You are on page 1of 11

Mobile Social Networks: Architectures, Social Properties and Key Research Challenges

Vysakh S, Student, MCA S5 Kristu Jyothi College Of Management and Technology


Vysakhs1990@gmail.com

AbstractMobile social networks(MSNs) are mobile communication systems focusing not only on the behaviour but also on the social needs of the users. In a broader view all mobile systems used by people in their everyday lives can be characterized as MSNs, since all interactions taking place follow social rather than random patterns. Whether the deployed communication system takes into account the social background of the underlying network, depends on its form and capabilities. This article presents a review of the relevant work published in MSNs. Initially the basic architectures and components are summarized. Then the basic social properties of the network, as found in the key literature, are extensively examined. These properties are the main source of inspiration for new MSN protocols and applications, especially for non-centralized systems. Finally the key research problems and the open issues in the area are presented, including future applications and privacy concerns. Index TermsMobile social networks, social properties, social metrics, neighbours, communities, routing;

I. INTRODUCTION THE NATURE of human relationships has troubled mankind for centuries if not for thousands of years. Even so, we can still not quite understand and perceive the way people interact with each other. What we do know though, is that our social relationships, although they cannot be accurately described, do have some specificandvery distinctive properties. One of the most obvious from our everyday experiences is that this is a Small World. Wattspresentsthis phenomenonand the experimentsprovingthat human social networks have very high clustering coefficient, while on the other hand the typical distance between people is as small as random. In simple words this means that in a social network, people belonging to a specific group tend to have close relationships with all the other members inside that group, while on the other hand, they also have a few relationships towards other groups in almost a random order. But before a further investigation on the area takes place, the notion of the Social Network should be defined. Social networks are sets or groups of people with some

patterns of contacts or interactions between them, forming meaningful social relationships. These relationships could vary from physical contacts to interest sharing and financialinterdependencies. Following the rapid evolution of the internet the first web-based social networks were created. Initially they had the form of forums and chat rooms. The first mainstream form of social networking though, came with centralized file sharing networks. When these were brought down by litigation, peer-to-peer networks appeared and afterwards revolution came through the dedicated social websites. Social networking has attracted billions of active users under major Online Social Network (OSN) systems such as Facebook. These systems allow people with common interests to come together and form virtual social communities. Nowadays, these social networks are increasingly going mobile, i.e., being used on mobile devices by end users thus rendering the so-called Mobile Social Networks (MSNs). Most OSNs regard their advancement into mobile as one of the key initiatives and key to their growth. Apart from OSNs, other major IT companies such as Google and Microsoft have also made strategic moves to embrace MSN technologies. Zivand Mulloth, present the evolution of social networking platforms, distinguishing them into three major categories. The purely web-based, the mobile-based mentioned previously and a hybrid one, combining the global and local nature of social interactions. A very clear visual representation of this distinction, along with examples of social networking platforms belonging to each architecture, is given in Figure 1. Various social networking platforms are placed on a 2D area based on their focus and type. The horizontal axis represents the focus of each platform, whether this is more social, content development or commercial. Socially-focused platforms just enable users tocontact each other, but provide no content development abilities. However, the more users share information which can be recalled by others, the more the content is developed. If the use of this content can be used for commercial reasons, then the platform can be seen as commercially focused. The vertical axis categorizes the social networks accordingto theirtype, whether mobile-based, hybrid or web-based. As a result, nine distinctive areas can be identified. Each platform can only have a specific

type, since none can be at the same time mobile-basedand webbased, but not hybrid. On the other hand, as far as the focus is concerned, social networking platforms can focus more or less on each area, combining them. The iniquitousness of mobile devices today is such that there is one mobile device for every two people in the world , making the networks created by the mobile users themselves, ideal for exploitation. These networks have some distinctive properties that differentiate them from other forms of dynamic networks. The knowledge gained from the observation and research of human social networks provides many useful tools in the attempt to efficient utilize MSNs. Consequently, MSNs can be defined as user-centric mobile communications systems, in which the ability of the users to exchange data is enhanced through socially-aware algorithms. It has to be made clear though, that the social properties of the network are mainly utilized in cases of limited or no infrastructure, such as ad hoc or opportunistic networks. In the field of mobile communications, because of the availability of technologies such as cellular networks, service providers mainly concentrated on providing users with access in the web-based social network platforms. This approach largely ignored the social properties of the underlying network, covering it behind a curtain of extensive coverage. This however came at the cost of huge infrastructure expenses. Recently, another form of communication attracted the interest of researches, previously used mainly in military activities, Delay Tolerant Networking DTN). This enables users toexchange data and communicate with each other without the need for centralized serverfacilities, at the cost though, of high latency. Physical human networks present a challenge for every communication system, since they are constituted by sparsely located disconnected groups of users. DTN follows a store-carry-and-forward mechanism to allow inter-communication between users belonging to physically distant groups. II. ARCHITECTURES AND COMPONENTS OF MSNS Online social networking services such as Facebook, LinkedIn and Twitter have become very popular in the last few years. Their huge success, but also the fact that these web-based centralized applications have lead to a shift from physical locations to virtual communities , pushed researches to employ similar functionality to mobile networks. The main advantage of this attempt is the consequent connec-tion between the physical location of the node and its social interactions. However, a mobile network consisting of mobile devices carried by humans is by nature a sparse dynamic environment, where some nodes or even groups might be temporarily or constantly disconnected. Bringing this huge number of individuals together and allowing them to socialize and exchange information, is no easy task. According to the way users are able to inject and access

information, MSN architectures can be distinguished into three categories, the Centralized, the Distributed and the Hybrid one. A. Centralized MSNs This is the most common architecture used for the deployment of MSNs. All the information concerning the members of the social networksis preserved in remote servers belonging to a service provider. End nodes use the deployed wireless infrastructure of a cellular network, Wi-Fi, or other similar technologies, to gain access to the remote service providers, in order to communicate with each other or to access and change their personal information. This client-server structure can be further enhanced by the use of third party application servers (SMTP Mail server, VoIP Servers etc.). The benefits of the centralized architecture are obvious and indisputable, as it can provide high quality services through the currentlyavailable infrastructure. Moreover, it bridges the gap between the physical location of the user and the global social community. Kemp and Reynolds argue though, that this hub and spoke model of clients and servers may cause bottlenecks, since all traffic should pass throughthe hub.Additionally, each transaction should pass up one spoke to the hub and then down another,evenwhenthe two endpointsofit are physically close. One more major disadvantage has to do with the huge amount of personal data stored in a few physical locations, under the authority of an alien entity. Centralized MSNs are mainly an extension of their web-based predecessors. By using specific mobile applications or simply by using a mobile browser, users are able to connect to a centralized server for sharing and exchanging information. In that sense, when a mobile user accesses his MySpace account from his mobile he is a part of a greater MSN. However, there are plenty more new mobile applications, targeting exclusively on mobile users, such as Juice Caster.Researches in this area mainly focus on creating a unified middleware, able to provide the basis of future mobile social networking applications. Such an attempt was made by Borcea et al. 1. www.facebook.com 2. www.linkedin.com 3. Twitter.com 4. www.myspace.com 5. www.juicecaster.com who present MoBiSoC, a mobile social computing middleware providing the necessary functionality for the deployment of mobile social computing applications (MSCAs).

The main goal of this attempt is to provide a unified solution for both people-centric and place-centric applications and is divided into several modules able to be distributed in multiple servers. Another interesting proposal was made by Ma et al., who introduced an application called MoViShare. This centralized application is able to provide a location-aware video sharing platform. A survey on the most important middleware projects so far is presented by Karam et al. Amongst the interesting features of this survey are the middleware challenges in the MSN environment as well as a framework for comparing the various projects. One moremethodofdeploying centralized MSNs is through the use of wireless sensor networks (WSNs). Sensors have al-ready infiltrated daily life and there is a plethora of advantages fromthe integrationof mobile social and sensor networks. By utilizing contextual information, WSNs can provide much more detailed and personalized social services. One prominent example is presented by Milosz et al. [13]. The authors designed and developed a smart phone application able to gatherinformationthroughthe sensors of a mobile device. The information is then transferred to a backend software. This is able to support push as well as pull-based data publishing via a HTTP transported markup language and a standard web-service based API respectively. A general centralized architecture for MSN services in the context of social services, is given by Chang et al. The main advantage of this proposal is its ability to provide location-based interactive services, without the

installation of an additionallocationmodule.An overviewof this architecture is presented in Figure 2. In that figure, users are split into two different social groups and are able to access the social information through the deployed wireless infrastructure. A hand-over mechanism can be employed to provide connection continuity. B. Distributed MSNs The key feature of the distributed MSNs is the total absenceof centralized servers. Mobile users are able to communicate and access social information only by connecting to other users. Therefore, the network devices themselves have to store and route the social data until the correct destination is found. However, this does not necessarily mean that there is no infrastructure at all. Cellular networks and access points can be used to provide inter-connection between two nodes. Nevertheless, the mobile terminals nowadays are further equipped with multiple wireless interfaces such as Bluetooth and WiFi, providing them with the capability of connecting in an ah hoc manner, with no need for any kind of infrastructure. The various ways in which distributed MSNs can be de-ployed are depicted in Figure 3. In thatfigure, every node is characterized by an integer ID while inter-communication is provided by three different techniques, namely a WiFi access point, a 3G network base station and an ad hoc network. The fact that nodes are

mobile and can roam through the three offered communication technologies is also depicted. Three non-overlapping social groups are being considered, characterized by the color of each user. Having in mind their capabilities, mobile devices all over the world can be seen a vast mobile ad hoc network with enormous potentials. In an approach by Pietilainen et al. introducing a middleware called MobiClique, the utilization of such a network is presented. MobiClique is able to create and preserve a mobile ad hoc social network, providing content exchange by utilizing the user mobility. However, it lacks the ability to predict user contacts and therefore performs a form offlooding as a means of content dissemination, leading to low efficiency and very high resource usage. As a general idea though, it may provide a valuable insight on how distributed mobile social networks will be in the near future. Although the distributed MSNs provide many advantages, such as low cost deployment and maintenance, they also pose many challenges. These include high latency, low delivery rate comparing to the centralized architecture, aswell as privacy concerns. Distributed MSNs can further be divided into two different categories. In the first one users share content directly or by using some deployed infrastructure (but non central in any way), such as EyeVibe In the second one however, which is the most challenging, intermediate nodes might be required to provided inter-communication between two end users. Because no infrastructure is considered, the mobile nodesthemselvesare designatedto route or carry the messages until thefinal destination. This is why, most of the research in the field of MSNs targets this area and the inherited problems appearing due to the dynamic nature of the underlying ad hoc networks. A very characteristic example of a middleware belonging in this

category is given by Sarigol et al. [15], called AdSocial. This is a software platform able to support social networking applications such as video calls and gaming over ad hoc networks. Nevertheless, it is widely accepted that delay tolerant networking (DTN) is preferable to overcome the dynamic and sparse nature of the underlying network, and provide intercommunication in delay tolerant networks. This is because in DTN it is not mandatory to have an a priori knowledge of the network topology. C. Hybrid MSNs Hybrid MSNs are an extension of the two major archi-tectures mentioned previously. They are created by a com-bination of both in the sense that despite the existence of centralized servers containing all the social information, direct data exchange between nodes is also possible, as presented in Figure 4. This assumes the existence of the necessary data locally as well as remotely. An interesting description of a hybrid architecture is introduced by Molina et al. in, where nodes asking for specific information can access it via the centralized server using the cellular infrastructure and might as well share information with each other in an ad hoc manner. Although the introduction of the ad hoc communication reduces the load the backhaul network suffers,this approach considers neither user mobility nor disconnected areas in the network. One big advantage of the hybrid architecture is that it can be established very easily by extending almost every existing centralized or distributed MSN. Han et al propose the exploitation of opportunistic forwarding in MSNs as a method for offloading mobile data from 3G networks. By using complementary technologies already available in mobile devices, mobile data traffic is reduced up to

73.66%. Three algorithmic approaches are examined by the authors, a greedy, a random and a heuristic one, with only the later one being realistic though. However, the heuristic approach raises privacy concerns, since the everyday contact information of the users is required to be uploaded to the service providers. D. Components The main components of the MSNs architectures mentioned previously are no different from the necessary components for the deployment of the underlying network. Having that in mind, they can be narrowed down to: Mobile Devices: The mobile devices that users carry, allowing them to access the available content and form social groups. The can take any form, as long as they are mobile and might have multiple network interfaces. Mobile devices apparently are always present in all centralized and distributed and hybrid MSNs. Network Infrastructure: Fixed or mobile infrastructure, including cellular network base stations and access points, providing inter-communication between source and destination. It can be present in all centralized, distributed and hybrid MSNs. Content Providers: Centralized servers storing and distributing the social content, mainly using the network infrastructure to make it available to the mobile users. Socially-aware algorithms are used to address security and privacy concerns. Third-party servers can also be included in this category, although they might not provide the users with services. Obviously these components can only be present in centralized and hybrid MSNs. III. SOCIAL PROPERTIES AND METRICS OF MSNS It has been made clear so far, that although mobile social networks may include a variety of systems and technologies, not in all cases the properties of the underlying social network are being utilized. At least not in the same level.

For instance, a cellular network designer will take into account that a more populated area will require smaller cells of higher capacity, which is an adaptation of the system design to the social factors. It is not however applied directly and more importantly, it does not dynamically adopt to the changes of the underlying social network. This is why, the social properties mainly concern distributed MSNs, where limited or none infrastructure is available. Due to this limitation, identifying andutilizing the propertiesof peoples behavior plays a major role in improving the routing mechanisms effectiveness (e.g. number of hosts able to be reached) and efficiency (e.g. lower bandwidth usage and lower latency). Researchers have introduced graphs for studying social networks. This gives them the ability to use the tools provided by graph theory for network manipulation and formulation of the problem of predicting future encounters. The main consequence is the heavy dependence of the problem formulations on the way the social graphs are constructed. Self-reported and detected social networks are known to differ , but there are also studies, like the one from Eagle et al. claiming the opposite. On the other hand, even in detected social networks, Hossman et al. point the problem of contact aggregation .In other words, when the social graph is constructed by the observed contacts, the aggregation when taking place in a constantly growing time window, leads to a meshed network. They argue that the optimal aggregation window depends not on time, but on obtaining a constant social graph density. After a graph has been constructed based on a socialnetwork, the concepts of social ties, neighbours and communities can be introduced. Additionally, network analysis can be performed through metrics such as centre and clustering coefficient. At this point there are two different analysis approaches, the centralized and the distributed one. The difference between the twois the way researchers perceive the network. The centralized one regards the social graph as a whole and the assumption is made that the total amount of the social information is available to all nodes. In the distributed one, a more practical implementation is followed. Each node is regarded to have limited access to the social information,

therefore being able to perceive only a subset. For instance, Freeman introduced the notion of ego network, where a node has only knowledge about its neighbours. Consequently it cannot calculate its network betweenness centrality, but only another value, the ego betweenness, representing it centrality inside its ego network. An analytical presentation of the most important social properties and metrics considered in MSNs follows, while a summary is given in Table I. These properties affect the communication patterns of the nodes in a positive or negative way and they are taken into consideration by researches, in accordance to the desirable results of the proposed protocols.

IV. KEY RESEARCH PROBLEMS AND OPEN ISSUES


A. Mobility models & Social Awareness There is an increasing number of data sets capturing real mobility and collocation traces like CRAWDAD and there have been many suggestions, such as in , on how they could be used for the extraction of social ties. However, in most of the cases they deal with very specific scenarios, thus making generalization very hard. Consequently, synthetic traces are usually preferred to real traces. It has been made clear so far, that the opportunistic routing protocols in MSNs focus increasingly on the social aspects of the network. Therefore the mobility models have to have the same focus as the evaluated protocols, the social factor. The widely used Random Way-Point mobilitymodel has been proven to be unrealistic and cannot represent the social relationships that influence our everyday choices. Consequently, more comprehensive mobility models have been created to address this need. What is surprising in this area of research is that although there are many surveys on mobility models for ad-hoc networks, based on the Random Waypoint and Random Way, such as , up to our knowledge, there is no complete survey on social network based mobility models. Hossmanetal.argue that the social properties of mobility are crucial for the design of efficient DTN protocols and therefore the models should be able to accurately capture microscopicandmacroscopicsocial properties. Additionally, as Hsu et al. point out, a good mobility model should achieve three more goals, (1) be able to generate realistic mobility patterns, (2) be mathematically tractable andfinally (3) be flexible enough to provide qualitatively and quantitatively different mobility characteristics. Realistic mobility patterns refer to the ability ofa mobility model to reproduce traces with similar characteristics to those of real mobility traces. Although the conditions under which mobility traces are gathered, can never be precisely defined or repeated, it has been noticed that traces,no matter whether run correlated or not, have an astonishing number of common properties . Cho et al. propose thePeriodic & Social Mobility Model(PSMM), mainly targeting the macroscopic human mobility properties. This model is based

on thefindings fromthe analysis of three datasets, containing user check-ins inlocations during largeperiods of time. Some of the mostinterestingfindings are the fact that the effect of friendship sonmobility increases with their distance and the limited influence of social ties to trajectory similarity. PSMM defines a number of primary locations on a map for each user and then predicts the probability of checking in any location according to the Gaussian distribution. These probabilities are additionally modelled as a function of the day time. A social factor is also introduced, as a users check-ins affect his friends decisions. The major disadvantages of this model though, are that it is only able to generate large scale check-in traces and that it needs location-specificconfiguration. A model of major interest is the Community-basedMobility Model (CMM) . According to CMM, nodes are organized into clusters, the so called communities, which are placed on a map grid. Nodes move in straight lines, choosing a next target whenever they reach the current one. These targets are chosen randomly inside a non-randomly chosen grid square (cell). This choice is made by measuring the social attractivity of each grid cell towards the deciding node. Social attractivity is defined as the aggregate attraction of all the nodes that are currently residing inside that cell. Apart from the attraction between nodes (e.g. between friends) though, in real life there are also physical locations attracting people such as a cafeteria or a school. However, this mobility model is unable to capture the attraction exerted on users by physical locations. To address this issue Boldrini et al.proposed the Home-Cell Community-based mobility model (HCMM). According to this model, each cell exerts a constant attraction to nodes, according to the social ties towards the initial nodes assigned inside it. One disadvantage of HCMM though, also present in CMM , is its inability to closely reproduce the available data traces, as it fails to capture basic mobility characteristics such as group mobility and temporal regularities. The Working Day Movement model is one of the newly published models and attempts to offer a new insight, combining three major human activities called sub models, being home, working and evening activities. It also incorporates a transport sub model to mimic the movement between two locations of interest, representing city traffic along streets and highways. Although it tries to capture some aspects of human life, it lacks the social structure of CMM and therefore is not appropriate for community based routing evaluation, since it does not create mobility traces according to a social network. In simple words, it does not consider any form of social attractively between the participating nodes. The Time-Variant Community mobility model (TVC), captures non-homogeneous dependencies in both space and time and is able to generate a large variety of synthetic traces. The similarity between the TVC generated traces and real mobility traces, makes it a quite interesting model. TVC also in-corporates the notion of communities. In that context however, communities correspond more to locations visited frequently by the nodes, than groups of

nodes themselves. Consequently, the model provides no correlation between the social network and the generated traces, unlike the community-based models mentioned previously. The simulation time on the other hand, is divided in periodic epochs of variable length, generating periodical mobility behaviours. SLAW is a another mobility model, able to produce very realistic synthetic human walk traces, but applies only to very specific scenarios. It attempts to capture the Levy flight property of human displacements, which follow a Pareto distribution. One of thefinest social mobility models presented is Ge-SoMo by Fisher et al. The basic concept of this model is defined by a direct connection between the underlying social ties defining the mobility and the resulting amount of contacts between the nodes. It has three different types of attraction, namely the location attraction,the nodeattraction and the node repulsion. Concerning temporal regularities, a metric called network similarity index is introduced to study the tendencyof users to reappear in a location after a certain amount oftime. Themost important feature of GeSoMo however, is the introduction of the conformance requirement. According to it, the connectivity graph, constructed from the frequency of this article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.contacts between nodes, should conform to the initial social network used for defining the mobility patterns. Although GeSoMo provides many of the desirable properties and it can provide inter-contact time distributions very close to real ones, the paper in which it is presented does not include any results of the distribution of contact durations, another very important tool for mobility model analysis. Moreover, although the model incorporates a number of social networks such as the Caveman model, the Holme-Kim model and the Toivonen model , it does not allow an arbitrary social network input. Having reviewed some of the most interesting socially-aware mobility models so far, it can be argued that some of the most desirable properties of a mobility model can be identified. First of all, a communitybased model seems like the best option for generating macroscopically realistic human traces, as they are the only ones that can capture the Small World phenomenon .Besides, as Hossman et al.argue, synthetic models usually fail to accurately capture the inter-community linkage. This is only possible if the trace generation process takes into account the social relationships between the nodes and the individual communities. The conformancefeature, introducedby Fisher , is another desirable property. Providing a mobility model with the ability to produce realistic traces, characterizing a single, unique social network, would constitute the collection of real world traces almost obsolete. Finally, apart from spatial regularities, present in most of the mobility models mentioned earlier, Fisher et al. point that the human habitual behaviour is characterized also by temporal regularities. All the mobility models mentioned previously along with the most important features are presented in Table II. As explained above, these features include a

community-based structure, the influence of social attraction in user mobility choices, the conformance to the underlying social network and the ability to also capture temporal correlations, instead of only spatial ones. The main purpose of that table is to show that no model so far provides all the necessary features. From the regarded mobility models, only CMM, HCMM and TVC incorporate the notion of communities. On the other hand, TVC as mentioned also by Hossman et al.doesnot take into account the social attraction between nodes for calculating the transition probabilities. Apart from CMM and HCMM only PSMM and GeSoMo have such a feature. GeSoMo on the other hand, is the only one that generates traces conformingto the initial social network.No othermodel has proven such a property. Finally, temporal regularities have been introduced in most of the models apart from CMM and HCMM. Therefore, nodes have the tendency to reappear to previously visited locations after certain periods of time. It is obvious that none of the above models is able to meet all the four requirements. This mainly has to do with the inherited complexity of the human mobility and the mathematical tractability of the mobility models. B. Socially-aware Routing Metrics In the previous section, the basic social properties of MSNs were presented. These can be utilized as metrics to improve the performance of various functions during the operations of an MSN. For example, clustering coefficient and tie strength could be employed to help mobile nodes detect their communities and organize themselves. On the other hand centrality could used to detect the most central nodes, for routing purposes, but also to avoid congestion. Apart from the presented ones though, there are many other propertiesthat could utilized as metrics, depending on the specificationsof the scenario investigated. Additionally, researchers tend to make changes, minor or not, to the previously applied metrics in order to maximize the efficiency of their algorithms. The most active part of research in MSNs is found in this area. The amount of publicationsdealing with opportunisticrouting, targeting mainly distributed mobile social networks, reflects the interest of the academic community. In ,Dhakan and Menezes propose a social structure scheme in order to provide intercommunication over a mobile network. Nodes are organized into clusters (communities), with a specific node in change (head). The metrics though, used for the hierarchical organization of the nodes, are not social at all. Instead the transmission range and Mobilitynormalized values are employed to distinguish the most suitable nodes to become cluster heads. Mobility plays an important role in another two approaches, namely CAR and SocialCast . In these proposals, the change rate of connectivityis used as a metric to identify the best carrier for a message towards a specific destination. The change rate of connectivity describes the mobility differences between the present node and

all the others. Additionally, another metric is used, which is influenced by the social behaviour of the nodes,colocation. This metric describes the probability of a node to be located in the same cloud as the destination.In PROPHET, a probability metric, called delivery predictability, is computed. This metric represents the probability for each node to contact all the known destinations, based on the contact history. Upon eachcontact, the delivery pre-dictability correspondingto the encountered node is increased. This metric also ages in time, making the algorithm able to adopt possible future changes of node ties. Trying to maximize the utilization of the social knowledge available, Haahr and Daly investigated the metrics of betweenness centrality,similarityandtie strength to determine the best carrier. Each of these metrics is related to a different stage of the routing process, trying to locate initially the best carrier towards the destination community and afterwards locating the proper node. Similarity is calculated according to the number of common neighbours two nodes have, while the tie strength is defined by multiple utility values, such as frequency, intimacy and recency of the connections between two nodes. Gao et al.proposea multicast routing scheme that uses social metrics in favour of latency and the number of relays. Instead of using the traditional definition of centrality, the authors introduced a new metric called cumulative contact probability, that represents thethe average probability a node ito meet a random node within a certain time periodT. Bulutand Szymanski [30] also introduce a new metric on their Friendship-based routing algorithm, called social pressure metric. Friendship-based routing offers users the ability to exchange messages with the other members of the MSN, by tracking their past contacts and defining the social pressure metric for every social relationship. This metric takes into account the frequency, the total contact period and the average separation period in order to characterize a link. Furthermore, in order to address issues arising from the ageing process of the metric, depending on its value for specified time periods, friendship groups are formed. Zhang et al. also use the notion of friendship as the basic idea behind their diffusion algorithm. Their goal was to create a data diffusion scheme with reduced delay. The main issue they tackled was the limited contact time available in mobile wireless connections, in comparing to the time needed to transmit all the necessary information. Therefore, the authors employed the concepts of friendship and homophily to determine the data propagation order. According to the homophily phenomenon, friends tend to share more common interests than strangers. Consequently, when two nodes meet, if they share a friendly relationship, the data of their common interests are diffusedfirst. If they are strangers, the data of theinterests they dont share are transmitted first. No informationon how friends are detected is given though. Hui et al.useFreemans betweenness and degree centralitymetrics to route messages towards the most popular nodes, before thefinal destination is located. The novelty of this scheme is the differentiation made between global and local

centrality. In this approach, a community structure is assumed, and each node has a global centrality value, and a local one, for every community it belongs to. Messages are forwarded to more popular nodes until the destination community is found. Afterwards, it bubbles up the most popular nodes locally, until the destination is reached. Li and Wu in Local Com epidemic forwarding scheme use a metric calledcloseness, which is defined as the Gaussian similarity function of the average separation period between two nodes. The variance of the separation period is also recorded to reflect theirregularity in the relationships. From closeness and irregularity,similaritycan be extracted, that is used to characterize a link. By representing links using the similarity metric, the neighboring graph of the network can be created and used for community detection. C. Privacy Mobile social networks enable users to share not only personal messages, but information on their current location, activities and personal interests. Transmitting all this sensitive information through the wireless channel posses major privacy challenges. Especially ininfrastructure less distributed MSNs, intermediate nodes (carriers) are required to store and forward the personal data of others. If no action is taken, this information will be visible by everyone. Moreover, even in centralized MSNs, as Chen and Rahman point, both the feedback and the control of the information accessibility are unreliable and weak. In their analysis, commercial mobile social applicationsin iPhones were studied following the Bellottiand Sellen feedback and control framework [78]. The major weakness of these systems is the extended interconnectivity between the multiple social networking applications that lead to inconsistent access policies, easily breached. Preibusch and Beresford [4] also deal with privacy concerns in centralized architectures. More specifically, they argue on the significance, in terms of privacy concerns, of uni-lateral friendship disclosures in mobile social sites. Since social relationships themselves constitute a very sensitive piece of information enclosed in the social network database, they should be protected. However, if both parties dont hide this connection, this constitutes anunintended exposure. They proposed a lightweight solution, using hashed identifiers in FOAF (friend-of-a-friend) files to overcome this problem. For the same issue, but this time targeting purely distributed MSNs, Parris [79] proposes a scheme under which nodes will make changes on their user profiles (by adding or removing nodes). Although the friendship list in these cases is commonly used for routing purposes, it is shown that up to 60% of the nodes from the social network can be removed, while still maintaining a delivery ratio of90%. In [80], Beach et al. propose a method enabling anonymous social network information exchange, using anonymous identifiers (AID), generated by a central identity server (IS). This approach is able to protect the user identities and solve problems such as direct anonymity, the

K-anonymity problem, as well as spoofing and eavesdropping attacks. One drawback of this approach though, is that it requires an internet connection with the centralized identifier server for authentication and verification purposes. This is something that cannot be guaranteed in distributed MSNs, while it can be proven to be quite costly even in centralized ones. Krishnamurthy and Wills [81] examine the leakage of private information in mobile online social networks (mOSNs).Their study focuses mainly onapplications and OSNs createdlargely to deal with the new mobile context. The new con-cepts concerning privacy in MSNs arepresenceandlocation. These two notions, although related, should not be confused. Location is defined by the accurate GPS coordinates of onesposition, while presence deals with the general presence of an individual in a larger area (such as a cafeteria or a sports event). A number of social platforms were examined for possible privacy leakages. The examined issues were 1) the availability of user information within mOSNs, 2) the location and presence, 3) the interconnections of mOSNs, 4) the leakage to third-parties and 5) leakage of personal identifiable information (PII) to third parties. The results show leakages in all of these issues by multiple social platforms, something that cannot go unnoticed. D. Applications Regardlessof the challengesin deployingMSNs, the mobile social applications have attracted a lot of attention because of their indisputable advantages, such as easy and cheap deployment and the connection provided between the physical location of the user and his social interactions. The variety of applications based on mobile social networks is impressive and the only certainty is that it will keep on increasing. Some of the most popular application categories are the following: Social Services: This is the most common category of MSN applications. They include the extensions of the web-based social network platforms, such as LinkedIn, as well as the purely mobile ones, like Dodgeball [5]. In these kind of applications, members have the ability to share their information and views, as well as personal files. Although in a way, all MSN applications can be regarded as social services, in this context, they are considered to include only those providing purely social networking services. Vehicular Networks: A greatly unexploited source of interactions in social networking is the daily roadway commute. People tend to use the same roadways to travel to specific destinations, with a high periodicity. A system attempting to utilize this phenomenonis Road Speak, presented by Small donate al.. This system allows users to automatically join communities called Voice Chat Groups (VCGs) along popular highways and roadways. Participants have the ability to exchange voice messages belonging to their VCG through a central Road Speakserver, acting as a coordinator.

Healthcare Services: MSNs applications are also able to assist people with special needs or those that require medical support. For instance, Chang et al. [83] propose a general architecture using GPS-enabled PDAs that report the location of patients with severe mental illness to a central server via GPRS. This system is able to match the patients to caregivers based on service request and availability. Wearable MSNs: This is another form of social applica-tions. People wear devices able to collect and exchange information related to their social interactions. Wearable MSN can be used not only for behaviour modeling and entertainment purposes but also to aid users in their day-to-day activities . An interesting wearable MSNdevice was presented byKanis et al. , in the form of a bracelet called iBand. This device uses an infra-red transceiver for communication purposes and every new user starts by entering some contact data and a personal LED logo. Two iBands exchange this information during a handshake. Social Learning Networks: Huang et al. [52] conceived a novel application of MSNs. By identifying the hu-man as an intelligent resource, the sharing of collective intelligence can be realized. Users are given the ability to search for resources or other users, in order to form communities of practice, such as reader communities. Techniques such as the vector space model are employed to categorise the searched items according to their similarity. Recommender Systems: A novel application of mobile social networking targets recommendation systems. These systems, by tracking the behaviour of the user and mining the context information available in the mobile devices, can provide recommendations on a variety of topics. Quercia et al. proposed FriendSensing and Sensing Happiness, two frameworks that aim to locate social contacts and nurture them respectively. Both are based on logging past encounters using short range radio technologies. Jung [87] focused on providing personalized content to mobile users. He proposed an interactive approach based on social network ontology, in order to detect social relationships. The main assumption of this Work was that personal contextis dependedon the context of other users. Min et al on the other hand, used Bayesian networks to recommend candidate callees. V. LESSONS LEARNED One thing the reader should keep in mind is that MSNs are not just another form of netwo rking, such as lets say ad-hoc networks. Theyare the result of a new era, wheresocial human relationships and digital communications merge. The fact that MSNs are mobile has more to do with the fact that people and thus their social relationships are mobile, rather than the mobile capabilities of some wireless devices. This is one of the reasons this paper concentrates on the social aspects of MSNs, rather

than the network or the application layers. The social properties of MSNs do not belong to one of the layers of the networking model, but they strongly influence the operation of each layer. Besides, it is the properties of the social network that dictate the behaviour of the users. Consequently, social-awareness in all the layers is needed for the successful implementation of MSNs in the future. Re-searchers are keen on incorporating the extracted knowledge from the underlying social network in the routing decisions and the privacy protection mechanisms. The interesting part is that, since all layers are affected by the social factors, social knowledge can be extracted from every layer. For instance, the physical layer can detect other users in proximity in order to categorize them as friends and at the same time social applications can detect friends based on the users input. Crosslayer communication can be used for transferring and utilizing the extracted knowledge. The MSN systems presented inSection II, especially the centralized ones, usually are quite monolithic and do not follow the guideline mentioned above. They just exchange user input or GPS data with the online services, following the mechanisms of online social services and failing to utilize useful information available in the mobile devices. Apart from that, social neighbor discovery is performed through the energy consuming location detection method via GPS or using the mobile communication network . This could easily be avoided by the use of short range radio technologies like Bluetooth, as well as utilization of social information found in the address book. The most prominent social properties, as found in the bibliography, were presented in this paper. In most cases though, researchers focus on a small subset of properties, related to their goal. This can be used as a guideline to avoiddealing with a huge amount of information. This does notnecessarily mean worse results. For instance, theFriendshipbased Routing algorithm performs much better than SimBet although the later involves a higher number of social metrics such as similarity, between and tie strength. Additionally, social networks are very complex structures and irrelevant information will only cause confusion and difficulties in handling the data, not to mention privacy issues. On the other hand, certain social properties cannot be considered separately. For example, GeSoMo mobility model, considers contact duration but not inter-contact times. These results in good results as far as the contact times are concerned, but the resulting mobility traces cannot be considered as realistic. On the contrary, community-based models such as CMM may generate less accurate traces in terms of contact duration, but they are more valuable as a whole. This comes as a result of the incorporation of all the important aspects of mobility. One more important observation made while performing this survey, has to do with the key areas of research. These areas concentrate not on developing new theories on social networks, but on utilizing the knowledge produced by so-cial sciences onto networks. For instance, the new metrics suggested for routing do not focus on developing new social metrics, but on identifying what type of

metrics could better recognise the patterns dictated from social sciences. The same applies for the MSN applications. Researchers are keen on developing new applications, in order to provide better services for already existing social relationships and needs. VI. THE FUTURE OF MSNS The majority of applications running on mobiles devicesNowadays pay little regard to the underlying mobile network. Work on MSNs so far is constraint on how to utilize the data available in the mobile phones. These data, mined by special data-mining or pattern recognition engines, are used to benefit the further development of OSNs rather than develop new mobile social networking services. The wireless mobile networks on which these mobile phones operate are not considered, although they play an equally important role to the success of MSNs. Thus, it is crucial that MSNs are regarded as a marriage of the traditional wired-network-based social networks and mobile wireless networks. This vision though, calls for cross-layer information exchange between social networks at the application layer and mobile wireless networks at the network and transmission layer. The architecture of the underlying wireless mobile network could also play an important role in the success of the MSNsThe belief of the authors is that only through a hybrid approach can the immense capabilities of MSNs be fully exploited. The importance of the infrastructure is indisputable for mobile networks, but on the other hand, the ubiquity usnessof the mobile devices opens new horizons for exploitation. Fox argues that the Internet provides the means to store information outside our bodies, beyond the capabilities of our brain to memorize. These kind of mechanisms consequently allow the expansion of intelligence beyond our bodies. MSNs move one step forward and allow the extension of social relationships beyond the physical borders of the past. VII. CONCLUSION MSNs pose a new challenge for the research community and industry,that if addressed could provide people with numerous novel ways of social interactions. The main advantage is the members of these social networks themselves, as they are able not only to use the available services but also to improve and expand them. After all, MSNs are user-centric communications systems, regardless of the deployed architecture. In this paper, the basic architectures and components of MSNs were presented and the major social properties that can be utilized in favour of the system performance were analyzed. Finallythe most key research problems and open issues in MSNs were summarized ,including the mobility modeling, the selection of social metrics, the privacy issues and the current applications available.

ACKNOWLEDGMENTS This work was funded by the Engineering and Physical Sciences Research Council (EPSRC). The authors would also like to thank Andreas Bontozoglou for his useful comments. REFERENCES [1] http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnum ber=6226796 [2] D.J.Watts,Small worlds: the dynamics of networks between order and randomness, 3rd ed. Princeton University Press, 1999. [3] M. E. J. Newman, The Structure and Function of Complex Networks,SIAM Review, vol. 45, no. 2, pp. 167256, 2003. [4] W. Stanley and F. Katherine, Social Network Analysis: Methods and Applications. Cambridge University Press, Nov 1994

You might also like