MURI 07

MURI 07
Information Sharing in Social Media
Xiao Wei
in collaboration with Lada Adamic & NETSI Group
School of information, University of Michigan
MURI 07
Research Projects
Ingredients in facilitating information sharing and trust building
1. Incentivizing individuals to meet each other’s information needs:
User Behavior Dynamics: A Sustainable Mechanism Works for Baidu Knows
2. Building reputation and trust through online and offline interactions:
Reputation and Reciprocity on CouchSurfing.com
3. Who is likely to contribute more valuable information?
Individual Focus and Knowledge Contribution
4. Should one build on “local” information & knowledge or draw on other
communities?
Information Diffusion in Citation Networks
MURI 07,
of Michigan
MURI
07 University
– 1. User Behavior
Dynamics: A Sustainable Mechanism Works for Baidu Knows
Seeking and Offering Expertise across Categories: A Sustainable Mechanism
Works for Baidu Knows, ICWSM 09, San Jose.
Baidu Knows:
• Largest Chinese QA site
• Virtual-point knowledge market
• Built-in community tools
Motivations & Research Questions:
• First study on this successful QA site
• How virtual points can help incentivize knowledge sharing
• Users’ adaptive behavior patterns
Dataset:
• Full history of Q&A during Dec, 2007~May, 2008
• 9.3 million questions (5.2 million of them are solved)
• 2.7 million users with 17.2 million answers
MURI 07,
of Michigan
MURI
07 University
– 1. User Behavior
Dynamics: A Sustainable Mechanism Works for Baidu Knows
Major Findings:
Answerers are incentivized by points, thus expertise can be better allocated to
more important questions
Users allocate points differently among questions:
e.g., different categories
Askers adjust price from initial question, and they can slightly improve the ability
of buying answers per point
In order to ask, users are driven to answer. Users who both ask and answer
contribute most.
MURI 07,
of Michigan
MURI
07 University
– 1. User Behavior
Dynamics: A Sustainable Mechanism Works for Baidu Knows
Conclusion:
A reinforcement cycle forms: people contribute
more, are rewarded, gain more experience,
improve their performance
MURI 07,
of aMichigan
MURI
07 University
– 2. Surfing
Web of Trust: Reputation and Reciprocity on CouchSurfing.com
Surfing a Web of Trust: Reputation and Reciprocity on CouchSurfing.com, SIN
09, Vancouver
Previous works on group level:
• Bialski & Batorski (2006) examined which factors contribute to higher trust between
CouchSurfing friends.
• Molz (2007) examined the sociological meaning of reciprocity in the context of
hospitality exchanges.
Research Questions:
Trust:
oWho is doing the vouching?
oWho is being vouched for?
oCan we predict which connections are vouched?
Dataset:
600,000+ users, 1.5 million+ friendship connections
MURI 07,
of aMichigan
MURI
07 University
– 2. Surfing
Web of Trust: Reputation and Reciprocity on CouchSurfing.com
Major findings:
Friendship degree:
1= Haven’t met yet
2= Acquaintance
3= CouchSurfing friend
4= Friend
5= Good friend
6= Close friend
7= Best friend
A high number of vouches are between “CouchSurfing friends”.
MURI 07,
of aMichigan
MURI
07 University
– 2. Surfing
Web of Trust: Reputation and Reciprocity on CouchSurfing.com
Major findings:
Results from logistic regression for each variable alone:
Global measures are poor predictors of whether an edge is vouched
Variable
Predictive accuracy:
Friendship degree
67.7%
Shared friend
55.8%
2-step vouch propagation
54.2%
PageRank
50.6%
Conclusion:
Friendship degree information is beneficial
Global measures may be useful in assigning overall reputation scores, but not for predicting if
a specific person will vouch for another or not
Further work is needed to determine if vouches are given too freely
MURI 07,
of Michigan
MURI
07 University
– 3. Individual
Focus and Knowledge Contribution
Individual focus and knowledge contribution, working paper
Previous works on group level:
• J. Katz, D. Hicks, Scientometrics 40, 541 (1997)
• B. Jones, S. Wuchty, B. Uzzi, Science 322, 1259 (2008).
• S. Wuchty, B. Jones, B. Uzzi, Science 316, 1036 (2007).
• R. Guimera, B. Uzzi, J. Spiro, L. Amaral, Science 308, 697 (2005).
• I. Rafols, M. Meyer, Scientometrics (2008).
• S. Page, The difference: How the power of diversity creates better groups, firms,
schools, and societies (Princeton University Press, 2007).
Motivations & Research Questions:
• First study on individual level
• To study whether an individual’s diversity is beneficial.
MURI 07,
of Michigan
MURI
07 University
– 3. Individual
Focus and Knowledge Contribution
Goal:
To measure the relationship between the narrowness of focus and the quality of
contribution of individuals across a range of knowledge sharing systems.
Approach:
•Focus (Stirling measure):
•Quality:
oPatents and Research Articles: Normalized citation count
oWikipedia: New word contributed that survive revisions
oQ&A forum participant: Win rate
Datasets:
• JSTOR: 2 million articles plus 6.6 million citations
•Patents: 5.5 million patents filed between 1976~2006
•Q&A forums: Crawled data from Yahoo! Answers, Baidu Knows, Naver
KnowledgeIN
•Wikipedia: Meta-history dump file of the English Wikipedia generated on Nov. 4th,
2006, parsed 7% pages
MURI 07,
of Michigan
MURI
07 University
– 3. Individual
Focus and Knowledge Contribution
Major findings:
Conclusion: Across all systems we observe a small but significant positive correlation
between focus and quality.
MURI 07,
of Michigan
MURI
07 University
– 4. Information
Diffusion in Citation Networks
Information Diffusion in Citation Networks.
Previous works:
• Visualization and quantification of the amount of information flow between
different areas in science [Boyack, 2005], [Bollen, 2009].
• Features of information flows in citation networks [Borner, 2004], [Rosvall, 2007].
• Effects of collaborations across different universities, and team collaborations [Katz,
1997], [Wuchty, 2007].
Research Questions:
What happens once information has diffused across a community boundary?
Shi X, Adamic LA, Tseng BL, Clarkson GS, 2009 The Impact of Boundary Spanning Scholarly Publications and Patents.
PLoS ONE 4(8): e6547. doi:10.1371/journal.pone.0006547
MURI 07,
of Michigan
MURI
07 University
– 4. Information
Diffusion in Citation Networks
Goal:
To study information diffusion within vs. across communities and its subsequent
impact.
Approach:
•Studying citation networks: the social ecology of knowledge – where
information is shared and flows along co-authorship and citation ties.
•Articles/patents -> nodes; citations -> directed edges, from cited to citing
•Communities: JSTOR -> Journal discipline; Patents -> Categories
•Community proximity:
nij  E[nij ]
Z ij 
E[nij ]
Datasets:
 IBM patent citation network and JSTOR citation network
MURI 07,
of Michigan
MURI
07 University
– 4. Information
Diffusion in Citation Networks
Major findings:
If we focus on patents and natural science publications that have had at least a given
level of impact, we consistently observe that citing across community boundaries leads
to slightly higher impact.
Correlations between impact and community proximity
Patent
Natural
science
Social science
Arts &
humanities
Overall
correlation
0.062***
-0.027***
0.033***
0.044***
Correlation
after removing
0 impact
-0.047***
-0.072***
0.040***
-0.011*
*** and * denote significance at < 0.001 and > 0.05 level respectively.
Conclusion: A publication’s citing across disciplines is tied to its subsequent impact.
 While risking not being cited at all, patents and publications in the natural sciences are
more likely be higher impact when they cite across community boundaries
There is no such effect in the social sciences and humanities.
Thank you!
further information:
http://www-personal.umich.edu/~ladamic
http://weixiao.us/projects.html