Tracking Political Bias in U.S. News Media: A Daily Sentiment Analysis of the 2024 Presidential Election Using RoBERTa

The Influence of political bias in U.S. news media has been a persistent factor in shaping public discourse and electoral outcomes. This study investigates the political bias within the 2024 U.S. presidential election on a daily scale from 15th October 2024 to 15th November 2024. This is achieved by applying Sentiment Analysis with Natural Language Processing (NLP) to news articles. Three U.S. news outlets are subjected to sentiment analysis, which are Reuters, CNBC, and FoxNews. Through those three news outlets, we managed to collect a total of 722 articles, with 270 being from Reuters, 205 from CNBC, and 247 from FoxNews. Articles are sorted based on their published date, and a longitudinal analysis is done to capture the trend of political bias in media as the election developed. By fine-tuning the RoBERTa model with a labeled dataset obtained from Kaggle, an accuracy score of 92.16% and F1- score of 91.81% is achieved. The biases of articles obtained from Reuters, CNBC, and FoxNews are then classified using the fine- tuned RoBERTa model. Through the analysis, we were able to find a significant amount of bias, with CNBC and FoxNews leaning left, while Reuters remained neutral. Media polarization was seen to peak during election day. On the other hand, the overall amount of bias each day is calculated. High fluctuations are seen, with peaks around the final campaign time and after Trump’s victory. The way where the bias leaned toward is also investigated. It is visible that despite them being left-leaning, there was a stable increase towards the right, which might correlate to Trump’s victory.