Distinguishing Political Parties From Inaugural Speeches Using Machine Learning

A massive amount of text data is available online; applying machine learning techniques to analyze the data helps human beings to grasp the status of previous and current eras. We analyze lexically all US presidential inaugural speeches from 1789 until now and use machine learning algorithms to learn patterns to categorize the authors of the speeches into their respective parties. We apply four different supervised learning approaches (Multinomial Naive Bayes, SVMs, Random Forest, and Logistic Regression) and evaluate their classification performance. The outcome is a 100% accurate classification based on our choice of lexical features (such as unigram and stopwords removal) and the parameters of SVM with linear, polynomial, and RBF kernels. Our study shows that the selected supervised machine learning algorithms can produce highly accurate association between political parties inaugural speech text without requiring any additional information such as metadata for authors or parties or semantics.