Identifying Controversies in Wikipedia using Support Vector Machines

Wikipedia is a very large and successful Web 2.0 example. As the number of Wikipedia articles and contributors grows at a very fast pace, there are also increasing disputes occurring among the contributors. As a result of disputes, many articles in Wikipedia are controversial. In this project, I propose a supervised learning model using Support Vector Machines (SVMs) to identify controversial articles in Wikipedia. The idea is to represent each article by a bag-of-word feature vector. Each value in this vector is a raw count of a word type appearing in the article. Experiments on real articles from Wikipedia show that the proposed approach can effectively identify controversial articles. Introduction and Motivation Using open source Web editing software (e.g. wiki), online community users can now edit, review and publish articles collaboratively. Among the large and more successful wiki sites is Wikipedia (Voss July 2005), the online encyclopedia which covers 16.6 million articles (both English and nonEnglish), 9.5 million users and 200 languages (Wikipedia a). As Wikipedia is growing very fast in both number and size, there is also a higher likelihood for disputes to occur among contributors. Disputes often happen in articles with controversial content, in which contributors have different or even opposite opinions. For example, “Iraq War” is one of the most controversial articles in Wikipedia. It attracts a lot of disputes among contributors because they have different standing points about the war. Some people support it while some others strongly oppose to it. They also have difficulties in agreeing on different facts of the war. Additionally, disputes can be caused by “defensive” contributors who always argue to defend their ideas even when they are incorrect. As a result of disputes, many articles in Wikipedia are controversial. In this project, I aim to identify controversial articles (controversies for short), which is important due to the following two reasons. First, controversies appearing in Wikipedia articles are often a good reflection or documentation of the real world. Finding controversies in Wikipedia can therefore help the general public and scholars to understand the corresponding real world controversies better. Second, It allows moderators and contributors to quickly identify highly Copyright c © 2009, Association for the Advancement of Artificial Intelligence (www.aaai.org). All rights reserved. controversial articles, thereby improving the effectiveness of the dispute resolution process by reducing the amount of effort searching for such articles. However, determining controversies in Wikipedia is a great challenge. First, there is a huge number of articles, which makes it impossible to manually look at each article and identify controversies. Second, the articles cover a wide range of topics that require a lot of background knowledge for identifying controversies. Third, articles are growing very fast and that makes any results ever obtained be soon outdated. There have been several approaches related to determining controversies in Wikipedia. Wikipedia currently lets users to assign controversial tags (Wikipedia b) to articles to signal its controversies. Some other works, including (Lim et al. 2006; Hu et al. 2007b; 2007a; Vuong et al. 2008; Adler and de Alfaro 2007; Kittur et al. 2007) focus on developing unsupervised models to rank articles’ qualities and controversies in Wikipedia. However, these approaches are either inefficient or achieving very low accuracy. In this project, I propose an automatic approach for identifying controversies in Wikipedia. In particular, I represent each article as a bag-of-word vector in which each value is the raw count of a word type. I then apply a supervised learning model based on SVMs to learn models for detecting controversies. These learned models are subsequently used to classify unlabeled articles. Experiments on real articles from Wikipdia show a great promise of this solution in detecting controversies.