sbalaji1996 / trump-sentiment-analysis

About

In this study, we investigate the rhetorical effects that Donald Trump’s Facebook statuses have on his social media followers. We begin with sentiment analysis of posts published over the course of his 2016 presidential campaign and train a classifier to determine natural language sentiment within close proximity of human performance. We then create a logistic regression model to prove that the presence certain keywords in a status can predict the overall valence of that post. In order to confirm our prediction that the emotional rhetoric of a status corresponds with users’ reactions to that post, we conduct permutation tests to ascertain the correlation between sentiment and particular reactions. Then, using the highvalence indicative words yielded by our classifier, we produce visualizations that relate those keywords to users’ reactions on posts made by the Facebook pages for Donald Trump, CNN, and the New York Times. We then analyze that visual data to draw conclusions about Trump’s rhetoric and the effects it has on the response of his followers. Ultimately we find that, similar to their associations with sentiment, certain keywords can predict the reaction distribution of a post. We also find that irregular formatting, such as special punctuation and entire-word capitalization, can have a magnifying effect on certain reactions.

Results

Our findings support our 3 hypotheses and allow us make strong assertions about the relationship between rhetoric and response in social media. We can confidently say that the sentiment of one’s posts will have a predictable effect on the reactions to those posts, thus confirming our framework hypothesis. We were able to select features from the text in such a manner that our logistic regression model achieved an accuracy rating of 78.66%, thereby confirming our statistical hypothesis. Furthermore, we can predict that posts with special formatting will have a modulating response on the reactions to those posts, thus supporting our cognitiveaffective hypothesis. Finally, our findings give us insight into the effect that Trump’s rhetoric has on his followers as well as the nature of their support. The most significant finding is that Trump supporters are intensely fond of him, displaying their overwhelming “love” on his Facebook posts. Another interesting finding is that objectively unassuming language, such as “our” and “make”, is made exceptional by Trump’s simple yet absolutist rhetoric, which elicits a powerful response from his followers. Lastly, we can say that Trump’s rhetoric on social media is deeply entrenched in emotional sentiment, and thus a potent tool for galvanizing his conservative base and rallying millions of supporters.

Acknowledgements

Professor Yang Xu, U.C Berkeley

Max Woolf: https://github.com/minimaxir/facebook-page-post-scraper

Packages Required

To run these scripts, you need to have the following packages installed:

1. nltk: www.nltk.org

2. pandas: pandas.pydata.org

3. numpy: www.numpy.org

4. matplotlib: matplotlib.org