27
Social Interactions and Commerce (in Second Life) April 5, 2011 (catchup)

Social Interactions and Commerce (in Second Life) April 5, 2011 (catchup)

Embed Size (px)

Citation preview

Page 1: Social Interactions and Commerce (in Second Life) April 5, 2011 (catchup)

Social Interactions and Commerce (in Second Life)

April 5, 2011

(catchup)

Page 2: Social Interactions and Commerce (in Second Life) April 5, 2011 (catchup)

Last week: How does the social network affect economic decisions?

• Observing other decisions• Coordination decisions (external costs)• Simple models and how to exploit models (viral marketing)

• The real world is a lot more complicated and harder to observe…

Page 3: Social Interactions and Commerce (in Second Life) April 5, 2011 (catchup)
Page 4: Social Interactions and Commerce (in Second Life) April 5, 2011 (catchup)

Today: How does the social network affect economic decisions?

• Observation study• A “real world”-ish setting

– Large observable social network– Diverse types of economic activity– Real money involved

• But:– Production costs are zero– Transportation costs are zero– No “mass market” advertising – instead events

and groups

Page 5: Social Interactions and Commerce (in Second Life) April 5, 2011 (catchup)

Niche markets? Users interacting with a seller are likely to share a group

Page 6: Social Interactions and Commerce (in Second Life) April 5, 2011 (catchup)

What contributes to revenue?

signif

not signif

Associative sorting? Partnerships?

Diversity of your “niche”, in groups

Page 7: Social Interactions and Commerce (in Second Life) April 5, 2011 (catchup)

Social network of 100 boutique sellers

Green links are transaction networks; blue links are chat networks; red links are friend networks. Node size represents total revenue made in April 2009. Edge weight is number of transactions and chat frequency.

Page 8: Social Interactions and Commerce (in Second Life) April 5, 2011 (catchup)

Social network of 100 largest sellers

Green links are transaction networks; blue links are chat networks; red links are friend networks. Node size represents total revenue made in April 2009. Edge weight is number of transactions and chat frequency.

Page 9: Social Interactions and Commerce (in Second Life) April 5, 2011 (catchup)

Shopping districts in SL

Page 10: Social Interactions and Commerce (in Second Life) April 5, 2011 (catchup)

What contributes to revenue?

signif

not signif

Associative sorting? Partnerships?

Diversity of your “niche”, in groups

Page 11: Social Interactions and Commerce (in Second Life) April 5, 2011 (catchup)

What contributes to repeat business?

signif

not signif

Diversity of your “niche”, in groups

%customers in a group with you

Page 12: Social Interactions and Commerce (in Second Life) April 5, 2011 (catchup)

Conclusions

• Goal: empirical foundation for understanding the relationship between social behavior, communication, and economic activity

• Social ties and group membership are significant factors in SL trade

• Social ties are more associated with transfer of free goods than commercial transfers

Page 13: Social Interactions and Commerce (in Second Life) April 5, 2011 (catchup)

CNN (2008). A British couple who married in a lavish Second Life wedding ceremony are to divorce after one of them had an alleged "affair" in the online world. Amy Taylor, 28, said she had caught husband David Pollard, 40, having sex with an animated woman. The couple, who met in an Internet chatroom in 2003, are now separated.

"I went mad -- I was so hurt. I just couldn't believe what he'd done," Taylor told the Western Morning News. "It may have started online, but it existed entirely in the real world and it hurts just as much now it is over.“ Second Life allows users to create alter egos known as "avatars" and interact with other players, forming relationships, holding down jobs and trading products and services for a virtual currency convertible into real life dollars

Taylor said she had caught Pollard's avatar having sex with a virtual prostitute: "I looked at the computer screen and could see his character having sex with a female character. It's cheating as far as I'm concerned.”

The couple's real-life wedding in 2005 was eclipsed by a fairy tale ceremony held within Second Life. But Taylor told the Western Morning News she had subsequently hired an online private detective to track his activities: "He never did anything in real life, but I had my suspicions about what he was doing in Second Life." iReport.com: Anger in a virtual worldPollard admitted having an online relationship with a "girl in America" but denied wrongdoing. "We weren't even having cyber sex or anything like that, we were just chatting and hanging out together," he told the Western Morning News.

Taylor is now in a new relationship with a man she met in the online roleplaying game World of Warcraft.

Page 14: Social Interactions and Commerce (in Second Life) April 5, 2011 (catchup)

Anonymity and Privacy in Networks

4-07-2010

Page 15: Social Interactions and Commerce (in Second Life) April 5, 2011 (catchup)

SL Dataset 1: 2008-2009

• Anonymized dataset from Linden Labs– SL users may want privacy!

• Friend lists (4.2M users, 43M relations)– Clustering coefficient, etc look like real social nets

• Groups (520K groups, 23M affiliations)– Some O(10,000) but median size 7

• Chat volume between users• Transactions

– Time, type, amount transferred

– Avg revenue for sellers is 100k/m (L$)

Page 16: Social Interactions and Commerce (in Second Life) April 5, 2011 (catchup)

Landmarks in Privacy• Numbers 1:1-1:2. O(103 years BC)

– God spoke to Moses in the Sinai Desert, in the Communion Tent on the first [day] of the second month in the second year of the Exodus, saying: Take a census of the entire Israelite community….

• Article I, Section 2, Clause 3 of the US Constitution 1787.– Census every 10 years

• … Large datasets about individual people

• AOL Search log release August 2006– 20M queries, 650k users

Page 17: Social Interactions and Commerce (in Second Life) April 5, 2011 (catchup)

Landmarks in Privacy

• AOL Search log: user 17556639

• 17556639 how to kill your wife17556639 how to kill your wife17556639 wife killer17556639 how to kill a wife17556639 poop17556639 dead people17556639 pictures of dead people17556639 killed people17556639 dead pictures17556639 dead pictures17556639 dead pictures

• 17556639 murder photo17556639 steak and cheese17556639 photo of death17556639 photo of death17556639 death17556639 dead people photos17556639 photo of dead people17556639 www.murderdpeople.com17556639 decapatated photos17556639 decapatated photos17556639 car crashes317556639 car crashes317556639 car crash photo

Page 18: Social Interactions and Commerce (in Second Life) April 5, 2011 (catchup)

Landmarks in Privacy• Numbers 1:1-1:2. O(103 years BC)• Article I, Section 2, Clause 3 of the US Constitution 1787.• AOL Search log release August 2006

– 20M queries, 650k users– Chain of command from CTO down fired or quit

• No user names but….– Lots of social security numbers, credit card numbers, …– Sometimes I search for “William W. Cohen”– If you know a few public things about me you can confirm which

user I am…who else is searching for Kleinberg papers, bluegrass tabs, CMU snow closing announcements ?

Page 19: Social Interactions and Commerce (in Second Life) April 5, 2011 (catchup)

Landmarks in Privacy• Numbers 1:1-1:2. O(103 years BC)• Article I, Section 2, Clause 3 of the US Constitution 1787.• AOL Search log release August 2006

– 20M queries, 650k users– Chain of command from CTO down fired or quit

• Netflix Competition round 2

• When can you uniquely re-identify someone in a dataset using only publically available information?

Page 20: Social Interactions and Commerce (in Second Life) April 5, 2011 (catchup)

WWW 2007

Page 21: Social Interactions and Commerce (in Second Life) April 5, 2011 (catchup)

Can you identify nodes in an unlabeled graph?Will someone at FaceBook be fired if they release just node ids and edges?

Page 22: Social Interactions and Commerce (in Second Life) April 5, 2011 (catchup)

Can you identify nodes in an unlabeled graph?

1. Decide who you want to identify: w1,…,wk

2. Create k dummy users (nodes) w1,…,wk and link them to each other in a unique way

3. Create edges from x1 to w1, … xk to wk

Before the release:

Yes!

Page 23: Social Interactions and Commerce (in Second Life) April 5, 2011 (catchup)

Can you identify nodes in an unlabeled graph?

1. Decide who you want to identify: w1,…,wk

2. Create k dummy users (nodes) x1,…,xk and link them to each other in a unique way

3. Create edges from x1 to w1, … xk to wk

Before the release:

After the release:

1. Find your subgraph and identify x1,…,xk

2. Use this to identify nodes w1,…,wk

Page 24: Social Interactions and Commerce (in Second Life) April 5, 2011 (catchup)

Can you identify nodes in an unlabeled graph?

1. Decide who you want to identify: w1,…,wk

2. Create k dummy users (nodes) x1,…,xk and link them to each other in a unique way (not in G)

3. Create edges from x1 to w1, … xk to wk

Before the release of G:

After the release:

1. Find your subgraph and identify x1,…,xk

2. Use this to identify nodes w1,…,wk

• Uniquely linking up x1,…,xk is easy: connect nodes randomly, and with high probability it will be unique, if k >> 2logn

– There are 2k*k ways of connecting the nodes

– There are only nk actual size-k subgraphs in G

• Finding the subgraph is easy too….

Insights

Page 25: Social Interactions and Commerce (in Second Life) April 5, 2011 (catchup)

Can you identify nodes in an unlabeled graph?

1. Decide who you want to identify: w1,…,wk

2. Create k dummy users (nodes) x1,…,xk and link them to each other in a unique way (not in G)…also linking xi to xi+1

3. Create edges from xi to wi,

Before the release of G:

After the release:

1. Find your subgraph

2. Use this to identify w1,…,wk

• Uniquely linking up x1,…,xk is easy: connect nodes randomly, and with high probability it will be unique, if k >> 2logn

– Especially if you vary the node degrees from xi to G as well

• Finding the subgraph is easy too….– Look for paths of length k rooted at

a node y1 (=x1)

– Prune at step j if it is connected to the wrong subset of y1,…,yj

Insights

Page 26: Social Interactions and Commerce (in Second Life) April 5, 2011 (catchup)

Can you identify nodes in an unlabeled graph?

• Experiments: on LiveJournal friendship graph (4.4M nodes, 77M edges):– Anonymize by discarding

name of each user– Run the attack with k

nodes, degree varying from 10…20 or 20..60.

– Measure Pr(success)– Seven nodes seems to be

enough to target 34-70 nodes (depending on degree).

Page 27: Social Interactions and Commerce (in Second Life) April 5, 2011 (catchup)

Other work

• Do passive attacks work?– Yes: a small coalition of users can identify their

collective friends with high probability (Bakstrom et al WWW 07)

• Can you make networks more resistant to re-identification attacks?– Yes: you can add noise (by randomly adding edges),

or aggregate the network by replacing subnets with nodes (Hay et al VLDB 2008, …)

• Can you hide or partition the network and allow an analyst to make (useful) queries?

– ….