Submitted to University of Konstanz
Date: February2019
Abstract: The term media bias denotes the differences of the news coverage about the same event. Slanted news coverage occurs when journalists frame the information favorably, ie, they report with different word choice about the same concept, thus leading to the readers’ distorted information perception. A word choice and labeling (WCL) analysis system was implemented to reveal biased language in news articles. In the area of Artificial Intelligence (AI), the WCL analysis system imitates well-established methodologies of content and framing analyses employed by the social sciences. The central thesis contribution is a development and implementation of the multistep merging approach (MSMA) that unlike state-of-the-art natural language preprocessing (NLP) techniques, eg, coreference resolution, identifies coreferential phrases of a broader sense, eg,“undocumented immigrants” and “illegal aliens.” An evaluation of the approach on the extended NewsWCL50 dataset was made achieving the performance of 𝐹1= 0.84, which is twice higher than a best performing baseline. Finally, to enable visual exploration of the identified entities, a four-visualization usability prototype was proposed and implemented, which enables exploring entity composition of the analyzed news articles and phrasing diversity of the identified entities.