Computational scaling of political positions from textual data using word embeddings

Abstract

Political positions can be scaled from textual data using word embeddings. This extended abstract proposes a method for automatically scaling political positions using word embeddings from political speeches. The method is based on the the idea of using association-based scores with two dictionaries, like in the word embedding association test (WEAT). I conducted computational experiments to show that the method works in principle and give ideas on how I want to improve the method in the future.