1. The Collaborative Organization of Knowledge D. Spinellis and P. Louridas Strong Regularities in Online Peer Production D. Wilkinson Ziyad Aljarboua Monday, November 10, 2008 1 Harvard University

2. Intro - Wikipedia Free multilingual encyclopedia launched in 2001 Operated by the non-profit Wikimedia Foundation Contains 2,610,291 articles in English and 10 million in total 236 active language editions Content written by volunteers 2

3. Intro - Wikipedia Developed by Jimmy Wales and Larry Sanger Time’s 2006 list of the world’s most influential people “Largest and most popular general reference work on the internet”. Wikipedia 3 Source: Wikipedia

4. Intro - Wikipedia No formal peer-review and changes take effect immediately New articles are created by registered users but can be edited by anyone Redistribution, creation of derivative works and commercial use of content is permitted 25,000 to 60,000 page request per second 50% of traffic to Wikipedia comes from Google 4

5. Intro - Wikipedia Wikipedia contributors by country Source: Wikipedia 5

6. Intro - Wikipedia Article Count from Jan, 2001 to Sep 2007 Source: Wikipedia 6

7. Michael Scott from the office:"Wikipedia is the best thing ever. Anyone in the world can write anything they want about any subject, so you know you are getting the best possible information". Quality of articles undermined Bias: Content reflects contributors’ interest Wikipedia - Concerns 7

8. Wikipedia - vandalism 8

9. The Collaborative Organization of Knowledge Attempts to study Wikipedia’s growth: how human knowledge is recorded and organized through an open collaborative process (in Wikipedia) Examines relationship between existing and referenced nonexistent articles How existing entries foster development of new entries? 9

10. The Collaborative Organization of Knowledge Examines the recorded evolutionary development of Wikipedia's structure through article revisions and contributions Motivation: Wikipedia’s coverage has not declined while its scope sharply increased. 10

11. Growth Technologies and open participation policy behind rapid growth – Edit with no prior authorization – Edit history for all pages – Watchlist for users to alerts them for changes in their selected pages – Ability to revert changes if page is vandalized – Ability to lock entries against revisions – Easiness to link to other articles – Categorizing articles using markup tags 11

12. The Study Study processed all material on Wikipedia as of February of 2006 (485GB worth of xml documents) examined all recorded changes (28.2 million revisions on 1.9 million pages) and how entries were created and linked 12

13. General findings Reverting is returning page to previous version most of the time to undo vandalism 4% of article revisions were reverts Average time to revert a vandalized page is 13 hours 11% of pages that were reverted at least once had been vandalized at least once Most reverted and revised: George W. Bush with 28,000 revisions (2*9,300 reverts and vandalism) 2,441 entries (0.13%) locked 20% of articles were stubs 13

14. Conclusion 1 Creation of new Wikipedia entries is not a random process but is related to the references to nonexistent articles “what drives Wikipedia growth is the inclusion of red links, ie references to articles that do not exist yet.” Wikipedia 14

15. Conclusion 1 15

16. Conclusion 1 Mena number of references to a nonexistent article raised exponentially until the article was created. Once article is created, mean rises linearly or levels. 16

17. Inflationary/deflationary hypothesis Inflationary hypothesis: number of links to nonexistent articles increase at a higher rate than that of the new article creation Wikipedia is located in a midpoint between the two scenarios (thin coverage vs. decline in growth rate) 17

18. Wikipedia growth *Incomplete include nonexistent articles and stubs 18

19. Wikipedia growth Between 2003 and 2006, number of entries increased from 140,000 to 1.4 million and ration of complete/incomplete remained roughly the same Growth of Wikipedia partly attributed to splitting of articles (depth in articles translate into breadth) Rate of article creation vs rate of knowledge expansion ? 19

20. Wikipedia content Process of adding new articles that depends on current nonexistent referenced articles leads to content balance Articles are more likely to be written because they are popular (have many references leading to them) that because contributor is interested Are not most references originating from an articles will link to an article similar in subject? (assumes knowledge is a fully connected graph) 20

21. Finding 1 Process of referencing an nonexistent article and subsequent definition of that article seemed to be a collaborative effort. The person who referenced a nonexistent article and the person who started the referenced article was the same in only 3% of the cases Wikipedia growth is limited by number of contributors not individual contributors! 21

22. Conclusion 2 Wikipedia is a scale-free network 22

23. Scale-Free Network Degree of a node = number of connections to other nodes Degree distribution: probability distribution of degrees over entire network For degree j: P(j) = # nodes with degree j / # nodes Fraction of nodes with degree j to all nodes 23

24. Scale-Free Network A network where degree distribution follows a power law i.e. degree distribution approaches 1/j^s as j increases Fraction of nodes with degree j decreases as j (number of connections) increases 24

25. Scale-Free network 25 Source: Wikipedia

26. Building the network Models explaining why Wikipedia is scale-free: – Power laws result of an optimization process – Power laws result of growth model (preferential attachment model) Simple network: Wikipedia: Expected #reference: 26

27. Building the network 27

28. Strong Regularities in Online Peer Production D. Wilkinson 28

29. Introduction Open source software development, blogs, wikis, social networks… Some of most visited website … and continue to grow Online peer production share common macroscopic properties? 29

30. Objective Describe strong macroscopic regularities in people’s contributions to PPS (distribution of user participation and activity per topic) Examine basic dynamical rules guiding evolution of PPS Why distribution of levels of user participation is power law? Not a psychological analysis of contributors 30

31. Methodology Examines 4 different PPS: Wikipedia, Bugzilla, Digg, Essembly Data analyzed are exhaustive; involves all users and contributions SystemTime spanUsersTopicscontributions Wikipedia6y, 10m5.07M1.5M50M Bugzilla6y, 7m111K357k3.08M Digg3y1.05M3.57M105M Essembly1y, 4m12.04K24.9K1.31M 31

32. PPSs Wikipedia Essembly: social network for individuals to discuss and vote on political matters and organize to take action Bugzilla: bug-tracking system where developers report and collaborate to fix bugs Digg: news aggregator 32

33. User Participation Power law distribution: few dedicated members account for most activity Focus on inactive users (generality) % of Inactive: – Wikipedia: 71% of editors – Bugzilla: 95% of commentors – Digg: 61% of voters ; 56% of submitters – Essembly: 83% of voters ; 53% submitters Inactive: – Digg & Essembly: 3 months – Wikipedia & bugzilla: 6 months 33

34. Essembly Votes Digg Votes Essembly Resolves Bugzilla comments Wikipedia edits Digg submissions User Participation 34

35. User Contributions Power law exponent is strongly related to the system’s barrier to contribution (cost of contributions) Both active and inactive users have distribution of contributions that follows a power law 35

36. Participation Momentum When people stop participating? Momentum associated with user’s participation Probability of stop is inversely proportional to # of contributions 36

37. Participation Momentum 37

38. Exponent Significance Probability to contribute proportional to contribution cost (exponent) Power law exponent reflects cost to make a contribution 38

39. Distribution of count of all users (active+inactive) also follows power law but with smaller exponent User Participation % Inactive users All usersInactive users 39

40. Activity per topic # contributions/topic. (#edits/article) Popular topics attract more users  more edits. Results: – Distribution of contributions/topic is lognormal – Lognormal mean and variance depend linearly on time for topics where novelty decay is not a factor – Contributions to a topic increases its visibility and popularity. 40

41. Activity per Topic Contributions  popularity  more contributions (multiplicative reinforcement mechanism) Wikipedia Essembly Digg 41

42. Activity per Topic Number of articles Log(number of edits) Number of resolves Log(number of votes) 42

43. Activity per Topic Variance and mean depend linearly on age (t) of topic 43

44. Popularity factor – interface design Digg vs. Essembly vs. Wikipedia Small number of topics attracts vast majority of contributions (long-tail log dist. plots) 44

45. Discussion How size of a group coactively working together affect results? 45

46. Sources Wikipedia D. Spinellis and P. Louridas, “The Collaborative Organization of Knowledge” D. Wilkinson,” Strong Regularities in Online Peer Production” 46