كشف التخريب في نماذج التعهيد الجماعي

Detecting Vandalism in Crowdsourcing Models

Crowdsourcing is a common term for collaborative environments that include a group of workers participating in the content of these platforms to create knowledge. The most important and well-known crowdsourcing platforms are Amazon’s Mechanical Turk, Kaggle, and Wikipedia. The significant expansion of the encyclopedia made it one of the most visited sites for getting information or using it as a knowledge base for many applications, such as chatbots. This issue comes with another aspect through which users can sabotage the content of Wikipedia, which is known as vandalism.
Vandalism is any attempt to modify the article negatively affecting its quality, which is considered one of the five layers of the model constructed for cyber threats and is known as cyber vandalism. Several automatic detection techniques and related features have been developed to address this issue.
This thesis introduces a deep learning model with a new and light architecture to detect vandalism in Wikipedia articles. The proposed model employs a one-dimensional convolutional neural network architecture (1D-CNN) that can determine the type of modification in Wikipedia articles based on two main stages: the feature extraction stage and the vandalism detection stage. Features are extracted from edits and their associated metadata, as well as new features (reviewers’ trust), and then only the salient features are adopted to make a decision about the article; regular or vandalism can contribute to improving the accuracy of prediction. The experiments were conducted on a benchmark dataset, the PAN-WVC-2010 corpus, taken from a vandalism detection competition hosted at the CLEF conference. The proposed system, with the new features added, has achieved an accuracy of 96%.