August 25, 2003

OSOSS - Crisis / Response

Many people know that I am extremely interested in the growth and emergence of Slashdot. In addition to keeping the evolution of moderation on Slashdot archive, I try to be a student of the archive and what it can teach us about catalyzing the growth of really large, really active online communities. This short foray explores a crisis-response model of large community growth.

The foundation of the model of growth we’re working on is the tension between centralization and decentralization. In their early stages, (eventually self-organizing) communities are actually highly centralized. However, as the community grows in numbers and participation metrics increase, the capacity of centralization to manage the inherent complexity in the communication channels diminishes. If the community can respond in such a way as to remove the responsibility for providing a given service from the center and distribute the responsibility over the group, the community can continue to grow and thrive. Eventually all services can be decentralized and distributed over the members of the group, allowing the group to scale to any size. (Slashdot currently boasts over 2.9 million unique users.)

Table 1 presents the history of Slashdot from the crisis-response perspective. Note that each response is a net increase in decentralization of services, whether creating content, rating the quality of comments, or rating the quality of ratings.

Crisis In the summer of 1997, Rob Malda begins a personal webpage which he updates daily. The page is of the news or blog variety. Readership quickly climbs into the tens of thousands, and adding new content to the site becomes overly time consuming for Rob.
Response In late 1997, Rob institutes a form interface to make it easier for readers to submit stories to Slashdot. Using a database-backed system, Rob begins to distribute the work load over the readership.
Crisis In January 1998, Rob posts that the TalkBack section is becoming unreadable due to name-calling and other childishness. By summer, Slashdot is receiving 1,000,000 hits per day.
Response In August 1998, Rob proposes user accounts to create accountability for statements made and to allow individual users to store preferences regarding topics they like, etc. More of the work load is distributed over the readership.
Crisis In September 1998 it is clear that accountability alone will not solve the signal-to-noise ratio problem. Although Rob and friends have been deleting offensive comments individually, their efforts cannot scale to insure quality over the entire system.
Response In September 1998 a broader voting system is instituted by which a larger group of trusted moderators (~40) can anonymously assign scores to user comments. (Slashdot source code is updated so that users can filter the comments they view based on the scores assigned by moderators.) Yet more of the work load is distributed over the readership.
Crisis In February 1999 the number of users has grown so large that the fixed group of moderators cannot provide timely ratings of user comments.
Response In April 1999 moderation rights are made available to any reader who meets certain criteria: length of membership in the community, history of positive contributions, active readership. Yet more of the work load is distributed over the readership.
Crisis In September 1999 complaints regarding the fairness of scores assigned to comments reaches a critical level. Balanced, informative comments are occasionally being scored down and trolls occasionally being scored up.
Response In September 1999 meta-moderation is implemented so that scores assigned to comments are anonymously reviewed and weighted accordingly. Yet more of the work load is distributed over the readership.
Slashdot never reached full decentralization because the site still only accepts news stories through the vetting of a handful of fulltime editors. Other communities like k5 come closer to the ideal of decentralization as community members vote on what content will appear on the site in addition to the quality of user comments.

This material is based upon work supported by the National Science Foundation under Grant No. 0133246. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.

Posted by david at August 25, 2003 02:15 PM | TrackBack