The state of the art technology in natural language processing (NLP) is dominated by neural networks. On the one hand, neural networks are used to learn how to calculate word embeddings, some of which are particularly well-suited for representing meaning and relations between words. On the other hand, neural networks make use of these word embeddings as input for additional NLP-related tasks such as sentiment analyses. The goal of this paper is an incivility analysis of Austrian parliamentary speeches. Therefore, different neural network types in combination with different word embeddings are compared in terms of performance and suitability. The best model was chosen to classify the given data set and analyze how incivility changes over time.