Literature Review on How to Measure Illiberalism Using Text Data
D3.1 Deliverable for the AUTHLIB Project
The goal of this paper is to review and assess the most innovative methodologies available to conceptualise illiberalism using text data. To do so, the paper is structured in four parts: a summary of the main underlying concepts, a review of methodologies linked to Language Models, a review of data sources and data-related problems, and a roadmap of the models we are planning to build. AUTHLIB will use a MdBERTa multilingual model trained on the whole corpus of manifestos available for our countries and years of interest to classify texts characterised by illiberalism. Using this model will make us able to be methodologically innovative and sound while using a new approach to classify illiberalism. Despite the multiple problems and decisions that will have to be taken along the way, from the creation of the training data to the use of different data sources, this will still give us the options to explore new underlying dynamics and potential new associations of words and topics to illiberalism. This will in turn make us able to create a map of text concepts related to illiberalism and we will be able to measure how different political actors relate to it.
This report is a work in progress. Do not cite without the permission of the authors.
Please, contact Jan Rovny: jan.rovny@sciencespo.fr
📒 Download the report from HERE.
