Creating a vandal detector for a wiki

Not so long ago, the First International Competition for the Detection of Acts of Vandalism on Wikipedia (hereinafter the links lead to PDF documents) ended: 9 groups (five from the USA, one connected with Google) tried to develop a holistic system to detect cases of vandalism.

The winning approach helps to detect 20% of acts of vandalism without erroneous triggers for regular changes, or, with a different setting, 95% of vandalism with 30% of false positives.



Thus, using both configurations, only 34% of all edits will need to be rechecked manually. It’s not possible to say whether bots based on Wikipedia’s rules will compete with this self-learning strategy machine. In any case, there is potential for improvement, because two detectors use different principles for defining violations: the first analyzes the edited content, while the second analyzes the context using WikiTrust .



via slashdot.org