Topic and Viewpoint Extraction for Diversity and Bias Analysis of News Contents

摘要

News content is one kind of popular and valuable information on the Web. Since news agencies have different viewpoints and collect different news materials, their perspectives on news contents may be diverse (biased). In such cases, it is important to indicate this bias and diversity to newsreaders. In this paper, we propose a system called TVBanc (Topic and Viewpoint based Bias Analysis of News Content) to analyze diversity and bias in Web-news content based on comparisons of topics and viewpoints. The topic and viewpoint of a news item are represented by using a novel notion called a content structure consisting touple of subject, aspect and state terms. Given a news item, TVBanc facilitates bias analysis in three steps: first, TVBanc extracts the topic and viewpoint of that news item based on its content structure. Second, TVBanc searches for related news items from multi-sources such as TV-news programs, video news clips, and articles on the Web. Finally, TVBanc groups the related news items into different clusters, and analyzes their distribution to estimate the diversity and bias of the news contents. The details of clustering results are also presented to help users understand the different viewpoints of the news contents. This paper also presents some experimental results we obtained to validate the methods we propose.