we have three new techreps available:
- Robust Content-Driven Reputation shows that the content-driven reputation we proposed in a WWW 2007 paper can be made robust to Sybil ("sock-puppet") and other coordinated attacks. In WWW 2007, we proposed "content-driven reputation" for Wikipedia authors, where authors gain reputation if their contributions are preserved, and lose reputation if their contributions are quickly undone. The original algorithms were very prone to attacks; we show here that they can be made resistant.
In these months, we have been busy working at WikiTrust, an open-source tool for assigning reputation to wiki authors and trust to wiki content. We already have a batch (or "off-line") system, which can compute reputation and trust based on wiki dumps, such as the Wikipedia dumps made available by the Wikimedia Foundation. We are developing an "on-line" system, which can assign reputation and trust in real-time, as edits are made. One of our chief concerns in developing an on-line system was to ensure that it was robust to attack, and we believe we have made progress in this direction, as reported in the above techreps. We are now proceeding with the implementation; my guess is that we will have a prototype in a month or so.
- Assigning Trust to Wikipedia Content proposes computing the trust of Wikipedia text on the basis of the reputation of the author, and the reputation of the people who revised the text. We display text trust by coloring text background. Many of you have seen the on-line demo for the English Wikipedia, at http://trust.cse.ucsc.edu/ . This is an improved version of a November 2007 techrep on the same topic. In this improved techrep, we show how the trust system can be made resistant to attacks.
By the way, the "batch" part of WikiTrust can be easily adapted to carry out various analysis tasks. Basically, it walks over all revisions of every page of a wiki, and it contains an efficient text analysis engine that tells you precisely how text was changed between versions. So, it is easy to use WikiTrust as a platform to write analysis algorithms for wikis: you don't have to worry about the boring tasks of reading and parsing markup language, and computing text diffs in a reasonable way; you can concentrate on the details of the specific analysis you want to do. It is all open source, and we welcome developers or people interested in it.
All the best,
Luca (with Ian, Bo, and the other wikitrusters).