Background
This repository contains Jupyter Notebook files created for a study on the real-time reception of the 2018 State of the Union address on Twitter. The notebooks are presented to demonstrate code written for the project and the methods adopted. Files necessary to run these notebooks are not included.
Aims and Method
The study gathered roughly 275,000 unique, English-language tweets with the hashtag "#sotu" made during the State of the Union address.
Natural Language Processing techniques were used to infer the gender and location of users. Sentiment analysis was conducted to apply each tweet with a polarity and subjectivity score. Machine learning (K-Means clustering) and data analysis methods were used to determine how reception of the speech varied by gender and location.
Conclusions
Data exploration showed that Clinton-voting states tended to tweet more per-person and use less positive language than Trump-voting states. The language of the speech and the policy areas it addressed were broadly reflected in online discussion, however certain infrequently used but politically-charged key terms led to disproportionate responses from viewers ("wall", "beautiful clean coal", "stand", "anthem").
Men proportionally engaged more on political issues around finance, employment, defense, and infrastructure, while women focussed more on immigration, family, and indeed women. Users from Clinton-voting states mentioned terms relating to policy on immigration, race, energy, and defense more frequently than those from Trump-voting states, who preferred patriotic language. Both groups heavily mentioned political leaders from opposing parties. Additionally, discussion of the speech was often informed as much by visual aspects of the broadcast as by the messages it contained.
Contributors
The study was conducted by Alex Franklin and Rowena Jones. Alex Franklin contributed appendices 1-3, 8-9, 11-13. Rowena Jones contributed appendices 4-7, 10, 14-15.