Machine Learning for Claim Detection and Classification

Currently two general approaches try to tap into the enormous potential that large text corpora present for the analysis of political processes. One the one hand several methods try to directly extract useful information from raw text by counting word or n-gram frequencies or by attempts to detect latent structures through applying topic models and similar techniques. This approach can easily handle very large text corpora but often has only a limited value for understanding the actual content of political discourse. On the other hand various discourse analytical methods rely on manually annotated text. While these approaches provide superior insights into the meaning and structure of political discourse, their applicability is limited to relatively small text corpora as the task of manual annotation is very time- and labor-intensive. In my paper I will present first results from a research project (MARDY – Modeling Argumentation Dynamics in Political Discourse) in which we develop annotation tools that use machine learning and artificial intelligence to partially automate and thus significantly speed up the annotation process. Such a tool will offer new possibilities to levering the power of discourse network analysis to the analysis of discourse dynamics based on large, long-term data sets.