which user’s
social data are kept centralized. As discussed earlier this centralized nature has some
issues and as an alternative we can have OSN in decentralized architecture, where
users have control over their own social data. Even in decentralized OSN, due to the
Selective Propagation of Social Data in Decentralized Online Social Network 7
social nature of the network we will have to deal with propagation of social data to
reduce irrelevancy, redundancy, and latency. The research domain in this paper is
related to propagation of social data from a user to her neighbours so that they only
get data relevant to them.
3 Approach
An approach of selective propagation of social data (i.e. information shared by
users in social networks, such as status updates, shared links) by modeling interest of
neighbouring users in a social network is proposed next, that ensures that social data
reaches only the relevant users for whom it would be interesting.
3.1 Social P2P Network of Users
The system is a decentralized online social network implemented over P2P
network. For simplicity, we will be dealing with a social network which can be
represented by social graph. Social graph is a graph in which each user is a node and
relationships between users is edges. Let 𝐺 be a social graph represent by * +
where represents set of nodes (users) and represents set of edges (relationships)
between nodes. We can say have some relationship with iff there
exists * + * + .
To route relevant social data to users, each user or node in the graph will model the
interests of other users with whom she has relationships. From the point of view of a
given user, the model of interests of other users is considered as relationship model
since it signifies how many positive interactions have happened between the users in
the context of particular area of interest. Positive interaction between two users in a
given area of interest means that one user has sent social data related to the area of
interest to the second user, and the second user has given positive feedback after
receiving the social data. As a result of positive interaction, the strength of the
relationship between the two users in the area of interest increases. The relationship
model is used by the peer to adaptively disseminate social data related to a given
interest area I, by sending it to peers with whom the user has sufficiently strong
relationship in area I.
3.2 Relationship modeling
In an online world, relationships between users strengthen as the interaction
between them increases. In the proposed approach, not only interaction between users
in general but within certain subject or interest is taken into account, so that the
system can model the strength of relationship in an area of specified interest between
interacting users. To determine the area of interest of the social data, users have to
either tag their updates with the interest areas or the system has to extract semantics
from the data.
8 Udeep Tandukar and Julita Vassileva
Interaction between users within certain context is captured by tracking the
feedback of the shared content. Feedback from friends (users connected in social
graph) can be of different types. For the proposed system, feedback is categorized as
follows.
Table 1: Categorization of feedback
Type Action Value
Type 1 Comment / Share 0.9
Type 2 Rate / Like 0.7
Type 3 View / Open 0.5
Type 4 Ignored / Not open 0.3
The response value varies from 0.3 to 0.9 according to the type of action the users
takes. These feedbacks from the receiver of the social data depend on the level of
interest and relationship model with the sender for that context. The relationship
model depends on the previous interaction between two users, and priority of a new
social data is determined according to previous history of interaction in the system.
The relationship model consists of a list of areas of interest and the corresponding
strength of relationship between two users in each of these areas. Strength of
relationship between user and user for an interest area should increase with
stronger feedback and decrease with weaker feedback, therefore it is calculated using
following equation:
( ) ( ) ( ) ( )
Here, ( ) is the new strength of relationship, ( ) is the previous strength of
relationship for an interest area . The parameter , - is a linear function of the
number of social data produced by the user in particular interest area . Initially is
so that the latter half of the equation has very low effect on the new strength. The
feedback from the recipient is denoted by , and its value varies from to as
specified in Table 1. The increase and decrease of the strength of relationship
calculated according to equation (1) is at very minimal rate, so as to maintain the
relationship between the users as long as possible.
For the propagation of social data belonging to interest , the strength of
relationship between users should be more than a threshold value. Initially, this
strength of relationships among all users is set as and it will increase and decrease
according to the interactions between users.
This approach of propagating social data takes into account the feedback of friends
and uses this feedback to calculate strength of relationship, which is used in the future
as a filter while sending the data of similar topic. It is taken in consideration that if
information is relevant to a user, she will at least open that message and the feedback
value is 0.5 which is more than critical value. If the information is irrelevant to a user,
she will ignore it and hence the feedback value is 0.3 which is less than critical value
and reduces the strength of relationship. This relationship models are all stored in the
user’s device since our system follows decentralized architecture. The process of
filtration during propagation is done at the sender; therefore, some computation power
Selective Propagation of Social Data in Decentralized Online Social Network 9
is consumed at the sender side but network traffic is reduced and the friend’s node
does not have to do much of the filtration process.
As the strength of relationship for a particular interest I in user A for user B fades
away, user B will not get any social data related to interest I from user A. For user B
to get social data related to interest I from user A, she has to make relationship model
in user A between her and user A stronger. User B can send a social data to user A
related to interest I, this will show that user B is becoming interested in I and user A
will increase the strength of relationship as high as possible so that social data from
her can reach user B. To give more control to users over their relationships, it is also
possible to allow users to directly adjust the relationship strength with other users via
an appropriate GUI, similar to the interactive influence adjustment deployed in the
KeepUP Recommender System [21].
In this way, relationships between users will grow or fade away in context to
certain interest groups. This will allow better communication between users since
they do not have to deal with irrelevant social data.
4 Evaluation Plan
In order to evaluate the discussed approach of using relationship modeling to filter
out irrelevant social data in a social network, a simulation of the system will be
developed using synthetically generated social graph and real-world social graph from
StudiVZ and Facebook. A random social graph with small world properties will be
generated using JUNG1 (Java Universal Network/Graph) Framework as a synthetic
dataset. Afterwards, two different real datasets will be used to generate the network
and message streams – one from StudiVZ2 and one from Facebook3. Both have
around 1 million users or nodes.
4.1 Distribution of interest
Possible areas of interest can be defined for users in hierarchical way by
introducing general categories and sub-categories, so initially, to avoid widely
separated interests in population the system will only have one level of general
categories of interest, such as “sports”, “news & events”, “politics”, “personal status
updates”, “photos”, “videos”, “curiosities & jokes”. Interests are distributed
exponentially over users in the social network with most common interests (in the
currently most popular music, movies, etc.) taking large portion of population and less
common interest (e.g. local sport) popular among small portion. The mechanisms to
generate such skewed distributions are known: growth – people gain new interests
with time, and preferential attachment – areas of interest that are already popular
attract newcomers with a higher likelihood. The simulation will use these rules to
1http://jung.sourceforge.net/index.html
2http://studivz.irgendwo.org/
3http://odysseas.calit2.uci.edu/doku.php/public:online_social_networks#available_datasets
10 Udeep Tandukar and Julita Vassileva
generate a realistic distribution of interests for a fixed set of interest semantic
categories.
The system depends on the growth of relationship strength between users for
particular interest. The feedback of each shared data is important to calculate this
strength. This feedback depends on the interest level of the receiver of the social data.
The system will simulate around 25 interest categories and these will be distributed to
all of the nodes in the graph. The nodes will have different interest levels, , -,
which signifies how much the user is interested on each category. When the users
receives an update of an interest (semantic) category, the higher the interest level, the
likelihood of feedback from Type 1 to 3 as illustrated in Table 1 increases by the
users. The users who have lower interest levels in the category of the update will give
Type 4 feedback. The distribution of the interest levels initially will be random. Since
there will be some nodes which will have more connection that other nodes, there is
probability of these nodes being interested in more interest groups. With the
interaction on particular area, the value of interest level will also grow. These
considerations will be taken into account while designing the proposed simulation.
4.2 Propagation of social data
The propagation of social data depends upon the relationship model of each
simulated user. Initially the system will consist of equally distributed relationship
model (equal value of relationship strength) so that propagation of social data at the
initial stage of the system reaches all friends of the user. With the feedback from
friends, these relationship models will either strengthen or weaken according to
equation (1). The likelihood of type of feedback as illustrated in Table 1 depends on
the interest level of the friends as discussed in earlier section. For simplicity, each
social data which will propagate will also carry semantic meaning along with it. The
semantic meaning will consist of types of interest group the social data belongs to. To
simulate the phenomenon of users injecting new content in the system, each node in
the system is fed with a number of social data within random interval of time. The
behaviour of each node whether to forward (share) an incoming social data to its
neighbour (friend) depends on the interest level, strength of the relationship with
respect to the interest category of the social data represented in the relationship
model. All these changes of relationship model, forwarding of data, filtering of
irrelevant data and interest level will be recorded and used in future analysis.
4.3 Analysis
For each node, the following data will be recorded in the simulation:
Relationship model established between each node with its friends.
Interest level of each node.
The number of social data a node shared with its friends.
The number of social data filtered by relationship model in each node.
The number of social data that is forwarded by a node.
Selective Propagation of Social Data in Decentralized Online Social Network 11
The total number of nodes that forwarded social data received from its friends and
interest level at the moment of forwarding.
With the simulation of our proposed system, we hope to have the following
analysis and insights.
New social data generated by a user will get filtered and the propagation will be
limited by the relationship model. The rate at which the nodes evolve to filter out
irrelevant social data will be analyzed.
The level of interest in a user has direct impact on the type of feedback to a shared
data. This feedback is used to calculate the relationship strength. The level of
interest will change as the number of interaction increases or decreases. The
correlation between interest level and relationship strength will be analyzed.
A node can forward an incoming social data. This behaviour largely depends on
the interest level. The system will record this forwarding behaviour of each node to
analyze range of propagation of a social data. Range of propagation means how far
a social data is forwarded from its source.
Spreading or sharing of social data will largely depend on the interest level and
relationship model of each node in the system. Sharing of a social data stops where
node has low interest level or weak relationship model with her friends.
When the system reaches its maturity, it will have nodes interacting with each
other without concerns about getting irrelevant social data.
5 Conclusion
In this paper, we have discussed an approach of using feedback from interaction
between users as a relationship building mechanism to filter out irrelevant social data
in decentralized online social networks. Decentralized online social networks give
users the control of their social data with respect to privacy, data ownership and
information dissemination. A simulation will be implemented to analyze the discussed
approach. The simulation will also deal with the numbers of hops a social data
transverse so as to analyze its spread. The simulation will have nodes which are
always online and cooperative. But in the real system, there are always issues with
availability, free riders, latency, and other factors. The area of decentralized online
social networks holds exciting research questions, associated with storage of social
data, privacy issues of social data, searching and indexing of friends in the network,
and many more. We plan to implement a real decentralized OSN that follows P2P
architecture after the successful analysis of the simulation and evaluation.
References
1. Adar, E. and Huberman, B.A.: Free Riding on Gnutella. First Monday 5 (10), 2-13 (2000)
2. Anwar, Z., Yurcik, W., Pandey, V., Shankar, A., Gupta, I., and Campbell, R.H.:
Leveraging Social-Network Infrastructure to Improve Peer-to-Peer Overlay Performance:
Results from Orkut. ACM Computing Research Repository (2005)
12 Udeep Tandukar and Julita Vassileva
3. Berners-Lee, T., Hollenbach, J., Lu, K., Presbrey, J., and Schraefel, M.: Tabulator redux:
Browsing and writing linked data. In Proceedings of the 1st Workshop on Linked Data on
the Web (2008)
4. Berners-Lee, T.: Linked Data, http://www.w3.org/DesignIssues/LinkedData.html (2006)
5. boyd, d and Ellison, N.B.: Social Network Sites: Definition, History, and Scholarship.
Journal of Computer-Mediated Communication 13 (1), 210-230 (2008)
6. Brickley, D. and Miller, L.: FOAF Vocabulary Specification 0.98,
http://xmlns.com/foaf/spec/ (2010)
7. Buchegger, S., Schiöberg, D., Vu, L.H., and Datta, A.: PeerSoN: P2P social networking:
early experiences and insights. In Proceedings of the Second ACM EuroSys Workshop on
Social Network Systems, ACM, 46–52 (2009)
8. Cozzatti, J.P.: A Perfect Storm of Whales, http://engineering.twitter.com/2010/06/perfect-
stormof-whales.html (2010)
9. Datta, A., Buchegger, S., Vu, L.H., Strufe, T., and Rzadca, K.: Decentralized Online
Social Networks. In B. Furht, ed., Handbook of Social Network Technologies and
Applications, Springer, 349–378 (2010)
10. Datta, A. and Sharma, R.: GoDisco: Selective Gossip Based Dissemination of Information
in Social Community Based Overlays. Distributed Computing and Networking, 227–238
(2011)
11. Facebook: Statistics, http://www.facebook.com/press/info.php?statistics (2010)
12. Heckmann, D. and Krueger, A.: A user modeling markup language (UserML) for
ubiquitous computing. Lecture Notes in Artificial Intelligence 2702, 393-397 (2003)
13. Koubarakis, M., Tryfonopoulos, C., Idreos, S., and Drougas, Y.: Selective Information
Dissemination in P2P Networks: Problems and Solutions. ACM SIGMOD Record 32 (3),
71-76 (2003)
14. Perez, J.C.: Facebookʼs Beacon More Intrusive Than Previously Thought,
http://www.pcworld.com/article/140182/facebooks_beacon_more_intrusive_than_previou
sly_thought.html (2007)
15. Pouwelse, J.A., Garbacki, P., Wang, J., et al.: TRIBLER: a social-based peer-to-peer
system. Concurrency and Computation: Practice and Experience 20 (2), 127–138 (2008)
16. Ratnasamy, S., Francis, P., Handley, M., Karp, R., and Shenker, S. A scalable content-
addressable network. In Proceedings of conference on Applications, technologies,
architectures, and protocols for computer communications, ACM, 161–172 (2001)
17. Rhea, S., Godfrey, B., Karp, B., et al.: OpenDHT: A public DHT service and its uses.
ACM SIGCOMM Computer Communication Review 35 (4), 73–84 (2005)
18. Seaborne, A., Manjunath, G., Bizer, C., et al.: SPARQL Update,
http://www.w3.org/Submission/SPARQL-Update/ (2008)
19. Sun, L., Upadrashta, Y., and Vassileva, J.: Ensuring quality of service in p2p file sharing
through user and relationship modelling. In Proceedings of User Modelling UM03
Workshop on User and Group Models for Web-Based Adaptive Collaborative
Environments, Johnstown, 57-66 (2003)
20. Webster, A. and Vassileva, J.: Push-poll recommender system: Supporting word of mouth.
In Proceedings of User Modelling, UM2007, Corfu, Greece, Springer, 278–287 (2007)
21. Webster, A. and Vassileva, J.: The KeepUP recommender system. Proceedings of the
2007 ACM conference on Recommender systems, ACM, 173–176 (2007)
22. Whitehead Jr, E.J. and Goland, Y.Y.: WebDAV: A network protocol for remote
collaborative authoring on the Web. In Proceedings of the sixth conference on European
Conference on Computer Supported Cooperative Work, 291–310 (1999)
23. Yeung, C.A., Liccardi, I., Lu, K., Seneviratne, O., and Berners-Lee, T.: Decentralization:
The future of online social networking. In W3C Workshop on the Future of Social
Networking Position Papers (2009)