Towards Automatic Comparison of Online Campaign Versus Electoral Manifestos

Social media is an important information outlet and a new political landscape for politicians. In fact, politicians use social media to promote their candidacies while running for office. In this paper, we discuss about an application prototype built to measure the closeness of a candidate electoral manifesto to hers/his online campaign. We use four different similarity measures on the resulting frequency arrays obtained by processing manifestos and timelines. We show our results tracking the 2019 Ecuadorian Sectional Elections based on data collected from candidates’ timelines on Twitter during the campaign and their official campaign manifestos. We configured our application to gather information from Major candidates in the city of Quito during the 2019 Ecuadorian Sectional Elections. This prototype collected Tweets into a relational database based on each candidate’s Twitter account. For this campaign, 18 candidates run for office. From these, we gathered 17 electoral manifestos and fed them to our application. Both, tweets and manifestos were preprocessed in order to produce a high dimensional word vector describing the collected timelines of each candidate and his/her manifesto. Later, each similarity measure, i.e., L1-norm, L2-norm, distance correlation and the cosine similarity were used to compare a candidate political plan against hers or his digital campaign. Our results show agreement from three measures (L2-norm, distance correlation and the cosine similarity) and suggest that candidates drift from their electoral manifestos during social media campaigns. We introduce a small experiment using a designed word query to enhance the discussion of possible reasons for our results and pave the path for future research.