Wikipedia Vandalism Detection: Combining Natural Language, Metadata, and Reputation Features

Abstract

Wikipedia is an online encyclopedia which anyone can edit. While most edits are constructive, about 7% are acts of vandalism. Such behavior is characterized by modifications made in bad faith; introducing spam and other inappropriate content. In this work, we present the results of an effort to integrate three of the leading approaches to Wikipedia vandalism detection: a spatiotemporal analysis of metadata (STiki), a reputation-based system (Wiki-Trust), and natural language processing features. The performance of the resulting joint system improves the state-of-the-art from all previous methods and establishes a new baseline for Wikipedia vandalism detection. We examine in detail the contribution of the three approaches, both for the task of discovering fresh vandalism, and for the task of locating vandalism in the complete set of Wikipedia revisions. Authors appear alphabetically. Order does not reflect contribution magnitude.

Extraction of Hidden Social Networks from Wiki-Environment Involved in Information Conflict

“…WikiTrust was also focused to identify vandalized encyclopedic articles. Here the acts of vandalism are identified by the activity of anonyms (users acting without registration) and new users, who are just registered [16,17].…”

Profiling vandalism in Wikipedia: A Schauerian approach to justification

“…Algorithms of the kind calculate the probabilities for each edit that it is actually vandalistic. Four varieties have been developed so far (Adler et al 2011). As far as content is concerned, they may focus on language features (e.g., bad words, pronoun frequencies), or on language-independent textual features (e.g., use of capitals, changes to numerical content, deletion of text).…”

The use of software tools and autonomous bots against vandalism: eroding Wikipedia’s moral order?

“…Research in this area has crystallized into four categories of algorithm (Adler et al 2011). Algorithms may focus on language features (like bad words, pronoun frequencies), on language-independent textual features (like the use of capitals, changes to numerical content, deletion of text), on metadata of edits (like time and place of the edit, anonymity, absence of revision comment), or on the editor's reputation as a trustworthy contributor.…”

“…Computer scientists have been struggling with the question which approach to the detection of vandalism is the most fruitful. Based on a computer tournament with all approaches in the competition the provisional answer seems to be: a combination of all four algorithms works best (Adler et al 2011).…”