Biden vs Trump: Modeling US General Elections Using BERT Language Model

Rohitash Chandra,Ritij Saini

doi:10.1109/access.2021.3111035

Abstract

Social media has played a crucial role in shaping the worldview during election campaigns. It has been used as a medium for political campaigns and a tool for organizing protests; some of which have been peaceful, while others have led to riots. Previous research indicates that understanding user behaviour, particularly in terms of sentiments expressed during elections, can indicate the election outcome. Recently, there has been tremendous progress in language modelling with deep learning via long short-term memory (LSTM) models and variants known as Bidirectional Encoder Representations from Transformers (BERT). Motivated by these innovations, we develop a framework to model the US general elections. We investigate if sentiment analysis can provide a means to predict election outcomes. We use the LSTM and BERT language models for Twitter sentiment analysis leading to the US 2020 presidential elections. Our results show that sentiment analysis can form a general basis for modelling election outcomes where the BERT model indicates Biden winning the elections.

Highlights

Political forecasting is an area where analytical and statistical methods predict election outcomes mainly using surveys and qualitative approaches [1]
There is interest in the US elections from many different countries in the world with tweets from 40 different languages; a large proportion of the tweets are in English that originate from the US with 92,984 classified as English tweets using the Langdetect library [59]
In the long short-term memory (LSTM) model, we find that the number of negative and positive tweets is similar, but there is a large influx of neutral tweets, which is almost double compared with the Bidirectional Encoder Representations from Transformers (BERT) model

Summary

Introduction

Political forecasting is an area where analytical and statistical methods predict election outcomes mainly using surveys and qualitative approaches [1]. This includes analysis of manifesto of political parties while looking at the trend of the popular news media, which is known as political analysis [2], [3]. There are major challenges in getting a good representation of opposing political viewpoints when it comes to data collection [5]–[8] Social networks such as Facebook and Twitter have somewhat addressed limitations of representation in sampling via surveys. Social networks been at the forefront of political campaigns and activism during elections [9]–[11]

Objectives

Methods

Results

Discussion

Conclusion