Social Web 2.0 Class Week 4: Social Networks, Privacy

Social Web 2.0Implications of Social Technologies for Digital Media

Shelly Farnham, Ph.D.

Com 597 Winter 2007

Week 4 Social Networks Privacy

Social Networks Defined

Set of pair-wise relationships vs. individual relationships or groups

Social network analysis examine density, boundedness, size,

heterogeneity Study flows of influence, information, social

supportNetwork vs. group social structure

Social Networks

Measuring Ties Ties: friendship, role sharing, common event,

common property, co-occurrence Tie strength: affection, frequency of interaction, trust,

frequency of co-occurrence

Network structures singletons, stars, middle region, giant Tend to be reciprocal (84% in yahoo, 70% in flickr)

Social Networks Today

Modern transformation in how things get done communication, collaboration, information flow

Social networks most appropriate model Lateral, not hierarchical command chains Ad hoc teams, not static groups

Developing social capital – the value of social networks

NetWORK It’s not what you know, it’s who you know Intensional networks

Building Maintaining Activating when need to get work done

Social Networks Online

100 million on MySpace, 80% of total

Why Articulate Social Networks Online

Social networks contextualizes information and behavior Increase accountability by putting people back in social

context Informational context, understanding author Exploit transitive trust (implicit and explicit referrals) Increase relevance via similarity/affiliation Provide access to weak ties Define access/sharing/subscription (filter out as much as

increase access)

Navigation tool Browse network vs. directed search Info transfer horizontal, across hierarchical boundaries

Social Networks Online cont’d

Processing outside user awareness: Alternative “similarity” measures aside from explicit friend/family lists:

Cross-links Communication patterns Co-occurrence in groups Co-occurrence of semantic tags

Prioritize match-making by distance in network “We recommend you check out Jon’s story…” Closer is better, # of overlap is better, etc.

Network/cluster analysis, use for prioritizing search results developing semantic hierarchies Extraction of groups (dense, tightly bounded networks)

Isolate connectors Identify people connected across network clusters, able to transfer info/trust

Online Social Networking Issues

Often binary (friend/no friend) with friend list glut Assume one network per person, no subnets causing role conflict Social capital of “connector” lost Systems do not expect social networks to be dynamic, become out

of date “Now what” – ok so I built my network, now what? No cross-property integration, building network over and over Developing critical mass Visualizations often outside comprehension of average user

Teens use

55% of online teens use social networks 66% of those have private profile 48% visit daily More common in older teen girls (70%) than boys (54%)

Why? Friends as center of life, 91% say to keep up with friends Stage of life: expanding network Do not have face-to-face access (parental control) Manage communications outside email (less spam)

Implicit Social Networks

Based on who’s interacting with whom Provide sense of who’s important to whom Dynamic, changes as levels of interaction

change Minimal maintenance required

Personal Map

Point to Point

Inner Circle

Wallop

managing, knowledge seeking, communication and sharing managing, knowledge seeking, communication and sharing

Personal MapAutomatically organize contacts in a way that is meaningful/intuitive to user

Infers implicit social groups from communication behavior in email

Provide sense of who’s important

Dynamic, changes as levels of interaction change

Minimal maintenance required

Shelly Farnham::Will Portnoy

Similarity (A B) = (sum (AB * significance))/sqrt(A * B)Grouped using hierarchical cluster analysis

Personal Map User StudyPersonal Map

Not at All Extremely So

765432

Fre

quency

8

6

4

2

0

Accuracy 15 MS employees

85% spent no time organizing contacts

contacts not very organized (M = 2.3)

They liked the Personal Map (M = 5.1) “wow” “that’s cool” “makes more sense sooner

than the contact list” They did not find it confusing

or difficult to use (M = 2.9) They rated it as very accurate

(M = 5.7)

on scale of 1 = not at all to 7 = extremely so

Point to Pointfacilitate knowledge exchange by exploiting corporate social network information

0

500

1000

1500

2000

2500

3000

3500

4000

4500

5000

1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49

Size of Distribution List

At Microsoft:75,000 mailing lists,each person belongs to on average 11 mailing lists

Social network info presented relative to selfShelly Farnham::Will Portnoy

Point to Point User Study I

Rank of Similarity to User (1 = Most Similar)

383430262218141062

Prop

orti

on o

n L

ist

.8

.7

.6

.5

.4

.3

.2

.1

0.0

People most similar to the user tended to also be on the user’s list of coworkers.

Rank of Similarity to User (1 = Most Similar)

383430262218141062

Prop

orti

on C

ross

ed O

ff M

ap

.8

.7

.6

.5

.4

.3

.2

.1

0.0

People most similar to the user were not crossed off map

as not belonging.

39 employees completed task Participants listed 15 closest co-workers, used to assess

accuracy of point to point map

Point to Point User Study II 17 employees completed 16 choices using Point to Point Decide between two randomly selected people whom you would like to

meet for knowledge exchange

Relative Status

0

2

4

6

8

10

12

Overlapping People

0

0.5

1

1.5

2

2.5

3

Unchosen

Chosen

Organizational Distance

10.8

11.1

11.4

11.7

12.0

12.3

12.6

12.9

network information affected decision-making

Point to Point User Study II

2 3 4 5 6 7

Job typeSimilarity to me in job title

Job Status (e.g., PM vs. Lead PM)Whether I know their team

Whether I know themWether they know someone I know

Whether they appear to be an expertNumber of reports

Nearness to me in corporation

% People Use Method

How Often Use Method*

Methods used to learn about a person:Company's email address book 100% 4.5Internal/external web searches 59% 3.1Ask co-workers 35% 2.8Company's web-based org chart 24% 3.0

Methods used to learn about a project or group:Internal/external web searches 82% 4.0Ask co-workers 47% 3.7

*where 1 = Never and 5 = Always

How People Find InformationHow people currently find information about people

and groups within their corporation.

Self-reported importance of features in deciding whom participants would meet (where 1 = not at all and 7 = extremely so).

Wallop embed interactions in social context to activate prosocial norms

Sean Kelly :: Shelly Farnham :: Alwin Vynmeister :: Richard Hughes :: Will Portnoy :: Ryszard Kott :: Lili Sean Kelly :: Shelly Farnham :: Alwin Vynmeister :: Richard Hughes :: Will Portnoy :: Ryszard Kott :: Lili ChengCheng

Blog, share media, build conversations in context of social network

Use communication and sharing behavior to build implicit network

Use network to define scope of search, notifications, sharing

Wallop:Large Scale Deployment

August 01 2004 – present

Over 47,000 registered users

26% become active, logging in and adding content at least once a week

72% users in Chinese Time zone

Communication and Sharing Behavior of Active Users

0 2 4 6 8 10

Mp3s

Blog entries

Images

Comments

Logins

Activity Count per Week

8.1

4.7

2.4

1.9

.6

Wallop Basic Usage Statistics People building conversations,

responding to each other’s content Total threads with > 1

messages: 47,074 ~38% of blog entries have a

threaded conversation Average thread length: 2.98 Average # participants: 2.53 Longest thread: 40

People customizing look of blog 48% active users have selected

profile image 23% have selected a background

image

Wallop Basic Usage StatisticsSocial Network of Active Users

Average number of people in visible network: 8.25

People wanted ability to explicitly add/remove people, but did not use too heavily Each user explicity include 1.7

people on their network Each user explicitly excluded .7

people from their network

Size of network largely determined by invite quota: r = .56

Total Network Size

363228242016128.04.0.00

Count

4000

3000

2000

1000

0

Managing Viral Growth Invite only

membership Tiered invite process

Limited invites by generations to optimize “seed” person’s network.

1st Generation: 10 invites2nd Generation: 5 invites3rd Generation: 5 invites4th Generation 0 invites

Managing Viral GrowthGoal: linear system growth Daily/lifetime activity

quotas

Daily recapture of invite quota from inactive users

Prioritize and promote healthy users for granting invite quota requests

Invite allocation = Function (System cap - Current registered users - Outstanding Invitation Liability)

Promoting Healthy UsersHealth = Function ((logins, content creation, commenting) * recency *

longevity)

Characteristics of a Healthy User Many active contacts in network Daily posts with pictures and

music Multiple comments from contacts

on each post Rich customization of profile and

blog Visits lots of other people’s

pages Long discussion threads

Wallop Users Feedback Users value the visual appeal of the user interface The interface is very cool!

The incredible ease of use and just plain "coolness" factor that allow me and friends who do not live close to interact on a daily basis

“Expressing myself” rated most important reason for use of Wallop, over sharing with friends and meeting new people I found we Chinese are really poor at expressing our passion, especially for

our family members, we love them indeed, but we are not able or dare not to speak it out. Fortunetely, wallop has provided me such a chance to record my feeling down. It's a great tool, sometimes i think it's amazing, disantce don't exist here, we can go everywhere in the community.

Needed better tools for managing bad behavior Implemented blocking, and protecting network: public but read only On average, 8% of active people have blocked someone (3 times each), and

8% have protected their network

Chinese users, UI problems and language problems

Wallop Deployment Lessons

Implicit network effective for bootstrapping, low maintenance Communication and invitations most useful measures

of connection People still want ability add/remove/pin people

People valued identity play and social interactions Personalization and value expressive features rated

most important by users Conversation around blogs and media actively used

feature

Design Implications Building social network should not be an end in

itself -- for users task is not grow network but define who I share with define who I watch Share who I know to help others find info/support

Users want to see how people are connected, provides context

Network info should be used to prioritize/structure information

Build in referrals, intros through network Reflect dynamism of relationships, multiplicity of

networks Simplify user interface to relevant data

Friendster Home Page

Dating:HookupsDirect PesteringFamiliar Stranger

FakestersCulturalGroupsPassing

FaceBook Home Page

FaceBook Home Page

Wallop Home Page

MySpace Home Page

Tribe Home Page

Tribe Tribe page

MySpace Invite

Build through invitations

Myspace Add Friend

Add person when viewing profile

Network used Primarily to find similar others(in same crowd,Same age etc)For dating

LinkedIn Invite

Building businessRelationships, Transitive trust important

LinkedIn – Social Network

Users want to see how connected

Friend of a friend meaningful,not beyond that

LinkedIn -- Add

LinkedIn Introduction Chain

Facebook

Emphasis on Similarity by Org, used to define access etc.

Facebook

Evite – managing network

Evite -- inviting

Gmail – managing network

Email still the “killer app”

Inner Circle (from MSR)Goal: provide easy

access to communication history and shared documents according to important people and groups

Infers importance from transaction history

Authorship and sharing history natural way to organize information

Sharing Models

PrivacyThe right and desire of a person to control

disclosure of personal health information Sharing models

Models for defining who has access to what information

Opposing tensions Desire to learn and share vs. privacy concerns

Privacy

Personally identifying information Identifies online persona with real world you

First name, last name Pictures SS #, credit card numbers Re-identification

Can piece together is neighbor Joe because he’s the only one with prostrate cancer in small town of springfield

Must be Jane because she’s the only person with breast cancer who’s a patient of Doctor Smith in Cincinatti

Sensitivity of privacy info Potential for abuse by industries, employers Shame, embarrassment

Design Implications -- privacy Identity

Anonymity Pseudonimity Within-system identity

(need transition to out of system) Privacy through aggregation Access controls

Perimeter definition By class, by person, by similar others, everyone, by data type, by

organization Access levels

password protected, unlisted, public/searchable Level of detail by distance in network

Strong tie, weak tie, 1st, 2nd 3rd degree Default settings: site should be more conservative than users

Design implications -- sharing

When sharing, people often less concerned with privacy than they say People tend to go with default settings Favor easy options over wise

Users want sense of control, sense of who sees what Make audience known, concrete Want to see what audience sees

Requires social intelligence Plausible deniability about status, not in your face you are not

categorized as “friend” Do not “delete” people, remove them from sharing list

MySpace privacy settings

FaceBook Privacy settings

Facebook – privacy settings

Facebook -- privacy

LinkedIn Privacy Settings

See friend of friend but not beyond

Profile fields, none basic or full

Social Networks in an Age of Web 2.0 FOAF