The algorithms underlying WikiTrust

WikiTrust implements an author reputation system, and a text trust system.

Content-driven reputation for wiki authors

The goals of the reputation system are two:
  • Authors who do good work should gain reputation.
  • It should be difficult for malicious users to fool the system, and gain reputation quickly without doing any real work.
WikiTrust uses a content-driven reputation system: authors gain reputation when their contributions are preserved by subsequent authors, and they lose reputation when their contributions are reverted.  The reputation system judges every revision on the basis of many subsequent ones, and it has been built in such a way as to be relatively resistant to spam and vandalism.  For instance, users lose only a negligible amount of reputation when vandals remove their contributions, much less than they gain from the subsequent revisions once their work is restored. 

Text trust

The trust of a portion of text is computed according to the reputation of the author, and the reputation of the people who subsequently revised the portion of text, and the text immediately surrounding it.  Thus, what WikiTrust calls "text trust" is an indication of the degree with which the text has been revised.  The trust is computed in such a way that every text change -- insertions, deletions, displacements -- leaves a low-trust mark, which alerts visitors to the modification.   It is possible to compute text trust also based on a mix of text age, and number of revisions for which the text has been present; in fact, it is straightforward to modify WikiTrust to do so.  The reason WikiTrust uses also the reputation of authors is to prevent a well-organized set of new users from cheating the system, creating content that gains full trust due to their coordinated revisions.  Since new users have low reputation, this type of attack cannot be carried out.


The algorithms underlying WikiTrust are described in various papers.