Harmful Content Detection

As part of the German GermEval 2025 competition series in the field of Natural Language Processing (NLP), we are organising a competition (shared task). The goal is to develop advanced NLP methods that automatically analyse and classify harmful and problematic statements in social media. The automatic identification of problematic content can, for example, support moderation processes in social media or law enforcement agencies in analysing mass communication data in social networks.

The shared task is divided into three subtasks that cover different aspects:

1. Recognising calls to action: The aim is to identify commands or calls to action, such as demonstrations or even criminally relevant actions.
2. Classification of attacks on the free basic democratic order: A distinction is made between neutral statements, legitimate criticism, inflammatory content and calls for the overthrow of the government.
3. Recognising problematic attitudes towards violence:  This task is designed to identify content that trivialises, condones or even glorifies violence.

The participants are provided with an extensive data set from an extremist movement’s environment, comprising more than 11,500 German posts and comments created by students and staff at the University of Applied Sciences Mittweida.

Participation: Every competition thrives on active participation. We hope to see many enthusiastic participants and interesting approaches. If possible, please register before the official start of the competition on 8 March 2025.

Detailed information on the task and further deadlines can be found here. Questions can also be addressed to the following email address.