Sentiment of tweets and socio-economic characteristics as the determinants of voting behavior at the regional level. Case study of 2019 Polish parliamentary election

Krochmal, Grzegorz

This work is dedicated to finding the determinants of voting behavior in Poland at the poviat level. 2019 parliamentary election has been analyzed and an attempt to explain vote share for the winning party (Law and Justice) has been made. Sentiment analysis of tweets in Polish (original) and English (machine-translations), collected in the period around the election, has been applied. Amid multiple machine learning approaches tested, the best classification accuracy has been achieved by Huggingface BERT on machine-translated tweets. OLS regression, with sentiment of tweets and selected socio-economic features as independent variables, has been utilized to explain Law and Justice vote share in poviats. Sentiment of tweets has been found to be a significant predictor, as stipulated by the literature of the field.