Automatic Manifesto Comparison using NLP Techniques and The Manifesto Project Domains - Case Study: 2021 Ecuadorian Presidential Elections

Abstract

Democracies rely on the capability of a country to conduct fair elections. And, fair elections rely on an open participation of candidates and general public. In particular, we explore a way to compare campaign proposals aiding general public to make an informed decision while choosing candidates. This document explores a way to compare campaign proposals through each candidate manifest using natural language processing techniques (i.e. Doc2Vec algorithm). As a linguistic corpus we used all the articles written in Spanish from Wikipedia and we used two models of neural networks, Distributed Bag of Words (DBOW) and Distributed Memory Model (DM). We chose the 2021 Ecuadorian Presidential Elections in its second round and tagged each manifesto paragraph (from the runoff candidates) into the seven domains according to the Manifesto Project. Finally, we compute manifesto comparisons by topic and also as a whole for different vector configurations. Our results indicate that Doc2Vec produces reasonable results while comparing documents but the DBOW model provides better results while dealing with larger documents and the DM model while dealing with smaller ones.