The Brexit Classifier: How Helpful is Text Vectorization for Mapping the Political Value Space on Twitter

• In Event: Methodological Challenges of Hateful and Problematic Content Detection Online

Abstract

In this talk we detail the research design employed in a proof-of-concept study using social media signal to identify ideological shifts in the British society. We rely on geographically-enriched Twitter data and a purpose-built, deep learning algorithm to map the political value space of users tweeting the UK EU membership referendum onto Parliamentary Constituencies. Unlike frequency-based approaches to text classification, which simply compute the number of positive and negative words and draw a conclusion based on the final sum, text vectorization is a deep learning algorithm that draws context from phrases. The algorithm was trained with a set of 10,000 tweets manually coded as representatives of nationalist, globalist, populist, and economist sentiments thus allowing to observe the temporal and spatial fluctuation of the British political value space expressed on Twitter. The talk unpacks the Brexit classifier, the tests performed to train the algorithms, and its performance in identifying the four key ideological coordinates explored in the study.

Authors

• Marco Bastos, City, University of London

• Dan Mercea, City U London

• Andrea Baronchelli, City, U of London