Creating and Comparing Dictionary and Word Embedding Models to Measure Discrete Emotional Appeals in German Political Text

Abstract

Previous research on emotional language relied heavily on off-­the-­shelf sentiment dictionar­ies that only measure negative and positive tone. These dictionaries are often tailored to non­-political domains and use bag­-of-­words approaches which come with a series of disadvantages. This paper creates and validates a (1) novel emotional dictionary specifically for political text and (2) word embedding models combined with neural­-network classifiers that overcome lim­itations of the dictionary approach. Both tools can measure emotional appeals associated with eight discrete emotions. The different approaches are validated on 10,000 crowd­-coded train­ ing sentences. The results highlight the strengths of word embeddings, but also important differences between both approaches. Furthermore, both approaches outperform widely used off­-the-­shelf dictionaries in measuring emotional language in German political communica­tion.