Tie Strength in Online Social Networks and its Applications: A Brief Study
TTie Strength in Online Social Networks and itsApplications: A Brief Study
Chandni Saxena [email protected]
Tanvir Ahmad [email protected]
Department of Computer Engineering,Jamia Millia Islamia,New Delhi-25, IndiaMarch 13, 2020
Abstract
In online social network (OSN), understanding the factors bound tothe role and strength of interaction(tie) are essential to model a wide va-riety of network-based applications. The recognition of these interactionscan enhance the accuracy of link prediction, improve in ranking of dom-inants, reliability of recommendation and enhance targeted marketing asdecision support system. In recent years, research interest on tie strengthmeasures in OSN and its applications to diverse areas have increased,therefore it needs a comprehensive review covering tie strength estima-tion systematically. The objective of this paper is to provide an in-depthreview, analyze and explore the tie strength in online social networks. Amethodical category for tie strength estimation techniques are discussedand analyzed in a wide variety of network types. Representative applica-tions of tie strength estimation are also addressed. Finally, a set of futurechallenges of the tie strength in online social networks is discussed.
A social network is a social organization of social actors and interactions amongthese actors as its integral parts. With the surge of information technology andonline social networks (OSN) platforms such as Facebook, Instagram, Twitter,Linkedin, YouTube, Snapchat, Pinterest and ibo [43]; the collaboration betweenpeople have taken new dimensions and rise of online social networks. Meanwhile,splendid developments in mobile communication and wireless technology haveimproved the operations and services used by OSN, this resulted into emergenceof another genre of social network called mobile social network (MSN). Essen-tially, MSN or OSN, have unique features where individuals with commonalitiesand alike interests can possibly connect people with one another. Social net-works in these social context can be mapped into graphs, where nodes stand for1 a r X i v : . [ c s . S I] M a r ndividuals and the edges layout the relationships among them. The character-istics and features from the graph and other metrics related to such networkscan be utilized to study the behavior of individual. Finally, such networkscan leverage to define many interesting applications including link prediction,recommendation system, ranking groups and individuals, pattern analysis andinformation diffusion [48, 29, 20]. The interactions and relationships amongnetwork structure can offer access to target information effectively. Knowledgeof the strength of these relationships; furthermore the social dynamics providenew insights and shown to enhance correctedness of link prediction, reliabilityof recommendation for collaboration, improve ranking for influence and frauddetection [4, 39], make better model for information and disease spread andlead to enhance targeted marketing as decision support system. The propertyof relationship/tie strength ascertains to investigate and measure the interactionbased on the knowledge extracted from the graph, social context, communica-tion patterns and other related features. In the recent years, research intereston tie strength measure in online social networks and its applications to vari-ous areas has increased. Fig 1 shows the importance of area field observed interms of the number of published papers from three important digital librariesof computer science from 2007 to 2018 with search keywords “tie strength onlinesocial networks”. The research trend on the problem of ties strength in onlinesocial networks has shown a rise during these years as increase in number ofpublications can be seen on this topic. In the light of this notable growth inarea and to understand the wide variety factors related with tie strength, itis essential to cover the reviews study of the topic. This paper contributes anoverview study of tie strength in online social networks and its applications tosort the solutions in various dimensions; including management science, busi-ness analytic, social science and computer science. First it gives the tie strengthstatement which includes formal definitions, types of ties and a generic frame-work for this metric. Then it presents strength estimation techniques from fivemajor aspects such as; node based, communication pattern based, multimediacontents and learning based, geo-information based and temporal informationbased. It majorly describes typical approaches based on topological featuresand multimedia contents based machine learning approaches with tie strengthapplications. Finally, beside the current research, it outlines the future chal-lenges for tie strength estimation. The rest of the paper is organized as follows.The problem statement and importance of tie strength is explained in section 2.Various techniques to estimate tie strength are classified and discussed in section3. Section 4 covers range of applications of the tie strength metric. Sections 5and 6 draw the future challenges and conclusion from this study. Consider a social network ( G, V ) , where V denotes the set of nodes and E denotes the set of links. The tie strength estimation aims to indicate the value( nominal or discrete) of an edge between two nodes which has a spectrum of2igure 1: Publications growth for research on tie strength estimation in digitallibraries.Figure 2: Example of strong and weak ties due to Granovetter’s hypothesis.widely accepted definitions. These ties in social networks have been categorizedin the seminal work by Granovetter [13] as weak ties and strong ties based on:interaction time, level of intimacy, emotional intensity and reciprocity. Thestrong ties are functional in a variety of situations with close friends and familymembers. Similarly, weak ties have less intense relationship between nodesaccording to above factors. Other class of ties such as latent ties, dormant tiesand intermediate ties have also been defined by authors. The tie strength notionaccording to Granovetter’s hypothesis (fig. 2) can be explained from schematicnetwork of nodes with related features to define weak and strong ties. The user in OSN can have weak or strong ties within the network, whichmaintains different levels of shared interest and belief system according to itsstrength. In the detailed work of its kind, Onnela et al. [28] studied the relation-3igure 3: Category of tie strength estimation techniques on OSNs.ship between tie strength and structure of mobile networks on phone logs data.Authors found that strong ties cover prime role in maintaining local subgroups,however weak ties appear to cover important part for maintaining the networkscohesive strength. Weak ties are also related to the Burt’s concept of structuralhole [8] and Granovetter’s concept of local bridge [13] which play vital roles inproviding novel information and information propagation on network. Strongties are described to have strong control methods and tend to have trusted re-lationship which proved to target for network related decision support systems[32].
There are numerous fundamental approaches to estimate tie strength.The ap-proaches mainly employ network topology, community, node and link level infor-mation. Approaches which look into the information related to social media datasuch as location and geographical information, co-location statistics, mobilitybehavior, communication patterns and temporal information, rely on the theseplatform and peer-to-peer communication patterns. Moreover, latent variablesand learning based methods are more complex as authors incorporate additionalexternal information for the estimation of this metric. This section tracks to-gether a systematic review for the techniques of estimation of tie strength (fig. 3)in online social networks.
Computing the tie strength of a node pair based on nodes information is anintuitive solution. According to “ the weak tie hypothesis ” [13], local networkstructure around pair of nodes exhibits important correlation with tie strength4nd relative overlap [28] of common neighbors for nodes is defined as: O xy = n xy ( k x −
1) + ( k y + 1) − n xy (1)Where x , y connected nodes pair have n xy common neighbors and k x , k y are individual neighbors of nodes x , y respectively. Highly established nodesimilarity indices such as path distance, number of common neighbors that twonodes share and proximity of nodes pair [22, 48] are used to infer link weightand tie strength measures which have further application to the link predictionproblem. Newman [27] suggested a measure of tie strength of collaborationnetwork which considers number of papers of two scientists in collaboration andnumber of coauthors for those collaborations as index of tie strength betweenthem. Nowell and Kleinberg [22] has reported a review of tie strength measuresbased on local proximity of node pairs. Brandão and Moro [7] investigated co-authorship tie strength using common neighborhood overlap in the scholarlynetwork. Node based metrics are basic and easy to incorporate as authorsmainly use nodal information and actions, however additional attributes to thiscategory leads to enhance accuracy of estimating this metric [18]. Online social networks provide IT-enabled communication environment, wheresocial networks such as communities, society and organizations engage the in-dividuals to communicate for the purposes and have certain interaction pat-terns. The communication in the forms of messages, emails and calls and allthe contents of such interactions can be presented in terms of edge attributes.Therefore, analyzing the patterns of communication provides an interesting in-sight to estimate tie strength in such settings. Onnela et al.[28] investigated tiestrength metric by call duration and cumulative number of such calls betweenindividuals along with its local topology in the mobile communication network.Pappalardo et al. [29] explored a multidimensional network built over Facebook,Twitter and Foursquare to introduce tie strength estimation based on numberon interactions of nodes. Other studies [41, 24] have investigated accuracy ofhaving communication patterns correlated to the tie strength. Furthermore,incorporating the contents information of communication links can improve theprecision of this metric.
Developement of web-based centralized applications and similarly functional-ity to mobile networks provide online social networks a deliverance of gettinggeographical. The evolution of geosocial networking [42] allows user to com-municate relative to their locations. Amazon, Facebook, Twitter, Google andeBay are among such platforms to include social APIs and expand geoloca-tion technology. The study of online social networks in this dimension hasacknowledged property of these networks in relation to geographical attributes5f social media users along with their topological structure [37]. The locationdata can be shared by users voluntarily or it can be based on geotagged con-tents such as tweets, flickr, Instagram or location based services like Foursquarecheck-ins. Geographical information provides additional features which is use-ful in applications such as location aware recommendation, marketing and salesanalysis. Jeferrey et al. [25] investigated the positive correlation between tiestrength and distance between two users on twitter. Pharm et al. [30] proposedan entropy-based model for estimating tie strength based on co-occurence ofindividuals. Sadilek et al. [34] conferred location prediction approach basedon social tie strength. Authors combined location-based features along withtopology of friendship graph information to determine the social tie strength.Further related studies [14, 17] of measuring tie strength based on informationsuch as distance between users, location and co-occurrence and their structuralfeatures, also confirm the positive dependency of the metric on location-baseduser information. The geolocating applications and emerging volume of userslocation-based data demonstrate an obvious threat to the privacy issue in thisregard. However, efficient and reliable use of such information can be utilizedfor cutting-edge business applications.
Static interpretations of online social networks usually fail to embrace tempo-ral aspects of human activity. Social relationships vary with time and tendto show human actions as Markovian and randomly distributed events againsttime [26]. The majority of OSNs show up this temporal pattern of chang-ing edge sequences, varied intimacy of relations, emergence of new ties, timespans of communications, timelines of relationships and continuous changingactivities over time. This temporal conduct of social networks can affect otherfactors applicable to tie strength determination in online social networks. Inresult, structural property (clustering coefficient, neighborhood overlap) andgeographical property (community structure,distance, homophily) also receiveconditional upshots [26]. With increasing availability of massive temporal infor-mation of how people act and communicate, facilitate to include temporal prop-erty of OSN to accurately characterize underlying characteristics of tie strengthestimation and dynamic process of such networks. Karsai et al. [19] conferredamount and time of interaction to define link strength in large scale mobilephone network. Laurent et al. [21] defined strong tie strength based on networkstructure of mobile communication and frequency of interactions among intra-community nodes. Brãndo et al. [6] identified tie strength in temporal networkof co-authorship relations by measuring edge persistence and evolution of tiesover time. To encode temporal network is a non-trivial task because it requiresto maintain temporal ordering of edges and compute time with varying networkproperties. However, beyond this restraint of maintaining temporal features ofOSN, tie strength estimation with temporal attributes could help in prominentapplications such as co-authorship analysis and collaboration prediction, crimeprediction in radically influenced networks [4], stock market fluctuation [9], also6anking research[11].
A cascade growth in online plaforms of information system and social networkingsites has shown a consequential surge in online content generation and sharingon such social networks. The contents generated by users include online com-ments on social networking sites such as Facebook and Twitter, sharing digitalimages (eg. on flickr), creating and working blogs, participating and posting onwebsites, creating tags, sharing ideas and sharing interests [42]. These contentsare also called as linkage features or interaction contents of individuals, whichprovide schematic dimensions to the nodal attributes for a social network. Theinteraction contents of individuals on online social networks have been identifiedas strong factors for analyzing tie strength estimation [46, 40]. A model view ofreal time update for online social networks and their tie strength estimation isrepresented in fig. 4. The multimedia contents of user profiles are provided bysocial media through their APIs, where users participate for various purposessuch as media posting, messaging, business, health, hobbies, shopping and geo-tagging. The attributed graphs with node’s features vector are obtained fromsuch user related contents. Further, applying various machine learning meth-ods can determine the estimation of relationship strength on these attributedgraphs. The learning methods for categorical node contents are provided furtherin this section.Figure 4: Tie strength estimation model on online social networks with multi-media contents. 7 .5.1 Multivariate features-based learning methods
Several recorded properties of online social networks play vital roles to under-stand various factors associated with tie strength estimation. Researchers haveemployed various methods of statistical machine learning for feature selectionand prediction of key features for tie strength estimation [24, 10, 40, 12] . Ka-handa et al. [18] exploited 50 transnational features of friendship network frommultidimensional graph for relationship estimation based on Naive Bayesianclassifier, logistic regression and bagged decision tree models. Giebert and Kara-halios [12] presented multivariate static model, based on classes of user featuresto estimate tie strength on Facebook. Mattie et al. [24] used random forest re-gression and classification to predict tie strength based on categorical featuresfrom CDR call network. He et al. [16] employed multivariate step-wise regres-sion to determine key features for tie strength estimation and verified model onNaive Bayesian classifier.
The relationship strength among users in online networks varies in differentactivity fields therefore incorporating interaction activities with profile infor-mation can enumerate another dimension to relationship estimation. Zhao etal. [47] proposed a framework to measure tie strength on Facebook network in-corporating interaction among users in activity fields on OSN. Xiong et al. [44]presented a model to measure tie strength between users in Sina network byconsidering similarity among profiles and interaction activity in various fieldswith co occurrence of users. Authors used graphical model for the approach.Abufouda and Zweig [1] addressed link estimation on Facebook network, basedon machine learning model on a set of associated interacting networks withFacebook’s friends network.
There is a wide spectrum of learning based link strength estimation models dueto high variability in node features and data availability. Tang et al. [36] pro-posed tie strength prediction task for partially labeled network data of mobile,email and publication networks. Authors proposed an algorithm to learn modelparameters and to predict unknown relationships in a semi-supervised frame-work. Rotabi et al. [33] presented a supervised learning method of relationshipstrength estimation on twitter network. Authors used structural graph featuresfrom the presence and frequency of small graph motifs on combined weak andstrong ties. Abufouda and Zweig [2] estimated social relationship strength basedon friendship social network with external interaction using supervised learningmodel. Authors employed machine learning classification techniques to conductthe link evaluation via edge-proximity based label classification method.8 .6 Social theories-based metrics
There exists numerous studies based on social theories for target of relationshipstrength estimation. A few reported studies are covered in this section. Sintosand Tsaparas [35] exploited “Strong-Triadic-Closure”
Principle to characterizerelationship strength on 4 different OSNs. Liberatore and Quijano-Sanchez [23]explored a questionnaire based validation and computational framework for tiestrength determination using non-linear factor analysis models for various com-ponents related to social media. Pi et al. [31] considered mobile social networkand proposed a unified framework to individual’s inter relationship strength.Authors investigated time-aware mobility behavior, co-occurrence diversity, lo-cation semantic information, location significance and correlative impact on tiestrength estimation. Adali et al. [3] studied statistical features of communicationpatterns between users on social media. Authors considered reciprocity, atten-tion, latency and assortativity as determinant features to capture contextualinformation as compared with textual features using Twitter data. Volkovich etal. [38] examined association between interaction strength, spatial inter-spaceand structural position of users relationships from
Google’s Tueti communi-cation network. Authors observed variation of tie strength with nodes k-coreindex, considering core and periphery structure of studied networks.
The tie strength metric can be used for diverse range of applications. Thissection remarks some typical applications including link prediction, recommen-dation system, ranking groups and individuals, pattern analysis and informationdiffusion as follows. • Group recommendation:
The group recommendation is knowledge basedsystem which provides recommendation to the group having social rela-tionship to support decision making. The tie strength of group membersare key factors to characterise and improve the quality of group recom-mendation system[32]. • Link prediction applications:
The social relationship strength estimationrelated to link prediction are among prominant techniques for such appli-cations [32, 48, 29]. • Strategic game theoretic studies:
The weak ties play a prominant role indetermining collective behaviour and maintaining cooperation in evolu-tionary process of prisoners dilema game [45]. • Location based friend prediction:
Increased availability to geographic so-cial networks data and tie strength estimation techniques have made itstraightforward to exploit location based social networks for its applica-tion to friendship prediction, user behaviour and location recommendation[37]. 9
Routing selection:
Social tie strength between nodes in mobile ad-hocnetwork can be exploited for improvement of optimized link state routing[15]. • Collaboration prediction in scientific networks:
Tie strength estimation forcoauthorship relationship network is a determinant factor for collaborationprediction in such networks [27, 7]. • Information diffusion and Network control:
The inclusion of weak ties insocial networks can disrupt the spreading extent and speed of information,also exploited for network control in mobile social networks [20, 28]. • Other range of applications based on this metric includes crime prediction[4], stock market fluctuation study [9], friendship prediction [37, 1] andpattern analysis [29].
Future work is imperative to improve and extend the link estimation precision.Online social networks are highly dynamic and a very few studies [5] has beenreported to deal with this issue. OSN tie strength prediction can be modeledusing and incorporating dynamic online social networks. A large number ofmethods for tie strength metric in OSNs consider only structural features andattributes, however social theory based methods have been investigated to lim-ited number. More comprehensive work based on both approaches [3, 38] canenhance the accuracy of the metric. The benchmark data sets for the study areinadequate to systematically appraise the performances and outline limitationsof the models. The availability of benchmark data-sets for fair evaluation ofmodel is next important step. In online social networks the incomplete or noisydata are usual conditions, therefore more accurate tie estimation methods arerequired in such settings. Further, practical social networks with multiple rela-tions types and heterogeneous nature of network need to challenge for the tiestrength determination goals [1, 31].
Tie strength estimation has been a primitive problem in the field of social or-ganization and its applications are emerging with the advent of time due toadvancement in technologies and availability of system data on online social net-works. This paper attempts to deliver a brief review of tie strength estimationtechniques with variety of solutions from related researches. The fundamentalconcept of interaction strength is categorized and classified related to the state-of-the-art in tie strength computing. Tie strength metric estimation based ontopological features and users context related machine learning techniques are10ainly presented in the paper. Finally, the real time applications are acknowl-edged and an excellent direction to future challenges are addressed which couldresolve the current research issues related to tie strength computing.
References [1] Mohammed Abufouda and Katharina A Zweig. Are we really friends?:Link assessment in social networks using multiple associated interactionnetworks. In
Proceedings of the 24th International Conference on WorldWide Web , pages 771–776. ACM, 2015.[2] Mohammed Abufouda and Katharina A Zweig. Link classification andtie strength ranking in online social networks with exogenous interactionnetworks. arXiv preprint arXiv:1708.04030 , 2017.[3] Sibel Adali, Fred Sisenda, and Malik Magdon-Ismail. Actions speak as loudas words: Predicting relationships from social behavior data. In
Proceedingsof the 21st international conference on World Wide Web , pages 689–698.ACM, 2012.[4] Somayyeh Aghababaei and Masoud Makrehchi. Mining social media con-tent for crime prediction. In , pages 526–531. IEEE, 2016.[5] Ravi Bapna, Alok Gupta, Sarah Rice, and Arun Sundararajan. Trust andthe strength of ties in online social networks: An exploratory field experi-ment.
MIS Quarterly , 41(1):115–130, 2017.[6] Michele A Brandão, Pedro OS de Melo, and Mirella M Moro. Tie strengthdynamics over temporal co-authorship social networks. In
Proceedings ofthe International Conference on Web Intelligence , pages 306–313. ACM,2017.[7] Michele A Brandão and Mirella M Moro. Analyzing the strength of co-authorship ties with neighborhood overlap. In
International Conference onDatabase and Expert Systems Applications , pages 527–542. Springer, 2015.[8] Ronald S Burt.
Structural holes: The social structure of competition . Har-vard university press, 2009.[9] Chen Chen, Wu Dongxing, Hou Chunyan, and Yuan Xiaojie. Exploit-ing social media for stock market prediction with factorization machine.In , volume 2, pages142–149. IEEE, 2014.[10] Chien-Chou Chen, Terry Hui-Ye Chiu, Yuh-Jzer Joung, and Shy Min Chen.An examination of online social networks properties with tie-strength. In
PACIS , page 41, 2011. 1111] Vinícius P Freire and Daniel R Figueiredo. Ranking in collaboration net-works using a group based metric.
Journal of the Brazilian ComputerSociety , 17(4):255–266, 2011.[12] Eric Gilbert and Karrie Karahalios. Predicting tie strength with socialmedia. In
Proceedings of the SIGCHI conference on human factors in com-puting systems , pages 211–220. ACM, 2009.[13] Mark S Granovetter. The strength of weak ties. In
Social networks , pages347–367. Elsevier, 1977.[14] Georg Groh, Florian Straub, Johanna Eicher, and David Grob. Geo-graphic aspects of tie strength and value of information in social networking.In
Proceedings of the 6th ACM SIGSPATIAL International Workshop onLocation-Based Social Networks , pages 1–10. ACM, 2013.[15] Riten Gupta, Niyant Krishnamurthi, Uen-Tao Wang, Tejaswi Tamminedi,and Mario Gerla. Routing in mobile ad-hoc networks using social tiestrengths and mobility plans. In , pages 1–6. IEEE, 2017.[16] Yaxi He, Chunhong Zhang, and Yang Ji. Principle features for tie strengthestimation in micro-blog social network. In , pages 359–367. IEEE, 2012.[17] Hsun-Ping Hsieh, Rui Yan, and Cheng-Te Li. Where you go reveals who youknow: Analyzing social ties from millions of footprints. In
Proceedings ofthe 24th ACM International on Conference on Information and KnowledgeManagement , pages 1839–1842. ACM, 2015.[18] Indika Kahanda and Jennifer Neville. Using transactional information topredict link strength in online social networks. In
Third International AAAIConference on Weblogs and Social Media , 2009.[19] Márton Karsai, Nicola Perra, and Alessandro Vespignani. Time varyingnetworks and the weakness of strong ties.
Scientific reports , 4:4001, 2014.[20] Jennifer M Larson. The weakness of weak ties for novel information diffu-sion.
Applied network science , 2(1):14, 2017.[21] Guillaume Laurent, Jari Saramäki, and Márton Karsai. From calls to com-munities: a model for time-varying social networks.
The European PhysicalJournal B , 88(11):301, 2015.[22] David Liben-Nowell and Jon Kleinberg. The link-prediction problem forsocial networks.
Journal of the American society for information scienceand technology , 58(7):1019–1031, 2007.1223] Federico Liberatore and L Quijano-Sanchez. What do we really need tocompute the tie strength? an empirical study applied to social networks.
Computer Communications , 110:59–74, 2017.[24] Heather Mattie, Kenth Engø-Monsen, Rich Ling, and Jukka-Pekka Onnela.Understanding tie strength in social networks using a local âĂIJbow tieâĂİframework.
Scientific reports , 8(1):9349, 2018.[25] Jeffrey McGee, James A Caverlee, and Zhiyuan Cheng. A geographic studyof tie strength in social media. In
Proceedings of the 20th ACM internationalconference on Information and knowledge management , pages 2333–2336.ACM, 2011.[26] Giovanna Miritello.
Temporal patterns of communication in social net-works . Springer Science & Business Media, 2013.[27] Mark EJ Newman. Scientific collaboration networks. ii. shortest paths,weighted networks, and centrality.
Physical review E , 64(1):016132, 2001.[28] J-P Onnela, Jari Saramäki, Jorkki Hyvönen, György Szabó, David Lazer,Kimmo Kaski, János Kertész, and A-L Barabási. Structure and tiestrengths in mobile communication networks.
Proceedings of the nationalacademy of sciences , 104(18):7332–7336, 2007.[29] Luca Pappalardo, Giulio Rossetti, and Dino Pedreschi. " how well dowe know each other?" detecting tie strength in multidimensional socialnetworks. In , pages 1040–1045. IEEE, 2012.[30] Huy Pham, Cyrus Shahabi, and Yan Liu. Ebm: an entropy-based model toinfer social strength from spatiotemporal data. In
Proceedings of the 2013ACM SIGMOD International Conference on Management of Data , pages265–276. ACM, 2013.[31] Ting Pi, Lingwei Cao, Pin Lv, Zhili Ye, and Hao Wang. Inferring implicitsocial ties in mobile social networks. In , pages 1–6. IEEE, 2018.[32] Lara Quijano-Sánchez, Belén Díaz-Agudo, and Juan A Recio-García.Development of a group recommender application in a social network.
Knowledge-Based Systems , 71:72–85, 2014.[33] Rahmtin Rotabi, Krishna Kamath, Jon Kleinberg, and Aneesh Sharma.Detecting strong ties using network motifs. In
Proceedings of the 26thInternational Conference on World Wide Web Companion , pages 983–992.International World Wide Web Conferences Steering Committee, 2017.[34] Adam Sadilek, Henry Kautz, and Jeffrey P Bigham. Finding your friendsand following them to where you are. In
Proceedings of the fifth ACMinternational conference on Web search and data mining , pages 723–732.ACM, 2012. 1335] Stavros Sintos and Panayiotis Tsaparas. Using strong triadic closure tocharacterize ties in social networks. In
Proceedings of the 20th ACMSIGKDD international conference on Knowledge discovery and data min-ing , pages 1466–1475. ACM, 2014.[36] Wenbin Tang, Honglei Zhuang, and Jie Tang. Learning to infer social tiesin large networks. In
Joint european conference on machine learning andknowledge discovery in databases , pages 381–397. Springer, 2011.[37] Jorge C Valverde-Rebaza, Mathieu Roche, Pascal Poncelet, and Alneude Andrade Lopes. The role of location and social strength for friend-ship prediction in location-based social networks.
Information Processing& Management , 54(4):475–489, 2018.[38] Yana Volkovich, Salvatore Scellato, David Laniado, Cecilia Mascolo, andAndreas Kaltenbrunner. The length of bridge ties: structural and geo-graphic properties of online social interactions. In
Sixth International AAAIConference on Weblogs and Social Media , 2012.[39] Peng Wang, BaoWen Xu, YuRong Wu, and XiaoYu Zhou. Link predic-tion in social networks: the state-of-the-art.
Science China InformationSciences , 58(1):1–38, 2015.[40] Xinyu Wang, Chunhong Zhang, and Li Sun. An improved model for de-pression detection in micro-blog social network. In , pages 80–87. IEEE,2013.[41] Jason Wiese, Jun-Ki Min, Jason I Hong, and John Zimmerman. Assessingcall and sms logs as an indication of tie strength. 2014.[42] Wikipedia contributors. Geosocial networking — Wikipedia, the free en-cyclopedia, 2018. [Online; accessed 27-February-2019].[43] Wikipedia contributors. Social media — Wikipedia, the free encyclo-pedia. https://en.wikipedia.org/w/index.php?title=Social_media&oldid=885509949 , 2019. [Online; accessed 1-March-2019].[44] Liyan Xiong, Yin Lei, Weichun Huang, Xiaohui Huang, and MaoshengZhong. An estimation model for social relationship strength based on users’profiles, co-occurrence and interaction activities.
Neurocomputing , 214:927–934, 2016.[45] Bo Xu, Lu Liu, and Weijia You. Importance of tie strengths in the prison-erÊijs dilemma game on social networks.
Physics Letters A , 375(24):2269–2273, 2011.[46] Xiaohua Zeng and Liyuan Wei. Social ties and user content generation:Evidence from flickr.
Information Systems Research , 24(1):71–87, 2013.1447] Xiaojian Zhao, Jin Yuan, Guangda Li, Xiaoming Chen, and Zhoujun Li.Relationship strength estimation for online social networks with the studyon facebook.
Neurocomputing , 95:89–97, 2012.[48] Tao Zhou, Linyuan Lü, and Yi-Cheng Zhang. Predicting missing links vialocal information.