“Got You! You!”: Automatic Vandalism Detection Wikipedia with Web-based Shallow Syntactic-Semantic Modeling

摘要

Discriminating vandalism edits from non non-vandalism edits in Wikipedia is challenging task, as ill ill-intentioned edits can include a variety of content and be expressed in many different forms and styles. Previous studie studies are li limited mited to rule rule-based method methods an and learning based on lexical features features, lacking in , linguistic analysis analysis. In this paper paper, we propose novel Web Web-based shallow syntactic syntactic- semantic modeling method, which utili utiliz- zes es Web search results as resource and train trains topic opic-specific n-tag and syntactic n-gram language models to detect va van- ndal dalism ism. By c combining basic task ombining task-specific and lexical features, we have achieve achieved high F F-measures using logistic boosting and logistic model trees classifiers classifiers, su sur- rpass passing ing the results reported by major Wikipedia vandalism detection systems systems.