social and techological networks

8/8/2019 Social and Techological Networks

http://slidepdf.com/reader/full/social-and-techological-networks 1/8

66

review articles

The past decade has witnessed a coming-together o the technological networks that connect computerson the Internet and the social networks that havelinked humans or millennia. Beyond the artiacts that have sprung rom this development—sites such asacebook, LinkedIn, MySpace, Wikipedia, digg, del.icio.us, YouTube, and fickr—there is a broader processat work, a growing pattern o movement throughonline spaces to orm connections with others, build

virtual communities, and engage in sel-expression.Even as these new media have led to changes in our

styles o communication, they have also remained

governed by longstanding principles o human social

interaction—principles that can now

be observed and quantied at unprec-

edented levels o scale and resolutionthrough the data being generated by these online worlds. Like time-lapse

video or photographs through a micro-

scope, these images o social networksoer glimpses o everyday lie rom

an unconventional vantage point—

images depicting phenomena suchas the fow o inormation through an

organization or the disintegration o

a social group into rival actions. Sci-ence advances whenever we can take

something that was once invisible andmake it visible; and this is now taking

place with regard to social networksand social processes.

Collecting social-network data has

traditionally been hard work, requir-ing extensive contact with the group

o people being studied; and, given

the practical considerations, researcheorts have generally been limited to

groups o tens to hundreds o indi-

viduals. Social interaction in onlinesettings, on the other hand, leaves ex-

tensive digital traces by its very nature. At the scales o tens o millions o in-

dividuals and minute-by-minute timegranularity, we can replay and watch

the ways in which people seek out con-

nections and orm riendships on a sitelike Facebook or how they coordinate

with each other and engage in creative

expression on sites like Wikipedia andfickr. We can observe a news story sud-

denly catching the attention o millions

o readers or witness how looming clouds o controversy gather around

a community o bloggers. These arepart o the ephemeral dynamics o or-

dinary lie, now made visible throughtheir online maniestations. As such,

we are witnessing a revolution in the

measurement o collective human

t

cvrg sl dtlgl

nwrk

Doi:10.1145/1400214.1400232

Internet-based data on human interactionconnects scientifc inquiry like never beore.

BY Jon KLeinBeRG

The Nexus friend grapher application, created

by Ivan Kozik, allows Facebook account holders

to generate graphs illustrating their social

network of friends. The resulting spheres not only

demonstrate how friends are connected, but also

indicate the interests shared by different groups

of friends. For more information, or to create a

graph, visit http://nexus.ludios.net.



67



68

review articles

vation period.26 The researchers ound

the average length o the shortest path

between any two people on this system

to be around 6.6—a number remark-ably close to Milgram’s, and obtained

by utterly dierent means.

Modeling the Phenomenon. Math-ematical models o this phenomenon

start by asking why social networksshould be so rich in short paths. Inan infuential 1998 paper, Watts and

Strogatz sought to reconcile this abun-

dance with the seemingly contrasting observation that the world is highly

clustered, consisting o acquaintanc-

es who tend to be geographically and

socially similar to one another.40 They

showed that adding even a small num-ber o random social connections to

a highly clustered network causes a

rapid transition to a small world, withshort paths appearing between most

pairs o people. In other words, the

world may look orderly and structured

to each o us—with our riends andcolleagues tending to know each other

and have similar attributes—but a ew

unexpected links shortcutting throughthe network are sucient to bring us

all close together.

There is a urther aspect to theMilgram experiment that is striking

and inherently algorithmic: the ex-

periment showed not just that theshort paths were there but that people

were able to nd them.20 When youask someone in Omaha, Nebraska,

as Milgram did, to use his or her so-

cial network to direct a letter halway across the country to Sharon, Massa-

chusetts, that person can’t possibly

know the precise course it will ollow or whether it will even get there. The

act that so many o the letters zeroed

in on the target suggests something powerul about the social network’s

ability to “unnel” inormation towardar-o destinations. The U.S. Postal

Service does this when it delivers aletter, but it is centrally designed and

maintained at considerable cost to do

precisely this job; why should a socialnetwork, which has grown organically

without any central control, be able to

accomplish the same task with any re-liability at all?

To begin modeling this phenom-

enon, suppose we all lived on a two-dimensional plane, spread out with a

roughly uniorm population density,

behavior and the beginnings o a new

research area—one that analyzes and

builds theories o large social systemsby using their refections in massive

datasets.

This line o investigation repre-sents a fow o ideas between comput-

ing and the social sciences that goes

in both directions. Using datasets oncollective human behavior, together

with an algorithmic language or mod-

eling social processes, we can begin to

make progress on undamental social-science questions, inormed by a com-

putational perspective. Meanwhile,

social scientists’ insights into theseproblems, which predate the Inter-

net, are essential to understanding the

current generation o computing sys-tems. Indeed, most o the high-prole

Internet applications to emerge overthe past hal-decade are governed not

just by technological considerationsbut also by recurring and quantiable

principles o human social interac-

tion; both technological and socialorces, working together, shape the

inherent operating constraints in such

systems.The resulting research questions

arise rom a coming-together o di-

erent styles o research, and it is im-portant to recognize that analyses o

truly massive social networks provideus with both more and less than we

get rom detailed studies at smallerscales. Massive datasets can allow us

to see patterns that are genuine, yet

literally invisible at smaller scales. But working at a large scale introduces its

own diculties. One doesn’t necessar-

ily know what any one particular individual or social connection signies;

and the riendships, opinions, and

personal inormation that are revealedonline come in varying degrees o reli-

ability. One is observing social activity in aggregate, but at a ne-grained level

the data is more dicult to interpret.The true challenge is to bridge this gap

between the massive and the detailed,

to nd the points where these lines o research converge.

With that goal in mind, we discuss

two settings where this research strat-egy is being pursued. We begin with

the “small-world phenomenon” in so-

cial networks—the principle that we

are all connected by short chains o acquaintances—and then look at the re-

lated problem o how ideas spread con-

tagiously through groups o people.

t sll-Wrld P d

Drlzd sr

When the playwright John Guarecoined the term “six degrees o sepa-

ration,”15 describing the notion that

we are all just a ew steps apart in theglobal social network, he was reerring

to a series o experiments peormed

by the social psychologist Stanley Mil-

gram in the 1960s.38 Milgram’s work

provided the rst empirical evidence

or this idea, and it is useul to consid-er the structure o his experiments and

their signicance.

Inspired in part by the work o the

political scientist Ithiel de Sola Pool with the applied mathematician Man-

red Kochen,

7

Milgram asked each o aew hundred people in Boston and the

Midwest to try directing a letter toward

a designated “target” in the network—a stockbroker who lived in Sharon,

Massachusetts. The participants in the

experiment were given basic personalinormation about the target, includ-

ing his address and occupation; but

each participant could only mail the

letter to someone he or she knew on arst-name basis, with the instructions

to orward it on in this way toward the

target as quickly as possible. The mailthus closed in on the town o Sharon,

moving rom riend to riend, with the

successul letters reaching the targetin a median o six steps.38 This kind

o experiment, constructing pathsthrough social networks to distant tar-

get individuals, has been repeated by a

number o other groups in subsequentdecades.9, 12, 23

Milgram’s experiment and its ol-low-ups come with many caveats. In

particular, they have tended to be most

successul when the target is afuentand socially accessible; and even then,many chains ail to complete. Never-

theless, the striking act at the heart o

the results—that such short paths canbe discovered in social networks—has

been borne out by many subsequent

analyses o large-scale network data.Quite recently, Leskovec and Horvitz

built a social network rom nearly a

quarter-billion instant-messaging ac-counts on MSN Messenger, connect-

ing two individuals i they engaged in a

conversation over a one-month obser-



review artic

69

and that we each knew our next-door

neighbors or some distance in each

direction. Now, ollowing Watts and

Strogatz, we add a small number o random connections—say, each o

us has a single additional riend cho-

sen uniormly at random rom theull population. Short paths appear,

as expected, but one can prove thatthere is no procedure the people liv-ing in this world can perorm—using

only local inormation and without a

global “bird’s-eye view” o the socialnetwork—to orward letters to araway

targets quickly.20 In other words, in astructured world supplemented with

purely random connections, the Mil-

gram experiment would have ailed:the short paths would have been there,

but they would have been unndable

or people living in the network.By extending things a little bit, how-ever, we can get the model to capture

the eect Milgram saw in real lie. To

do this, we keep everyone living on atwo-dimensional plane but revisit the

random connections, which are sup-

posed to account or the unexpectedar-fung riendships that make the

world small. In reality, o course, these

links are not completely random;they too are biased toward closer and

more similar people. Suppose, then,

that each person still has a randomar-away riend, but that this riendis chosen with a probability that de-

cays with the individual’s distance

in the plane—say, by a “gravitationallaw” in which the probability o being

riends with a person at a distance d

decays as d −r or some power r . Thus,

as the exponent r increases, the worldgets less purely random—the long-

range riendships are still potentially

ar away, but overall they are moregeographically clustered. What eect

does this have on searching or targetsin the network?

Analyzing this model, one nds

that the eectiveness o Milgram-style

search with local inormation initially gets better as r increases—because the

world is becoming more orderly and

easy to navigate—and then gets worse

again as r continues increasing—because short paths actually start be-

coming too rare in the network. The

best choice or the exponent r , whensearch is in act very rapid, is to set it

equal to 2. In other words, when the

probability o riendship alls o like

the square o the distance, we have a

small world in which the paths arenot only there but also can be ound

quickly by people operating without

a global view.20 The exponent o 2 isthus balanced at a point where short

paths are abundant, but not so abun-

dant as to be too disorganized to use.Further analysis indicates that this

best exponent in act has a simple

qualitative property that helps us un-

derstand its special role: when riend-ships all o according to an inverse-

square law in two dimensions, then on

average people have about the sameproportion o riends at each “scale

o resolution”—at distances 1–10, 10–

100, 100–1000, and so on. This prop-erty lets messages descend gradually

through these distance scales, nd-ing ways to get signicantly closer to

the target at each step and in this waycompleting short chains, just as Mil-

gram observed.

Validating and Applying the Model. When such models were rst proposed,

it was unclear not only how accurate

they were in real lie but also how to goabout collecting data to measure the

accuracy. To do so, you would have to

convince thousands o people to report where they lived and who their riends

were—a daunting task.But o course, the public proles

on social-networking sites readily do just that, and as these sites began to

grow explosively in 2003 and 2004, Li-

ben-Nowell et al. developed a rame- work or using this type o data to test

the predictions o the small-world

models.30 In particular, they collected

data rom the riendship network o

the public blogging site LiveJournal,

ocusing on hal a million people whoreported U.S. hometown locations

and lists o riends on the site. Theythen had to extend the mathematicalmodels to deal with the act that real

human population densities are high-

ly nonuniorm. To do so, they denedthe distance between two people in an

ordinal rather than absolute sense:

they based the probability that a per-son v orms a link to a person w on the

number o people who are closer to vthan w is, rather than on the physical

distance between v and w. Using thismore fexible denition, the distribu-

tion o riendships in the data could

a rr, pllg, r

lk lvd— rll xpl r prd rpr pr,gly, yl pd.



70

review articles

are built on the principle that there

should not be a central index o the

content being shared (in contrast, or

example, to the way in which searchengines like Google provide a central

index or Web pages). As a result, look-

ing up content in a peer-to-peer sys-tem ollows a Milgram-style approach

in which the hosts participating in thesystem must orward requests withonly a local view.31 Mathematical mod-

els o small worlds—originally built with human networks in mind—can

provide insights into the design o e-

cient solutions or this distributed

search problem as well. We’ve thus seen how viewing such

models in the online domain can help

us understand the global layout o so-cial-networking sites, the fow o com-

munications within organizations,and the design o peer-to-peer systems. We now look at how the insights we’ve

gained here can provide perspective

on an important related problem—thespread o inormation through large

populations.

sl cg d

sprd id

Milgram’s experiment was about o-

cusing a message on a particular target,but much o the inormation that fows

through a social network radiates out- ward in many directions at once. A ru-mor, a political message, or a link to an

online video—these are all examples

o inormation that can spread romperson to person, contagiously, in the

style o an epidemic. This is an impor-

tant process to understand because itis part o a broader pattern by which

people infuence one another over lon-

ger periods o time, whether in online

or ofine settings, to orm new politi-cal and social belies, adopt new tech-

nologies, and change personal behav-ior—a process that sociologists reer toas the “diusion o innovations.”35 But

while the outcomes o many o theseprocesses are easily visible, their inner

workings have remained elusive.

Some o the basic mathematical

models or the diusion o innova-tions posit that people’s adoption o

new behaviors depends in a proba-

bilistic way on the behaviors o theirneighbors in the social network: as

more and more o your riends buy a

new product or join a new activity, you

then in act be closely approximated

by the natural generalization o theinverse-square law.

It was dicult not to be a bit sur-

prised by the alignment o theory andmeasurement. The abstract models

were making very specic predictions

about how riendships should depend

on physical distance, and these predic-tions were being approximately borne

out on data arising rom real-worldsocial networks. And there remains a

mystery at the heart o these ndings.

While the act that the distributions

are so close does not necessarily imply the existence o an organizing mecha-

nism (or example, see Bookstein5 or

a discussion o this general issue inthe context o social-science data), it is

still natural to ask why real social net-

works have arranged themselves in apattern o riendships across distance

that is close to optimal or orwarding

messages to araway targets. Further,

whatever the users o LiveJournal aredoing, they are not explicitly trying to

run versions o the Milgram experi-

ment—i there are selective pressuresdriving the network toward this shape,

they must be more implicit, and it re-

mains a ascinating open question whether such orces exist and how they

might operate.

Other research using online datahas considered how riendship and

communication depend on nongeo-

graphic notions o “distance.” For ex-

ample, the probability that you know someone is aected by whether you

and they have similar occupations,

cultural backgrounds, or roles withina large organization. Adamic and Adar

studied how communication depends

on one such kind o distance: they measured how the rate o email mes-

saging between employees o a corpo-

rate research lab ell o as they lookedat people who were arther and arther

apart in the organizational hierarchy.1

Here too, this rate approximated ananalogue o the inverse-square law—

in a orm adapted to hierarchies21,

39—although the messages in the re-searchers’ data were skewed a bit more

toward long-range contacts in the or-

ganization than short-range ones.Finally, these models can rapidly

turn into design principles or distrib-

uted computing systems as well. Mod-

ern peer-to-peer le-sharing systems

are more likely to do so as well.13 Re-

cent studies o online data have pro-

vided some o the rst pictures o what

this dependence looks like over largepopulations. In particular, Leskovec,

Adamic, and Huberman studied how

the probability o purchasing books,DVDs, and music rom a large online

retailer increased with the number o email recommendations a potentialcustomer received.25 Backstrom et al.

determined the probability o joining groups in a large online community

as a unction o the number o riends

who already belonged to the group.4 And Hill, Provost, and Volinsky 16 ana-

lyzed how an individual’s adoption

o a consumer telecommunicationsservice plan depended on his or her

connections to prior adopters o the

service. While the probability o adopting a behavior increases with the number

o riends who have already adopted it,

there is a “diminishing returns” pat-tern in which the marginal eect o

each successive riend decreases.4,25

In many cases, however, an interest-

ing deviation rom this pattern is ob-

served—a “0–1–2 eect,” in which theprobability o joining an activity when

two riends have done so is signicant-

ly more than twice the probability o

joining when only one has done so.4

The structure of cascading behavior.

Beyond these local mechanisms o so-cial infuence, it is instructive to trace

out the overall patterns by which infu-

ence propagates through a large socialnetwork. In recent work, David Liben-

Nowell and I investigated such global-

scale processes by gathering data onchain-letter petitions that had spread

widely over the Internet.29 A particular-ly pervasive chain letter, which spread

in 2002 and 2003, purported to orga-

nize opposition to the impending inva-sion o Iraq. Each copy o the petitioncontained the list o people who had

received that particular copy, in the

order in which they added their namesand then passed it on to others in their

email address books. In the process,

several hundred o these copies hadbeen sent to Internet mailing lists; by

retrieving them rom the mailing lists’

archives, we could reconstruct a largeragment o the branching tree-like

trajectory by which the chain letter

had spread.



review artic

71

The structure o the tree was sur-

prising, as it challenged our small-

world intuitions. Rather than anning

out widely, reaching many people withonly a ew degrees o separation, the

chain letter spread in a deep and nar-

row pattern, with many paths consist-ing o several hundred steps. The short

chains in the social network were stillthere, but the chain letter was getting to people by much more roundabout

means. Moreover, we ound a very sim-

ilar structure or the one other large-scale chain letter on which we could

nd enough mailing-list data, this one

claiming to be organizing support or

National Public Radio. Why this deep and narrow spread-

ing pattern arises in multiple settings

remains something o a mystery, but

there are several hypotheses or recon-ciling it with the structure o a small

world. In our work on chain letters, weanalyzed a model based on the natural

idea that people take widely varying

amounts o time to act on messagesas they arrive: when recipients orward

the chain letter at dierent times to

highly overlapping circles o riends,

it can in eect “echo” through denseclusters in the social network, ollow-

ing a snaking path rather than a direct

one. Simulations o this process on

real social networks such as the onerom LiveJournal produce tree struc-

tures very similar to the true one weobserved.29

It is also plausible that the nature

o social infuence—properties suchas the 0–1–2 eect in particular—play

an important role. Suppose that most

people in the social network needto receive a copy o the letter at least

twice beore actually signing their

name and sending it on. As Centolaand Macy have recently argued, our

long-range riendships may be muchless useul or spreading inormationin situations such as these: you can

learn o something the rst time rom

a ar-fung riend, but to get a second

conrmatory hearing you may needto wait or the inormation also to ar-

rive through your more local contacts.6

Such a pattern could slow down the

progress o a chain letter, orcing it toslog through the dense structure o our

local connections rather than exploit

the long-range shortcuts that make

the world small.

Contagion as a design principle. As with the decentralized search prob

lem at the heart o the small-world

phenomenon, the idea o contagion innetworks has served as a design princi-

ple or a range o inormation systems

Early work in distributed computing

proposed the notion o “epidemic al-

gorithms,” in which inormation up-dates would be spread between hosts

according to a probabilistic contagionrule.8 This has led to an active line oresearch, based on the act that such

algorithms can be highly robust and

relatively simple to congure at eachindividual node.

More recently, contagion and cas-

cading behavior have been employed

in proposals or social computingapplications such as word-o-mouth

recommendation systems,

25

incentivemechanisms or routing queries to in-

dividuals possessing relevant inorma-

tion,22 and methods to track the spreado inormation among Weblogs.2, 14

Large-scale social contagion data

also provides the opportunity to iden-tiy highly infuential sets o people

in a social network—the set o people

who would trigger the largest cascadei they were to adopt an innovation.11

The search or such infuential sets

is a computationally dicult prob-

lem, although recent work has shownthat when social infuence ollows the

kind o “diminishing returns” patterndiscussed here, it is possible to nd

approximate methods with provable

guarantees.19, 32

frr Dr

Research on large-scale social-networkdata is proceeding in many urther di-

rections as well. While much o what

we have been discussing involves thedynamic behavior o individuals in so-

cial networks, an important and com-plementary area o inquiry is how thestructure o the network itsel evolves

over time.

Recent studies o large datasets haveshed light on several important princi-

ples o network evolution. A central one,

rooted in early work in the social sci-

ences, is the principle o “preerentialattachment”—the idea that nodes that

already have many links will tend to ac-

quire them at a greater rate.33 An active

line o research has shown how preer-ential attachment can lead to the highly

t vlbly r d

pll d r ld pr dbklp, llwg dvlp dvl dl l p

lrg l d dl dg w pgppl.



72

review articles

American Statistical Association 97 (2002).18. K, m., s, s., m, n. a

cg jcwk. Science 313 (2006), 824–827.

19. K, d., Kg, J., t, e. mg fc c wk. i Proceedingsof the 9th ACM SIGKDD International Conferenceon Knowledge Discovery and Data Mining (2003),137–146.

20. Kg, J. t -w : agc c. i Proceedings of the 32nd ACM Symposium on Theory of Computing (2000),

163–170.21. Kg, J. s-w

c . i Proceedings of the 14thAdvances in Neural Information Processing Systems (2001), 431–438.

22. Kg, J. rg, p. Q cwk. i Proceedings of the 46th IEEE Symposiumon Foundations of Computer Science (2005), 132–141.

23. K, C. mg, s. acqc wkw c g: ac w. J. Personality and Social Psychology 15 (1978).

24. K, G., W, d. ec g c wk. Science 311 (2006), 88–90.

25. lkc, J., ac, l., h, b. tc kg. ACM Transactions on theWeb 1, 1 (m 2007).

26. lkc, J. h, e. Ww : p-c w -gg wk. i

Proceedings of the 17th International World Wide WebConference, 2008.27. lkc, J., Kg, J.m., , C. G

: dc w, kg . i Proceedings of the 11thACM SIGKDD International Conference on KnowledgeDiscovery and Data Mining (2005), 177–187.

28. l-nw, d. Kg, J. t k-c c wk. J. American Society for Information Science and Technology 58 , 7 (2007),1019–1031.

29. l-nw, d. Kg, J. tcg fw g c g i c- .i Proceedings of Natl. Acad. Sci. 105 , 12 (m. 2008),4633–4638.

30. l-nw, d., nk, J., K, r., rg, p., tk, a. Ggc g c wk.i Proceedings of Natl. Acad. Sci. 102, 33 (ag. 2005),11623–11628.

31. l, e.K., Cwc, J., p, m., s, r., l,

s. a c -- wk c. IEEE Commun. Surveys and Tutorials 7 , 2 (2005), 72–93.

32. m, e. rc, s. o fc c wk. i Proceedings of the39th ACM Symposium on Theory of Computing, 2007.

33. nw, m.e.J. t c c cwk. SIAM Review, 45 (2003), 167–256.

34. r, a. s g w c-c i: a . Bulletin of Mathematical Biophysics 15 , 4(dc. 1953), 523–533.

35. rg, e. Diffusion of Innovations, 4th Edition. p, 1995.

36. sgk, m., d, p., W, d. e q c cc k. Science 311 (2006), 854–856.

37. sk, p. m, a.W. dc c wk g c . i Advances in

Neural Information Processing Systems, 2005.38. t, J. mg, s. a w . Sociometry 32, 4 (1969),425–443.

39. W, d.J., d, p.s., nw, m.e.J. i c c wk. Science 296 (m 2002),1302–1305.

40. W, d.J. sg, s.h. Cc c ‘-w’ wk. Nature 393 (1998), 440–442.

Jon Kleinberg ([email protected].) c cc C u, ic, ny.h wk c c wk , w c wk W . h c mca, pck, s w, n p i mc u.

© 2008 aCm 0001-0782/08/1100 $5.00

skewed distributions o links that one

sees in real networks, with certain nodes

acting as highly connected “hubs.”3 Another principle, also a key issue

in sociology, is the notion o “triadic

closure:” links are much more likely

to orm between two people when they have a riend in common.34 Recent

work using email logs has providedsome o the rst concrete measure-

ments o the eect o triadic closure in

a social-communication network.24

Further principles have begun to

emerge rom recent studies o social and

inormation networks over time, includ-ing “densication eects,” in which the

number o links per node increases as

the network grows, and “shrinking di-

ameters,” in which the number o stepsin the shortest paths between nodes can

actually decrease even as the total num-ber o nodes is increasing.27

It is also intriguing to ask whether

machine-learning techniques can beeective at predicting the outcomes

o social processes rom observations

o their early stages. Problems here in-clude the prediction o new links, the

participation o people in new activ-

ities, the eectiveness o groups at

collective problem-solving, and thegrowth o communities over time.4,

16, 17, 18, 28, 37 Recent work by Salganik,

Dodds, and Watts raises the interest-ing possibility that the outcomes o

certain types o social-eedback eectsmay in act be inherently unpredict-able.36 Through an online experiment

in which participants were assignedto multiple, independently evolving

versions o a music-download site—

essentially, a set o articially con-structed “parallel universes” in which

copies o the site could develop inde-

pendently—Salganik et al. ound that when eedback was provided to users

about the popularity o the items being downloaded, early fuctuations in thepopularities o dierent items could

get locked in to produce very dierent

long-term trajectories o popularity.

Developing an expressive computa-tional model or this phenomenon is

an interesting open question.

Ultimately, across all these do-mains, the availability o such rich

and plentiul data on human interac-

tion has closed an important eedbackloop, allowing us to develop and evalu-

ate models o social phenomena at

large scales and to use these models

in the design o new computing ap-

plications. Such questions challenge

us to bridge styles o scientic inqui-ry—ranging rom subtle small-group

studies to computation on massive

datasets—that traditionally have hadlittle contact with each other. And they

are compelling questions in need o answers—because at their heart, they are about the human and technologi-

cal connections that link us all, and

the still-mysterious rhythms o thenetworks we inhabit.

akwldg

I thank the National Science Founda-

tion, the MacArthur Foundation, the

Cornell Institute or the Social Sci-

ences, Google, and Yahoo or their

support and the anonymous reviewerso this manuscript or their comments

and eedback.

References1. ac, l. a, e. hw c c wk.

Social Networks 27 , 3 (2005), 187–203.2. a, e., zg, l., ac, l.a., lk, r.m.

ic c c gc.i Proceedings of the Workshop on the WebloggingEcosystem, 2004.

3. a, r. bó, a-l. sc cc c wk. Reviews of Modern Physics 74(2002), 47–97.

4. bck, l., hc, d., Kg, J., l, x. G g c wk:m, gw, . i Proceedings of

the 12th ACM SIGKDD International Conference onKnowledge Discovery and Data Mining, 2006.5. bk, a. ic , p ii:

rc g. J. American Society for Information Science 41, 5 (1990), 376–386.

6. C, d. mc, m. C cg wk g . American J. of Sociology 113(2007), 702–734.

7. s p, i. Kc, m. Cc fc.Social Networks 1, 1 (1978) 5–51.

8. d, a.J., G, d.h., h, C., i, W.,l, J., sk, s., sg, h.e., sw, d.C., t, d.b. ec g c c. i Proceedings of the 6th ACM Symposium on Principles of Distributed Computing(1987), 1–12.

9. d, p., m, r., W, d. a c g c wk. Science 301(2003), 827–829.

10. d, p. W, d. u

g cg. Physical Review Letters 92 (2004).

11. dg, p. rc, m. mg wk c. i Proceedings of the 7th ACM SIGKDD International Conference on KnowledgeDiscovery and Data Mining (2001), 57–66.

12. G, e. i’ w . Current Contents 43 (1979), 5–10.

13. G, m. t cc. American J. Sociology 83 (1978), 1420–1443.

14. G, d., l-nw, d., G, r.v., tk,a. i g gc. iProceedings of the 13th International World Wide WebConference, 2004.

15. G, J. Six Degrees of Separation: A Play. vgbk, 1990.

16. h, s., p, ., vk, C. nwk-kg: ig k cwk. Statistical Science 21, 2 (2006), 256–278.

17. h, p.d., r, a.e., hcck, m.s. lc c c wk . J.

social and techological networks

Documents