Developing an Incivility Dictionary for German Online Discussions – a Semi-Automated Approach Combining Human and Artificial Knowledge

Incivility in online discussions has become an important issue political communication research. Instruments and tools for the automated analysis of uncivil content, however, are rare, especially non-English user-generated text. In this study, we present a) extensive dictionary (DIKI - Diktionär für Inzivilität, English: Dictionary Incivility) to detect incivility German-language discussions, b) a semi-automated two-step-approach that combines manual content with keyword collection using pre-trained word embedding model. We show DIKI clearly outperforms comparable dictionaries have been used as alternative instruments measure (e.g., LIWC) well basic machine learning approaches text classification. Further, provide evidence embeddings can fruitfully be employed explorative phase creating dictionaries. Still, evaluation confirms detecting complex context-dependent forms remains challenging constant update would needed maintain performance. Finally, detailed documentation developing process may serve guideline further therefore freely available instrument also will applicable web interface drag-and-drop data (diki.limitedminds.org).