[PAST EVENT] Honors Thesis Defense - Grace Smith
Title: Investigating Text Mining Techniques Within the Context of Politicized Social Media Data
Abstract: Social media data has recently been looked to as a source of public opinion for elections, public policies, and the economy. In order to use this data effectively, natural language processing (NLP) techniques have been developed. Topic modeling, one branch of NLP, works to uncover latent topics within a large collection of tweets. Many topics modeling methods such as LDA and k-medoids clustering are unsupervised. We propose adding a supervised Random Forest layer before performing topic modeling in order to incorporate externally known topics. We find that implementing this layer helps increase the interpretability of topics as well as uncover unique topics. Sentiment analysis, another branch of NLP, measures the polarity of a tweet in order to gain insight into the author’s opinions. We apply several sentiment analysis methods to our dataset and examine the results; we then identify weaknesses in these methods and propose steps for improvement.