Detection of conspiracy theories and antisemitic speech in short political texts using language models
Speaker: Helena Mihaljević, HTW Berlin/ ECDF
The modern digital landscape allows conspiracy theories and hate speech to spread with unprecedented speed and reach. This makes it increasingly difficult for NGOs, journalists, and researchers to monitor, analyze, or even counter the spread of these phenomena. AI models can be a useful tool in this regard. In this talk, I will show how Masked (Small) Language Models and Autoregressive Large Language Models can be used to classify short texts from online platforms for the presence of conspiracy theories or anti-Semitic content. We will discuss the specific challenges of different platforms and linguistic expressions for algorithmic recognition, as well as the opportunities and hurdles for a meaningful use of such models in a social context.
