Levels of Hate in Online Environments

Hate speech in online environments is a severe problem for many reasons. The space for reasoning and argumentation shrinks, individuals refrain from expressing their opinions, and polarization of views increases. Hate speech contributes to a climate where threats and even violence are increasingly regarded as acceptable. The amount and the intensity of hate expressions vary greatly between different digital environments. To analyze the level of hate in a given online environment, to study the development over time and to compare the level of hate within online environments we have developed the notion of a hate level. The hate level encapsulates the level of hate in a given digital environment. We present methods to automatically determine the hate level, utilizing transfer learning on pre-trained language models with annotated data to create automated hate detectors. We evaluate our approaches on a set of websites and discussion forums.