[ieee 2010 international conference on advances in social networks analysis and mining (asonam 2010)...

Enriching and Simplifying Communication bySocial Prioritization

Juwel RanaCSEE Department

Lulea University of TechnologySE-971 87, Lulea

Email: [email protected]

Johan KristianssonEricsson ResearchSE-971 28 Lulea


Kare SynnesCSEE Department

Lulea University of TechnologySE-971 87, Lulea


Abstract—This paper presents a framework for AggregatedSocial Graphs, the ASG Framework, for developing applicationsthat harness information available on social networking sites andtelecom platforms to make communication simpler and richer.The framework aggregates users social graph by harnessingcontact and social information from multiple communicationenvironments and then calculating the social strength between theusers, based on the users communication history. The frameworkdiscovers communication patterns in different level which iseventually used to simplify and enhance communication servicesby inviting appropriate users for collaboration, selecting suitabletools for communication and prioritizing information flows overthe cloud. The framework also enables new types of applicationssuch as social search clients, smart dialers or contact applications.

I. INTRODUCTION

The convergence of the web and telecom platforms israpidly evolving into a truly ubiquitous environment for anew generation of services that often utilize social networkingtechniques for making communication richer and easier. Userscan now utilize multimedia messaging to blog or share photos,multi-part conference calls to do online collaboration, andgroup-enabled contact applications to manage their socialnetworks.Social networking services (such as Facebook, MySpace andTwitter) have recently become tremendously popular by con-necting users and communities while exploring the opportu-nities for advance communication and content sharing mech-anisms. Harnessing data from these services while utilizingthe user’s communication history makes it possible to buildeven better services and tools, for instance, by utilizing socialranking of contacts and content through data mining of user’son-line interaction. Moreover, in designing services that easilyform micro communities of relevant users (dynamic groups)by studying call logs, blogs or other on-line communicationpatterns.However, it has been observed that users are overloaded due tothe large quantity and popularity of social networking services[1], where the amount of information easily can becomestaggering. To enhance and simplify communication in social

DISCLAIMER: The work has been carried out as part of an academicresearch project and does not necessarily represent Ericsson views andpositions.

networks and to reduce the cognitive overload is thereforea very important challenge. This paper therefore addresseshow to analyze user’s communication history obtained fromdifferent social networking and communication services tobuild a social recommender system, that may recommendor filter users for group communication based on previousinteraction patterns and the social strength between users.The social strength highly reflects daily life [2], where on-linesocial networking and communication services have naturaldependencies with the social strength between users [3]. Thesocial strength between users can be expressed as a socialgraph, that aggregates interaction patterns from multiple ser-vices and that enable a new type of richer and easier commu-nication services. The aggregated social graph can hence beutilized for recommending or filtering users and informationin a personalized manner. This can be conceptualized by thefollowing scenario.Amanda uses several communication and social networkingservices and she has a large number of contacts. Most ofthese contacts are family, friends and colleagues, while someonly share common interests or goals. She interacts with themon Facebook, by twittering or through mobile calls, but shestruggles with keeping herself updated with the most importantactivities in her social networks. Her 30th birthday is comingup, so she plans on having a party together with Melvin andEbba, two of her friends at work that also have birthdaysthe same month as her. She gives them a call and they startplanning what to do and whom to invite.They first open up their dynamic group tool that first filters outall their shared contacts that they have no personal relationto and then prioritize the remaining contacts based on howmuch they interact and how strong their social relation is. Theydecide to invite 50 persons and create a dynamic group for theparty. Amanda then sends out invitations using Facebook tousers frequently there and by SMS to the others, while startinga Twitter feed about the ongoing party activities. She ends thecall by promising to organize the food for the party.Amanda likes Italian food so she searches the new dynamicgroup for persons interested in cooking and invites a handfulof them to come early to the party to help with the preparationof food against free availability of beer - and all accept. Shemakes a shopping list and emails that to her two friends. The

2010 International Conference on Advances in Social Networks Analysis and Mining

978-0-7695-4138-9/10 $26.00 © 2010 IEEE

DOI 10.1109/ASONAM.2010.74

336


978-0-7695-4138-9/10 $26.00 © 2010 IEEE

DOI 10.1109/ASONAM.2010.74

336


978-0-7695-4138-9/10 $26.00 © 2010 IEEE

DOI 10.1109/ASONAM.2010.74

336

party is on! :-)To face the challenges expressed above the following re-

search questions needs to be addressed:

• How can personialized social graphs be built automati-cally by harnessing information from the social network-ing services and communication services?

• How can a personalized service automatically calculatethe social strength between contacts based on the inter-action logs?

This rest of this paper is organized as follows: Section 2 dis-cusses related work, Section 3 discusses a social recommender,Section 4 concludes the paper together with the Section offuture work.

II. RELATED WORK

Social networking analysis and methods for social commu-nity discovery have been discussed in [4], [5], [6], [7], wherediscovery methods generally use static attributes like location,profession or interests for identifying communities. Anupriyaet. al proposes to use communication history for findingthe strongest communities [2]. However, methodologies andalgorithms for capturing and representing the communicationhistory in an unified form is not a well understood researcharea. Discovering communities based on real-time interactionslogs is also something has not been completely solved.Systems supporting activity-oriented social networking cancollect and aggregate real-time presence status together withinterests from different communication services [8], [9], whichenable social search, filtering and recommendation by analyz-ing activity messages. CenseMe, the personal sensing system,automatically shares presence and activity information withthe contacts of different social networks, which can be utilizedfor managing communication services by social filtering [10].However, managing contacts from different social networkswhich in addition to utilizing different platforms is becomingvery hard, as the number of social networking and commu-nication platforms increase rapidly [11]. Instead, informationneeds to be automatically aggregated from the different socialnetworks and communication platform, aggregated and storedin a format that enables more powerful algorithms for socialsearch, filtering and recommendation. This paper proposes aaggregated social graph to solve the problem and by doingso also enable efficient calculation of social strength andoptimization of social ranking.It has been proposed some new social search techniquesin addition to Google’s page-rank algorithm, that typicallyoperate on a social network and performing ranking byanalyzing social data [11], [12]. These techniques exposetags, bookmarks and taggers (users) as social data, while theproposed aggregated social graph expose the communicationhistory. Integration of social networks, such as Del.icio.us1

and Flickr2, and Web 2.0 applications can improve accuracy insocial search and recommendation systems [13], [14]. Martin

1http://delicious.com/2http://www.flickr.com/

Fig. 1. Conceptual Model of an Aggregated Social Graph

et. al proposes a approach of aggregating social data whereuser profiles are integrated by a service provider, where thechances of privacy leakage is considered to be very high[13]. In contrast, aggregation in the proposed framework isfundamentally based on users’ own access to social data,possibly in conjunction with access to social data throughservice owners.COI (conflict of interest) is measured by scientific communi-ties mostly based on an aggregated social networks populatedby FOAF (friend-of-a-friend) based dataset (collecting datafrom Swoogle in 2005-2006) and DBLP data source [15].In this work, social relationship is considered based on thepublication relationship, which is domain specific. To measureCOI, our method can provide more neutral decision by consid-ering not only professional sources of communication, such asDBLP, but also considering all on-line based communicationhistory.The main contribution of this paper are (1) algorithms forcalculating social strength, (2) how to use the social strengthfor social ranking and dynamic grouping, and (3) a study thatevaluates social strength from multiple perspectives.

III. SOCIAL RECOMMENDATION

This section discusses and proposes a theoretical frameworkfor calculating social strength and finally, provides an approachof social ranking of contacts. In general, social ranks enablerecommender systems for filtering social information andfor optimizing search techniques considering user’s socialinvolvement in social networking services. For example, usersmay have access in different social networking services (thatgenerates lots of social data), therefore, for better precisionfor a particular search content, an unified search on the userssocial Web could be one of the best option [12]. In this paper,social ranking deals with ranking contacts beyond particularcommunication services. In addition to this, the intendedrecommender system is aimed for providing social rank in thelevel of the contacts, communication tools and communicationservices. The proposed ASG framework provides a new wayof social ranking.

337337337

In the framework as depicted in Figure 1, social strength isused as the main parameter of measuring deepness of con-nectivity of a specific user along with the contact individuals.All the contacts are discovered from different communicationsservices and form an aggregated social graph of contacts. Nextsub-section provides high-level view of the overall frameworkwith brief a introduction of the main building blocks.

A. Conceptual Overview

Figure 1 shows a conceptual overview of the AggregatedSocial Graph (ASG) service framework. The framework con-sists of five components, the Aggregated Social Graph (AGS)service; the Communication History Aggregator and the SocialRanker are the main components while the Web applicationand third party communication services (e.g., Myspace, Twit-ter) are external components. Web Application sends queries tothe ASG service and receives recommendations for prioritizinginformation or ranking contacts. In the framework, both ofthe ASG service and Social Ranker utilize communicationhistory log for building the Aggregated Social Graph (i.e.global contact lists of specific user) and offers the serviceframework to the 3rd party applications (i.e. Web applications).Here, the Social Ranker does the analytics part and mainlyresponsible for ranking contacts and prioritizing information.For instance, the Web Application usually generates a queryrequest for inviting a targeted community through the ASGframework. Then the framework provides those third partyWeb applications (or recommender applications) receiving thesocial strength of the contacts in the community in terms ofthe Social Ranker. Consequently, the social rank is utilizedfor prioritizing and filtering information. The Social Rankeranalyzes the communication history log and preferences takenas input from the Communication History Aggregator service.The Communication History Aggregator component collectsthe communication logs from the different communicationservices. In the next subsection, social ranking is discussed inmore details and how it supports the Social Ranker componentof the framework.

Fig. 2. Formal Model of an Aggregated Social Graph

B. Social Ranking

Figure 2 shows the general topology of AggregatedSocial Graph (ASG) and high level abstraction of so-cial strength in ASG. ASG contains a set of entities(actors/users/nodes/contact individuals) and provides inter-relationships among those entities. For example, U, n1, p1 are

the nodes in the ASG and the arrow in-between U and n1indicates connectivity/relationship. Here, direct connectivityprovides direct contact (e.g., first degree friends, U and n1)between two entities, while indirect connectivity providesindirect contact in particular, the case of U and p1, whichare connected through another node n1 (e.g., second or moredegree friends, U is connected to p1 through n1).An algorithm is proposed to ranking contacts which uses com-munication history (see Table 1) as main input. To describethe algorithm, user U is considered as a user who is theowner of social graph (i.e., all the contacts in the social graphrelates with user U ). For computation of social strength, firstdegree contacts the individuals who are connected directly tothe user are considered in the algorithm. For searching andranking, the algorithm uses contextual data. For example, tofiltering presence information, the algorithm rank the contactsand then prioritize or filter the presence messages based onthat. However, the outputs of the algorithm are ranks ofsocial contacts utilizing the context and interactions. It isalso possible to generate the rank of contacts in differentgranularity. For example, a user has different preferences ofcommunication with the contacts in different communities. AFacebook contact may not never be enlisted in mobile phonecontact due to not being called over phone for that Facebookcontact. Therefore, ranking for filtering presence and rankingfor prioritizing phone calls can be different. Because of this,the proposed algorithm are designed in a way that could beuseful for those purposes as well.

Fig. 3. Frequency Data Type

Formally, C is a set of contact individuals {n1, n2, n3...nn}representing the social graph of user U , uses a set of communi-cation services S (e.g., social networks, Web 2.0 applications)s1, s2, s3...sp and generates bi-directional interactions. Forsimplicity, the heterogeneity problem of contacts in differentcommunication services of the user U is avoided. Heterogene-ity problem arises where the synchronization of contacts havenot been done periodically among different communicationservices. For example, the assumption is that aggregatedcontacts of user U are available to all the services which arebeing used by the user U . Another fact is that, a single com-munication service may provide one or more communicationtools (e.g., tools/Facebook/Message, Facebook/Photo). There-fore, the set T contains the number of tools {t1, t2, t3...tp}

338338338

TABLE IFREQUENCY LOGS

Name FB/Message(i=1,j=1) FB/Photo(i=1,j=2) FB/Comment(i=1,j=3) Mobile-phone/Call(i=2,j=1)Johan(k=1) 3 0 0 5Stefan(k=2) 2 2 1 1Kare(k=3) 1 0 6 2Josef(k=4) 2 2 3 3Andrea(k=5) 4 1 4 0Basel(k=6) 3 4 2 2

used in different communication services. For examples, if s1represents Facebook service, then t1 is the number of tools(in Table 1 in column Facebook/Message, Facebook/Photo,Facebook/Comment). Finally, frequency fi,j,k (see Figure 3),contains the number of interactions in between user U and con-tact k on the usage of communication tool j in communicationservice i. For example, f1 ,1 ,1 = 3, which means that contact1 using the communication tool 1 of commutation service 1has made 3 interactions (see Table 1). In this algorithm, theoverall social strength of the user U is,

S =

n∑

k=1

sk = 100

As the social strength is relative, therefore the sum of thesocial strength for all the contacts is always 100. The sumof strength is considered 100 due to ease of calculation butit is possible to scale the number for increasing (e.g., 1000)or decreasing (e.g., 1) the value of the sum of strength. Theproposed algorithm is discussed into more detail in the nextsection.

Algorithm 1 Social Strength Calculation AlgorithmINPUT: contacts from aggregated social graph, frequenciesof interactions using each of the communication tools usedamong the user and contactsOUTPUT: Rk, rank of the contacts

Require: fi,j,k and Ck

Ensure: k = 1...n, n is the number of contactswhile k �= n dosk =

∑pi=1 thresi si,k

=∑p

i=1 ti∑q

j=1fi,j,k thresi,j∑n

k=1 fi,j,kRk = rank(sk)end while

C. Utility Functions for Strength Calculation

si,a =

thresi∑

j=1

fi,j,a thresi,j∑nk=1 fi,j,k

(1)

The proposed algorithm is based approach in which two mainequations are used for calculating social strength. The firstequation calculates the strength of contacts on the basis of aparticular communication service. That is, the strength of userU among with the contacts C for a particular communication

service si (e.g., Twitter). The second equation provides over-all strength considering all communication services S. Theparameter si,a in Equation 1, provides the strength of user Uwith the contact a, considering all the interactions happenedin communication service i. In addition, n is the number offriends, ti is number of communication tools for the servicei and thresi,j is the threshold value of tool j in commu-nication service i. For instance, if the user has 4 contacts, 4communication services, in which each services has 2 differentcommunication tools and each of the communication tools hasthreshold (the sum of thresholds is 1) to signify the effectsof that particular communication tool, then the equation willprovide 4 different strengths for the 4 contacts. The overallsum of the strength will be 100, as mentioned before.

sa =

p∑

i=1

thresi si,a ∗ 100. (2)

Equation 2 accumulates strengths collected from S to getglobal ranking of the contacts. Here, the parameter sa providesthe overall strength of the user U with the contact a. Theparameter thresi, is the threshold value estimated over theusage on communication service i. Another important factorof utility based approach is tuning threshold values for thedifferent communication tools and services. In this case, usageof the tools and services are taken into consideration for tuningthis parameter dynamically. There are two functions (Equation3 and Equation 4) which, helps tuning thresholds. The equa-tions ensure that the tools which are used more, get higherthreshold values comparing with the less frequently usingtools. Therefore, both of the functions counts the frequenciesof usage of tools and services (see Table 1) and prioritizescommunication tools and services. In the Equation 3, thresi,jprovides percentage of usage of the tool j in communicationservice i by the user U . The rank function is responsible forranking the contacts according to the calculated social strength.

thresi,j =

∑nk=1 fi,j,k∑q

m=1

∑nk=1 fi,m,k

. (3)

For instance, if there are four different tools in communicationservice i, then thresi,j measures ratio of usage of a particulartool. Yet thresi,j is not enough to estimate overall effect ofa particular communication services. Therefore, the secondfunction measures threshold for the communication sources.

thresi =

∑qj=1

∑nk=1 fa,j,k∑p

m=1

∑qj=1

∑nk=1 fm,j,k

. (4)

339339339

In the Equation 4, thresi provides percentage of usage ofcommunication service a with respect to all interactions in allkind communication sources of the user U . In the utility basedmethod, calculated strength simplify the ordering for rankingthe contacts. For example, if user U has three contacts, whichare A, B and C with relative strength of 20, 50 and 30, thenthe rank of these contacts in descending order is B, C and A.That is, user U has the strongest connectivity with contact Bcomparing with A and C. In addition, it is possible to identifysuch ranking by consider all communication services or anyparticular communication services.

IV. CONCLUSIONS

The proposed ASG framework for utilizing aggregatedsocial graphs to enrich and simplify communication servicesmeets an emergent need due to an immensive growth ofWeb-based communication services. The framework is basedon personalized social graphs that harness communicationpatterns from Web environments, communication platformsand mobile devices and thus enable communication services tobenefit from the social strength between persons. In compar-ison to existing social ranking methods, the ASG frameworkuses social strength as main metric of ranking, groupingand recommending contacts. User-specific social data areharnessed to calculate the social strength between users andto build aggregated social graphs, that enable predictionof user’s preferences and prioritization of information frommultiple sources (micro-blogs, news feeds, etc). This alsofacilitates user-centric discovery of user preferences regardingcommunication, content sharing and collaboration with groupsand communities. Social strength is an important feature forsocial searching, ranking and filtering, especially since socialstrength calculation algorithms can be simplistic in design yetvery effective in practice for applications made aware of thesocial strength between users.

V. FUTURE WORK

Additional parameters such as the duration of calls or thelocation of users will be considered for enhancing the per-formance of social strength calculation algorithms. Studyingemail communication, or even physical meetings, is also possi-ble and would increase the accuracy of the aggregated socialgraphs. However, privacy concerns needs to be weighed inmore conclusively before that as complexity of the frameworkincreases together with risks of misuse.The plan is finally also to conduct more tests using real usersand data-sets, as user-evaluations of prototype applications cangive very valuable feedback from users on the acceptance,usability, reliability and accuracy of the framework.

VI. ACKNOWLEDGMENT

This work was funded by the Research Area MultimediaTechnologies of Ericsson Research, where Stefan Hakanssonhas been vital for the direction and execution of the research.The work was also supported by the Centre for Distance-spanning Technology (CDT) at Lulea University of Tech-nology and by the Satin-II research project, funded by the

European Regional Funds (mal-2), the Swedish Agency forEconomic and Regional Growth, the County AdministrativeBoard of Norrbotten, Norrbotten County Council, and the Cityof Lulea.

REFERENCES

[1] Q. Jones, M. Moldovan, D. Raban, and B. Butler, “Empirical evidence ofinformation overload constraining chat channel community interactions,”in CSCW ’08: Proceedings of the 2008 ACM conference on Computersupported cooperative work. New York, NY, USA: ACM, 2008, pp.323–332.

[2] A. Ankolekar, G. Szabo, Y. Luon, B. A. Huberman, D. Wilkinson,and F. Wu, “Friendlee: a mobile application for your social life,” inMobileHCI ’09: Proceedings of the 11th International Conference onHuman-Computer Interaction with Mobile Devices and Services. NewYork, NY, USA: ACM, 2009, pp. 1–4.

[3] P. R. Messinger, E. Stroulia, K. Lyons, M. Bone, R. H. Niu, K. Smirnov,and S. Perelgut, “Virtual worlds - past, present, and future: Newdirections in social computing,” Decis. Support Syst., vol. 47, no. 3,pp. 204–228, 2009.

[4] M. E. J. Newman and M. Girvan, “Finding and evaluating communitystructure in networks,” Physical Review E, vol. 69, p. 026113, 2004.[Online]. Available: doi:10.1103/PhysRevE.69.026113

[5] M. E. J. Newman, “Fast algorithm for detecting community structurein networks,” Physical Review E, vol. 69, p. 066133, 2004. [Online].Available: doi:10.1103/PhysRevE.69.066133

[6] J. Heer and D. Boyd, “Vizster: visualizing online social networks,” inInformation Visualization, 2005. INFOVIS 2005. IEEE Symposium on,Oct. 2005, pp. 32–39.

[7] Y. Dourisboure, F. Geraci, and M. Pellegrini, “Extraction and classifi-cation of dense implicit communities in the web graph,” ACM Trans.Web, vol. 3, no. 2, pp. 1–36, 2009.

[8] N. Banerjee, D. Chakraborty, K. Dasgupta, S. Mittal, S. Nagar, andSaguna, “R-u-in? - exploiting rich presence and converged communica-tions for next-generation activity-oriented social networking,” in MDM’09: Proceedings of the 2009 Tenth International Conference on MobileData Management: Systems, Services and Middleware. Washington,DC, USA: IEEE Computer Society, 2009, pp. 222–231.

[9] J. Rana, J. Kristiansson, H. Josef, and K. Synnes, “Challenges for mobilesocial networking applications,” in EuropeComm 2009: Proceedings ofthe First International ICST Conference Communications Infrastructure.Systems and Applications in Europe. Communications Infrastructure.Systems and Applications in Europe: Springer Berlin Heidelberg, 2009,pp. 275–285.

[10] A. T. Campbell, S. B. Eisenman, K. Fodor, N. D. Lane,H. Lu, E. Miluzzo, M. Musolesi, R. A. Peterson, and X. Zheng,“Cenceme: Injecting sensing presence into social network applicationsusing mobile phones (demo abstract),” 2009. [Online]. Available:http://citeseerx.ist.psu.edu/viewdoc/summary?doi=?doi=10.1.1.136.6257

[11] S. Bao, G. Xue, X. Wu, Y. Yu, B. Fei, and Z. Su, “Optimizing websearch using social annotations,” in WWW ’07: Proceedings of the 16thinternational conference on World Wide Web. New York, NY, USA:ACM, 2007, pp. 501–510.

[12] E. Amitay, D. Carmel, N. Har’El, S. Ofek-Koifman, A. Soffer,S. Yogev, and N. Golbandi, “Social search and discoveryusing a unified approach,” in 18th International World WideWeb Conference (WWW2009), April 2009. [Online]. Available:http://data.semanticweb.org/conference/www/2009/paper/186

[13] M. N. Szomszor, I. Cantador, and H. Alani, “Correlating user profilesfrom multiple folksonomies,” in HT ’08: Proceedings of the nineteenthACM conference on Hypertext and hypermedia. New York, NY, USA:ACM, 2008, pp. 33–42.

[14] B. Fazzinga, S. Flesca, and A. Pugliese, “Retrieving xml data fromheterogeneous sources through vague querying,” ACM Trans. InternetTechnol., vol. 9, no. 2, pp. 1–35, 2009.

[15] B. Aleman-Meza, M. Nagarajan, L. Ding, A. Sheth, I. B. Arpinar,A. Joshi, and T. Finin, “Scalable semantic analytics on social networksfor addressing the problem of conflict of interest detection,” ACM Trans.Web, vol. 2, no. 1, pp. 1–29, 2008.

340340340

[ieee 2010 international conference on advances in social networks analysis and mining (asonam 2010)...

Documents