Automatic Detection of Point of View Differences in Wikipedia

Proceedings of COLING 2012: Technical Papers, pages 33–50, COLING 2012, Mumbai, December 2012. Automatic Detection of Point of View Differences in Wikipedia Khalid Al Khatib1 Hinrich Schütze1 Cathleen Kantner2 (1) Institute for Natural Language Processing University of Stuttgart (2) Institute for Social Sciences University of Stuttgart ❦❤❛❧✐❞❅✐♠s✳✉♥✐✲st✉tt❣❛rt✳❞❡ ABSTRACT We investigate differences in point of view (POV) between two objective documents, where one is describing the subject matter in a more positive/negative way than the other, and present an automatic method for detecting such POV differences. We use Amazon Mechanical Turk (AMT) to annotate sentences as positive, negative or neutral based on their POV towards a given target. A statistical classifier is trained to predict the POV score of a document, which reflects how positive/negative the document’s POV towards its target is. The results of our experiments on a set of articles in the Arabic and English Wikipedias from the people category show that our method successfully detects POV differences. KEYWORDS: sentiment analysis, content analysis, natural language processing. 33 1 Introduction In many areas of public discourse, content creators strive for objectivity. Reporters try to report the facts without bias; judges are expected to write opinions that are not influenced by personal views; encyclopedias are committed to what Wikipedia calls a “neutral point of view” (NPOV), defined as “. . . representing fairly, proportionately, and as far as possible without bias, all significant views that have been published by reliable sources”.1 Even though objectivity is an important ideal, content creators cannot avoid being influenced by their background and context. There are two main reasons for this (Scheufele, 1999; Habermas, 2006; Littlejohn and Foss, 2010; D’Alessio and Allen, 2000). First, there are always many different non-equivalent ways of conveying a given piece of information. By choosing one vs. the other, the content creator introduces part of his/her point of view (POV) into the discourse. For example, “his wars caused the death of more than a million civilians” (a translation of a sentence in the French Wikipedia) puts Napoleon in a more negative light than “more than a million civilians died in his wars”, which in turn is more negative than “more than a million civilians died in the wars fought between him and his enemies”. This is so because the chain of causality between Napoleon and people being killed is more explicit in the first sentence than in the third. None of these sentences is a violation of Wikipedia’s neutral point of view policy. The second reason for different objective points of view is that the selection of what information to present is also influenced by background and context. If for space reasons only one of two equally relevant facts about a politician – one positive, one negative – can be added to a newspaper article, then this choice impacts how positive/negative the article is. This is an unavoidable dilemma journalists face on a daily basis. It is usually impossible to include all available information. We call the difference between two objective …

Content truncated to 500 words