Text Classification and Topic Modeling on Presidential Candidates’ Tweets

For my final project, I decided to explore the nature of Walter Benjamin’s “politicized art” as it relates to social media of a political variety. I downloaded every Tweet from the main Twitter accounts of presidential candidates Hillary Clinton, Donald Trump, Bernie Sanders, Gary Johnson, and Jill Stein from the dates each individual announced their running for president, until November 8th, 2016, when Donald Trump was elected. I applied different text classification algorithms to this dataset with the intention of developing a probabilistic distribution of textual similarity for any given text. For user input, five percentages appear and are graphed as a pie chart in order to represent this distribution.

I also chose to employ topic modeling on each candidate’s corpus of Tweets in order to potentially extract precisely which kinds of ideas each candidate expressed online. This topic model is then represented as a probabilistic distribution which presents the likelihoods of each topic contained within the user input per candidate. Note that the topics for each candidate differ between them.

In the end, it’s fundamentally but intentionally flawed. These Tweets do not consistently represent the policies nor the feelings of the candidates, and were not even necessarily written by them. This aspect intentionally raises questions on the nature of the media participation and consumption of politicians and the citizens they represent. Just as the web application does not appropriately represent the interests or concerns of any specific candidate, our understandings of those aspects are necessarily informed by the social media representations they choose to put forth.

Perhaps this project could be received as an interesting glimpse into what can be gained from reading politicians’ Twitter pages. What are we to do when the people who seek to be elected to represent our interests as citizens inform their understandings of our interests through the minority audience using these platforms? Are the concerns of Donald Trump on Twitter informed by the social media platform and its users, or the concerns actually felt by Donald Trump the human being?

Above all else, this project makes me want to think critically about what I accept as truth on social media. I encourage others to do the same across all media representations of human intent. I am not the me you see on Facebook, and certainly when you feed the program quotes from Donald Trump which fail to resemble things he has said over this election cycle, the program fails to provide accurate analyses.

Thanks for reading, and please try it out! I didn’t spend over 40 hours making it for nobody to use it!

Leave a Reply

Your email address will not be published. Required fields are marked *