As part of the inaugural lectures of the Language Technology and Data Analysis Laboratory (LADAL
), Dr. Gregor Wiedemann
will give a presentation on automatic detection of hate speech.
The Event begins at 9:00 a.m. CET
About the Event
Social sciences have opened up to text mining, i.e., a set of methods to automatically identify semantic structures in large document collections. However, the methods have often been limited to a statistical analysis of textual data, strongly limiting the scope of possible research questions. The more complex concepts central to the social sciences such as arguments, frames, narratives and claims still are mainly studied using manual content analyses in which the knowledge needed to apply a category (i.e. to “code”) is verbally described in a codebook and implicit in the coder’s own background knowledge. Supervised machine learning provides an approach to scale-up this coding process to large datasets. Recent advantages in neural network-based natural language processing allow for pretraining language models that can transfer semantic knowledge from unsupervised text collections to specific automatic coding problems. With deep learning models such as BERT automatic coding of context-sensitive semantics with substantially lowered efforts in training data generation comes within reach to content analysis. The talk will introduce to the applied usage of these technologies along with two interdisciplinary research projects studying hate speech and counter speech in German Facebook postings, and information extraction for the analysis of the coverage of protest events in local news media.