Extracting Topic-Specific Ideological Positions from Articles

• In Event: Text Analysis: Applications

Abstract

Social media is increasingly transforming political information consumption from an outlet-based process to an article-based process. However, previously proposed methods for measuring media slant have focused on the average ideological position of media outlets, and perform poorly at the article, and even at the aggregated topic or issue level. They are also not suitable for more complex ideological environments, in which there are multiple salient, cross-cutting ideological dimensions. Collaborating with the IBM Europe Center for Advanced Studies, we build upon Watson Discovery News and Watson Language Understanding, as well as a new large corpus of coded texts from newspapers, links shared on social media, individual browsing histories, and texts generated by political actors, and introduce a semi-supervised algorithm which identifies the main topic of political articles, and places them on a topic-specific ideology scale. We also introduce a mobile application which allows UK users to find the ideological position of the articles they read, and suggests other articles on a similar topic, but which are situated at different ideological positions.

Authors

• Iulia Cioroianu, New York University

• Susan A. Banducci, University of Exeter