How is the News Feeling Today? Leveraging Emotional Flow in News Articles

Abstract

Emotion recognition tools are becoming widely used in social media, advertisement, feedback, in general, anywhere where understanding the user is of big significance. While those tools are used in different fields, these technologies have not yet been applied in the journalistic area. Moreover, NLP research has a clear focus on English language, in particular, increasing efforts are dedicated to experiments on English datasets and to development of different tools, which are initially created for English language, for example, emotion lexicons. In this work, we focus on detection and analysis of Emotion Flows from German news articles. We use multilingual embeddings to improve the coverage of words with emotions in German news articles. With an extensive exploratory analysis of emotion flows, we identify and then study prominent emotional patterns in German news articles. Finally, we identify key differences in emotion flows of fake and real German news articles writing about COVID pandemic.